The University of Mississippi, Easson: Rapid Prototyping Capability for Earth-Sun System Sciences (RPC) Subaward 191000-361604-01

The University of Mississippi

The University of Mississippi Geoinformatics Center University, Mississippi 38677

Subaward No. 191000-361604-01 Mississippi State University

National Aeronautics and Space Administration Prime Award No. NNS06AA65D

Project Investigator: Contractual Contact: Gregory L. Easson, Ph.D. Robin C. Buchannon, Ph.D. Associate Professor and Chair Director of Sponsored Programs Geology and Geological Engineering Administration Room 118E, Carrier 100 Barr Hall University, MS 38677 University, MS 38677 (662) 915-5995 (662) 915-7482

Project Title:

Rapid Prototyping Capability for Earth-Sun System Sciences (RPC)

July 31, 2008

FINAL REPORT

The University of Mississippi, Easson: Rapid Prototyping Capability for Earth-Sun System Sciences (RPC) Subaward 191000-361604-01

FINAL REPORT SUMMARY Submitted by the University of Mississippi Geoinformatics Center Project Title: Rapid Prototyping Capability for Earth-Sun System Sciences (RPC)

Submitted to: Mississippi State University (MSU) Budget: $1,626,674.00 MSU Investigator: Dr. Robert Moorehead UM Investigator: Dr. Gregory L. Easson, Department of Geology and Geological

Engineering, 118E Carrier Hall, University of Mississippi, University, MS 38677—Phone: (662) 915-5995, Fax: (662) 915-5998, E-mail: geasson@olemiss.edu

Subcontracts: Universities Space Research Association (USRA) Science Systems & Applications, Inc. (SSAI) Dartmouth College Tennessee Technological University (TTU) Institute for Technology Development CATHALAC Project Duration: January 1, 2006 through June 30, 2008 Total Expenditures: $1,551,224.69 Unexpended Funds: $75,449.31 Reports: Integration of NASA Global Precipitation Measurement Mission Data into The SERVIR Flood Decision Support System for Mesoamerica

Rapid Prototyping: Modeling Seagrass Community Change Using Remote Sensing and Real-time Instrument Packages Rapid Prototyping of NASA Next generation Sensors with the Nonpoint- Source Pollution and Erosion Comparison Tool Use of NASA Satellite Assets for Predicting Wildfire Potential for Forest Environments in Guatemala Date of Submission: July 31, 2008

1

 

Integration of NASA Global Precipitation Measurement  Mission Data into the SERVIR Flood Decision Support 

System for Mesoamerica

Joel S. Kuszmaul1, Elizabeth G. Johnson1, Greg Easson1, Faisal Hossain2, G. Robert Breckenridge3, Emil Cherrington4, Timothy Gubbels5

July 31, 2008 Preliminary report

1 University of Mississippi Geoinfomatics Center (UMGC)  2 Tennessee Technological University 3 Dartmouth Flood Observatory  4 CATHALAC  5 Science Systems and Applications, Inc., Lanham, MD  20706 

2

EXECUTIVE SUMMARY This Rapid Prototyping Capability (RPC) experiment explores the feasibility of using new and future satellites, namely the Tropical Rainfall Measurement Mission (TRMM) and the next generation Global Precipitation Measurement Mission (GPM), to support flood water measurement by SERVIR. Currently available SERVIR products depend upon the continued availability of the AMSR-E sensor, which will not be available long term due to the expected decommissioning of the associated satellite. This RPC experiment constitutes a first attempt to lead to that transition. We have two components of our experiment; each employed TRMM data as a proxy for the planned GPM sensor. Experiment 1 uses the current AMSR-E 36.5GHz channel from the radiometer to the equivalent 63.5GHz band of the TRMM sensor. The current Dartmouth Flood Observatory’s practices regarding a flood hydrograph measurement were applied to data from both sensors. In Experiment 2, the potential for using rapidly available or real time sensor data from TRMM (as a proxy for GPM) to estimate rainfall and river flooding estimates was examined. The results of Experiment 1 give encouraging results in the continued or improved offering of the Dartmouth Flood Observatory’s flood inundation map. Of the eighteen regional sites where AMSR-E and TRMM results were compared, the six sites with a moderate or strong flood signal revealed a strong correlation (rave= 0.80) with only slight variability between the correlation on these six sites (the standard deviation was 0.072). When the twelve sites without a strong flood signal are considered, the correlation is weaker (rave= 0.143) and much greater variability (the standard deviation was 0.414). The results of Experiment 2 give encouraging results in the improvement in the detection of flooding in Mesoamerica. Our estimates rainfall during 1 to 10 October 2005 at the two study sites revealed good agreement between measured and estimated rainfall at one of the sites and poor agreement at the other site. When we applied an empirical approach to estimate cumulative discharge associated with this rainfall, we found fairly good agreement with stream gage measurement of discharge and estimated discharge at one of our sites and poor agreement at another site. These results shows that the planned enhancement of TRMM-based real-time satellite rainfall estimation based on a regional and dynamic bias adjusting scheme may hold promise to overcome the shortcomings observed for the site with poorer agreement. The combined RPC results suggest that continued availability of the SERVIR flood measurement products will require calibration procedures with the transition to new NASA sensors. The potential for additional products, including the possibility of estimated flood “now casts” may be possible with the continued development of the dynamic bias adjusting scheme using real-time satellite products.

1. INTRODUCTION

This Rapid Prototyping Capability (RPC) experiment explores the feasibility of using new and future satellites, namely the Tropical Rainfall Measurement Mission (TRMM) and the next-generation Global Precipitation Measurement Mission (GPM), to support flood water

3

measurement within the SERVIR architecture. The current method of measuring floods in Mesoamerica uses Advanced Microwave Scanning Radiometer (AMSR-E) data providing measurement of flood water approximately every two days (see example flood map in Figure 1). SERVIR is a NASA managed program integrating satellite and geospatial data with the aim to develop scientific and decision making knowledge for issues affecting Mesoamerica. The issues addressed by this program include disasters, ecosystems, biodiversity, weather, water, climate, oceans, health, agriculture and energy of the region.

Figure 1: Shows example flood water mapping currently available from SERVIR using AMSR-E data.

(SERVIR, 2008)

1.1 RELEVANCE TO NASA

This RPC experiment is directed at sustaining and improving the performance of a NASA-managed decision support tool (DST) with a well established track record of credibility and utility. This RPC experiment directly contributes to the NASA Applied Sciences Programs mission to extend the results of NASA Earth Science Division’s (ESD) contribution to national priority applications by supporting the ongoing mission of SERVIR’s Flood Measurement as related to National Application Areas of disaster management, ecological forecasting, and water management. Within the six National Focus Areas established by NASA ESD, this RPC contributes to an improved understanding within the categories Surface & Interior, Weather, and Water & Energy Cycles.

