Mapping the major axis of tephra dispersion with a mesoscale atmospheric model

Mapping the major axis of tephra dispersion with a mesoscale atmospheric model

Material Information

Mapping the major axis of tephra dispersion with a mesoscale atmospheric model Cerro Negro Volcano, Nicaragua
Byrne, Marc A
Place of Publication:
[Tampa, Fla.]
University of South Florida
Publication Date:


Subjects / Keywords:
Volcanic ash
Atmospheric diffusion
Mesoscale modeling
Computer modeling
Dissertations, Academic -- Geography -- Masters -- USF ( lcsh )
government publication (state, provincial, terriorial, dependent) ( marcgt )
bibliography ( marcgt )
theses ( marcgt )
non-fiction ( marcgt )


ABSTRACT: Models of tephra fallout are used to assess volcanic hazards in advance of eruptions and in near-real-time. Current models often approximate the wind field using simplistic assumptions of the atmosphere that cannot account for typical variations in wind velocity that occur in time and three-dimensional space. Here, a widely used mesoscale atmospheric model is used to improve forecasts of the location of the major axis of dispersion for erupting plumes. The Pennsylvania State University-National Center for Atmospheric Research fifth-generation Mesoscale Model (MM5) specializes in atmospheric prediction for regions on the order of ten to hundreds of kilometers on a side. MM5 generates realistic wind fields based on the laws of conservation of mass, energy, and momentum, along with land surface data and atmospheric forecasts and observations.
Thesis (M.A.)--University of South Florida, 2005.
Includes bibliographical references.
System Details:
System requirements: World Wide Web browser and PDF reader.
System Details:
Mode of access: World Wide Web.
General Note:
Title from PDF of title page.
General Note:
Document formatted into pages; contains 117 pages.
Statement of Responsibility:
by Marc A. Byrne.

Record Information

Source Institution:
University of South Florida Library
Holding Location:
University of South Florida
Rights Management:
All applicable rights reserved by the source institution and holding location.
Resource Identifier:
001681098 ( ALEPH )
62769971 ( OCLC )
E14-SFE0001016 ( USFLDC DOI )
e14.1016 ( USFLDC Handle )

Postcard Information



This item has the following downloads:

Full Text
xml version 1.0 encoding UTF-8 standalone no
record xmlns http:www.loc.govMARC21slim xmlns:xsi http:www.w3.org2001XMLSchema-instance xsi:schemaLocation http:www.loc.govstandardsmarcxmlschemaMARC21slim.xsd
leader nam Ka
controlfield tag 001 001681098
003 fts
005 20060215071234.0
006 m||||e|||d||||||||
007 cr mnu|||uuuuu
008 060103s2005 flu sbm s000 0 eng d
datafield ind1 8 ind2 024
subfield code a E14-SFE0001016
G116 (Online)
1 100
Byrne, Marc A.
0 245
Mapping the major axis of tephra dispersion with a mesoscale atmospheric model
h [electronic resource] :
b Cerro Negro Volcano, Nicaragua /
by Marc A. Byrne.
[Tampa, Fla.] :
University of South Florida,
Thesis (M.A.)--University of South Florida, 2005.
Includes bibliographical references.
Text (Electronic thesis) in PDF format.
System requirements: World Wide Web browser and PDF reader.
Mode of access: World Wide Web.
Title from PDF of title page.
Document formatted into pages; contains 117 pages.
3 520
ABSTRACT: Models of tephra fallout are used to assess volcanic hazards in advance of eruptions and in near-real-time. Current models often approximate the wind field using simplistic assumptions of the atmosphere that cannot account for typical variations in wind velocity that occur in time and three-dimensional space. Here, a widely used mesoscale atmospheric model is used to improve forecasts of the location of the major axis of dispersion for erupting plumes. The Pennsylvania State University-National Center for Atmospheric Research fifth-generation Mesoscale Model (MM5) specializes in atmospheric prediction for regions on the order of ten to hundreds of kilometers on a side. MM5 generates realistic wind fields based on the laws of conservation of mass, energy, and momentum, along with land surface data and atmospheric forecasts and observations.
Adviser: Dr. Graham Tobin.
Co-adviser: Dr. Arlene Laing.
Volcanic ash.
Atmospheric diffusion.
Mesoscale modeling.
Computer modeling.
Dissertations, Academic
x Geography
t USF Electronic Theses and Dissertations.
4 856


Mapping the Major Axis of Tephra Dispersion with a Mesoscale Atmospheric Model: Cerro Negro Volcano, Nicaragua by Marc A. Byrne A thesis submitted in partial fulfillment of the requirements for the degree of Master of Arts Department of Geography College of Arts and Sciences University of South Florida Co-Major Professor: Graham A. Tobin, Ph.D. Co-Major Professor: Arlene G. Laing, Ph.D. Charles B. Connor, Ph.D. Steven Reader, Ph.D. Date of Approval: April 12, 2005 Keywords: volcanology, meteorology, volcanic ash, climatology, atmospheric diffusion, mesoscale modeling, computer modeling Copyright 2005, Marc A. Byrne


Acknowledgements I would like to thank Dr. Arlene Laing, Dr. Charles Connor, Brian Smith and Dustin Binge at the Research Computing Core Facility, Wei Wang and Cindy Bruyre at the National Center for Atmospheric Resear ch, Dr. Graham Tobin, Dr. Steven Reader, Laura Connor, Shelly Happel, and my family and friends. I would also like to acknowledge the use of the services provided by the Research Computing Core at the University of South Florida. This thesis was supported by a grant from the U.S. National Science Foundation [EAR-0130602]. Additional funding was provided by the National Center for Atmospheric Research and the University of South Florida. Data products provided by the NOAA-CIRES Climate Diagnostics Center were used in this thesis.


i Table of Contents List of Tables iv List of Figures v Abstract viii Chapter One Introduction 1 Tephra Fall Hazards 1 Tephra modeling 3 Project Goals 6 Apply models to well studied eruption 6 Determine if winds mimic deposit 7 Determine necessary sophistication of atmospheric model 7 Establish a link between Volcanology and Meteorology 8 Chapter Two Literature Review 9 Tephra Modeling 9 Suzuki Model 9 Aviation 11 VAFTAD 11 CANERM 12 PUFF 12 Other Tephra Models 13 RAMS/HYPACT 14 ASHFALL 16 Mesoscale Atmospheric Models applied to non-tephra dispersion 16 CANERM 16 RAMS 17 LEDI 17 MM5 17 Desription 17 Case study prediction 22 Modeling particles and pollutants 24 Sand 24 Gas pollutants 24 Particulates 25


ii Chapter Three Methodology 27 Initial Setup 27 Data Acquisition 27 Gridded surface and atmospheric data 28 Soundings 29 Model Setup 32 TERRAIN 32 REGRID 34 LITTLE_R 35 INTERPF 36 MM5 36 INTERPB 37 GRAPH and RIP 37 Post processing 38 Reanalysis winds 40 Particle Fall Model 41 Three-Dimensional Grid 42 Particle Fall Program 43 Fall velocity 43 Model Validation 45 Sounding data 46 Root Mean Square errors 47 U-winds 48 V-winds 52 Geopotential Height 53 Summary 54 Chapter Four Descriptive Results 56 Cerro Negro 56 1995 Eruption Parameters 57 Tephra Deposit Pattern 59 Results 59 Wind shifts over time 62 Bilobate Deposit 62 Wind shifts with altitude 62 Wind model comparison – particle trajectories 63 Wind shifts over x-y space 66 Wind model comparison Satellite Imagery 67 Summary 74 Chapter Five Discussion 76 Why should volcanologists use these models? 76 Model calibration 76 Should this be used for probabilistic work? 77 Should mesoscale models be used to forecast in near-real time? 77


iii Bilobate tephra deposition 78 Wind shifts 79 Chronology of deposition 79 Small lobe 79 Large lobe 79 Model Limitations 80 Resolution 80 Particle aggregation 80 Interaction with eruption column 81 Turbulence 82 Fall velocity 83 Particles fall to zero 83 Sources of Error 83 Chapter Six Conclusions and Recommendations for Future Work 85 Major Findings 85 Future Research Improving the particle fall model 86 Release Heights 86 Effects on Column Rise 87 Diffusion term 87 Topography 88 Wind / friction effects 88 Additional future research 89 Apply to other eruptions 89 Calibrate extent of diffusion 89 TEPHRA Model incorporation 90 References 91 Bibliography 96 Appendices 98 Appendix A: MM5 Specifications 99


iv List of Tables Table 3.1. Domain parameters used in the Cerro Negro model run. 34 Table 3.2. RMS errors for the 19 meteorological stations. 48 Table 3.3. Root Mean Square erro r for the White et al. (1999) study vs. this project. 54


v List of Figures Figure 1.1. The 1995 eruption of Cerro Negro, Nicaragua, as viewed from the northwest. 2 Figure 1.2. A typical tephra plume encountering wind. 4 Figure 1.3. A weak tephra plume advected by wind. 5 Figure 2.1. The MM5 vertical coordi nate system, or sigma levels. 19 Figure 2.2. The nested grid system of MM5. 20 Figure 2.3. A mesoscale model domain. 21 Figure 3.1. A flow diagram for the Ce rro Negro tephra modeling project. 28 Figure 3.2. The 19 meteorological sta tions used for model validation. 30 Figure 3.3. An atmospheric s ounding and the co rresponding MM5 output sounding generated for the same observation time. 31 Figure 3.4. The MM5 modeling system flow chart. 33 Figure 3.5. The three-domain se tup vs. the five-domain setup. 34 Figure 3.6. The 40 km by 32 km grid used in the particle fall model, showing mapped deposit isopachs in centimeters. 39 Figure 3.7. A comparison of grid ce ll resolution for the wind models. 41 Figure 3.8. Geographical distribu tion of station RMS errors for the 850 hPa level of the atmosphere. 49 Figure 3.9. RMS error over time for u -winds. 50 Figure 3.10. RMS error over time for u -winds, 0000 UTC only. 51 Figure 3.11. RMS error over time for u -winds, 1200 UTC only. 51


vi Figure 3.12. RMS error over time for v -winds. 52 Figure 3.13. RMS error over ti me for geopotential height. 54 Figure 4.1. Tectonic set ting of Cerro Negro. 56 Figure 4.2. The location of Cerro Negro within the Americas. 57 Figure 4.3. The 1995 bent-ove r plume of Cerro Negro. 58 Figure 4.4. Tephra deposit dist ribution from Cerro Negro. 60 Figure 4.5. MM5-forecast trajectori es with 1995 deposition pattern. 61 Figure 4.6. MM5-forecast trajec tories for 0.26 mm particles. 61 Figure 4.7. 3-dimensional view of MM5 -predicted particle trajectories. 63 Figure 4.8. 9-km resolution MM5-forecast trajectories with 1995 deposition pattern. 64 Figure 4.9. NCEP reanalysis wind traj ectories with 1995 deposition pattern. 65 Figure 4.10. Uniform wind-predicted traj ectories with 1995 deposition pattern. 65 Figure 4.11. A wind field comparison. 67 Figure 4.12. 1-km resolution MM5-forecast tr ajectories for 0.01 mm particles. 68 Figure 4.13. 9-km resolution MM5-forecast tr ajectories for 0.01 mm particles. 69 Figure 4.14. NCEP reanalysis wind-predict ed trajectories for 0.01 mm particles. 69 Figure 4.15. Satellite Imagery for November 27th at 1315 UTC. 71 Figure 4.16. Satellite Imagery for November 27th at 2215 UTC. 71 Figure 4.17. Satellite Imagery for November 28th at 1315 UTC. 72 Figure 4.18. Satellite Imagery for November 28th at 2215 UTC. 72 Figure 4.19. Satellite Imagery for November 29th at 1315 UTC. 73 Figure 4.20. Satellite Imagery for November 30th at 2215 UTC. 73


vii Figure 4.21. Satellite Imagery for December 1st at 1315 UTC. 74 Figure 4.22. Satellite Imagery fo r December 1st at 1745 UTC. 74 Figure 5.1. How features are repr esented in a mesoscale model. 82


viii Mapping the Major Axis of Tephra Dispersion with a Mesoscale Atmospheric Model: Cerro Negro Volcano, Nicaragua Marc A. Byrne ABSTRACT Models of tephra fallout are used to assess volcanic hazard s in advance of eruptions and in near-real-time. Current m odels often approximate the wind field using simplistic assumptions of the atmosphere th at cannot account for t ypical variations in wind velocity that occur in time and threedimensional space. Here, a widely used mesoscale atmospheric model is used to impr ove forecasts of the location of the major axis of dispersion for erupting plumes. The Pennsylvania State University-National Center for Atmospheric Research fifth-gene ration Mesoscale Model (MM5) specializes in atmospheric prediction for regions on the order of ten to hundreds of kilometers on a side. MM5 generates realistic wind fields based on the laws of conservation of mass, energy, and momentum, along with land surf ace data and atmospheric forecasts and observations. It is particularly effective at resolving circulation patterns in areas with sparse meteorological observations, and/or m ountainous terrain. MM5 is applied to the 1995 eruption of Cerro Negro, Nicaragua. Estimat es of particle settling velocities are used in conjunction with MM5-derived wind fiel ds to forecast the plume track. Particle trajectories generated from MM5 winds are compared to those produced by other wind models. The complex wind fields ge nerated by MM5 explain non-linear plume


ix dispersion. MM5 winds are shown to be mo re accurate than the other models in reproducing the observed tephra deposit, and sa tellite imagery is us ed to confirm the accuracy of MM5 winds. The appropriate appl ication of meteorological data sets and mesoscale models should ultimately improve the utility of tephra fallout hazard assessments, especially in the absence of abundant meteorological observations.


1 Chapter One Introduction Tephra, commonly known as volcanic ash, is defined as airborne volcanic ejecta of any size (Tilling et al. 1987). It is a frequent product of volcanic eruptions. Because tephra poses numerous hazards to human in terests, volcanologists have modeled its motion in the atmosphere and ground deposition for over two decades. Although these models have increased in complexity over time, the meteorological component of these models was conventionally simple, often out of necessity. The incorporation of winds from a high-resolution atmospheric model c ould make tephra dispersion modeling more robust. This thesis takes winds produced by the atmospheric model MM5 and applies them to the 1995 eruption of Cerro Negro, Nicar agua (Figure 1.1). Particle trajectories are generated using MM5 winds as well as wi nds from other sources used in dispersion modeling. These trajectories are compared to observed deposition patterns and satellite imagery, in order to determine if MM5 pr ovides an improvement over more commonlyused wind fields. Tephra fall hazards Volcanoes create numerous natural hazards th at pose a risk to society. Lava flows and ballistics generally affect the area pr oximal to a volcano, and are for the most part easy to avoid. Vent collapses, meanwhile, can result in massive rock avalanches that are


2 extremely destructive. Pyroclastic flows ofte n form when part the eruption column itself collapses and are primarily composed of volcan ic ash (tephra), and heated gases. Moving down slope due to gravity, pyroclastic flows can travel over 40 meters per second, and cover great distances. They represent a par ticularly deadly volcanic hazard. Volcanic gases can also pose a threat to surrounding populations, but deaths generally are rare (Latter 1989). Figure 1.1. The 1995 eruption of Cerro Negro, Ni caragua, as viewed from the northwest. Courtesy of Dr. Brittain Hill. It is wind-driven tephra, though, that thre atens far larger areas than any other volcanic hazard. Tephra has been observed to completely circle the globe after some eruptions. When tephra, often colloquially termed volcanic ash, settles from the atmosphere onto EarthÂ’s surface, it creates a number of consequences. Following an