4

1.2 INTRODUCTION TO THE DECISION SUPPORT SYSTEM

SERVIR is an internationally supported visualization and monitoring system for improving environmental decision making in Mesoamerica. SERVIR has created modules which address nine areas to benefit society under the Global Earth Observation System of Systems (GEOSS): disasters, ecosystems, biodiversity, weather, water, climate, oceans, health, agriculture, and energy. The currently available visualization of flooding is prepared in cooperation with the Dartmouth Flood Observatory (DFO) using the AMSR-E data to yield a web-based product describing floodwater measured every two days. This product is included within the disaster, water and weather modules. While flooding data is valuable, it lacks the real time quality that could potentially be used more effectively in disaster and water management. In particular, the time required for this data to become available limits the value of the data for decision making in disaster management. The first goal of this experiment is to examine how alternative imagery might be substitute for current methods to provide more real time measurement of flood water. Such data could be more useful for disaster management response. In other areas of its mission, SERVIR seeks to provide short-term forecasts (or now casts) of significant weather events. The second goal of this experiment examines the potential usefulness of rainfall measurements by current and future NASA missions in combination with knowledge local drainage basins to provide predictive flooding information. Such now casts of flooding that would be more useful not only in disaster response, but also during emergency conditions.

1.3 PARTNERS IN THE INTERDISCIPLINARY RPC EXPERIMENT

Our partners in this RPC Experiment bring a diverse set of skills. We have partnered with experts in the science and application of the SERVIR flood mapping tools. These include Dan Irwin of NASA’s Marshall Space Flight Center (NASA-MSFC) and project manager for SERVIR, to provide an understanding of how SERVIR is applied. NASA-MSFC is partnered with CATHALAC to manage the SERVIR project and serve information products to users in Central America. For understanding of the current AMSR-E-based flood modeling system we have relied on Dr. Bob Brackenridge of the Dartmouth Flood Observatory and Tim Gubbels, of SSAI. We also incorporated experts in the NASA next-generation sensors including Dr. Faisal Hossain of Tennessee Technical University who works extensively with TRMM and simulated GPM data.

A. Science Systems and Applications Inc. at National Aeronautics and Space Administration (NASA) – Goddard Space Flight Center (GSFC).

Timothy Gubbels – Tel.: 301-867-6260 Email: timothy_gubbels@ssaihq.com Address: 10210 Greenbelt Road, Suite 600 Lanham, MD 20706

Dr. Gubbels is a senior scientist with SSAI and his areas of expertise include Earth system science, remote sensing technology and applications, and information systems.

B. Tennessee Tech University (TTU).

Faisal Hossain – Tel.: 931-372-3257, Email: fhossain@tntech.edu Address: Department of Civil and Environmental Engineering, Tennessee Technological University - Prescott Hall

5

332, 1020 Stadium Drive Box 5015, Cookeville TN 38505-001 Dr. Hossain and his research group have agreed to contribute to the project by post processing the data provided by GSFC and convert it to simulated GMI rainfall estimates.

C. Water Center for the Humid Tropics of Latin America and the Caribbean (CATHALAC)

Emil Cherrington – Tel.: (507) 317-0053 Email: cathalac@cathalac.org Address: City of Knowledge, Clayton, Panama. P.O. Box 873372, 7 Panama

Dr. Cherrington and his group provide data and support for research based in Guatemala. D. Dartmouth Flood Observatory (DFO).

Robert Brakenridge – Tel.: 603-646-2870; Email: G.Robert.Brakenridge@dartmouth.edu Address: Dartmouth College, Fairchild 121, Hanover, NH 03755 Dr. Brakenridge has agreed to provide technical support for AMSR-E data processing to measure discharge and runoff.

E. Mesoamerican Regional Visualization and Monitoring System Decision Support System (SERVIR) – Marshal Space Flight Center (MSFC).

Dan Irwin, Marshall Space Flight Center-Tel.; 256-544-0034 Email: Daniel.irwin@nasa.org Address: Marshall Space Flight Center, 320 Sparkman Derive, Huntsville, AL 35805 Mr. Irwin has agreed to provide access to the SERVIR Flood Decision Support Tool to integrate experiment results.

2. THE SERVIR DECISION SUPPORT SYSTEM FOR FLOOD

MONITORING AND PREDICTION

2.1 THE FLOOD MAPPING DECISION SUPPORT TOOL

SERVIR integrates satellite and geospatial data to improve scientific knowledge and better decision making. SERVIR addresses nine societal benefit areas of the Global Earth Observation System of Systems (GEOSS): disasters, ecosystems, biodiversity, weather, water, climate, oceans, health, agriculture, and energy. The SERVIR DSS for floods provides several products for both monitoring existing and analyzing historical events. Some of these products include discharge measurements and runoff from watersheds, and flood maps of river reaches in Central America (Figure 2). This SERVIR DSS has been developed to aid the work of environmental monitors and disaster management personnel.

Flooding, or changes in the ground surface water condition, is detected by satellite microwave

6

radiometry. The flood tool from DFO ratios the radiance from a calibration pixel on dry land against radiance for a pixel centered over targeted portion of river to estimate change in discharge. Ground based gauging station data provide for conversion from remotely sensed signal to water discharge values. The method is sensitive enough to measure 1.3 year and 5 year floods as well as daily fluctuations. To date, DFO has only implemented this method using microwave responses from AMSR-E data at 36.5GHz.

Figure 2: Flood status in several watersheds in Central America (Source: SERVIR) The DFO processes the discharge measurements and watershed runoff data to observe ongoing flooding. Flooding information is transferred to the SERVIR DSS and thereby delivered to the public. The revisit time of Aqua limits AMSR-E data to approximately every two days. The limitation of the flood inundation map product is that its works best over large, clearly delineated river reaches in relatively flat terrain where swollen rivers can be easily detected by satellite imagery. As seen in figures 1 and 2, the vast majority of visible flooding is on the coastal plains.

2.2 FUTURE SERVIR FLOOD MAPPING

Currently, the immediacy of the flood inundation maps is limited by the availability of microwave data. One objective of this study investigates the applicability of alternate sources of similar microwave data so that flood map is 1) insured by more plentiful sensors providing a data product that can be integrated and 2) improved by data sources that provide input data more frequently. To meet this objective, the TRMM Microwave Imager 37 GHz product will be tested as a possible alternate product to replace AMSR-E 36.5 GHz data. TRMM data could augment and improve the existing AMSR-E data. Additionally, TMI is due to be superseded by the GPM Microwave Imager so the ability to integrate TRMM data would ensure a continuation to the

7

flood product in SERVIR.

2.3 FUTURE SERVIR “NOW CASTING”

The current flood decision support system is only capable of mapping the extent of flooding in low relief areas for clearly delineated river reaches. If satellite-derived rainfall rates could be constantly monitored and converted into flood potential, then a predictive system could be developed. As an initial step, ground-based and satellite-based rainfall estimates are compared with ground-based river gage data to determine if the rainfall is predictive of discharge.