3 eruption, large tephra deposits often become remobilized by precipitation. The lahars made of tephra and water often pick up more water and rock debris as they travel, and can move at speeds of over 10 meters per s econd. The deadliest lahar on record, formed after the eruption of ColombiaÂ’s Neva do del Ruiz in 1985, killed over 25,000 people (Sigurdsson and Carey 1986). Tephra depos ited on roofs can easily cause building collapse, which accounted fo r most of the 300 deaths caused by Mt. PinatuboÂ’s 1991 eruption (Wolfe 1992). Tephra can clog airplane engines, and then melt inside due to high heat, destroying the mechanisms (Scott et al. 1995). Air traffic is advised to completely avoid tephra clouds, which can lead to disruption of transportation routes, resulting in massive delays and wasted funds (Turner and Hurs t 2001). When settled in populated areas, tephra can cause similar problems with autom obiles. Vehicles frequently stir up the recently settled particles, ex tending the problem for even longer. Tephra can also cause power outages when settling on electrical equipment such as transformers (Scott et al. 1995). Tephra often causes water contamination, wh ich can kill land animals and fish and destroy crops. Health consequences of tephra for humans include adverse effects on respiration, and irritatio n of eyes and nose. Direct e xposure to tephra in the air is obviously best avoided (Latter 1989, Baxter 1990). Tephra modeling In order to better understand tephra plumes and hazards, many volcanologists have engaged in numerical simulation of volcanic eruptions (Suzuki 1983, Carey and


4 Sigurdsson 1982, Barberi et al. 1990, Connor et al. 2001, Bonadonna and Phillips 2003). In every model, some physical assumpti ons are made, and eruption parameters are estimated. These assumptions apply not only to the characteristics of the eruption and the eruption column, but also th e tephra plume (Figure 1.2). Figure 1.2. A typical tephra plume encountering wind. From Gardner et al. 1995. Tephra is forced up through th e eruption column, and is advected downwind in the plume. Wind continues to affect part icle movement throughout tephra fall. The movement of tephra in the atmos phere is crucial for modeling tephra deposits. Once tephra particles in the erupt ion column have achieved their maximum height, their movement depends upon their characteristics (d ensity, shape, terminal fall velocity) and very importantly, wind. Erup tions with little or no wind can produce bullseye tephra deposits centered on the volcan ic vent (Bonadonna and Phillips 2003). However, much more commonly, tephra plum es are advected by winds, and become more diffuse with time and distance downwind. In general, ballistic ejecta reach the


5 ground very quickly, smaller tephra fragments fa ll out more slowly, and the finest tephra settles last (Figure 1.3). Cha nges in wind speed and directio n with distance or height, known as wind shear, can create irregular deposition patterns. While many tephra deposits exhibit axisymmetrical patterns, is opach maps of tephra deposits often reveal non-linear axes, multiple deposition axes, and multiple thickness maxima (Ernst et al. 1994). Figure 1.3. A weak tephra plume advected by wind. From Bonadonna and Phillips 2003. In general, large particles fall close to the vent, and smaller particle are deposited farther away. Xw represents the distance that a partic le is transported by wind during particle fall. The last 25 years of plume modeling ha ve used wind assumptions ranging from the most basic and static to relatively complex and dynamic (Woods et al. 1995, Bursik 1998, Turner and Hurst 2001, Bonadonna et al. 2005). Despite these observations of complex tephra deposits, winds, more often than not, are assumed to be uniform (Hill et al. 1998). This assumption may be made for simplicityÂ’s sake, or because wind data is


6 not easily accessible, difficult to retrieve, and in many cases, incomplete. It is not uncommon to find that wind data have been extrapolated from a single observation from a weather station tens of kilometers away (Barberi et al. 1990, Macedonio et al. 1994). This is a serious problem because the simplistic wind assumption neglects a major impact on deposition location. This infrequency of application of detail ed wind is symptomatic of the fact that there has been minimal collaboration betw een volcanologists and meteorologists in tephra prediction models. In addition, much of the research on tephra dispersion that has involved atmospheric models has addressed the issue as it rela ted to aviation interests. The role of wind in the surface accumulation of tephra can be explored in greater detail. Atmospheric models can fulfill this pressing need in volcanological models. While employed most commonly for forecasting weat her conditions, they have also been used extensively for tracking particles, pollutants, and for atmospheric chemistry. Mesoscale models would be of particular importance beca use they deal with the same spatial scales as a typical tephra plume. Mesoscale refers to weather phenomena that are on the scale of ten to hundreds of kilometers across (Brasseur et al. 1999). Project Goals Apply models to well studied eruption This project has several specific objec tives and goals. To determine if atmospheric models are appropriate for volcanological modeling, a mesoscale atmospheric model is applied to the tephra plume from a well-studied volcanic eruption, the November 19-December 6, 1995 eruption of Cerro Negro, Nicaragua. The


7 meteorological model employed for this projec t is version 3 of th e Pennsylvania State University-National Center for Atmosphe ric Research (NCAR) fifth-generation Mesoscale Model, or MM5. A particle fall prog ram is created to simulate the movement of tephra particles in the atmosphere, and the forecast wind fields produced from MM5 are used as input. Determine if winds mimic deposit To evaluate the utility of MM5 for predicting tephra dispersion and deposition, the model needs to be tested using actual er uption data. Because there are no consistent soundings available for the area during the pe riod of the eruption, global wind data and distant station data are used by MM5 in order to rec onstruct the wind field. The MM5 output is used in a particle trajectory mode l, and the predictions are compared to the tephra deposit. The extent to which part icle trajectories match the observed tephra deposition pattern will determine how accurate ly the MM5 predicted winds within the Cerro Negro study area. Determine necessary sophistication of atmospheric model MM5 is evaluated at two different spa tial resolutions, in order ascertain the minimum resolution needed in order to suffi ciently simulate winds for the region during the eruption. The results of the MM5 output winds are also compared to those generated from archived global reanalysis wi nds, and from a uniform wind field. This research tests whether the MM5 output winds are closer to the true local wind patterns, and therefore more useful to volcanologists, than are reanalysis wind data


8 or uniform winds. The assumption of uniform wind in a three-dimensional space over an extended time period is investigated to exhibit its inadequacies. The temporally variable, three-dimensional MM5 winds are used to im prove the model winds away from the vent, and possibly illuminating large downwind shifts that would go undetected in the uniform wind models. If MM5 performance sets itse lf apart from the conventional input winds for volcanologists, it will have clearly demons trated the need for a more advanced wind model. This project will al so test if low-intensity volcanic eruptions produce enough tephra in the atmosphere to require a complex wind model. Establish a link between Volcanology and Meteorology This research project provides a link between the disciplines of volcanology and meteorology, using methods that already have proven useful in their respective fields. The study creates a template that could be used to improve future models, or at the very least encourage more collaboration between the two fields. It is also hoped that the project marks an improvement in tephra deposition forecasting. Looking towards possible future researc h, the combination of a meteorological model with a true tephra dispersion model that can account for many features of a volcanic eruption should lead to a more comp lete methodology and more robust results. It is anticipated that the resu lts of this study could eventually lead to a re-examination of tephra and lahar hazards, and could serve as an important step in improving the methodology of tephra fall prediction, hazard as sessment, and issuing warnings in the event of an imminent eruption.


9 Chapter Two Literature Review Tephra Modeling This chapter first reviews some of th e past work conducted in tephra dispersion modeling, focusing on the wind fields used in each study. Following that, examples of dispersion modeling using atmospheric models ar e explored. Finally, several case studies involving MM5 in particular are presented. Suzuki Model Tephra dispersion models have been in ex istence for over twenty years. Suzuki (1983) was the first to numerically model the di spersion of tephra in the atmosphere. The model uses a line source for particles, w hose maximum height is calculated from the energy and mass flow of an eruption (Connor et al. 2001). The particlesÂ’ settling velocities are a function of particle shape, size, and dens ity, while horizontal movement is governed by the diffusion-advection equation using a uniform wind field, where direction and magnitude are held constant in three-dimensional space and in time. Glaze and Self (1991) adapted Suzuki Â’s (1983) dispersion model for an investigation of the 16 September 1986 eruption of Lascar, in Chile. Like the Suzuki (1983) model, tephra diffused horizonta lly while falling to the ground in a time t which is independent of horizontal motion. The mode l was modified so that it could account for


10 variations in wind speed and di rection with height above the vent. In this temporally uniform, stratified wind model, wind magnitude and direction were uni form within each atmospheric level, but varied over the verti cal domain. Variations in wind direction caused a bimodal tephra deposition pattern fo r Lascar in the model (Glaze and Self 1991). They found that the model could give a reasonable approximati on of distal tephra deposits for short-lived eruptions, and is best suited for instantaneous (vulcanian) blasts, rather than continuous eruptions. The resear chers suggested that future improvements to the model would include more attention to the motion of small particles, and capability to simulate ballistic fallout (Glaze and Self 1991). A model designed by Armienti et al. ( 1988) to describe tephra deposition following the 1980 eruption of Mt. St. Helens sl ightly modified the Suzuki (1983) model by using a vertically stratified wind field and a gradual (rather than instantaneous) release of particles. Using a wind field from 3 s oundings in Spokane, Wash ington, on the day of the eruption, the model, like Suzuki (1983) reproduced the double maximum that was observed. It was concluded that this phenom enon was created by th e stratified, varying winds, aggregation of the finest particles, and the distribution of mass in the eruption column. The researchers found that forecas t deposit was too wide, because particle aggregation that had taken place was not mode led. They also mentioned that topography and variation of wind velocitie s in the horizontal plane are not taken into account, which are both a source of model er ror (Armienti et al. 1988). The Suzuki model was further adjusted to account for the conservation of mass within the eruption column (Connor et al. 2001, Bonadonna et al. 2002). The model was integrated into a probabilistic hazard asse ssment for Cerro Negro, Nicaragua, taking into


11 account a variety of possible eruption characte ristics and atmospheric conditions (Connor et al. 2001). Other modifications to the Suzuki mode l included parameterizations for winds, particle aggregation, and vari ations in particle density (Bonadonna and Phillips 2003). Using two different vertica lly stratified wind models, th eir study acceptably modeled the spreading current of the 1980 Mt. St. Helens eruption, as well as several other historic eruptions. Aviation Alternative approaches to tephra disp ersion modeling are dedicated to the interests of aviation and commerce. The following section describes some of these models. VAFTAD The Volcanic Ash Forecast and Trans port And Dispersion model (VAFTAD) was developed by the National Oceanic and Atmospheric Administration (NOAA) specifically to track ash and its hazards affecting aviation inte rests in real-time (Heffter and Stunder 1993). For their study, input winds were acquired from the NOAA National Meteorological Center (NMC) at 12-hour interv als. The model required some input on eruption characteristics, but assumed that pa rticles were all spherical, with uniform density, and diameters ranging from 0.3 to 30 m These fine-grained particles generally have considerable residence times in the atmo sphere. In the model, tephra was advected horizontally and vertically, descending accordi ng to StokesÂ’s Law with a slip correction.


12 A unit source for the eruption mass was used, because actual erupti on masses are rarely available during eruptions. Eruption duration was assumed to be no more than three hours, at which point the erup tion column was expected to have achieved its maximum height. Model verification was conducted usi ng satellite imagery, as large-scale ground observations were constrained by time and funds for the eruptions that were investigated. Although the model moved partic les vertically, it did not track tephra accumulation on the ground, because it focused on hazards to aircraft. CANERM CANERM (CANadian Emergency Respons e Model) is a three-dimensional Eulerian model that has been used extensiv ely for forecast threats to aviation interests (Simpson et al. 2002). It uses terrain-following z coordinates, ha s a maximum wind resolution of 25 kilometers, and like the VAF TAD model, its purpose is to track tephra particles within the atmosphere. Although the model accounts for wet and dry deposition, it is not programmed for es timating tephra deposi tion (D’Amours 1998). PUFF PUFF is a high-resolution mode l designed specifically to track volcanic ash in the atmosphere (Searcy et al. 1998). It was de veloped to aid real-time hazard warnings, especially in cases where visual observa tion of the ash cloud may be hampered by weather or other factors. PUFF is most adep t at predicting dispersa l for ‘young’ particles, within the first 48 hours after an eruption. It acts as a pollu tant tracer model, using a three-dimensional Lagrangian formulation of pollutant dispersion, using a finite number


13 of tephra particles to represent the entire te phra plume. Separate trajectories are then calculated for each particle, one by one, from one time step to the next, using a “random walk” formulation. Because PUFF requires ap proximately real-time wind forecasts, the input data used were 4-dimensional wind fi elds from UCAR's twice-daily forecast runs. These data are available from the surface up to the 100 hectopascal (hPa) level, and have a grid resolution of 5 degrees longitude by 2.5 degrees latitude. Although PUFF allows particles to settle out of the atmosphere, it does not predict tephra deposition on Earth’s surface (Searcy et al. 1998). Other Tephra Models Probabilistic tephra hazard modeling has a heightened importance in areas where large populations lie close to volcanoes. Barber i et al. (1990) created probabilistic tephra hazard maps for a theoretical large-scale erupt ion of Mt. Vesuvius in Italy. By using historic eruption data to cal culate the annual magma suppl y rate, the likelihood of an extreme eruptive event was determined. With 15 years of archived wind profiles from the meteorological station in th e city of Brindisi, 300 kilome ters to the east, the group determined the historical distribution of wi nd velocity, to calculate the probability of tephra being transported to a partic ular area (Barberi et al. 1990). A similar project was undertaken by Mace donio et al. (1994). In this study, the hazard posed by tephra plumes on the dense aircraft transit in the vicinity was determined. The researchers modeled theore tical plumes for each wind profile from a 10year archive from Brindisi. It should be noted that Vesuvi us borders the Tyrrhenian Sea, on Italy’s west coast, while Brindisi lies on th e Adriatic, on Italy’s ea st coast. Local wind


14 trends will not be captured when using wind da ta from a distant source. Results of the study indicated that there were two seasonal sh ifts in wind patterns: at the beginning and the end of summer, when weaker and variab le winds prevail (Mace donio et al. 1994). The rest of the year has stronger winds, uni formly out of the west. The highest plume hazards, then, were located di rectly east of the volcano. Tephra modeling is also commonly used to reconstruct the dynamics of long-past eruptions, such as the A.D. 79 eruption of Ve suvius (Macedonio et al. 1998). In this project, the model accounts for diffusion a nd advection by wind, and calculates settling velocities based on tephra part icle granulometric data. The researchers depict the terrestrial deposits from both phases of the eruption as relatively axisymmetric. This suggests that winds were relatively uniform in proximal regions for the estimated 19-hour combined duration of the two phases. Because a large percentage of the tephra was deposited in the sea, deposit thickness had to be interpolated from the small number of land-based sample points. The mapped deposit, therefore, has some level of uncertainty. To apply a wind field to their model, the re searchers found the mean wind velocity at different altitudes for summer m onths, as recorded from two st ations, Rome and Brindisi. Because mean winds were found to move from east to west, this result was rotated clockwise 60 degrees to bette r fit the observed deposit. Th e predicted model deposit was found to agree well with the results from interpolation. RAMS/HYPACT One of the most similar research project s to this study involve d a combination of meteorological and volcanological models to forecast tephra deposition for Mt. Ruapehu


15 (Turner and Hurst 2001). These researcher s used the Regional Atmospheric Modeling System (RAMS) along with the Hybrid Par ticle And Concentration Transport model (HYPACT). RAMS is a three-dimensional atmospheric model that is capable of simulating the effects of rough terrain and a reasonably high-resoluti on inner grid (2.5 km). HYPACT is a dispersion model that trea ts plumes as Lagrangian particles, which undergo advection and turbulence. The study assumed that tephra particles were entirely between 1 and 200 m in diameter, with a uniform distribution of sizes Fall velocities were approximated by Stokes’s law, although the authors acknowle dged that this assumption might be unsuitable for particles over 80 m in diam eter. Best results for the 1995 Ruapehu eruption occurred when the tephra plume was simulated to range from 7 to 10 kilometers in height. Certain forecasted isopachs we re found to exceed ground observations, but the researchers attribute this to overestimation of initial eruptio n volume. Other areas that observed deposition of tephra were not predicte d to have any by the models. Simulations also predicted that the main axis of the er uption plume to be 10 degrees rotated from its actual position. The researchers concluded that the initial eruption parameters were of paramount importance in determining tephra deposition location, because particles can encounter high wind variation depending on wher e they are “released” in simulations. Initial and lateral boundary conditions for RAMS were another limitation of model accuracy. It was also reasoned that while the ability of HYPACT in tracking tephra plume movement had merit, it was inappr opriate for quantifying tephra deposition, because of shortcomings related to particle fall velocities.