3. THE ONGOING NASA CONTRIBUTION TO THE SERVIR DECISION

SUPPORT SYSTEM

3.1 CURRENT SENSOR AND LIMITATIONS OF CURRENT SENSOR PRODUCT

At present, the DFO-generated flood maps rely upon AMSR-E data from Aqua. The Advanced Microwave Scanning Radiometer is a multichannel passive microwave instrument provided by the Japanese Aerospace Exploration Agency (JAXA) for the National Aeronautics and Space Administration (NASA) as part of the Aqua global hydrology mission. AMSR-E measures a number of geophysical parameters including: Sea Surface Temperature, Surface Wind Speed, Atmospheric Water Vapor, Cloud Liquid Water, and Rain Rate. Precipitation rates from AMSR-E data are generated at the Global Hydrology and Climate Center Science Investigator-led Processing System (GHCC-SIPS), in Huntsville Alabama. These data can be accessed at the National Snow and Ice Data Center Distributed Active Archive Center (NSDIC-DAAC) (Regner, et al, 2005). The data is generally available 9 hours after observation. The current sensor used for DFO flood inundation maps is the 36.5 GHz channel from the AMSR-E radiometer onboard the Aqua Satellite. The Satellite was launched in 2002 with a design life of 5 years; it has now exceeded its design life. Aqua is a sun-synchronous satellite in an orbit 705 km above earth at an inclination of 98.2°. The number of satellite passes per day is a function of latitude (NSIDC, 2008). In the area of Guatemala, Aqua’s revisit time is just over once a day (JAXA, 2003). Data from AMSR-E is processed through the National Snow and Ice Data Center and provided with both vertical and horizontal polarization at a resolution of 14.4 x 8.2 km (instantaneous field of view) for the 36.5 GHz frequency with a swath width of 1445 km. AMSR-E is a conical scanner with an incidence angle of 55°. The flood product and associated work of SERVIR is threatened with the future decommissioning of primary period missions such as Aqua. River measurement activities for SERVIR would have to be stopped if AMSR-E data discontinued and no replacement would be available. The need to ensure a continuous measurement of geophysical parameters through the use of new EOS increases. Our goal is to establish the suitability for the Flood Product of microwave information from current and future NASA missions (namely TRMM and GPM) for

8

the continuation of precipitation data collection.

3.2 TRMM, EXPECTED IMPROVEMENTS AND PROBLEMS

The TRMM provides sensor data similar to AMSR-E. The TRMM Microwave Imager (TMI) is a passive microwave sensor that measures the intensity of radiation at 37 GHz. NASA and JAXA launched the TRMM in 1997 as a three-year mission but TRMM has been collecting data for seven years. The main improvement of TMI over AMSR-E is improved ground resolution because TRMM occupies a lower altitude at 402 km compared to 705 km of AMSR-E. TMI has an 875 km wide swath on the surface with 10x7 km field of view at 37 GHz. Pixels are 5km. The TMI is similar to AMSR-E in that is also a conical scanner inclined at 52.8° (Lee, 2002). TRMM 37 GHz vertical and horizontal polarizations can be imaged without the need for further processing. One of the primary differences between TRMM and Aqua is their orbit. Most near-sun-synchronous like Aqua have polar orbits with orbital inclinations of about 98° and view spots on the earth at two specified local times 12 h apart. The TRMM orbit is circular, inclined 35° with respect to the equator, which allows excellent coverage in the tropics. TRMM was placed in a low-inclination orbit both to improve the sampling of tropical latitudes and to enable it to visit points on the earth at many different local times. TMI offers 15–16 orbits per day with a swath width of 875 km. The drawback is that there is a long interval between the TRMM visits to a specific region at a given hour of the day and its next observation at a similar hour (23 days at the equator and 46 days at the highest latitudes accessible to TRMM) (Imaoka, 2000). Climatological studies using satellite data must be done with appropriate averaging of the data in order to minimize biases in the averages due to varying sample sizes at different times of the day (Salby and Callaghan, 1997). Negri et al (2002) show that the optimum period of accumulation is 4 hours to minimize the potentially large variations in sampling error due to this pattern.

3.3 GPM, EXPECTED IMPROVEMENTS AND PROBLEMS

The value of testing the viability of 37 GHz data from TRMM is that it serves as a proxy for the future GPM with a 37 GHz channel. The GPM will be similar to the TRMM with frequent measurements (3-hourly). The GPM is another JAXA/NASA collaborative satellite mission and is set to launch in 2013 to replace the TRMM. It will be similar to the TRMM in its altitude (400km) but its orbit will be circular at 65° instead of non-sun-synchronous at 35°. A GPM microwave imager, GMI, will replace the current TMI and will function similarly with a conical scanning passive microwave radiometer inclined at 49°. GMI’s swath width will be 850 km and it will capture 37.0 GHz horizontal and vertically polarized radiation. Level 1 data (calibrated, geo-located, instrument values at the instrument field of view) will be made available to users within 20 minutes of collection.

9

4. THE RPC EXPERIMENT: PLANS TO UTILIZE NASA NEXT-

GENERATION SENSORS

Our goal is to test simulated GPM data and assess its ability to replicate or improve the flood products currently generated using AMSR-E data. Because GPM data will be very similar to TRMM data but with increased temporal and spatial resolution, we assessed whether TRMM data products could meet two objectives; 1) to enhance the current satellite-based inputs into SERVIR by replacing AMSR-E data used for SERVIR’s Flood Decision Support System and 2) to provide predictive information about increased flood risk by comparing storm related runoff discharge with stream discharge. To accomplish these objectives, we created two experiments related to a specific time and place, namely Hurricane Stan rainfall event in southwestern Guatemala

4.1 STUDY TIME FRAME AND SITE

Hurricane Stan (October 2005) was a relatively weak storm embedded in a larger non-tropical system that delivered torrential rains to SERVIR countries including Guatemala, El Salvador and southern Mexico. Storm related flooding and mudslides lead to between 1600 and 2000 deaths with damage of approximately $1 billion U.S. dollars. Figure 3 shows rainfall accumulations of up to 500 mm (20 inches) over parts of Central America.

Figure 3: NASA generated Rainfall accumulation map for Hurricane Stan

4.2 EXPERIMENT 1

The objective of Experiment 1 is to examine how alternative imagery types might be substituted for current methods to provide more real time measurement of flood water. Our method requires finding the closest proxy to meet DFO’s current model input requirements and testing DFO’s ability to a generate flood inundation map products from the new sensor. Our aim was to compare flood inundation products from two different sensors against ground data from river gauging stations.

10

Study Site   DFO generated river and calibration pixel measurements for both TRMM 37.0 GHz and AMSR-E 36.5 GHz sensors across 18 sites from September 25, 2005 through October 29, 2005. Figure 4 shows the locations of the sites. Figures 5 and 6 show AMSR-E and TRMM acquisitions respectively for the time of Hurricane Stan.

Figure 4: River locations for AMSR-E comparisons with TRMM (Image base map from GoogleEarth)

11

Figure 5: AMSR-E data for Hurricane Stan from NASA

Figure 6: TRMM data for Hurricane Stan from NASA

Flood Measurement Procedures  Microwave radiometers can measure flooding by differentiating brightness temperature in a given pixel over a reach of water compared with a calibration pixel over land. The amount of radiation emitted from the earth’s surface is a function of the temperature of the surface and also of the wavelength of the radiation. The brightness temperature of water covered surfaces is usually 50% of the actual temperature compared with 90% for land surfaces. The key to satellite-based flood measurements is the utilization of gauging reaches rather than gauging stations. River flow cross sections are difficult to measure from space, but measurements of changing water surface areas within carefully defined river reaches can be retrieved with relatively high precision using existing methods. Such gauging reaches range from ~10 km to 30 km in length (Brackenridge, 2005).