16 ASHFALL Turner and Hurst (2001) compared th eir methodology to that used by the New Zealand government for ash fall advisories, the ASHFALL model, which simplifies the Suzuki (1983) model. They found that ASHFALL had many of the same limitations regarding boundary conditions and eruption char acteristics, but its results were not as accurate at the RAMS/HYPACT suite. ASHF ALL presented several other drawbacks when the models were compared. It can acc ount for vertical and temporal variation in winds, but not for horizontal spatial variati on in winds. The maximum horizontal grid spacing is 5 kilometers, requires height leve ls without regard to local topography, and showed little sensitivity to va riations in particle fall velo cities (Hurst and Turner 1999). Mesoscale Atmospheric Models ap plied to non-tephra dispersion CANERM D’Amours (1998) used CANERM to simula tion dispersion of tracers for the 1994 ETEX (European Tracer EXperiment) releases of October 23 and November 14, 1994. The model uses three-dimensional Eulerian advection-diffusion to determine atmospheric concentrations of particles. The study us ed 6-hour analysis data provided by the European Centre for Medium Range W eather Forecasting (ECMWF), which was interpolated from a 73 by 55, 0.5-degree gr id to a 101 by 101, 25-kilometer grid. The simulated tracer cloud appeared “quite plau sible,” but surface concentrations were overestimated in all areas. The model appeared to be more accurate in predictions within the first 30 hours, and less accurate for times after that.


17 RAMS A 1999 study explored terrain influence on surface ozone co ncentration in a mountainous region of eastern Spain (Sal vador et al. 1999). Using the Regional Atmospheric Modeling System (RAMS), the re searchers tried to pinpoint atmospheric transport processes that contribute pollutants at four monitoring stations. The input data was one-degree resolution from the ECMWF (European Centre for Medium-Range Weather Forecasts). Large diurnal variations in wind were apparent. The researchers also discovered that certain wind patterns we re only captured at higher grid resolutions, and that some local effects could not be re produced by the model at all. Resolution had the biggest impact on modeled vertical winds. LEDI Mesoscale modeling has also been implemented for tracking radioactive contaminants. Wind and rainfall were simulated by LEDI (Lagrangian-Eulerian DIffusion model) for the 12 days followi ng the 1986 Chernobyl nuclear disaster, and were found to effectively describe 137Cs contamination due to both wet and dry deposition (Talerko 2005). The project used a time step of 10 minutes and a uniform particle size of 1 m. Topography and land us e data had a resolution of 10 kilometers. MM5 Description The MM5, like other mesoscale atmospheric models, simulates an atmosphere that evolves based on the physical laws of motion, and the laws of conservation of


18 energy, mass, and momentum. It is desi gned to forecast weather phenomena at a regional scale, on the order of ten to hundred s of kilometers. The current version is non-hydrostatic, removing the previous 10-k ilometer limit imposed by the hydrostatic assumption. MM5 uses a terrain-following ve rtical coordinate system, referred to as the “sigma” coordinate system (Figure 2.1), a 1or 2-way nested grid system (Figure 2.2), model physics, lateral and vertical boundary conditions. The upper radiative boundary condition allows the reflection of gr avity waves (Grell at al. 2000). The physics of the model, described in deta il by Dudhia (1993) can be summarized as: Cumulus parameterization represents sub-gr id scale vertical motions that are due to convective clouds. Boundary layer and vertical diffusion sc hemes handle friction and other processes such as turbulent motion in the lowest layer of the atmosphere. Microphysical schemes treat cloud and preci pitation processes on a resolved scale Horizontal diffusion represents sub-grid horizontal eddy mixing and serves as a horizontal filter. Radiative schemes represent radiative fo rcing in the atmosphere and at the surface, e.g., the reduction of so lar radiation due to cloud cover. Surface schemes represent the effects of land and water surfaces, e.g., sub-soil temperature profile or movement of water through root systems.


19 Figure 2.1. The MM5 vertical coordinate system, or sigma levels (courtesy NCAR/MMM). This coordinate system follows surface terrain. The gray object at the bottom represents a mountain. This example ha s 15 full vertical leve ls and 5 half-levels. Originating from models developed in the 1970s, MM5 is a well-known and often used mesoscale model, with a community of us ers, an online how-to tutorial, help guides, and a standby expert for user support. Vers ions of MM5 have been in use for over 20 years, and it has been updated and imp roved throughout its history of use. White et al. (1999) found that the MM5 wa s the most developed research model tested in their comparative study, and that it produced superior forecasts to other research models. MM5 accounted for more dynamical atmospheric processes than the other existing models that were tested, and provided the best forecasts at short-range timescales (White et al. 1999).


20 Figure 2.2. The nested grid system of MM5 (courtesy NCAR/MMM). Domain 1 is the “parent” domain; Domains 2 and 3 are nested inside 1, and Domain 4 is nested inside Domain 3. Domain 4 would have the hi ghest resolution in this configuration. The model has already been used to track gas and particle pollutants, including dust, demonstrating its utility to this projec t. A more extensive description of these studies will follow. MM5 is a relatively thorou gh model that considers many environmental factors, which are generally sp ecified by the user. These variables include topography, land use and land cover, vegetati on type, soil type, soil temperature, sea surface temperature, archived meteorologi cal forecasts, available surface weather observations recorded at st ations, and upper-air observat ions from weather balloons. The MM5 is effective at resolving circulat ion patterns in areas that have a dearth of meteorological observations an d/or rough terrain. This is not much of an issue in the United States and Europe, but less wealt hy countries often do not have plentiful resources, and may have even less archived weather information, going back in time a decade or more. Large bodies of water such as seas or oceans generally have even fewer


21 direct observations, so the importance of MM5 is elevated in studies involving such areas. It is also competent at simulating or ographically driven circulation such as gap winds, downslope winds, lee waves, and topogr aphically forced preci pitation (White et al. 1999, Zngl 2002). The MM5 produces realistic winds in four dimensions, varying in x y z space and in time. A typical model domain is depicted in Figure 2.3. Figure 2.3. A mesoscale model domain (fro m the COMET website “How mesoscale models work”). Features and parameterizati ons are depicted using the three-dimensional coordinate system. The horizontal and vert ical grids are spaced on different scales. This complexity is necessary for supplying a more accurate wind field for a tephra dispersion model. The MM5 is also well-equipp ed to handle small-scal e areas of interest, using nested grids. Up to 6 domains can be used, with one inside another. The maximum


22 resolution is a grid cell that is 1 kilometer on a side. The ability to generate wind patterns on large and small scales in the same model r un is particularly helpful for different types of volcanic eruptions. MM5 accounts for atmospheric parameters lacking in most tephra models, providing realistic atmospheric conditions for a wide range of spatial resolutions. Its handling of boundary layer mixing and eddy diff usivity make it applicable for use in particle dispersion. Case study prediction White et al. (1999) tested six atmos pheric forecast models, including MM5, for comparison over the western United States. Modeling involved a single 65 by 65 cell grid with a spacing of 27 kilometers, and 27 ha lf-sigma levels in the vertical dimension (see Figure 2.1). Half-sigma levels do not in clude all information that a full sigma level has. Input data were bilinear ly interpolated to fit the MM5 domain grid points. Grid nudging was also employed in the study. Model validation was determined by bias error and mean square error for variables incl uding wind and geopotential height, though in their study, the error for u and v -winds are combined into one value. Bias error shows whether values in the model are generally ove ror underestimated, while the mean square error provides the typical magnitude of error (W hite et al. 1999). Highly resolved models expressed high apparent error when compared to gridded analysis data, because of great differences in resolution between the two da ta sets. Comparison with point observation, as conducted here, resulted in smaller, more believable model errors (White et al. 1999). Topographic features seemed to influence not only the MM5, but all model forecasts.


23 A study by Cox et al. (1998) demonstrated that regions of complex topography have marked localized wind effects, and thus require high-resolution reconstructions in order to be realistic. During the Special Observing Period (SOP ) of the Mesoscale Alpine Programme (MAP), Ferretti et al. (2003) tested the ab ility of MM5 to forecas t heavy precipitation. Although they reported underest imated rainfall amounts and shortcomings regarding surface temperature prediction, especially in areas strongly influenced by marine air masses, the model showed better results wh en a higher resolution grid was used. The study employed MM5 Version 2 (MM5V2), an ol der version than used in this tephra deposition study. MM5 produced accurate re sults for surface pressure values, and performed well in regions of complex terrain (Ferretti et al. 2003). The researchers also noted that MM5 seemed to resolve circulati ons better during cloudy skies that clear ones. They found high RMS errors for zonal ( u ) and meridional ( v ) wind during the early stages of the forecast, but these gradually decreased. RMS errors for temperature and relative humidity, on the other hand, increased over the course of the forecast. MM5 has been applied for high-resoluti on forecasting in mountainous, tropical regions such as Colombia (Mapes et al 2003b). Using a 4-do main setup with a maximum grid resolution of 2 kilometers, the researchers f ound that MM5 could replicate diurnal shifts in precipitation, as well as other very localized phenomena (Warner et al. 2003).


24 Modeling particles and pollutants Not only has MM5 been used for forecasts specific for weather, it has also been used to track many types of particles in th e atmosphere. The following details several projects that applied MM5 in various locations. Sand A 1998 sand “event” in China was simula ted using winds generated by MM5 (In and Park 2001). The winds were then fed in to an aerosol transport model, which was capable of accounting for diffusion, advecti on, and wet and dry deposition. Aggregation of sand particles is not a common occurrence, so this was not considered in the study. The simulated particle sizes ranged from 0.1 to 1000 m. The study found that predicted arrival of suspended sand in Korea was we ll predicted by the model. Two large dust rises, spaced 4 days apart, were transported al ong different paths, and this was verified by the models (In and Park 2001). Gas pollutants The MM5 has been employed to help track anthropogenic air pollution as well. A recent study used MM5 winds in a transpor t-chemistry-deposition model (Kitada and Regmi 2003) for the mountainous region of Ka thmandu, Nepal. The primary pollutants tracked were Sulfur Dioxide (SO2) and Nitrogen Dioxide (NO2). The investigation used a high-resolution (1 km by 1 km) grid to map pollution levels and model winds. It revealed large diurnal changes in wind pattern s in the valley containing the city, as well as a double-layered flow. MM5 also helped elucidate seasonal peaks in air pollution.


25 Grell et al. (2000) applie d MM5 to the VOTALP (Vertical Ozone Transports in the ALPs) Valley campaign in order to model the wind flow and pollutant concentration in mountain valleys. They found that the MM5 simulated pollutant transport from the Po valley to smaller alpine valleys qui te well, at the highest resolution. Particulates Another application of MM5 took place in Berlin, Germany, where it was used in conjunction with a Lagrangian particle traj ectory model (Becker a nd Keuler 2001a). The study examined source attribution of pollutants in 4 dimensions, using probability density distributions. Rather than attempt a model using backwards trajectories, the researchers used millions of potential particles, then tracked them to determine which matched the locations of observed pollution within the cit y. The researchers found that heterogeneity in wind fields led to an increased cont ribution of nearby pollution sources, while homogenous winds tend to make distal pollu tion sources more important (Becker and Keuler 2001b). The research concluded that wind pattern changes from one day to the next caused a shift in pollution sources. Using a two-nest domain for the eastern United States, Seig neur et al. (2003), tracked the movement of anthropogenic Ben zene and Diesel part icles over a 5-day period. The grid resolution used was 12 a nd 4 kilometers, with a 48-hour period of model spin-up. Particles were released in to the MM5 wind field using multiple point sources, and rose as plumes according to the SMOKE modeling system. Simulations corresponded very well to observations, with urban concentrations of the pollutants much


26 higher than remote areas, and peak concentr ations occurring during the busiest vehicle commute hours. These examples strongly suggest that MM5 can provide realistic wind fields for a variety of different needs, conditions, and lo cations. Because it has already been proven adept at handling various particle dispersi on problems, it should prove quite useful for modeling tephra dispersion and deposition.


27 Chapter 3 Methodology This chapter is separated into two parts: 1) the methods used to set up and run the models applied in the study, and 2) met hods and results for MM5 model validation. Initial Setup Computational and data resources of seve ral institutions were used during this project. At the University of South Florida, accounts were set up on Linux servers dedicated to research usi ng parallel processing and administered by the Research Computing Core Facility. The MM5 suite of programs was downloaded from the MM5 website, and in stalled on the USF Linux servers mimir and wyrd. Accounts were also established at the NCAR Scientif ic Computing Division (SCD) in order to retrieve additional mete orological and environmental datasets. A significant part of the data processing and troubleshooting of the simulations was conducted while the author was a visitor at the NCAR Mesoscale and Microscale Meteorology Division in Boul der, Colorado. The methodology represented in Figure 3.1 is described in the following sections.


28 Data Acquisition MM5 can incorporate input data from a vari ety of different sources and formats. This section describes the data that were acquired specifically for this project. Acquire Reanalysis & Terrain Data, Observations Run MM5 Convert Output Statistical Error Tests Run particle fall model Compare results with satellite, ground observations High error values Interpolate Reanalysis winds Low error values Modify parameters/ troubleshoot Figure 3.1. A flow diagram for the Cerro Ne gro tephra modeling project. The modeling program steps are outlined in blue. Gridded surface and atmospheric data To initialize the mesoscale atmospheric m odel, a subset of retrospective global data on low-resolution 2.5-degree grids were collected. The surface and atmospheric data used in this study are calle d NNRP, which stands for the NCEP-NCAR Reanalysis Project (Kalnay et al. 1996). Th e NNRP data are 6-hourly (0000, 0600, 1200, 1800


29 UTC) at 17 different pressure levels: 1000, 925, 850, 700, 600, 500, 400, 300, 250, 200, 150, 100, 70, 50, 30, 20, and 10 hectopascals (hPa). The NNRP dataset that is archived by the NCAR SCD was formatted for direct input into the MM5 modeling system. The NNRP data were obtained by ed iting the script “get _nnrp.deck” for the desired dates, and then executing the program The retrieved data then had to be transferred by ftp to the local USF serv er and integrated into the MM5 setup. Surface and upper air observational data are also archived by the NCAR Scientific Computing Division. These data, required by MM5 to improve on initial forecast values, were retrieved by editing and running the “f etch-little_r-d” script from NCAR’s server. Soundings Atmospheric soundings, used in the valid ation of the MM5 simulations, were acquired from the University of Wyoming’s Department of Atmospheric Science website, found at http://weather.uw Although meteorological station data were incorporated as a part of the MM5 model initialization, data for each individual stati on were not readily available as plain text files. The University of Wyoming has a user-friendly Graphical User Interface (GUI) that allows the user to select the station from a map, or enter the 5-digit station number to retrieve the information in a text, image, or portable document format. Finding a comprehensive list of meteorological station number s requires a visit to another we bsite. A list of stations can be found at diosglobe/world.html, where the stations are grouped by continent.


30 MM5 output was tested against meteorologi cal station observations to determine if a model run provided realistic results. For the purposes of co mparison, all of these stations were located within the spatial domain of the model. A total of 19 stations in 9 different countries were chosen for model va lidation: Brownsville, Tampa, Miami, Cape Kennedy, Key West, and San Juan in the Unite d States, Veracruz, Merida, and Cancun in Mexico, Belize City in Belize, San Jose in Co sta Rica, Grand Cayman in the British West Indies, Kingston in Jamaica, Nassau in The Bahamas, Panama City in Panama, Bogot, Leticia, and San Andres Island of Colombia and Curacao in the Netherlands Antilles (Figure 3.2). Figure 3.2. The 19 meteorological sta tions used for model validation.


31 The files were saved as plai n text, for statistical test s, and in portable document format (.pdf), for visual comparisons An output “sounding” along with the corresponding atmospheric sounding is shown in Figure 3.3. Figure 3.3. An atmospheric sounding a nd the corresponding MM5 output sounding generated for the same observation time. Co mparisons were made for the temperature profile (right bold line), dew point profile (left jagged line), and wind velocity (wind barbs along right side of graphs).


32 Model Setup TERRAIN It has been shown that hi gher resolution grids are nece ssary to depict realistic wind patterns in regions of comp lex terrain (Salvador et al. 1 999). Local effects such as a mountain wake may not be simulated at resolu tions 6 kilometers and lower. However, a 2-kilometer resolution grid can account for 95 percent of the terrain variance, a 0.5 kilometer resolution grid can explain 99 pe rcent. Similar findings were found when simulating flow of atmospheric pollutants in the Alps. Only the highest resolution grid was able to reveal the fine-scale particle paths from valleys to mountains (Grell et al. 2000). Because this study modeled winds that are essentially in the wake of Cerro Negro, the highest possible resolution available in this version (1 kilometer) was utilized. The following sections deal with MM5 se tup as it pertains to the winds modeled for the 1995 Cerro Negro eruption. A flow chart of the MM5 modeling system is shown in Figure 3.4. The first program in the MM5 model is called TERRAIN. Here the user sets all the parameters of the model domains, and all the model terrestrial information that is used in the model is then generated. Data accessed by the TERRAIN step include USGS elevation, land use and vegetation, land/w ater mask, soils, and soil temperature. For a more thorough description of these data, the reader is directed to the MM5 online tutorial page, /mm5v3/tutorial/teachyourself.html or the MM5 technical memo (Grell et al.1995).


33 Figure 3.4. The MM5 modeling system flow char t. Modeling always begins with the TERRAIN step, and in this study, was followed by REGRID, LITTLE_R, INTERPF, MM5, and INTERPB. Separate MM5 trials were run using three nested domains (at a maximum resolution of 9 kilometers) and five nested domains (with a maximum resolution of 1 kilometer). The three largest domains in the 5-domain setup were identical to the 3 domains in the lower resolution run (Figure 3.5) The domains were set up one grid at a time, from largest to smallest. The first grid was 61 by 61 cells, centered at 12.0 degrees North, 85 degrees West (-85.0), and had grid cells 81 km on a side. A summary of the Cerro Negro domain setup is shown in Table 3.1. Detailed descriptions of all


34 modifications made to the default terrain.d eck file can be found in the appendix, along with the final version used for the project. Figure 3.5. The three-domain se tup vs. the five-domain setup. Domain 1 comprises the entire map area. Domain 2 is the largest white box. The five-domain setup has two additional high-resolution nested inner domains in the Cerro Negro region. Number of domains 5 Time steps 15 Vertical levels 23 Large-scale input dataNNRP Domain 1 2 3 4 5 Domain size 61x61 49x52 40x40 49x49 49x49 Grid cell size (side) 81 km 27 km 9 km 3 km 1 km Table 3.1. Domain parameters used in the Cerro Negro model run. The time steps were spaced 12 hours apart. Each parent domain must maintain a 3:1 grid cell resolution compared to its child domain. REGRID After the TERRAIN module has been succe ssfully completed, the next step is REGRID. REGRID accesses archived weathe r analyses and forecasts and trims and


35 interpolates them to fit the domains specified in TERRAIN. These fo recasts act as a first guess for the model. The output files gene rated here are used in the LITTLE_R and INTERPF steps that follow. The user input fo r pregrid is limited to specifying a choice for input data format, the beginning and end da tes of the model run, and the time interval that the output data will have. It is often necessary to allow for a 24 to 72-hour period of “spin-up,” during which the model is given ti me to create dynami cally and physically consistent meteorological fiel ds. The model is able to sufficiently develop mesoscale circulation patterns before the actual time pe riod of interest is simulated. Spin-up is especially necessary for the innermost dom ain, because initial c onditions are simply interpolated input data when the model is initiated (White et al. 1999). To account for spin-up, the start date of the model was se t to November 25, though the continuous phase of the eruption did not begin until the 27th. LITTLE_R The LITTLE_R step incorporates the fi rst-guess forecast information from REGRID, and then uses actual meteorological observations or indi rect measurements such as satellite-derived variables, to improve on them. This procedure is called objective analysis. Objective analysis techniques used by MM5 include Cressman-based analyses (distanceand prevailing wind-wei ghting options that are dependent on which meteorological field is being analyzed) and multi-quadratic analyses (Nuss and Titley 1994). Since poor observations can lead to bad analysis, the objective analysis step applies various tests to screen the data for spurious observations. The user has a choice of two programs for this task, RAWINS or LITTLE_R. For this study, LITTLE_R was


36 used, as it allows greater flexibility in the t ypes of observations that can be used as input. With RAWINS, only standard rawinsonde ( upper-air) observation formats are allowed. With LITTLE_R, users may create idealized da ta or use non-standard atmospheric data. The output information from LITTLE_R is used in the following interpolation step. INTERPF INTERPF simply prepares all the data that will be fed into the main MM5 program. The user input requ ired includes start and end dates, and listing the sigma levels that will be used as vertical coordinates. Sigma levels are calculated using both topography and air pressure, and serve as an alternate z -coordinate, as op posed to altitude or simple pressure levels, as s hown in the previous chapter. MM5 The numerical weather prediction step of MM5 is appropriately called MM5. It incorporates all the previous information generated and specified by the user, so user input at this stage is mini mal. Required entries include the number of domains to consider (this must agree with TERRAIN), maximum dimensions of any domain, and number of half sigma levels in the mode l. The model run uses the default physics options, except for the planetary boundary laye r, which was set to Blackadar (2) for all domains, instead of MRF (5). Anothe r option used was Four-Dimensional Data Assimilation (FDDA). This technique “nudges” forecast values towards meteorological reanalysis data to ensure th at output does not substantially diverge from realistic values over time. This was recommended for longer simulations by Grell et al. (2000).


37 The MM5 step of the modeling program is unquestionably the most complex and temperamental. This study utilized MM5Â’s multi-processor mode (MPP), which involved using a different set of programs than single-processor mode. Using MPP saves computing time by assigning the tasks of the program to several computers at one time. In fact, any model runs with more than 3 domai ns require the use of MPP. For this mode, the user must specify how many processors will be needed for the model run, and what the configuration of the processors will be. INTERPB INTERPB vertically inter polates the MM5 output from sigma levels back to pressure levels, so that it can be used for meteorological and statistical analysis. The output from INTERPB was converted to ASC II format for two purpos es: validation of the wind and geopotential height field at the upp er air stations, and as input for particle fall modeling. GRAPH and RIP The GRAPH program is a user-oriented vi sualization tool that provides quick plots of output from any of the MM5 steps. It can be used to look at topography, view skew-t plots, 2-dimensional plots of winds for any of the output domains, cross sections, and many other forms of output. To run GR APH, a table named g_plots.tbl must be edited so that settings match the MM5 model runs, and to specify which results of the model need to be plotted. Because GRAPH was used to visualize many different output


38 files, the entire process will not be discussed. While GRAPH is a very useful tool, it is not a required step in MM5 modeling. RIP is another plot utility available for MM 5, but it was not used for this project. There are two utility programs, readv3.f and ieeev3.csh, which are used to convert MM5 output data to manageable text files. Post processing After the MM5 step was complete, one of more of the individual domain output data files, named MMOUT_DOMAINx (Domain 5 in this case), were interpolated from sigma levels back to pressure level data, using INTERPB. The resulting file was named MMOUTP_DOMAINx. A FORTR AN script modified from the original readv3.f program provided with the MM5 package was next run to extract u and v -wind and geopotential height for a user-specified thre e-dimensional grid within the domain in question. For the 1995 eruption, this was a 40 km ( x -direction) by 32 km ( y -direction) by 20 vertical pressure level grid. The pre ssure levels ranged from 100 hPa, increasing incrementally by 50 hPa up to 1000, and 1001 hPa (essentially the surf ace). The program outputs 4 files: one for u -wind, v -wind, geopotential height, and one for corresponding pressure level. These output files were th en downloaded so that particle fall modeling could be performed. A PERL script was then run to prepare the output data for use. The script accessed the four MM5 files, then extracted a nd stored the information in an array. The script next used the geopotential height in formation to vertically interpolate both the u and v -winds, from the initial 20 pressure levels to 34 height levels. The height levels


39 have a 500-meter vertical interval, ranging from 0 meters altitude to 16,500 meters. In addition, the wind values were temporarily multiplied by 100 and converted to integers, for easier data storage. The horizontal grid spacing is approximately 1 kilometer, and its extent was based on the geometry of the 1995 tephra deposit. The grid was expanded after initial runs to track particles that ma y not be predicted to fall within the mapped deposit area (Figure 3.6). The out put file was formatted to input directly into the particle fall program. Figure 3.6. The 40 km by 32 km grid used in the particle fall model, showing mapped deposit isopachs in centimeters (from Hill et al. 1998). For the wind model comparison, MM5 winds were also extracted at 9-kilometer resolution in the same way. These data were then applied to the particle fall grid.


40 Reanalysis winds Because MM5 winds are compared to ot her wind models, archived global retrospective wind data was collected from the National Centers for Environmental Prediction (NCEP) at http://www.cdc.noaa .gov/cdc/data.ncep.reanalysis.html. The download requires the user to specify the da tes needed, the area of interest bounded by specified latitudes and longitudes, the type of data required (in this case u and v -wind and geopotential height), and the time interval of the data. Extraction of the data required a pre-processing step, using a command-line sc ript called ncdump provided with the data. The reanalysis wind data consists of a matrix of points spaced 2.5 degrees apart along both latitude and longitude. Each data point in the grid consists of 4 times daily observations of u -wind, v -wind, and geopotential height ( h ), at 0000 UTC, 0600 UTC, 1200 UTC, and 1800 UTC, at 17 different pre ssure levels, from 10 hectopascals (hPa) down to 1000 hPa. Using geopotential height data, the wi nd information was first interpolated vertically, to the particle fall grid used for MM5 winds. Using a 1/d distance-weighted mean, the wind data for 0000 UTC and 1200 UTC at each altitude were then interpolated horizontally, from the two closest NC EP grid points to each of the 1280 x y grid points in the particle fall grid. Figure 3.7 shows the significant differen ce in resolution between the MM5 output winds and the NCEP reanalysis winds. At 12 degrees north, 2.5 degrees of longitude are approximately equal to 270 kilometers. A grid cell from NCEP data is 900 times larger than the “low-resolution” 9-kilometer MM5 output, and 73,000 times larger that a 1 square kilometer MM5 grid cell.


41 Figure 3.7. A comparison of grid cell re solution for the wind models. MM5 highresolution winds have a cell size of 1 km, depicted as the tiny pink boxes. MM5 midresolution winds, at 9 km cell size, are the la rger black boxes. NCEP reanalysis winds have a resolution of 2.5 degrees, which co rresponds to 270 km at this latitude. Particle Fall Model The particle fall model used in this study uses a simple approach. The number of parameters involved in tracking particles as they erupted from a volcanic vent is enormous. To test the suitability of MM5 winds for tephra dispersion modeling, a


42 number of assumptions were made. First, tephra particles from the 1995 eruption are all assumed to be released into the atmosphere at the exact same location, at an initial height of 2400 meters above sea level, directly a bove the center of the volcanic vent. Ground observations during the erupti on estimated a maximum column height of 2000 to 2500 meters. This release assumes stationary par ticles, i.e. there is no upward or horizontal motion left over from the erupti on. In reality, particles emer ge from the buoyant eruptive column over a range of altitudes, and may have residual momentum. In addition, the program does not take into account any complex particle / atmosphere interaction. Although friction in the vertical direction is accounted for in calculating the terminal fall velocity of each particle, no friction in the horizontal direction is modeled. Turbul ence, eddies, and convection are also neglected in the model. Horizontal motion of tephra particle s is determined solely from MM5-predicted winds. Three-Dimensional Grid The three-dimensional grid that contains the wind fields wa s designed to enclose the 1995 tephra deposit out to the 0.1-centime ter isopach, as described by Hill et al. (1998). For reference, the grid extends r oughly 38 kilometers west of Cerro Negro, 2 kilometers to the east, 10 kilometers to the north, and 22 kilometers to the south. The x y dimensions of the grid correspond to the (r oughly) 1 kilometer by 1 kilometer spacing of the innermost domain in the MM5 setup. Th e domain grid centers were converted to latitude-longitude points within the model, and then these points were in turn projected into UTM Zone 16N, so that a metric scale co uld be used. This reprojection of the MM5


43 Domain 5 coordinates from deci mal degrees to UTM caused grid cell locations to shift slightly. The reprojection revealed that th e Domain 5 cells are actually closer to 978 meters east-west by 972 meters north-south, on average, over the 40 by 32 cell x y grid. Particle Fall Program The particle fall program stores u and v wind fields within the 40 x by 32 y by 34 z grid. The respective position of each released particle is advanced second by second, based upon which three-dimensional grid cell the particle falls within. When a particle crosses from one cell to another, it acquires th e horizontal velocity field of the new cell, while falling according to its calculated termin al velocity. The particle continues this motion until its z -coordinate reaches zero, wh ich is sea level. Fall velocity As a particle is “released” into the wi nd grid, it is adv ected in one second increments by the wind vectors it encounters based on its position. A positive value for u means winds are blowing from west to east, and a positive value for v means winds are blowing from south to north. If a particle is within a wind field with values of –5 meters per second for u and –2 meters per second for v the particle will move 5 meters to the west and 2 meters to the south each second, until the particle crosses into a new grid cell. Tephra particles are generally oblong and ca nnot be regarded is spheres in modeling (Armienti et al. 1988). The settli ng velocity of a particle is tied mainly to its dimensions and density. Particles between 1 m and 1 cm can reach their settling velocity in 5 meters or less, so they can be assumed to have this fall velocity upon release (Armienti et


44 al. 1988). Particles fall verti cally as determined by Suzuki ’s (1983) equation for settling velocity. (Eq. 3.1) f a t f f t op g p p g v 07 1 5 1 81 9 ) (3 64 0 2 232 0 where (Eq. 3.2) a c b fp p p p 2 where t = tephra density in kg/m3, a = air density in kg/m3 = air viscosity in Ns/m2, g = gravitational acceleration in m/s2, pf = particle shape factor, and = grain size. In Equation 3.1, however, most have the variable s are held constant, at assumed values. The tephra density t was set at 1100 kg/m3. Air density a is assumed to be 1.229 kg/m3, whereas the true air density is inversel y related to altit ude. Air viscosity is set at 1.73 x 10-5 Ns/m2, and gravitational acceleration g is 9.8 m/s2, a common assumption. The particle shape factor, pf, is a dimensionless parameter held at 0.5. This would occur for tephra particles where th e major axis is twice as long as either of the minor axes of the particle. Grain size is the one “variable” in the Suzuki e quation that was truly varied. Diameter was tested at 0.26, 0.5, 1, 2, and 5 mm. The 0.26 mm grain size was chosen because it produced a simple terminal fall ve locity of 1.0 meter per second, which made error checking easier. Becau se air viscosity and air density are held constant, this particle fall model should only be applied to the lower atmosphere. These variables change significantly with height.


45 Model Validation To check the accuracy of the MM5 predictions, the model output was compared with actual meteorological observations. Atmospheric soundings were generated by model runs by horizontally interpolati ng MM5 output to the latitude/longitude coordinates of each of the meteorological stations. The model predictions were compared both visually and statistically to the observations, to ensure that the MM5 forecast was sufficiently accurate to use for the particle fall model. Initially, large errors between model and observed values were found, but this was simply due to inconsistencies regarding ordering of x y coordinates. The variables tested statistically are twice daily winds ( u and v directions) and geopotential heights for the 850 hPa, 500 hPa, and 200 hPa levels. These heights are representative of the atmospheric boundary layer, the middle-tropospheric steering le vel, and the approximate height of the subtropical jet stream during winter, exhibiting upper atmo spheric flow. To quantify model error over time and space, the root mean square (RMS) errors for horizontal wind components u and v geopotential height h were computed for the available soundings in the outermost domain. In general, model simulations are not exp ected to be perfect, but rather to be reasonable. The observational data being used to initialize the model are scarce; all soundings are from more than 380 kilome ters from the volcano, and there are no soundings at all for Nicaragua. Because of the absence of soundings, it is impossible to directly validate MM5 in the Cerro Negro vi cinity. Furthermore, high-resolution models can resolve small circulation features that are sometimes not appa rent in observational data. In the study verifying pr ecipitation in the Mesoscale Al pine Program, Ferretti et al.