When comparing scenes, a calibration area over dry land (but within 50 km of the river) maintains a relatively constant brightness temperature while the brightness temperature of a measurement area over a river varies as a function of the water surface area. When total water surface area increases along the river, the mean radiance declines for the measurement sub-reach, but not outside it so the radiance ratio records the water area increase. Because of the large

12

difference in water and land radiances at this spectral band (841–876 nm), the ratio calibration reach/measurement reach is a sensitive and consistent measure of surface water (Brackenridge, 2005). A ratio of the brightness temperatures in the measurement and calibration areas yields a fairly reliable hydrograph of river levels which can then be calibrated with data from gauging stations. It should be noted however that space-based surface water observations are not as precise as gauging station data and the relationship of reach surface water area to discharge may be affected by errors such as hysteresis (Brackenridge, 2005).

Methods  DFO, in conjunction with Dr. Tom De Groeve of the Joint Research Center (European Union Commission research lab in Ispra, Italy), adapted a system created to access NASA AMSR-E hdf formatted swath data to access TRMM hdf formatted swath data. The system queries files using the latitude and longitude of the two targets per scene and returns only the signal numbers within 5 km radius of the target locations. A four-day running mean smoothing and data-filling algorithm is used because both sensors do not return daily data (AMSR-E returns data every 1 to 3 days while TRMM similar but slightly more frequent) (Breckenridge, Pers. Comm.).

4.3 EXPERIMENT 2

The objective of the second experiment is to test and validate Global Precipitation Measurement Mission data as: 1) an enhancement to current satellite-based precipitation inputs into SERVIR, and; 2) a replacement for the Advanced Microwave Scanning Radiometer Earth Observation System to improve the determination of ground surface water conditions. This experiment examines the potential usefulness of rainfall measurements by current and future NASA missions in combination with knowledge local drainage basins to provide predictive flooding information. Such “now casts” of flooding that would be more useful not only in disaster response, but also during emergency conditions.

Study Site  Guatemala was chosen as the study site for several reasons, 1) there is good on site support infrastructure and data through the National Institute of Seismology, Volcanology, Meteorology and Hydrology (INSIVUMEH) and 2) Guatemala suffered the largest loss of life associated with Hurricane Stan. The southwestern coast of Guatemala was focused on because it received the heaviest rainfall associated with Hurricane Stan and it includes the village of Panabaj, site of a devastating landslide which resulted in over 1000 deaths.

The climate of southwestern Guatemala is divided based on elevation into three zones that run parallel to the Sierra Madre mountains and the coastline: the coastal plain (0 – 300 masl), The Bocacoasta (300 – 1400 masl) and the high plain (>1400 masl). The Bocacosta receives the greatest rainfall in the country with maximum rainfall from June to September. It has two climate seasons (wet and dry), the vegetation is jungle and the temperature is a function of elevation. The high plains have a drier climate overall with two seasons. The wet season is from

13

May to October. Weather in this zone is highly variable with many microclimates and it is affected by dense population centers. The coastal plain is much hotter than the other two regions with significantly less rainfall. The vegetation is forest and grassland.

Guatemalan watersheds were chosen from those in the southwest with both stream gage and rain gage data collected by INSIVUMEH across the period from Oct 1-Oct 10, 2005. Of nine possible sites, two basins met that criteria: 1) Rio Coyolate en Puente Coyolate (in future referred to as Coyolate) and 2) Rio Villalobos at Villa Canales (referred to hereafter as Villa Canales). Table 1 lists the locations of these two basins according to USGS records. Table 1: Watershed Locations

USGS Site number 

Site name  Latitude (DMS) 

Longitude (DMS) 

Datum Elev.(ft ASL) 

Datum Climate Zone 

Drain. area (Km2) 

ID ‐INSIVUMEH 

50832‐01100102 

Rio Coyolate en Puente Coyolate 

14°22'39"N  91°08'12"W NAD27 698.4 NGVD29 Bocacosta  511.88  1.10

50832‐01130645 

Rio Villalobos at Villa Canales 

14°29'10"N  90°32'10"W  NAD27 3986.4 NGVD29 High Plain  311.52  1.13

The chosen watersheds are within 65 km of each other and are of similar size, but there are important differences between the watersheds. Figure 7 shows a DFO generated flood inundation map over southwestern coast of Guatemala at the time of Hurricane Stan and the location of the two watersheds. The Coyolate gauging station is at the base of a relatively steep-sloped basin in volcanic terrain that faces the Pacific Ocean along the windward side of the Sierra Madre Mountains. It is less than 20 miles southeast of the site of the landslide at Lake Atitlan which decimated the village of Panabaj. The Coyolate watershed is largely unpopulated. The Villa Canales watershed is almost due east of Coyolate, but it is on the leeward side of the mountains in the high plains climate zone. The catchment area is nearly flat and heavily populated, containing the western portions of Guatemala City.

14

Figure 7: DFO generated flood inundation map for Hurricane Stan showing two watersheds of interest

Method Description  We use an empirical method of converting the rainfall estimates to discharge estimates. Data from the chosen watersheds was difficult to obtain and incomplete. Data was insufficient estimate the arrival time of a pulse of water from a specific rainfall time. Instead, the Soil Conservation Survey’s Curve Number method (Hjelmfelt et al., 2004) was applied to both the rainfall measurements and the TRMM-based estimates of the rainfall accumulation. This method produces the total estimated discharge associated with a rainfall event. The curve number runoff equation is

where Q [L] is runoff, P [L] is depth of rainfall, [L] is initial abstraction ( ) and S [L] is potential retention

where CN is the curve number (taken to be 85).

15

In situ stream gage measurements were converted to discharge estimates using an empirical relationship

where Q is discharge, h is the gage height, h0 is the reference gage height and n and K are empirically determined constants. Using this method, we expect that the total discharge by all methods would be in agreement, but that our rainfall-based discharge estimates will rise more quickly than actual discharge measurements. This relates to the non-linearity and the thresholding of the rainfall-runoff process. In nature it takes time for the rainfall to reach the rain gage, but our method assumes that runoff is immediately measureable. So with this method, the definition for good agreement would be for the total discharge by all methods to be similar, while dismissing the premature discharge estimates that are based on rainfall total. The limitation of this method is that it does not account for concentrated pulses of water. A rainstorm that lasts 3 days and a rainfall that lasts 3 hours would both yield the same total discharge estimates but may have different flooding results.

Watershed Data  The Coyolate station recorded stream gage values every 15 minutes continuously from September 29 through October 6 when the gage broke. It recorded dramatically increased discharge associated with Hurricane Stan in two sharp pulses on Oct. 4-6, 2005. The Villa Canales station was monitored daily from Oct. 1 -10, 2005 and recorded a sharp pulse on Oct. 5, 2005. The empirical factors used to generate stream discharge are listed in Table 2. Table 2: Empirical Factors for Calculating Discharge from Gage Height Measurements

Watershed K H0 Z Coyolate 1.1826 1.1745 4.5567 Villa Canales 0.93 1.572 2.722

Both the Coyolate and Villa Canales sites recorded rain gage data approximately every 15 minutes from Sept. 29-Oct.10, 2005. Data was resampled to hourly measurements by using the first measurement of each hour to represent the full hour and converted to units of mm/hr. This method was chosen so that the data would closer reflect the TRMM data.

For the runoff/discharge calculation in the watersheds, a static curve number of 85 was assumed. This high CN is justified because there was heavy rainfall across the end of September leading up to Hurricane Stan and the ground was saturated. The landslide at Lake Atitlan occurred Oct 4, 2005 before the bulk of the rains fell.