46 (2003) found that, even with an exceptionall y high-density observation network, some observation sites missed nearby features. Their RMS errors for the horizontal wind components ranged from 2 to 9 meters per second during their intense observation periods. In addition, White et al. (1999) showed that when using gridded data, large errors could be found at high resolution. Howeve r, gridded data can be better analyzed at different vertical levels over a wide area. Consequently, in this study, gridded data from the model forecast were generated and analyzed at various levels and compared with horizontal NCEP reanalysis data Another method to verify the accuracy of the model is comparison of the predicted airborne tephra dist ribution and tephra deposition pa tterns with contour maps of tephra thickness and satellite images of th e tephra plume. The latter methodology was used by Turner and Hurst (2001) in their si mulation for Ruapehu, New Zealand. Similar methods were used in this study. Sounding data Before any statistical analyses could be conducted to check MM5 performance, the soundings that were downloaded for each of the 19 stations had to be formatted for statistical analysis. A PERL script was writte n that removes header information, extracts the 4 variables (pressure, geopotential height wind direction, and wind speed) from the 11 variables included in each sounding, calculates u and v -wind for each observation, and appends the soundings from November and December into one file for each station. Only three pressure levels are considered for model validation, 850, 500, and 200 hPa, so all observations at other pressu re levels were discarded.


47 Root Mean Square errors To compare MM5 output directly to ob servations, simulated soundings were created by horizontally interpol ating data from the nearest grid points in the outermost domain (Domain 1), and vertically interpol ating from terrain-following sigma levels to pressure levels. Observed sounding data was compared to MM5 predicted values at each of the 19 meteorological stat ions, at three levels, every 12 hours from 27 November 0000 UTC to 4 December 0000 UTC. The Root Mean Square (RMS) errors were calculated, with the goal of validatin g MM5 performance. Two separate sets of RMS errors were calculated for model validation. This was first done station by station. The difference between the model-pred icted value and the observation given by a sounding is defined as th e model error for a pa rticular variable ( u v or h ), at each time, pressure level, and stati on. Ideally, there would be an error for each of the 19 stations at each pressure level (850, 500, and 200 hPa) for each variable. This corresponds to 171 unique values. In the station-by-station me thod, the squared errors of each variable (i.e. u -winds at 850 hPa, height at 200 hPa, etc.) were summed over the 9day model run. This sum was then divided by the number of entries, and the square root of this quotient is the RMS error for the variable. The model performed significantly better at certain stations, but this may be re lated to the number of available soundings for each station. Some stations had sounding data available every 12 hours over the entire duration of the eruption for a total of 57, wh ile Leticia, Colombia had only 3 soundings in total. One complete sounding was removed from the data set because of anomalous values. Station 3, Bogot, had an unreasona ble geopotential height value at 500 hPa on November 28 at 1200 UTC. The u -winds were found to have an anomalously high error


48 for this sounding too, suggesti ng that the instrument used on this date was not working correctly. The RMS errors for each station are summarized in Ta ble 3.2, and Figure 3.8. Table 3.2. RMS errors for the 19 meteorological stations. “Total Obs” shows how many observations were available for each stat ion for the model simulation. “U 850” represents the RMS error, in meters per second, for u -winds at the 850 hPa level. RMS errors for geopotential height h are in meters. There are no observations for Bogot at the 850 hPa level because the surface pressure at 2,600 meters is lower that 850 hPa. The method for validating model consiste ncy compared the overall results hour by hour. This required summing the squares of the errors of all stations, for each sounding time. The RMS errors for the 850 and 500 hPa levels at most stations were less than 3 m/s for both wind variables, and 20 me ters for height (Figures 3.9, 3.12, and 3.13). U-winds For east-west ( u ) winds, the model seems to perform better closer to the surface. The RMS error at 850 hPa was the lowest for 13 of the 19 observation times. The 500 hPa u -winds had slightly higher e rrors than the 850 hPa, while the error for the 200 hPa winds were the highest and highly variable.


49 Figure 3.8. Geographical distribu tion of station RMS errors for the 850 hPa level of the atmosphere. Values for u -wind (red) and v -wind (yellow) are in meters per second, while values for geopotential height h (blue) are in meters, at a different scale. Looking at a plot of u -wind error vs. time (Figure 3.9), it is clear that the model performed better, at least in the lower atmos phere, for observations that were recorded at 0000 UTC (Hours 0, 24, 48Â…), as opposed to 1200 UTC (Hours 12, 36, 60Â…). Assuming all things in the model are relativel y consistent, the likely explanation for this artifact is that the model is simply more accura te at those stations that have consistent


50 soundings at 0000 UTC, while the model does not perform as well at certain stations that only reported at 1200 UTC. Differences may also be due to the MM5 not adequately representing changes in the boundary layer due to the diurnal cycle. During daylight hours, there is more convective turbulence, and a higher boundary layer, as heat, momentum, and moisture are mixed up from the surface. At nighttime, the convection diminishes dramatically, there is more lami nar flow, the boundary laye r is shallower, and mixing is more mechanical (Mapes et al. 2003a). RMS Error by Model Hour0 1 2 3 4 5 6 7 8 024487296120144168192216HourU-wind RMS error (m/s) 850hPa 500hPa 200hPa Figure 3.9. RMS error over time for u -winds. The 850 hPa le vel, lowest in the atmosphere, generally has the smallest erro r, shown in blue. The 200 hPa level, in yellow, consistently has th e highest errors over time. It is worthwhile, then, to compare the model consistency over time using all the 0000 UTC soundings and all the 1200 UTC soundi ngs separately (Figures 3.10 and 3.11). Here we can see that the u -wind errors are fairly consis tent through the duration of the model run at the two higher pressure levels ranging between 1 a nd 3 meters per second at the 0000 UTC observations, and between 2 and 4 meters per second at the 1200 UTC


51 RMS Error by Model Hour 0000 UTC0 1 2 3 4 5 6 024487296120144168192216 HourU-wind RMS error (m/s) 850hPa 500hPa 200hPa Figure 3.10. RMS error over time for u -winds, 0000 UTC only. RMS Error by Model Hour 1200 UTC0 1 2 3 4 5 6 7 8 12366084108132156180204 HourU-wind RMS error (m/s) 850hPa 500hPa 200hPa Figure 3.11. RMS error over time for u -winds, 1200 UTC only.


52 observations. At 200 hPa, however, th e 0000 UTC observations show a more pronounced up-and-down variation of error in the model. For the Cerro Negro eruption, though, it is important to remember that the eruption column did not exceed 2,500 meters in altitude, and the 200 hPa height is generally above 12,000 meters. V-winds For the north-south ( v ) winds, MM5 performed best at the 500 hPa level, with an RMS error mostly between 1 and 3 meters per second (Figure 3.12). The error for 850 hPa was only slightly higher, while the 200 hPa level again had highly variable RMS errors, mostly ranging from 3 to 6 meters per second. The up and down pattern from 0000 UTC to 1200 UTC that is present for the u -wind RMS error is not apparent here. RMS Error by Model Hour0 1 2 3 4 5 6 7 024487296120144168192216HourV-wind RMS error (m/s) 850hPa 500hPa 200hPa Figure 3.12. RMS error over time for v -winds. Here the lowest errors are found at the 500 hPa level, although all error is relatively consistent over time.


53 Geopotential Height Geopotential height RMS errors follow a dis tinct trend: the lowe st pressure level, 850 hPa, has the lowest errors and the lowest variation of error, ge nerally between 10 and 20 meters (Figure 3.13). As pressure decreases, both the error and th e variation in error increase. Errors at 500 hPa mostly range from 15 to 25 meters, while errors at the 200 hPa level vary from about 25 to 40 meters. De spite two peaks at all three levels at hour 72 and hour 216, there does not seem to be any distinct temporal tre nd that would suggest that the MM5 is inconsistent over the 9-day period. Looking at the time series for each of the three variables used in model validation, no dubious time-dependent trends of model erro r are apparent. No runaway processes are present in the model, meaning that the fo recast values do not become more and more unrealistic over time. This is significant cons idering error trends e xhibited in the study by White et al. (1999). RMS e rror increased in almost every subsequent model time step, for every atmospheric model tested, meaning th at model error at 0 < 12 < 24 < 36 hours. Higher errors after 36 hours were believed to be the result of air masses from the Pacific Ocean moving over the research domain, the western United States. The Pacific is essentially a data “void,” since very little at mospheric data can be collected in the region. For the Cerro Negro study, however, the volcan o is downwind of the Caribbean Sea, and there is a good network of weather stations both on is lands and along the mainland perimeters of adjacent water bodies. If th e prevailing wind in this area of the tropics moved west to east, this st udy would probably have encountered the same “data void” problems as the study by White et al. (1999).


54 RMS Error by Model Hour0 5 10 15 20 25 30 35 40 45 024487296120144168192216HourGeopotential Height RMS error (m) 850hPa 500hPa 200hPa Figure 3.13. RMS error over time for geopotential height. E rror increases with height, but is consistent over time. Geopotential height RMS error for this study was significantly lower on the whole than those reported by White et al. (1999) A comparison is shown in Table 3.3. Level Error (m)-White et al 1999Error (m)-Cerro Negro 700 / 850 26.3 12.9 500 30.6 20.6 300 / 200 42.4 27.0 Table 3.3. Root Mean Square error for the Wh ite et al. (1999) study vs. this project. White et al. (1999) tested the 700, 500, and 300 hPa levels, while this study used the 850, 500, and 200 hPa levels. Summary In summary, model RMS error increased w ith height, and error values compared favorably to similar studies. Mapping and cr eating a time series of model error did not reveal any spatial or temporal trends in the simulation. One factor that likely influenced station RMS error was availability of s oundings. Stations with fewer observations


55 generally had higher error. Model valida tion has demonstrated that MM5 produced realistic simulation of wind fields for this study.


56 Chapter Four Descriptive Results Cerro Negro Cerro Negro volcano is situated on th e northwest slope of Volcan El Hoyo, located among the central Marrabios Range, wh ich runs parallel to NicaraguaÂ’s Pacific coast. Other Nicaraguan volca noes located this chain are Ma saya, Telica, San Cristobal, and Momotombo, all of which are capable of tephra-producing eruptions. The chain runs northwest to southeast thr oughout Central America, marki ng a convergent plate boundary where the Cocos plate is subducted unde r the Caribbean plate (Figure 4.1). Figure 4.1. Tectonic setting of Cerro Negro. The Cocos Plate is moving to the NE as it subducts under the Caribbean Plate. Modifi ed from Scripps Institution of Oceanography.


57 Standing approximately 728 meters in el evation, Cerro Negro is located at 12.5 degrees north latitude, 86.7 degr ees west longitude (Figure 4.2) It sits 20 kilometers east-northeast of Len, which is NicaraguaÂ’s educational and cultural capital, and at approximately 150,000 residents, its second-largest city. Figure 4.2. The location of Cerro Negro within the Americas. 1995 Eruption Parameters A basaltic cinder cone, Cerro Negro has had 23 eruptions since its formation in April 1850. The 1995 eruption took place be tween November 19 and December 6. Beginning with a series of relatively weak Strombolian eruptions followed by a lava


58 flow, the event maintained a continuous eruption column between November 29 and December 2, the period of the most prolific tephra output. The total volume of tephra erupted was estimated at a dens e rock equivalent of 1.3 x 106 m3 (Hill et al. 1998). The eruption produced a bent-over plume (F igure 4.3), indicating that the vertical velocity of the eruption column was the same order of magnitude as the crosswinds it encountered. Figure 4.3. The 1995 bent-over plume of Cerro Negro. The plume is moving towards the west-southwest. The photo was taken at 9:50 am on December 1, 1995. Courtesy of Dr. Brittain Hill.


59 The maximum column height was calculate d at 2400 meters, and visual estimates support this calculation (Hill et al. 1998). Th e Volcanic Explosivity Index (VEI) for the 1995 eruption was 2. For comparison, the 1991 er uption of Mt. Pinatubo had a VEI of 6, and the 1992 eruption of Cerro Negro had a VEI of 3 (Connor et al. 2001). Although the 1995 eruption did not cause any human fatalities, it forced the evacuation of 1,200 people, and totaled $700,000 of damage to agriculture and infrastructure (Connor et al. 2001). Tephra Deposit Pattern The tephra deposit from the 1995 erupti on was sampled immediately after the eruption, and isopachs were calculated using a geogra phic information system, commonly known as GIS (Hill et al. 1998). Thicknesses of 30 centimeters extend 2 kilometers west of the vent, and the deposit was about 0.5 cm thick as far away as Len. The most striking feature of the deposit is th at it has two lobes. The major axis runs west-southwest directly from Cerro Negro to Len; the minor northern axis is directed west-northwest, towards the town of Telica (Figure 4.4). The deposit is only mapped to the 0.1-centimeter isopach, and th e exact location of this is opach is uncertain due to difficulties in sampling such a small thickne ss, even a short time after accumulation. Results The trajectories predicted by the high-re solution MM5 wind fi eld correspond very well to deposit measurements. Figure 4.5 s hows all trajectories generated for the 5 chosen grain sizes, from 27 November 0000 UTC to 4 December 0000 UTC. The


60 coordinate system used in all particle traject ory figures is Universal Transverse Mercator (UTM) zone 16 North (16N), North American Datum 1927 (NAD27). Figure 4.4. Tephra deposit distribution from Cerro Negro. From Hill et al. (1998). Isopachs are in centimeters. Cerro Negro itsel f is within the 50 cm isopach (upper right). Only 2 of the 75 particle paths terminat e outside of the 0.1-centimeter isopach, and the path locations appear to mimic the thickness contours rather well on the whole. These trajectories can account for deposition al ong both major axes of the deposit. The trajectories demonstrate that wind speed and direction chan ged significantly during this 216-hour modeled period. Figure 4.6 highlig hts the 0.26 mm diameter particles. The more distant paths that reach Len represen t stronger winds encountered by these smaller particles, nearing 9 meters pe r second. Some of the shor ter paths for small particles terminate in the middle of the mapped depos ition region, and represent mean horizontal wind speeds of 4 meters per second. The two trajectories that terminate beyond the outermost isopach do not deviate dramati cally from the major axes of deposition.


61 Figure 4.5. MM5-forecast trajectori es with 1995 deposition pattern. Figure 4.6. MM5-forecast trajec tories for 0.26 mm particles.


62 Wind shifts over time The major axis of dispersion shifts betw een the azimuths of 240 and 280 degrees over the course of the 9-day model run, a range of 40 degrees. For the most part, the winds encountered by the majority of erupted particles are just north of easterly (70 degree azimuth, towards 250 degrees), as evidenced by the deposit. Bilobate Deposit Figure 4.7 is a vertically ex aggerated three-dimensional view of the particle trajectories, exhibiting how the northern par ticle tracks highlighted in blue can account for the creation of the smalle r northern lobe of the 1995 deposit. These tracks were generated from wind fields on November 27th, 28th, and 30th, all at 1200 UTC. According to MM5 winds, there is clearly a diurnal wind shift during the first few days of the eruption. Morning winds (1200 UTC) have a greater tendency to have a southerly component, blowing tephra to the northwes t, while evening winds (0000 UTC) tend to have a northerly component, distribut ing tephra towards the southwest. Wind shifts with altitude Figure 4.7 also demonstrates how wind velocities change significantly with altitude. The most dramatic shifts in wind direction in th e entire wind grid domain are found just above the surface during most even ings (0000 UTC). It is most pronounced on November 30th, December 1st, and December 2nd. Several trajectories, especially when viewed in two dimensions, appear to take sharp turns back toward the vent near the ground surface. In three-dimensional view, it is clear that winds in the bottom 750


63 meters of the atmosphere generally decrease and may reverse directi on, such that in the very bottom layer of grid cells, these winds are slightly out of the west instead of easterly. In Figure 4.7, this is evidenced by the trajec tories in the foreground just east of Len. Cox et al. (1998) illustrated that weak wi nds are generally associated with higher deviations in direction than are strong winds, so we would expect the most variable winds in the model to come from the atmospheric boundary layer. Figure 4.7. Three-dimensional view of MM5-p redicted particle trajectories. Blue trajectories represent contribu tions to the northern lobe. 4 Isopachs are symbolized are as varying gray fill. The cities of Len and Telica are shown as orange and pink polygons. Wind model comparison – particle trajectories Figures 4.8, 4.9, and 4.10 show predicted part icle trajectories resulting from a 9kilometer resolution MM5 run, NCEP reanalysis winds, and uniform winds, respectively. For trajectories generated by the high-resolu tion MM5 run, refer to Figure 4.4. It appears that the 9-kilometer resolu tion MM5 output has a slight shift towards the southwest


64 compared to the 1-kilometer output. This could be due to a stronger influence of the four-dimensional data assimilation (FDDA) nudging and a decreased role for localized topographic effects. Figure 4.8. 9-km resolution MM5-forecast traj ectories with 1995 deposition pattern. The particle trajectories produced by th e reanalysis winds seem to follow two major axes, which are not aligned with the major axes of the deposit. One axis runs toward the southwest at 235 degrees, and the ot her runs nearly west at 265 degrees. The major axes of deposition are approximately 250 and 280-degree azimuths. The reanalysis winds, then, predict trajectories that ar e 15 degrees over-rotated, counterclockwise overall. Seven of the 75 forecast particle tracks fall outside the 0.1-centimeter isopach, still a reasonably accurate prediction. Studyi ng the geometry of all the trajectories, it appears that the reanalysis winds have an overestimated northerly component, which


65 Figure 4.9. NCEP reanalysis wind traj ectories with 1995 deposition pattern. Figure 4.10. Uniform wind-predicted traj ectories with 1995 de position pattern.