16

TRMM Data  The TRMM Multi-satellite Precipitation Analysis (TMPA) research product 3B42 Real Time (RT) was used for this experiment for dates from October 1 to 10, 2005. The chosen Guatemalan watersheds are smaller than the standard TMPA grid box size. For the time period, the 3B42RT product was Coyolate (185th row, 1074th column) and Villa Canales (185th row, 1076 column). 3B42 RT combines precipitation estimates in mm/hr from multiple satellites, as well as gauge analyses, where available, at a 3-hour time step and 0.25° degree spatial resolution. TMPA is available in real time and after real time products based on calibration by the TRMM Combined Instrument and TRMM Microwave Imager precipitation products, respectively. The after-real-time product (V6 3B42) incorporates gauge data but is delayed by approximately 1 month (Huffman, 2007).

The 3B42-RT product combines microwave, infrared and radar data. 3B42-RT has a temporal resolution of three hours and is published daily with a 6 hour and 40 minute delay. Daily totals compiled by the FEWS NET team at the U.S Geological Survey Center for EROS, are also available (Funk, 2006). The 3-hourly data was resampled at 1-hour increments to match the time step of the rain gage. Missing values were supplied by the previous value.

GPM Data  This experiment explores the potential to convert discharge measurements into a “now-cast” of expected discharge and make that information available to SERVIR following a TRMM pass. The time delay for TRMM would be close to 7 hours after receiving the 3B42RT data and applying an algorithm to generate discharge. GPM data is forecast to be available 20 minutes after acquisition at roughly 3 hour intervals.

4.4 RESULTS AND DISCUSSION

Experiment 1 Results and Discussion The eighteen regional sites where both AMSR-E and TRMM data, see Figure 3, can be compared in a number of ways. The critical measure described previously the ratio calibration reach/measurement reach, must be compared for both these sensors. While the exact ratio values need not be reproduced with the TRMM sensor, the ratios must be clearly related so that the TRMM data can be calibrated to enable it to make the same measurement of flood hydrographs. One way that these two data sets can be compared is on the basis of the simulated hydrographs. Because this is the first use of TRMM data in this application, the ratio values are compared directly prior to hydrograph calibration. An example of this type of comparison is shown in Figures 8 and 9, where results are shown for Site 1063 for the Hurricane Stan measurement.

17

Figure 8: AMSRE-derived flood hydrograph for Tropical Storm Stan at Site 1063: 18.1569 N, 95.3552 W, Rio

La Trinidad, Mexico.

Figure 9: TRMM-derived flood hydrograph for Tropical Storm Stan at Site 1063: 18.1569 N, 95.3552 W, Rio

La Trinidad, Mexico. The similarity of the two sensor measurements shown in Figure 7 and 8 make it clear that under some circumstances, TRMM-derived flood hydrographs may be as useful as the AMSR-E produced measurements. The timing and strength of signals are similar in Figures 7 and 8, but the magnitude of the flooding differs in the two results. This discrepancy demonstrates that the TRMM-based data must be properly calibrated to reproduce the correct flood description. One

18

useful way to compare these to signals is to use the correlation coefficient for the ratio values from the two sensors. For Site 1063, the site shown in Figures 7 and 8, the correlation coefficient for these ratios is r = 0.87 (out of a possible range of -1.0 to 1.0), demonstrating a strong correlation between the two sensor’s results for the time period 27 September to 25 October 2005.

To present the results of all eighteen regional sites for this same time period, there are a number of methods for presenting and discussing the results. For each site, we compare the band ratio values (using a four day running average). We compared them first as time series for a qualitative assessment of the similarity of the two sensor’s ratio values, where a qualitative comparison was made. We also calculated the correlation coefficient for each of the eighteen regional sites to obtain a qualitative assessment of the similarity of the two measurements. A wide range of results were found, ranging from strong correlation to no (or incorrect) correlation. Examples of this range of correlation results are shown in Figure 10, where Site 69 (r = 0.85), Site 277 (r = 0.5) and Site 2426 (r = 0.05) are shown as time series for the period 27 to 25 October 2005. Another way to show the results is shown in Figure 11, where the approximate linear alignment of the Site 69 results demonstrate the strong correlation, but the cloud of daily measurements reveal weaker correlations for Sites 277 and 2426.

(a) Site 69 (b) Site 277 (c) Site 2426

Figure 10: Shows the four-day running average of the M/C ratio for the AMSR-E and TRMM data sets for a)

Site 69, b) Site 277 and c) Site 2426 for the period 27 September to 25 October 2005.

(a) (b)

AMSR-E ratio AMSR-E ratio

TRMM ratio TRMM ratio

19

Figure 11: Shows the four-day running average of the M/C ratio for the AMSR-E and TRMM data sets for Sites 69, 277, and 2426. If results from the two sensors matched perfectly, they would fall along the line of equality. Observations that fall roughly along a line (such as the observations from Site 69 in Figure 10a.) indicate correlated results. Observations that fall in a random pattern (such as shown in the detailed view in Figure 10b) indicate weaker correlation. This range of results appeared at first to suggest that we obtained mixed or inconsistent success when assessing the potential of TRMM to replicate the AMSR-E measurement of flood hydrographs. However, we noticed a pattern to the results, when there was a clear flood pulse from Hurricane Stan we tended to obtain strong correlation. In the absence of a clear flood pulse, we had mixed or poor correlation. In an effort to quantify this observation we used three different measures of whether there was a clear flood pulse evident in the AMSR-E hydrograph. The first of these measures was a qualitative assessment of a strong, moderate, weak, or no flood pulse. The second measure of whether there was a flood observed in the hydrograph was the overall standard deviation of the AMSR-E ratio for this time period; the larger the standard deviation, the wider the hydrograph varied over this time. The third measure of whether there was a flood observed in the hydrograph was the overall range (maximum minus the minimum value) of the AMSR-E ratio for this time period; the larger the range of values the stronger the interpreted storm pulse. All three measures demonstrated consistent results (shown in Table 3), where these measures are provided along with the correlation coefficient for the two sensors. Table 3: The correlation coefficient between the M/C ratio for the AMSR-E and TRMM sensors for the time period 27 September to 25 October 2005. Also included are three estimates of the strength of the storm pulse in the AMSR-E signal (see discussion in text).

Site ID  r  Quality  σAMSRE  Range 

1060  0.81  Strong  0.393  1.235 69  0.85  Strong  0.254  1.045 70  0.68  Strong  0.227  0.703 

2569  0.83  Strong  0.224  0.733 67  0.75  Mod  0.137  0.480 

1063  0.87  Mod  0.115  0.420 1066  0.30  None  0.097  0.365 951  0.83  Weak  0.077  0.265 277  0.50  Weak  0.074  0.257 1061  0.60  Weak  0.074  0.282 1065  ‐0.42  Weak  0.073  0.242 1062  0.33  Weak  0.061  0.260 2501  ‐0.01  None  0.042  0.185 2427  ‐0.40  None  0.040  0.165 2500  0.20  None  0.040  0.185 2426  0.05  None  0.035  0.145 1036  ‐0.47  None  0.019  0.067 

1035  0.21  None  0.017  0.065 

20

While the correlation coefficient for all eighteen sites range from -0.47 to 0.87, the average value is 0.36 with a standard deviation for all values is 0.46. Such an outcome suggests modest correlation at best with the large standard deviation indicating very inconsistent results. When only the six sites with moderate or strong storm signals are considered to obtain these same measures (the shaded portion of Table 3), the average correlation coefficient is 0.80 with a standard deviation of 0.072. In contrast, the twelve sites with weak or no storm pulse (not shaded in Table 3) had an average correlation coefficient of 0.14 and a standard deviation of 0.414. Thus, when there was a significant flood pulse to measure (as assessed using the established AMSR-E measure), the analogous TRMM ratio is very likely to produce a useful result. In the absence of a significant flood signal, the correlation was inconsistent and commonly weak.

Experiment 2 Results and Discussion  We compared estimated watershed discharge calculated from rainfall accumulation with estimated watershed discharge from calculated from stream gage height measurements. Rainfall measurements came in two forms, 1) in situ rain gages and 2) TRMM real time measurement. Two locations were chosen where TRMM, in situ rainfall estimates and gage heights measurements were available with sufficient measurement for comparison purposes, 1) Rio Coyolate de Pensativo and 2) Rio Villalobos at Villa Canales. The chosen watersheds are located in the southwest of Guatemala, an area subjected to heavy rains during Hurricane Stan, are within 65 km of each other and are of similar size (511 and 311 km2, respectively).

Initially we compared in situ rainfall measurement versus rainfall estimates based on TRMM real time data (using the 3B42RT data product) at both locations. For each of the two sites, the comparison is shown in Figure 12.

Figure 12. A comparison of the rainfall guage data in red (measured 1 October 2005 to 10 October 2005) to rainfall estimated using the TRMM 3B42RT product in blue for a) Rio Coyolate de Pensativo and b) Rio

Villalobos at Villa Canales.

21

These results reveal good agreement at the Rio Coyolate site and poor agreement at the Villa Canales site. In both cases, the measured rainfall is higher than the estimated rainfall. For the Rio Coyolate site the estimated rainfall is 26 to 33% lower than the measured precipitation, while at the Villa Canales site the estimated rainfall is 80 % lower than the measured precipitation. There is reasonable agreement across all times at the Rio Coyolate site, with the time of greatest precipitation in agreement between the two rainfall estimates. The Rio Coyolate site supports earlier research that, once properly calibrated, the TRMM data can provide useful estimates of rainfall amounts (Harris and Hossain, 2008). At the Villa Canales site, however, the rainfall estimates do not agree across all time frames. The time of most rapid rainfall accumulation, however, was similar for both estimates. There are a number of human or natural factors that could have caused this discrepancy. Human factors include basin catchment or other parameters of the region that are not properly calibrated for the Villa Canales site or that there was equipment failure on site. The most probable natural factor causing the discrepancy is topography. The Villa Canales watershed is located immediately on the lee side of the coastal mountain range. It is possible that the storm clouds lost much of their moisture crossing the range as they moved inland, but the drop in moisture was not detected by TRMM because of the spatial resolution was too coarse. Rio Coyolate would have been unaffected by this topographic issue because it lies on the windward side of the coastal mountains.

For a time series of discharge based upon the rate of rainfall, we expect that the final total discharge by all methods would be in agreement, but that our rainfall-based discharge estimates will rise more quickly than actual discharge measurements. This is because, it takes time for the rainfall to reach the rain gage, but our method assumes that runoff is immediately measureable. So good agreement, in this case would be for the total discharge by all methods to be similar, while dismissing the premature discharge estimates that are based on rainfall total. When the results for our two sites are considered (see Figure 13), they do not compare as expected.

Figure 13. Discharge measurements and estimates at a) RioCoyolate de Pensativo and b) Rio Villalobos at Villa Canales. At each site the stream gage discharge measurement (golden) is compared to the empirical estimate of

discharge (using the procedure described above) based on the rainfall gage (blue) measurement or the TRMM 3B42RT product.

22

The comparison at Rio Coyolate de Pensativo was very good. The in situ discharge measurements did not cover the full span of the storm event, but were in good agreement for discharge magnitude. As expected, the discharge estimates based on the SCS method predict an earlier arrival of the discharge than was measured. In fact, it would not have been surprising to find a greater disagreement. After 7 October, 2005, in situ measures became unavailable, and the two rainfall-based estimates of discharge differ distinctly. The difference in rainfall estimations by the two methods is amplified in the discharge estimates. It is not clear why there is such a sharp difference estimated rainfall, and the source of this error is not immediately obvious. It is unlikely to be related to the application of the SCS equation for estimating discharge because the same curve number was used in both cases (CN=85).

When the discharge at the Villa Canales site is examined, a very different result is seen. Here, the in situ measures show much less rainfall and much smaller discharge values than at Rio Coyolate. The low discharge may relate to lowered rainfall (i.e. the in situ rain gage is more accurate than the TRMM) and may be further reduced by high agricultural usage. Experts familiar with the Villa Canales site report that the gauging station does not have reliable data, due to some external influences, such as: 1) this river receives urban drainage and 2) communities around the river get their water supply for agricultural purposes from this river in some seasons of the year. Agricultural use may lead to discharge measures that were much smaller than would be expected for the observed rainfall. At this point, the two different results at these sites suggest that the use of TRMM data to estimate rainfall and discharge may be a significant advance, but that the discrepancies at the Villa Canales site indicate that the calibration process may require that additional details.

Our study shows that the planned enhancement of TRMM-based real-time satellite rainfall estimation based on a regional and dynamic bias adjusting scheme (Huffman et al., 2007) may hold promise to overcome the shortcomings observed for Villa Canales

5. CONCLUSIONS AND RECOMMENDATIONS

The development of this methodology will allow for rapid streamlining of new GPM-generated flood products into an important regional decision support system, SERVIR. GPM rainfall products are expected to be more accurate and timelier than current products from either TRMM or AMSR-E. GPM will continue to provide the information needed for Dartmouth to generated flood inundation maps long after the AMSR-E satellite fails. Dartmouth’s discharge measurements are crucial for calibrating models for discharge of rainfall from select watersheds. The integration of rainfall into the decision support system represents an important first step in the shift from monitoring discharge to anticipating discharge and thereby forecasting flooding events. GPM integration into SERVIR will allow for disaster anticipation, rapid flood response and appropriate allocation of resources. The goal of this research is to mitigate injury and death due to flooding should another heavy rainfall event like Hurricane Stan occur in the future.

23

The results of Experiment 1 give encouraging results in the continued or improved offering of the Dartmouth Flood Observatory’s flood inundation map. Of the eighteen regional sites where AMSR-E and TRMM results were compared, the six sites with a moderate or strong flood signal revealed a strong correlation (rave= 0.80) with only slight variability between the correlation on these six sites (the standard deviation was 0.072). When the twelve sites without a strong flood signal are considered, the correlation is weaker (rave= 0.143) and much greater variability (the standard deviation was 0.414). These initial results give encouraging evidence that TRMM (and since TRMM is being used here as a GPM proxy) GPM data will give continued flood detection performance. With the transition to a new sensor, however, calibration will be required. While more frequent sampling that is expected to become available with the GPM sensor may be useful, our experiment was unable to address this issue.