66 results in tracks that push too far to the south. This appears true in particular on 27 November 0000 UTC, 29 November 0000 UTC, and 1 December 0000 UTC. The winds also produce fewer northerly tr acks that could account for the smaller lobe in the tephra deposit. The uniform wind field matched the large southern lobe of the deposit very well, but completely fails to account for the deposition in the northern lobe. Wind shifts over x-y space Wind shifts within any one layer of the three-dimensional grid for a particular time period were not severe. The grid domain in only 32 kilometers wide by 40 kilometers across, so sizeable changes in wi nd velocity should not be expected. In the case of a much larger eruption, a correspondi ngly larger wind grid would be necessary, and bigger changes in wind velocity would be expected. Figure 4.11 shows wind fields generated by MM5 and NCEP reanalysis da ta for 28 November 1200 UTC, at 500 meters altitude. The MM5 winds, shown in blue, have noticeable changes in direction, but very large changes in magnitude. In the northern central part of the gr id, MM5 winds exceed 20 meters per second, while in the northwest they are as low as 5 meters per second. NCEP winds are essentially uniform over the x y grid, with winds blowing from the northeast at just under 10 meters per second. The uniform wind assumption is similar to this observation. Standard devi ations of MM5 winds for every x y domain at each altitude level in the particle fall program we re calculated. Winds had the highest spatial variability closer to the ground, and temporally speaking, were more variable in the first few days of the simulation, November 27th – November 29th.


67 Figure 4.11. A wind field comparison. MM5 winds are shown in blue, and NCEP reanalysis winds are in red. Wind barbs are on the “tail,” indicating wind direction. Each small barb represents 5 meters per second, while long barbs are 10 meters per second. MM5 winds show much greater va riation in speed and direction. Wind model comparison Satellite Imagery In Meteorology, winds are conventionally described by the direction they are blowing from, i.e., a south to north wind would have an azimuthal direction of 180 degrees. In describing the tephra plume as it moves horizontally, it is more intuitive to describe it in terms of the direction to wh ich it is moving. Because the plume visible in these satellite images is primarily finer-grained tephra than what was selected in the particle fall model, a sma ller grain size of 0.01 mm diameter was tested, to simulate


68 particles that stay in the atmosphere longe r. These particles remain in the 2000-meter wind level for the duration of their paths in the wind grid, over 40 km from the vent. These trajectories should more closely resemble the visible plume on satellite imagery, as they stay in higher parts of the atmosphere longer. These higher-level winds have a greater variation in direction over the 9-day period. There is a more significant southerly component to these winds, as shown by 4 particle tracks moving northwest away from the volcano (Figure 4.12). Figure 4.12. 1-km resolution MM5-forecast tr ajectories for 0.01 mm particles. Particles 0.01 mm in diameter were run through the particle fall model using 9-km resolution MM5 winds (Figure 4.1 3) and NCEP reanalysis wi nds (Figure 4.14) as well. Trajectories from the three wind models we re then compared to satellite imagery.


69 Figure 4.13. 9-km resolution MM5-forecast tr ajectories for 0.01 mm particles. Figure 4.14. NCEP reanalysis wind-predict ed trajectories for 0.01 mm particles.


70 On November 27th at 1315 UTC (Figure 4.15), the tephra plume is barely visible on satellite imagery, and is appears very di ffuse. The plume is blowing to the west, towards the 265-degree azimuth. The MM5 (1 km) at 27 November 1200 UTC predicts winds blowing slightly to the west-northwest and then turning direc tly west. The 9-km winds blow the plume directly west. The NC EP reanalysis winds have the plume moving towards the 265-degree azimuth. On November 27th at 2215 UTC (Figure 4.16), cloud cover over the deposition area make discerning the tephra plume in visible satellite image extremely difficult. For 28 November 0000 UTC, the 1 km MM5 winds predict the high-level plume moving west-southwest directly over Len, while the 9 km winds take the axis just north of the city. The NCEP winds essentially mimic the 9 km MM5 winds at this time. On November 28th at 1315 UTC (Figure 4.17), the plume is clearly blowing to the north-northwest at an azimuth of about 320 degrees, and is visible to the Honduras border. MM5 (1 km) predicts its most nor thern track for 28 Nove mber at 1200 UTC, at an azimuth of 315 degrees. The 9 km winds ha ve the plume track slightly more to the west at 310 degrees, while the NCEP winds pred ict the plume axis at an azimuth of 285 degrees. On November 28th at 2215 UTC (Figure 4.18), the tephra plume appears to be blowing directly west from Cerro Negro, a nd is visible well over the Pacific Ocean, beyond the western extent of Nicaragua. For 29 November at 0000 UTC, both resolutions of MM5 predict tr acks moving to the west at 26 5 degrees, while NCEP winds take the plume over Len, at an azimuth of 245 degrees.


71 Figure 4.15. Satellite Imagery for November 27th at 1315 UTC (left). Figure 4.16. Satellite Imagery for November 27th at 2215 UTC (right). On November 29th at 1315 UTC (Figure 4.19), sate llite imagery shows that the plume is blowing towards the west-southwes t at an azimuth of about 235 degrees, is visible over Pacific Ocean, and appears to be sp reading laterally at a constant rate. The 28 November 1200 UTC MM5 forecast sends the plume to the west at 280 degrees at 1 km resolution, and 270 degrees at 9 km resolution. The NCEP winds move the plume southwest at 240 degrees, close to the satellite observation.


72 Figure 4.17. Satellite Imagery for November 28th at 1315 UTC (left). Figure 4.18. Satellite Imagery for November 28th at 2215 UTC (right). On November 30th at 2215 UTC (Figure 4.20), the plume is blowing towards west-southwest from the vent, towards approxi mately 250 degrees, and is clearly visible to the Pacific Ocean, at which point it appear s to bend towards the west-northwest. For 1 December at 0000 UTC, MM5 predicts the pl ume to move towards 250 and 245 degrees for 1 and 9 km resolution, respectively. NC EP winds bring the plume farther south, at 235 degrees.


73 Figure 4.19. Satellite Imagery for November 29th at 1315 UTC (left). Figure 4.20. Satellite Imagery for November 30th at 2215 UTC (right). On December 1st at 1315 UTC (Figure 4.21) the plume is not clearly distinct from clouds that are close to vent, but it appears to be moving towards the westsouthwest at 250 degrees. The 1 Decembe r 1200 UTC MM5-predicted winds take the plume towards azimuths of 255 and 260 degrees at 1 and 9 km resolution, while NCEP winds move the plume toward the 265 degree azimuth. On December 1st at 1745 UTC (Figure 4.22), the plume oscillates and fans between 265 and 235 degrees, and is visibl e over the ocean. This observation sits between two MM5 forecast times. For 2 December at 0000 UTC, MM5 high-resolution


74 winds move towards the west and then bend towards the west-southwest at 250 degrees. The 9 km MM5 winds take the plume toward s 260 degrees, while the NCEP winds move the plume towards the 240-degree azimuth. Figure 4.21. Satellite Imagery for December 1st at 1315 UTC (left). Figure 4.22. Satellite Imagery for December 1st at 1745 UTC (right). Summary In summary, the 1-kilometer MM5 winds exhibited the highest spatial variation and fit the tephra deposition pattern better than the 3 other wind models. MM5 windpredicted trajectories from both resoluti ons were found to be accurate within 10 azimuthal degrees of satellite observations, with the exception of one image. NCEP


75 reanalysis winds agreed with the deposit somewhat less, forecasting particle tracks slightly too far to the south. Errors between NCEP winds and satellite observations were significantly higher than thos e of MM5, on the whole. The uniform winds only agree with the southern lobe of the tephra deposit, and match satellite image very well in half of available images.


76 Chapter Five Discussion Why should volcanologists use these models? This study has shown that a mesoscale atmospheric model such as MM5 can provide a relatively accurate and high-resoluti on depiction of the winds that a volcanic plume will encounter after an eruption. A lthough this project focused on a historic eruption, mesoscale models are no less important as forecasting tools, and thus can be used in conjunction with eruption models to predict imminent tephra fall hazards for eruptions anywhere on Earth. MM5 would have heightened importance in regions where wind data are not readily available, such as isolated volcanic islands or impoverished nations, and where mountainous terrain causes localized wind effects for which a nearby sounding could not substitute. For large-scale eruptions, atmo spheric models, or gridded reanalysis data, at the very least, should be used as input for tephra plume models. With strong eruption columns that are able to reach the stratosphere, tephra can be advected thousands of kilometers by wind. Over such la rge areas, winds are more likely to vary in magnitude and direction, rendering the a ssumption of uniform wind inadequate. Model calibration The promising performance of MM5 suggests that the forecast winds can be used as a control in order to bette r calibrate eruption models. Ce rtain parameters describing an


77 eruption, as well as assumptions such as atmo spheric diffusion, can be adjusted, given the greater certainty MM5 provides in the exis tence of variable wi nd patterns over time. Should this be used for probabilistic work? As useful as atmospheric models can be for volcanic plume forecasting, a 5domain MM5 run, as used in this study, would be computationally expensive for probabilistic tephra hazard ma pping. Each additional domain (along with 2-way nesting) adds a large factor of co mputing time to the run. Th e setup as described under methodology required 2780 minut es of processing time, for a model run of 12960 minutes. Each minute of computing time accounted for 4 minutes and 40 seconds of forecast time. If one needed to generate 1000 possible scenarios for a weeklong model run of winds in order to determine the pr obability of a plume traveling in a certain direction, it could take over 4 years of computing time in 8-processor mode, which was found to be the most stable configuration on the cluster used. With some modifications such as reducing the size of each domain, li miting the number of scenarios required, and most importantly using 3 domains or fewer, th e enterprise becomes more practical. Should mesoscale models be used to forecast in near-real time? MM5 could be extremely useful for shor t-term forecasting of tephra dispersion for real-time eruptive events. Given ade quate input data for initial and boundary conditions, the MM5 is a dependable forecasti ng tool. Upper-air sounding data taken near an erupting volcano are certainly useful, but the data cannot be extrapolated very far


78 into the future. A forecast from MM 5, on the other hand, could provide a more comprehensive prediction of winds. Bilobate tephra deposition Ernst et al. (1994) studied the pheno menon of volcanic plume bifurcation resulting in a bilobate tephra deposit. They detail the 1981 eruption of Mt. Pagan in the Mariana Islands as the first we ll-documented bifurcated plume. Several eruptions within the last 30 years appeared to have had plum e bifurcation when inve stigated on satellite imagery. There are several possible contribu ting factors leading to plume bifurcation: buoyancy, release of latent heat, geometry and orientation of the plume source, and interaction with a density surface. A pl ume with two vortices and constant buoyancy should split into two distinct plumes that co ntinue to separate as they move downwind. Evaporation of water along the margins of a volcanic plume should increase the internal circulation in the plume and enhance bifur cation. If a double-vortex plume encounters a strong density interface such as the tropopause, the vortices ma y be forced to diverge. Ernst et al. (1994) asserted th at bifurcation is typical of bent-over and “straight-edged” plumes, but does not develop within strong plum es in weak crosswinds. Bifurcation in plumes can occur at a wide range of plume heights and wind velocities nonetheless. The paper investigates the bilobate 1968 Cerro Negr o tephra deposit, but concludes that this was likely caused by diurnal wind shifts rather than plume bifurcation (Ernst et al. 1994).


79 Wind shifts Ernst et al. (1994) stated that diffusion-advection mode ls that predict tephra deposition are applicable for eruptions with strong plumes, but not weaker plumes than can bifurcate. While tracking the central ax is of dispersion would be meaningless for a plume with two distinct axes, satellite imagery from the 1995 Cerro Negro eruption essentially dismisses this possibility. Imag ery indicates that the weak 1995 Cerro Negro plume did diffuse significantly, bu t did not bifurcate. Shifts in wind direction therefore account for the deposition pattern. Chronology of deposition Small lobe The output MM5 winds suggest that the smaller, northern lobe of the 1995 Cerro Negro tephra deposit was created primarily be fore and at the beginning of the major phase of the eruption, on November 27th and 28th. Satellite imag ery confirms that winds had a southerly component during this time, as evidenced by the tephra plume extending off to the west-northwest (Figure 4.17). Highlevel (fine) tephra cont inued to move to the northwest, while the larger particles settled and began to move more towards the west. Large lobe The larger southern lobe of the de posit was then created after November 28th through the remainder of the te phra-generating phase of the er uption. In several satellite images, tephra is seen reaching all the way to the Pacific Ocean. Thes e particles are finer


80 than most of the particles modeled in this project, and did not result in a permanent deposit. Model Limitations Resolution One of the drawbacks of a complex at mospheric model is the computing time required to make a reasonable, high resolution forecast. While the nested grid setup maximizes the efficiency of the model, spatial and temporal resolution problems still persist. The innermost model domain has a gr id size of one kilometer, which is not small enough to accurately describe fine topographi c features such as Cerro Negro itself. Although the volcano is identifiable in the MM5 Â’s digital elevation model, the entire height of the structure is not accurately depicted. Because the volcano is only about 1 kilometer in diameter, its topography is e ffectively smoothed over in this elevation model. A 30-meter DEM would probably be su fficient to represent the volcano, but this would certainly increase the overa ll computing time significantly. Particle aggregation A major drawback in the particle fall model, and indeed many volcanic plume models, is that it neglects of the effects of particle aggr egation. Although there was no significant evidence of particle aggrega tion during the 1995 eruption of Cerro Negro (Connor, personal communication), this phenomenon is believed to take place in many tephra-producing eruptions, and can occur due to electric charge at traction, precipitation, or condensation (Brasseur et al 1999). When particle aggreg ation does occur, clumps of


81 fine tephra are removed from the atmosphere mo re quickly than they otherwise would be, because the cluster essentially has a larger di ameter, and thus a higher fall velocity. Veitch and Woods (2001) detailed a number of examples of eruptions where particle aggregation significantly changes the pattern of tephra deposit. Fine-mode particles have been found to have constant diameter versus distance from the vent, suggesting that their distributi on was dictated by the size of the aggregate with which the joined. Liquid water in the atmosphere is thought to be the pr imary binding agent for these aggregates (Veitch and Woods 2001). Carey and Sigurdsson (1982) postulated that the double maximum created from the 1980 Mt. St. Helens deposit co uld be explained by an aggregation of particles smaller than 63 m in diameter clustering and dropping out of the atmosphere 325 km east-northeast of the volcano. Interaction with eruption column Although the MM5 digests many types of land surface information, including topography, vegetation, soil moisture, sea surf ace temperature, and others, it does not take into account the effects of the actual volcanic erupti on. Heat transfer, convection, vertical momentum, and density differences are essentially unknown quantities as far as MM5 is concerned. For this reason, there is no attempt in this proj ect to re-create the eruption column, instead, the te phra particles are “dropped” from the top of the column. There is no feedback from the eruption into the atmospheric model.


82 Turbulence While MM5 produces higher-resolution wi nds than most wind fields used in volcanology, it cannot recreate atmospheric pheno mena smaller than a few kilometers in width, exhibited in Figure 5.1 (Mann 2002). For this reason, small-scale eddies and turbulence within the plume cannot be simulated. Because the particle fall model only tracks particles along the major axis of dispersion, however the diffusive effects of turbulence are not of primary concern. Most plume models that attempt to consider turbulence use a parameterization such as a coefficient of diffusion. Figure 5.1. How features are represented in a mesoscale model. Only features of sufficient size can be represented.


83 Fall velocity The particle fall program, as stated befo re, makes many simplifying assumptions. The fall velocity of each particle is calculated solely as a function of its size, because all particles are assumed to have the same shap e and density. Changes in air density and viscosity are not considered. For this particul ar eruption, variability of the air column is not as crucial, because the particles are only falling thr ough the bottom 2400 meters of the atmosphere. For more powerful eruptions especially those whose columns reach the tropopause, changes in the air density beco me much more important to consider. Particles fall to zero To simplify the particle fall program, the tephra particles in the model are all assumed to fall to sea level (0 meters altit ude). True elevations in the region of the deposit range from 50 meters to wards the Pacific up to the ba se of the volcano, at around 450 meters altitude. Sources of Error There are several sources of error that can compli cate a model run. Certain atmospheric features will be lost if grid resolution is not high e nough. Interpolation from a low grid resolution to a high one may misrepre sent data, just as using a low resolution grid for fine features can resu lt in loss of data. Using th e wrong coordinate systems may distort features in the model. Physics schemes that may be applicable to some regions may not be appropriate for all. Many processes that need to be accounted for in mesoscale models are too


84 complex or are too small to be represented numerically. These phenomena are therefore parameterized, given best estimate s based on existing knowledge. These parameterizations can create grave consequen ces if mis-specified. Forecasts can end up being changed drastically as a result. Initial and boundary conditions of the m odel can also play a major role in the results. Models generally require a peri od in which atmospheric motions can evolve within it. This requires observation or fo recast data, which can potentially propagate errors into subsequent model predictions. Boundary conditions can also impact forecasts. The domain edge should be placed a reasonable distance from features of interest. MM5 requires that nested domains have a buffer from the edge of its parent domain. Despite these possible so urces of error, model va lidation showed that MM5 produced sufficiently realistic forecasts. The results of the particle fall model confirm that MM5 winds were highly accurate for the eruption.


85 Chapter Six Conclusions and Recommendations for Future Work Major Findings The Pennsylvania State University-Nati onal Center for Atmospheric Research (NCAR) fifth-generation Mesoscale Model ( MM5) was applied to the 1995 eruption of Cerro Negro, Nicaragua. The major axis of dispersion for this eruption was mapped as a series of particle trajectories for the peri od of highest tephra output, and showed very good agreement with the observed tephra deposit The trajectories for particles down to 0.26 mm in diameter fell almost entirely w ithin the 0.1 cm isopach for the eruption. It was also found that the axis of dispersion sh ifted dramatically in space and time over the course of the eruption. Ninety -seven percent of the axis tracks fall within the observed pattern of tephra deposition for the 1995 eruptio n. This is very promising, considering that MM5 had no input station observations within 380 km of the volcano, and did not utilize any data extrapolated from the tephra deposit. Although the eruption parameters and at mospheric conditions seemed to be favorable for plume bifurcation (Ernst et al. 1994), MM5 winds and satellite imagery show that the bilobate tephra deposit fr om the 1995 eruption can be accounted for completely by variation of winds. In additi on, evidence shows that th e northern lobe of the tephra deposit was produced, in large part, at the outset of the cont inuous phase of the eruption, on the 27th and 28th of November.


86 Statistical analyses revealed consider able spatial and temporal differences between MM5-generated wind fields and NCEP reanalysis winds as well as uniform winds. The NCEP winds appear to move te phra farther to the south than MM5 winds, while the uniform winds match satellite obs ervations for about half of the time. The significant shifts in wind direction between each 12-hour simulation, verified by satellite imagery, show that the assumpti on of uniform winds, even for a relatively weak eruption such as this, is faulty. Model winds vary over x y space noticeably, but vary by much greater differences for small changes in altitude. The 9 km resolution MM5 wind field was found to be adequate resolution for producing trajectories that match the tephra depo sit pattern. Reanalys is winds applied to the small 40 by 32 kilometer domain were reason able, but forecast the plume too far to the south for most time periods. The uniform winds were adequate in producing the larger, southern lobe of the deposit, but could not account for the northern lobe. The MM5 reproduced an accurate 4-dimensi onal wind field for the 1995 Cerro Negro eruption, confirmed by tephra de position patterns, satellite imagery, and meteorological station soundings. Future Research Improving the particle fall model Release Heights While the MM5 is a very complex atmos pheric model, the particle fall program used to track tephra for this study was simplis tic. There are a number of logical updates to the model that would make it more robust and realistic.


87 One relatively simple addition would be to incorporate varyin g release heights for the tephra particles. This would ideally be based on visual eviden ce or satellite imagery of the eruption column and volcanic plume, as they evolved from day to day. Particles would likely need to be released from multiple heights for each release time, as a single point source is not a realisti c approximation of any erupti on. Release heights from the eruption column could be modeled as a pr obability density function, such as the procedure used in Suzuki et al. (1983). Because the maximum column height of the 1995 Cerro Negro eruption is not believed to ha ve exceeded 2400 meters in altitude, the variation in release heights would remain entirely below this “ceiling” value. The resulting trajectories, which w ould therefore start from height s less than or equal to those used in the study, would thus e xhibit more proximal patterns. Effects on Column Rise Another modification to the particle fall model would allow it to take into account the effects of wind during buoyant column rise. This is more applicable to bent-over plumes, where the vertical velocity of the er uption column is close to the magnitude of the crosswind. Most plinian eruptions, on the other hand, are sufficiently explosive that crosswind has little effect on the column un til it approaches its maximum height. Strong plumes in weak winds generally exhibit the same behavior (Ernst et al. 1994). Diffusion term The current particle fall model simply tracks particles along the major axis of dispersion by advancing them through the wind field. A logical progression of the model


88 would be to add a diffusion term, so that sp atial distributions of tephra on the ground can be forecast. Topography Unlike the MM5 model, which uses t opography in creating a wind field for a particular region, the particle fall model does not have a built-in digita l elevation model. Particles are assumed to fall to zero elevat ion, which is a simple approximation of the true surface. In the region where the majo rity of tephra fell in the 1995 eruption, slopes are gradual, dipping to the we st-southwest, with elevati ons mostly between 50 and 300 meters. Another logical advance, then, would be to assimilate topography into the fall model, so that particleÂ’s motion would cease wh en it reached the elevation of surface at a particular x y location. Wind / friction effects The current version of the fall model assumes a uniform shape for each particle, which only affects the downward motion th rough the air column. The model already accounts for different particle diameters, whic h is the primary factor, in this setup, for determining the final distance from the vent. The particles are then simply assumed to be advected horizontally at the same velocity as the wind. The fall model could be adjusted by taking into account the complex effects of friction from winds. This friction affects the horizontal motion of particles, and varies depending on the shape of each particle.

PAGE 100

89 Additional future research Apply to other eruptions To test their viability more thoroughl y, the MM5 and particle fall models should be run to simulate tephra plumes of other we ll-studied eruptions. Tw o of these that have tephra deposits that have already been mapped extensiv ely are the 1992 Cerro Negro eruption, and the 1996 eruption of Mount Ru apehu in New Zealand. The model has given accurate results for a relatively small-s cale (VEI = 2) eruption, so applying it to larger and more far-reaching eruptions would do more to prove its value to volcanologists. Calibrate extent of diffusion One benefit of having model tr ajectories that vary in space and time is that they can be used to calibrate the extent of diffusi on more realistically. Values for the diffusion term calculated from uniform wind certainly wo uld lead to overestimates, because tephra would have to travel farther la terally away from the central ax is of dispersion. With axes that move over time, the amount of diffusion required to produce the observed deposition pattern would be smaller and almost certainly more accurate. In other words, variable winds can deposit tephra across a wide area off the major axis of deposition, whereas a uniform wind could only accomplish this if te phra was spread far off axis by diffusion, due to high turbulence.

PAGE 101

90 TEPHRA Model incorporation Continuing research shall involve usi ng MM5 output as input for the TEPHRA dispersion model. Currently, TEPHRA requires a single stratified wind profile as input. Thus it assumes that winds are uniform over each x y domain for each altitude. Downwind changes in direction and speed ca nnot be accounted for using this method, so the model is inherently limited. Coupling th e MM5 to the TEPHRA model would likely yield more robust results, especially as the wind assumptions break down over great distances.

PAGE 102

91 References Aloisi, M., M. D'Agostino, K.G. Dean, A. Mo staccio, and G. Neri, Satellite analysis and PUFF simulation of the eruptive cloud ge nerated by the Mount Etna paroxysm of 22 July 1998, Journal of Geophysical Research-Solid Earth 107, B12, 2002. Armienti, P., G. Macedonio, and M.T. Pare schi, A numerical-model for simulation of tephra transport and depos ition applications to Ma y 18, 1980, Mount St. Helens eruption, Journal of Geophysical Research 93, B6, 6463-6476, 1988. Barberi, F., G. Macedonio, M.T. Pareschi, a nd R. Santacroce, Mapping the tephra fallout risk: an example from Vesuvius, Italy, Nature 344, 142-144, 1990. Baxter, P.J., Medical effects of volcanic eruptions. I. Main causes of death and injury, Bulletin of Volcanology 52, 532-544, 1990. Becker A.; E. Schaller, and K. Keuler, Con tinuous four-dimensional source attribution for the Berlin area during two days in July 1994. Part I: The new Euler-Lagrangemodel system LaMM5, Atmospheric Environment 35, no.32, 5497-5508, 2001. Becker A.; E. Schaller, and K. Keuler, Con tinuous four-dimensional source attribution for the Berlin area during two days in July 1994. Part II: Sensitivity studies, Atmospheric Environment 35, no.32, 5509-5523, 2001. Bonadonna, C., G. Macedonio, and R.S.J. Spar ks, Numerical modelling of tephra fallout associated with dome collapses and Vulcanian explosions: application to hazard assessment on Montserrat, in The eruption of Soufrire Hills Volcano, Montserrat, from 1995 to 1999 edited by T.H. Druitt, and B.P. Kokelaar, 517537, Geological Society, London, Memoir, 2002. Bonadonna, C., and J.C. Phillips, Sedime ntation from strong volcanic plumes, Journal of Geophysical Research 108, B7, 2003. Bonadonna, C., C.B. Connor, B.F. Houghton, L. Connor, M. Byrne, A.G. Laing, and T.K. Hincks, Probabilistic modeling of te phra dispersal: Hazard assessment of a multi-phase rhyolitic eruption at Tarawera, New Zealand, Journal of Geophysical Research 110, B03203, doi:10.1029/2003JB002896, 2005. Brasseur, G.P., J.J. Orlando, G.S. Tyndall, eds, Atmospheric Chemistry and Global Change 1999.

PAGE 103

92 Bursik, M., Tephra Dispersal, Geological Society Special Publication no.145, 115-144, 1998. Carey, S.N., and H. Sigurdsson, Influence of particle aggregation on deposition of distal tephra from the May 18, 1980, eruption of Mount St. Helens volcano, Journal of Geophysical Research 87, B8, 7061-7072, 1982. Connor, C.B., B.E. Hill, B. Winfrey, N.M. Fr anklin, and P.C. La Femina, Estimation of volcanic hazards from tephra fallout, Natural Hazards Review 2 no. 1, 33-42, 2001. Cox, R.M., J. Sontowski, R.N. Fry, Jr., C.M. Dougherty, and T.J. Smith, Wind and diffusion modeling fo r complex terrain, Journal of Applied Meteorology 37, 9961009, 1998. D'Amours, R., Modeling the ETEX plume di spersion with the Canadian emergency response model, Atmospheric Environment 32, no. 24, 4335-4341, 1998. Ernst, G.G.J., J.P. Davis, and R.S.J. Sparks, Bifurcation of volcanic plumes in a crosswind, Bulletin of Volcanology 56, 159-169, 1994. Ferretti, R., T. Paolucci, G. Giuliani, T. Cherubini, L. Bernardini, and G. Visconti, Verification of high-resolution real-time forecasts over the Alpine region during the MAP SOP, Quarterly Journal of the Royal Meteorological Society 129, 587– 607, 2003. Gardner, C.A., K. M. Scott, C. D. Miller, B. Myers, W. Hildreth, and P. T. Pringle, Potential Volcanic Hazards from Future Activity of Mount Baker, Washington : USGS Open-File Report 95-498, 1995. Glaze, L.S., and S. Self, Ashfall dispersal for the 16 September 1986, eruption of Lascar, Chile, calculated by a turbulent-diffusion model, Geophysical Research Letters 18, no. 7, 1237-1240, 1991. Grell, G. A., J. Dudhia, and D.R. Stauffer, A description of the fifth-generation Penn State/NCAR Mesoscale Model (MM5), NCAR Technical Note NCAR/TN-398 + STR 122 pp, 1995. Grell, G.A., S. Emeis, W.R. Stockwell, T. Schoenemeyer, R. Forkel, J. Michalakes, R. Knoche, and W. Seidl, Application of a multiscale, coupled MM5/chemistry model to the complex terrain of the VOTALP valley campaign, Atmospheric Environment 34, no.9, 1435-1453, 2000. Heffter, J.L., and B.J.B. Stunder, Volcan ic Ash Forecast Transport and Dispersion (VAFTAD) Model, Weather and Forecasting 8, no. 4, 533-541, 1993.

PAGE 104

93 Hill, B.E., C.B. Connor, M.S. Jarzemba, P.C. La Femina, M. Navarro, and W. Strauch, 1995 eruptions of Cerro Negro volcano, Nicar agua, and risk assessment for future eruptions, Geological Society of America Bulletin 110, no. 10, 1231-1241, 1998. Hurst, A.W., and R. Turner, Performan ce of the program ASHFALL for forecasting ashfall during the 1995 and 1996 er uptions of Ruapehu volcano, New Zealand Journal of Geology and Geophysics 42, no. 4, 615-622, 1999. In, H., and S. Park, A simulation of long -range transport of Yellow Sand observed in April 1998 in Korea, Atmospheric Environment 36, no. 26, 4173-4187, 2002. Kalnay, E., M. Kanamitsu, R. Kistler, W. Collins, D. Deaven, L. Gandin, M. Iredell, S. Saha, G. White, J. Woollen, Y. Zhu, A. Leetmaa, B. Reynolds, M. Chelliah, W. Ebisuzaki, W. Higgins, J. Janowiak, K. C. Mo, C. Ropelewski, J. Wang, R.Jenne, and D. Joseph, The NCEP/NCAR 40-Year reanalysis project, Bulletin of the American Meteorological Society 77, 437–471, 1996. Kitada T., and R.P. Regmi, Dynamics of air pollution transport in late wintertime over Kathmandu Valley, Nepal: As reveal ed with numerical simulation, Journal of Applied Meteorology 42, no.12, p. 1770-1798, 2003. Latter, J.H. (ed.), Volcanic Hazards : Assessment and Monitoring 1989. Macedonio, G., M.T. Pareschi, and R. Santacroc e, A numerical simulation of the Plinian fall phase of 79 AD eruption of Vesuvius, Journal of Geophysical Research-Solid Earth and Planets 93, B12, 14817-14827, 1988. Macedonio, G., R. Santacroce, P. Papale, M. T. Pareschi, and M. Rosi, A statistical approach to the assessment of volcanic hazard for air traffic: application to Vesuvius, Italy, US Geological Survey Bulletin 2047, 245-252, 1994. Mann, G., Mesoscale models issues of resolution, presentation for COMET Mesoscale Analysis and Prediction (COMAP) training course, 2002. Mapes, B.E., T.T. Warner, and M. Xu, Diur nal patterns of rainfall in Northwestern South America. Part III: Diurnal gr avity waves and nocturnal convection offshore, Monthly Weather Review 131, no. 5, 830-844, 2003b. Mapes, B.E., T.T. Warner, M. Xu, and A.J. Negri, Diurnal patterns of rainfall in Northwestern South America. Part I: Observations and context, Monthly Weather Review 131, no. 5, 799-812, 2003a. Nuss, W.A. and D.W. Titley, Use of multiqua dric interpolation for meteorological objective analysis, Monthly Weather Review 122, 1611-1631, 1994.