The results of Experiment 2 give encouraging results in the improvement in the detection of flooding in Mesoamerica. Our estimates rainfall during 1 to 10 October 2005 at the two study sites revealed good agreement between measured and estimated rainfall at one of the sites and poor agreement at the other site. When we applied an empirical approach to estimate cumulative discharge associated with this rainfall, we found fairly good agreement with stream gage measurement of discharge and estimated discharge at one of our sites and poor agreement at another site. These results shows that the planned enhancement of TRMM-based real-time satellite rainfall estimation based on a regional and dynamic bias adjusting scheme (Huffman et al., 2007) may hold promise to overcome the shortcomings observed for the site with poorer agreement.

Together both experiments suggest that continued availability of the SERVIR flood measurement products will require calibration procedures with the transition to new NASA sensors. The potential for additional products, including the possibility of estimated flood “now casts” may be possible with the continued development of the dynamic bias adjusting scheme using real-time satellite products.

6. REFERENCES

Brackenridge, G. Robert, 2005. Space-Based Measurement of River Runoff. EOS, Transactions American Geophysical Union Vol. 86 (11)

Funk, C., Ederer, G. and D. Pedreros, 2006. Tropical Rainfall Measuring Mission, UCSB Climate Hazard Group.

Harris, A., S. Rahman, F. Hossain, L. Yarborough, A.C. Bagtzoglou and G. Easson (2007).Satellite-based Flood Modeling using TRMM-based rainfall products , Sensors, vol. 7: 3416-3427.

Hjelmfelt, Allen T., Mockus, Victor and Helen Fox Moody, 2004. Estimation of Direct Runoff from Storm Rainfall. Part 630 Hydrology National Engineering Handbook, Chapter 10.

24

United States Department of Agriculture, Natural Resources Conservation Service. 210-VI-NEH, July 2004 url address: www.wsi.nrcs.usda.gov/products/W2Q/H&H/tech_refs/eng_Hbk/ chap.html.

Huffman et al., 2007, The TRMM Multi-satellite Precipitation Analysis (TMPA): Quasi-Global, Multi-Year, Combined-Sensor Precipitation Estimates at Fine Scales. Journal of Hydrometeorology 8, 38-55

Imaoka K., and R. W. Spencer, 2000: Diurnal variation of precipitation over the tropical oceans observed by TRMM/TMI combined with SSM/I. J. Climate, 13, 4149–4158.

Japan Aerospace Exploration Agency. 2003, updated daily. AMSR-E/Aqua L1A Raw Observation Counts V001, September to October 2003. Boulder, CO: National Snow and Ice Data Center. Digital media.

Lee, Tom, 2002. TRMM Tropical Cyclones - 37 GHz Tutorial. NRL Monterey, Marine Meteorology Division. Digital Media. url address: http://www.nrlmry.navy.mil/sat_training/ tropical_cyclones/trmm/tmi_37v/

Negri Andrew J., Bell Thomas L., and Liming Xu, 2002 Sampling of the Diurnal Cycle of Precipitation Using TRMM. Journal of Atmospheric and Oceanic Technology, Vol. 19 (9), pp. 1333–1344.

NSIDC, National Snow and Ice Data Center, 2008. AMSR-E/Aqua L1A Raw Observation Counts Digital Media. url address: nsidc.org/data/docs/daac/amsrel1a_raw_counts.gd.html

Regner, K., Conover, H., Beaumont, B., Graves, S., Hawkins, L., Parker, P., et al., 2005, Flexible Processing at the AMSR-E SIPS, Proceedings of the Geoscience and Remote Sensing Symposium, IGARSS 05’, IEEE International.

Salby M. L., and P. Callaghan, 1997: Sampling error in climate properties derived from satellite measurements: Consequences of undersampled diurnal variability. J. Climate, 10, 18–36.

SERVIR, 2008 url address: www.servir.net. Tropical Rainfall Measuring Mission Earth Science Reference Handbook url address:

www.jaxa.jp/missions/projects/sat/eos/trmm/

1

Rapid Prototyping: Modeling Seagrass Community Change Using Remote Sensing and Real-time Instrument Packages

Marc Slattery, Deborah Gochfeld, Greg Easson University of Mississippi, University MS 38677

Anne Boettcher

University of South Alabama, Mobile AL 36688

Executive Summary The purpose of this Rapid Prototyping Capability (RPC) experiment is to determine the capability of the Moderate Resolution Imaging Spectroradiometer (MODIS) to assist biologists by providing necessary information to accurately model the growth of particular species of seagrass. Seagrass beds are one of the most important coastal ecosystems in the northern GOM and are subject to multiple interactive anthropogenic and natural disturbances. The decline of this crucial nursery habitat has implications for several important fisheries, as well as coastal environmental and human health. The current method of obtaining the necessary information for seagrass models involves collecting the data manually at the site location. This experiment will compare results between the model using data collected in the field and the model that uses information provided by MODIS, and to assess the strengths and weaknesses of these approaches in an effort to provide the most reliable tools to local resource managers. 1.0 Introduction Seagrass beds are one of the more important marine ecosystems worldwide (Carruthers et al. 2002; Duffy 2006). They provide essential coastal habitat that acts as a refuge for potential prey species, and feeding grounds for a variety of high trophic level consumers including birds, finfish, and many commercially-important species (Thomas & Stratis 2001). Seagrass beds in the GOM are vital nursery grounds for a variety of fish and invertebrate early life history stages (Heck et al. 2003; Adams et al. 2006), and also serve as crucial habitat for threatened and endangered species, including manatees and green sea turtles (Dawes et al. 2004). Seagrasses provide oxygen, filter nutrients and contaminants, and cycle essential inorganic and organic minerals for ecosystem consumer guilds. Seagrasses reduce coastal erosion by acting as baffles for wave energy and sediments, and are an important economic resource supporting fisheries, tourism and biodiversity (Zieman & Zieman 1989). The major physiological factors that influence seagrass dynamics in the northern GOM include: 1) light intensity and quality, 2) nutrient and CO2 availability, 3) water motion, 4) temperature, and 5) salinity (Dawes et al. 2004). Light intensity and quality is influenced by water depth and clarity; Dennison et al. (1993) noted that Thalassia testudinum required about 20% of surface photosynthetically-active radiation (PAR) to survive. Recent research has shown that filter-feeding by oysters and mussels helps to clear the water surrounding seagrass beds, resulting in healthier seagrass communities (Peterson & Heck 1999; Newell & Koch 2004). However,