PAGE 105

94 Salvador, R., J. Calbo, and M.M. Millan, Horizo ntal grid size selec tion and its influence on mesoscale model simulations, Journal of Applied Meteorology 38, no.9, 13111329, 1999. Searcy, C., K. Dean, and W. Stringer, PU FF: A high-resolution volcanic ash tracking model, Journal of Volcanology and Geothermal Research 80, no.1-2, 1-16, 1998. Scott, W.E., R.M. Iverson, J.W. Vallance, and W. Hildreth, Volcano Hazards in the Mount Adams Region, Washington : USGS Open-File Report 95-492, 1995. Seigneur, C., B. Pun, K. Lohman, and S.-Y Wu, Regional Modeling of the Atmospheric Fate and Transport of Benzene and Diesel Particles, Environmental Science and Technology 37, no.22, 5236-5246, 2003. Sigurdsson, H., and S. Carey, Volcanic di sasters in Latin America and the 13th November 1985 eruption of Nevado del Ruiz volcano in Colombia, Disasters 10, no. 3, 205-216, 1986. Simpson, J.J., J.S. Berg, C. Bauer, G.L. Hufford, D. Pieri, and R. Servranckx, The February 2001 eruption of Mount Clevel and, Alaska: Case study of an aviation hazard, Weather and Forecasting 17, no.4, 691-704, 2002. Suzuki, T., A theoretical model for dispersion of tephra, in Arc Volcanism, Physics and Tectonics edited by D. Shimozuru, and I. Yokoyama, pp. 95-113, Terra Scientific Publishing Company (TERRAPUB), Tokyo, 1983. Talerko, N., Mesoscale modelling of radioact ive contamination formation in Ukraine caused by the Chernobyl accident, Journal of Environmental Radioactivity 78, no. 3, 311-329, 2005. Tilling, R.I., C. Heliker, and T.L. Wright, Eruptions of Hawaiian Volcanoes: Past, Present, and Future : USGS Special Interest Publication, 1987. Turner, R., and T. Hurst, Factors influenc ing volcanic ash dispersal from the 1995 and 1996 eruptions of Mount Ruapehu, New Zealand Journal of Applied Meteorology 40, no. 1, 56-69, 2001. Veitch, G. and A.W. Woods, Particle aggregation in volcanic eruption columns, Journal of Geophysical Research B: Solid Earth 106, no.11, 26425-26441, 2001. Warner, T., Mapes, Brian E., Xu, Mei. Diur nal patterns of rainfall in Northwestern South America. Part II: Model simulations. Monthly Weather Review 131, no. 5, pp. 813-829, 2003.

PAGE 106

95 White, G. B., Peagle, J., Steeburgh, W. J., Ho rel, J. D., Swanson, R. T., Cook, L. K., Onton, D. J. and Miles, J. G., Shortterm forecast validation of six models. Weather and Forecasting 14, 84–108, 1999. Wolfe, E.W., The 1991 eruptions of Mount Pinatubo [Pamphlet], Earthquakes and Volcanoes 23, no. 1, US Department of the Interior, US Geological Survey, 1992. Woods, A.W., R.E. Holasek, and S. Self, Wind-driven disper sal of volcanic ash plumes and its control on the thermal structure of the plume-top, Bulletin of Volcanology 57, no. 5, 283-292, 1995. Zngl, G., Stratified flow over a mountain with a gap: Linear theo ry and numerical Simulations, Quarterly Journal of the Royal Meteorological Society 128, no.581, 927-949, 2002.

PAGE 107

96 Bibliography Bonadonna, C., G.G.J. Ernst, and R.S.J. Sparks, Thickness variations and volume estimates of tephra fall deposits: the im portance of particle Reynolds number, Journal of Volcanology and Geothermal Research 81 (3-4), 173-187, 1998. Bursik, M.I., S.N. Carey, and R.S.J. Spar ks, A gravity current model for the May 18, 1980 Mount St. Helens plume, Geophysical Research Letters 19, no. 16, 16631666, 1992a. Bursik, M.I., R.S.J. Sparks, J.S. Gilbert, and S.N. Carey, Sedimentation of tephra by volcanic plumes: I. Theory and its comparison with a study of the Fogo A plinian deposit, Sao Miguel (Azores), Bulletin of Volcanology 54, 329-344, 1992b. Carey, S.N., and R.S.J. Sparks, Quantitative mode ls of the fallout and dispersal of tephra from volcanic eruption columns, Bulletin of Volcanology 48, 109-125, 1986. Ernst, G.G.J., R.S.J. Sparks, S.N. Carey, a nd M.I. Bursik, Sedimentation from turbulent jets and plumes, Journal of Geophysical Research-Solid Earth 101, B3, 55755589, 1996. Heffter, J.L., Volcanic ash model veri fication using a Klyuchevskoi eruption, Geophysical Research Letters 23, no. 12, 1489-1492, 1996. Koyaguchi, T., and M. Ohno, Reconstruction of eruption column dynamics on the basis of grain size of tephra fall deposits. 1. Methods, Journal of Geophysical Research 106, B4, 6499-6512, 2001a. Mass, C. F., D. Ovens, K. Westrick, and B. A. Colle, Does increasing horizontal resolution produce more skillful forecasts?, Bulletin of the American Meteorological Society 83, 407-430, 2002. Phadnis, M.J., G.R. Carmichael, Y. Ichika wa, and H. Hayami, Evaluation of long-range transport models for acidic deposition in East Asia, Journal of Applied Meteorology 37, no.10, 1127-1142, 1998. Souto, J.A., J.J. Casares, T. Lucas ; V. Pe rez-Munuzuri, M. deCastro, and M.J. Souto, Forecasting and diagnostic analysis of plume transport around a power plant, Journal of Applied Meteorology 37, no.10, 1068-1083, 1998.

PAGE 108

97 Sparks, R.S.J., M.I. Bursik, G.J. Ablay, R.M.E. Thomas, and S.N. Carey, Sedimentation of tephra by volcanic plumes. 2. Controls on thickness and grain-size variations of tephra fall deposits, Bulletin of Volcanology 54, no. 8, 685-695, 1992.

PAGE 109

98 Appendices

PAGE 110

99 Appendix A: MM5 Specifications TERRAIN Changes to the original terrain.deck file are described here; th e default settings were used for anything that is not mentioned. The largest grid has a resolution of 81 km on a side, while the next smallest grid had a resolution of 27 kilometers (the “parent” to “child” resolution ra tio must be 3:1 for two-way nesting), starting at the point (24, 21) on the larg er nest. Two-way nesting means that the parent and child domains are able to interact in both directions, so values from one are passed to the other and vice ve rsa. The process of defining domains was continued until all 5 were set up properly. The TERRAIN output is used in all subsequent model programs, all using the domains described here, and some also requiring the terrestrial information. The TERRAIN setup takes place by editing th e terrain.deck file in the TERRAIN directory. The number of domains was changed from 2 to 5, and Domain 1 was recentered to 12.0 North, 85.0 West. The domain resolutions were changed to 81, 27, 9, 3, and 1 kilometer. IIMX and JJMX were each increased fr om 100 to 150. This pertains to the maximum size, in cells, of any domain in the setup. ITRH and JTRH were both increased from 500 to 1000. These higher values are necessary because of higher resoluti on input data than the default.

PAGE 111

100 PHIC was changed from 36.0 N to 12.0 N, to center Domain 1 on Central America. IPROJ was changed from ‘LAMCON’ to ‘MERCAT’, because Mercator is a more appropriate projection than La mbert Conformal at 12.0 degrees North latitude. MAXNES increased from 2 to 5, as a 5-domain setup was necessary. NESTIX was changed from 35, 49, 136, 181, 211 to 61, 49, 40, 49, 49 and NESTJX was changed from 41, 52, 181, 196, 211 to 61, 52, 40, 49, 58 to better fit the study area. DIS was changed from 90, 30, 9, 3.0, 1.0 to 81, 27, 9, 3, 1, in order to maintain the 3 to 1 ratio from “parent” to “child” domain. NESTI was moved from 1, 10, 28, 35, 45 to 1, 24, 17, 14, 17 and NESTJ was moved from 1, 17, 25, 65, 55 to 1, 21, 17, 12, 16 to better fit the volcano location. These values describe the position of a domain within its parent’s grid. When the following command is executed: “./terrain.deck”, the nested grids are set up, and all the surface data is compiled and interpolated to the grids. A successful run will create a log file, terrain.print.out, a nd files named TERRAIN_DOMAINx, where x represents the numeric order of the domain, 1 through 5. These files are used in all the subsequent steps of the MM5 modeling program.

PAGE 112

101 REGRID The pregrid setup required the following changes to the original pregrid.csh file, found in the /REGRID/pregrid/ directory: The “Data” directory was set to /MM5V3/REGRID/. The data source format was changed from ON84 to GRIB (gridded binary). The “InFiles” were changed from /NNRP_GRIB* to /nnrp/pgb* and /nnrp/SFCNNRP*, as these were the locat ions of the downloaded input data. The start date was moved from 3/13/1993 0000 UTC to 11/25/95 0000 UTC, and the end date was changed from 3/14/1993 0000 UTC to 12/04/95 0000 UTC, to bound the major phase of the Cerro Negro eruption. The pregrid step culminates in issuing the command “pregrid.csh >& log”, which generates a log file and a pletho ra of gridded data files. Ne xt, in the sub-step regridder, the data is converted to two-dimensional grids at each pressure level. This is virtually automatic, unless the user intends to add a dditional pressure le vels into the threedimensional grid. The REGRID output is pass ed on to the objective an alysis step and to INTERPF. The regridder step requires editing the na melist.input in the /PREGRID/regridder/ directory. Changes to the original template file are as follows: The beginning and end dates were changed to match pregrid.csh.

PAGE 113

102 The “root” files were pathed for GRIB fo rmat as is directed in the MM5 online tutorial: '../pregrid/FILE' '../preg rid/SST_FILE' '../pregrid/SNOW_FILE' '../pregrid/SOIL_FILE'. “constants_full_name” was changed to '../pregrid/SST_FILE:1995-11-25_00'. Issuing the command “regri dder >& log” creates a log file and the file REGRID_DOMAIN1, which is accessed in LITTLE_R and INTERPF. LITTLE_R The LITTLE_R namelist.input file was edited as follows: The beginning and end dates were ch anged to match previous steps. “obs_filename” was pathed to /FETCH/a dp_upa/ and pointed to a list of files named in a sequence from “obs:1995-11-25_00” to “obs:1995-12-04_00”, with the observation time increasing at a 6-hour interval. “obs_filename” also pointed to the sequence of files from “ upper-air_obs_r:1995-11-25_00” to “upperair_obs_r:1995-12-04_00”, also in creasing at a 6-hour interval. “sfc_obs_filename” was pathed to ./FETCH/a dp_sfc/ and pointed to a sequence of files from “surface_obs_r:1995-11-2 5_00” to “surface_obs_r:1995-12-04_00”, which increased at a 1-hour interval. “f4d” was switched from TRUE to FALSE, to turn off the creation of surface FDDA files.

PAGE 114

103 After these changes were made, the step is completed by executing “little_r >& log”. In addition to the l og file, the file LITTLE_R_DOMAIN1 is produced. This is used in the next step, INTERPF. INTERPF Changes to the namelist.input file for INTERPF involved simply changing the beginning and end dates to match the other step s of the model. The step is run by typing “interpf >& log”, which generate s a log file, and three files th at will be used in the MM5 step: BDYOUT_DOMAIN1, LOWBDY_DO MAIN1, and MMINPUT_DOMAIN1. MM5 After creating the mm5.deck in this folder by typing “make mm5.deck”, the file is ready to edit. Below are the steps taken to prepare the mm5.deck for the simulation. The identical changes were made to the “mmlif” file, to ensure agreement between the two files. Model runs were done within the /MM5V3/MM5MPP/ direct ory, rather than the /MM5V3/MM5/ directory. TIMAX was increased from 720 to 12960 mi nutes, indicating a 9-day model run. TISTEP was reduced from 240 to 180, for better model stability and temporal resolution. SAVFRQ was changed from 360 to 720 mi nutes, to match the 12-hour output interval. CDATEST, the date of the starting fi le, was moved from 1993-03-13_00:00:00 to 1995-11-25_00:00:00.

PAGE 115

104 LEVIDN was changed from 0,1,2,1,1Â… to 0,1,2,3,4Â… to represent all 5 model domains. NUMNC was changed from 1,1,2,1,1Â… to 1,1,2,3,4Â… to represent the parent of each domain. NESTIX, NESTJX, NESTI, NESTJ were moved according to the domain locations specified in terrain.deck. XSTNES was changed from 0,0,900,0,0Â… to 0,0,0,0,0 so that all domains are initiated at the outset of the model run. XENNES was increased from 1440,1440,1440,720,720Â… to 12960,12960,12960, 12960Â… so that all domains terminate when the model run is complete. IOVERW was changed from 1,2,0,0,0Â…to 1,2,2,2,2Â… so that the highestresolution topography is used for the nested domains, instead of being interpolated from the parent domain. FDAEND was increased from 780,0, 0,0,0Â… to 12960,0,0,0,0Â… so that gridded 4dimensional data assimilation will be applied for the entire model run. I4D was changed from 0,0,0,0,0Â… to 1,0,0,0,0Â… to turn on the gridded 4dimensional data assimilation. DIFTIM for surface analysis nudging was increased from 180,180,0,0,0Â… to 720,720,0,0,0Â… to match the interval for three-dimensional analysis nudging. Because MM5 requires MPP for any runs with more than three domains, a separate MM5MPP directory was created to ho use all the files for this mode. To run MM5 in either mode, the mm5.deck and mm lif files need to be edited, but multi-

PAGE 116

105 processor mode requires the spec ifications of a third file, co nfigure.user. The following steps detail how the configure.user file was edited specifically for this project. FDDAGD was switched from 0 to 1, indica ting that gridded 4-dimensional data assimilation will take place. MAXNES was increased from 2 to 5, because there are 5 domains in this setup. MIX and MJX were increased from 49 and 52 to 61 and 61, respectively, matching the maximum grid cells of any domain. IMPHYS was changed from 4,4,1,1,1… to 4,4,4,4,4… because the simple ice moisture scheme is needed for all 5 domains in the model. ICUPA was changed from 3,3,1,1,1… to 3,3,3,3,3… so that the Grell Cumulus scheme is utilized in all 5 domains. IBLTYP was switched from 5,5,0,0,0… to 2,2,2,2,2… meaning that the planetary boundary layer changed from MRF to High-resolution Blackadar. PROCMIN_NS was increased from 1 to 4. PROCMIN_EW was increased from 1 to 2, giving 2 x 4 = 8 total processors solving the MM5 for ecast in parallel. After all the changes had been made, th e MPP step was enacted by typing “make mpp” and then submitting the executable to a queue on the mimir server. The command “qsub mpp-parallel8.sub” calls up a file that te lls the server to run the model on 8 nodes. Many model runs needed to be performed in order to determine the correct settings and produce reasonable results. Afte r each model run, output files we re archived, or else they would get overwritten during the subsequent run. In addition to “make mpp”, which

PAGE 117

106 needed to be submitted before each run, tw o additional commands had to be executed before the next run could commence: “make mpclean” and “make uninstall”, both of which essentially clean up old files from the previous run. INTERPB The steps listed below were show change s to the namelist.input file in INTERPB: The input file to interpolate was re-p athed from /MM5/Run/ to /MM5MPP/Run/. MMOUT_DOMAIN1 was interpolated for the purposes of model validation, while MMOUT_DOMAIN5 was interpolated to produce the wind grid for the particle fall model. Start and end dates were changes to reflect the dates of the model run. The output interval was increased from 21600 to 43200 seconds (12 hours), to match the interval of available sounding data.


Download Options

Choose Size
Choose file type
Cite this item close


Cras ut cursus ante, a fringilla nunc. Mauris lorem nunc, cursus sit amet enim ac, vehicula vestibulum mi. Mauris viverra nisl vel enim faucibus porta. Praesent sit amet ornare diam, non finibus nulla.


Cras efficitur magna et sapien varius, luctus ullamcorper dolor convallis. Orci varius natoque penatibus et magnis dis parturient montes, nascetur ridiculus mus. Fusce sit amet justo ut erat laoreet congue sed a ante.


Phasellus ornare in augue eu imperdiet. Donec malesuada sapien ante, at vehicula orci tempor molestie. Proin vitae urna elit. Pellentesque vitae nisi et diam euismod malesuada aliquet non erat.


Nunc fringilla dolor ut dictum placerat. Proin ac neque rutrum, consectetur ligula id, laoreet ligula. Nulla lorem massa, consectetur vitae consequat in, lobortis at dolor. Nunc sed leo odio.