2

epibiosis can have significant effects on seagrasses as fouling species prevent uptake of nutrients and exposure to PAR (Dixon & Leverone 1997), moreover, they can provide drag on plant fronds resulting in detachment (sensu Dixon et al. 1981). Seagrasses appear to be CO2 and nutrient limited, with uptake primarily from pore water (Fourqurean et al. 1992), and they grow best at temperatures between 20-30°C and salinities of about 34-36 ppt (Barber & Behrens 1985; Czerno & Dunton 1995). In addition to abiotic factors, seagrasses are heavily influenced by “top-down” herbivory effects; a variety of invertebrates and fishes feed primarily or exclusively upon seagrasses, including sea urchins, molluscs, crustaceans and fishes (Valentine & Heck 1999), albeit at different times in their life histories. Protist pathogens of seagrasses, Labyrinthula spp., have been found to cause significant wasting in turtlegrass, Thalassia testudinum, with loss of vast tracts of Florida seagrass beds (Robblee et al. 1991, Zieman et al. 1999, Blakesley et al. 2002). 1.1 Relevance to NASA The NASA Applied Sciences Program has a mandate to provide results from the NASA Earth Science Division (ESD) to national priority applications, including our coastal EEZ. This RPC experiment contributes to ecological forecasting in coastal GOM seagrass communities. The fact that these communities are important nursery habitat for commercially fished species and baffles to storm energy, demonstrates the utility of this program to NASA’s regional public health and disaster management. 1.2 Introduction to the Model Fong & Harwell (1994) provided detailed scientific evidence that 5 environmental factors play a major role in controlling biomass and abundance of seagrasses in time and space. These factors include temperature, salinity, light, water-column nutrient concentrations, and sediment nutrient concentrations. Significantly this model has been “ground-truthed” multiple times since its development and it is still considered the best available predictor of seagrass community change. The MODIS sensor can provide most of the factors needed for this model. The three main variables that MODIS provides are temperature, light and nutrients. Data retrieved from the sensor will provide direct information on temperature and light, and indirect information on nutrients. The nutrient data can be derived via correlations of Chla, detected by the sensor, to nutrient data collected by the field team. The two species of seagrass studied in Grand Bay NERR are extremely halotolerant so this factor can be dropped from the model with no effect on the predicted productivity.

1.3 Partners in the Interdisciplinary RPC Experiment

This RPC represents a partnership between NASA, NOAA, and the UMGC. The core scientists in these groups bring unique skills and expertise to an important problem for regional resource managers. Drs. Slattery, Gochfeld, and Boettcher have worked extensively in GOM seagrass communities, and have developed a collaborative network with the regional natural resource managers. Slattery serves as director of the NOAA NIUST Ocean Biotechnology Center &

3

Repository, an institute mandated to apply novel technologies to issues relevant to oceans and human health. In collaboration with the scientists of NOAA’s Grand Bay National Estuarine Research Reserve, he and the coPIs have addressed seagrass community stressors in single and multifactorial approaches. This RPC provided an opportunity to further support NOAA’s regional goals using unique satellite-based approaches. Dr. Easson and his crew at the University of Mississippi Geoinformatics Center (UMGC) have provided the expertise to acquire and analyze MODIS imagery products from the NASA Goddard Space Flight Center. The resulting scientific synergy represents a novel approach to identifying strengths and weaknesses of current management strategies.

2.0 In Situ Sampling: Populating the F&H Model with Real Ground-truthed Data

We initiated our sampling regime during this quarter. Temperature was sampled continuously using HOBO® Water Temp Pro v2 data loggers and these data were downloaded in January 2008. Salinity was monitored every 4 weeks during water sample collections using a handheld refractometer. Data to date indicate that this approach was the weak link in our “in situ generated model” as the salinity was clearly changing over tidal temporal scales (however, as noted, salinity is never limiting so the data can be overlooked for the purposes of the model). Light was measured continuously using HOBO® Pendant Temp/Light data loggers and these were downloaded in January 2008. However these loggers only provided a limited measurement

Grand Bay NERR Seagrass Ecosystem

Grand Bay

Middle Bay

JoseBay

Pont AuxChenes

4

of lumens; to gain a more thorough understanding of the light field we also monitored PAR, UVA, and UVB spectra using the International Light IL1400 Spectroradiometer every 4 weeks during water sampling. These data were then calibrated to the HOBOs in order to provide estimated daily light fields for the sampling period. Nutrients were also sampled monthly (by our NOAA collaborators), in conjunction with seagrass biomass estimates, but one important point is that GBNERR seagrass biomass and abundance dropped to virtually zero by November/December due to senescence. In addition we collected data on endocrine disruptors and mercury to determine whether anthropogenic variables can alter the model. To date those data are low background values that would probably not override natural seagrass community cycles. 2.1 Remote sensing: Populating the F&H Model with MODIS Data An evaluation of the capabilities of MODIS data for use in investigating the changes in seagrass communities provides researchers with an extensive dataset as the MODIS sensor collects information for a target location daily. As noted, MODIS has some inherent limitations including the lack of a direct measurement for nutrients, and potential data loss due to clouds. Nonetheless, over the course of any given month we were able to pull at least 10 reasonable “signals”, so MODIS still offers increased monitoring relative to the common monthly ground-based sampling strategies. 2.2 The F&H Model

The F&H model predict that seagrass productivity is a function of 4 key environmental parameters: salinity, temperature, light, and nutrients (seawater and sediment). Biomass of the seagrass can be predicted based upon levels of these 4 parameters. Since seagrass productivity has implications to the community health, modeling this parameter is crucial for seagrass resource managers. 3.0 Application of the F&H Model to Grand Bay NERR Seagrass Monitoring In Grand Bay, MS (GB NERR), the seagrass beds are dominated by Ruppia marina and Halodule wrightii. These primary producers compete with a variety of ephiphytic microalgae, macroalgae and periodic phytoplankton blooms for sunlight and nutrients; recent evidence suggests that epiphytes are responsible for most of the primary production within the system (Moncreiff & Sullivan 2001). Meso-grazers, such as the amphipod Gammarus mucronatus, are an important trophic link between resources within the ephiphytic community and the

P. Fong & M. Harwell, 1994. Modeling seagrass communities in tropical and subtropical bays andestuaries: a mathematical synthesis of current hypotheses. Bulletin of Marine Science 54:757-781.

Productivity seagrass = Pmaxseagrass (Salinity seagrass x Temperature seagrass x Lightseagrass x nutrients seagrass )[productivity asscw/ salinity]

[productivity asscw/ temperature]

[productivity assc w/ light]

[productivity asscw/ nutrients]

• Biomass seagrass[t +1] = Biomassseagrass[t] + Productivity seagrass - Lossseagrass [Loss f (senescence)]

5

consumers. Seagrasses share substrate with a variety of filter feeders, including the oyster Crassostrea virginica and the mussel Modiolus americanus, which remove phytoplankton and seagrass-derived DOM (Valentine & Heck 1993); their metabolic waste provides nutrients for the seagrass and epiphytic communities. An important primary consumer in these seagrass beds is the sea urchin Lytechinus variegatus, and while seagrass is not its most preferred food, it is the most dominant (Beddingfield & McClintock 1998; Rose et al. 1999). Other herbivores also utilize seagrasses, including various crustaceans, molluscs, and fishes. The indirect effects of seagrass productivity are apparent in a number of secondary consumers that feed on seagrass herbivores. For example, juvenile pinfish Lagodon rhomboides are predators of G. mucronatus, as well as the killifish Fundulis grandis, which also feeds on the amphipod. Higher trophic level interactions are exhibited by the commercially-important blue crab Callinectes sapidus; this omnivorous scavenger feeds on meso-grazers, bivalves, small fishes, and occasionally algae and seagrasses. The blue crab, sea urchins, and pinfish represent the most obvious linkages with large finfish and seabirds that forage in seagrass communities. 3.1 Plans to Utilize NASA Next-Generation Sensors The use of Moderate Resolution Imaging Spectroradiometer (MODIS) for this project was due to the availability of data collected by MODIS to replace in-situ data to be used in the Fong & Harwell (1994) model. The MODIS products used for this project were OC3 chlorophyll a concentration, sea surface temperature and normalized water-leaving radiance at 412nm. These products were derived from the Ocean Color Data Processing System (OCDPS) at Goddard Space Flight Center (GSFC). Performance of the MODIS sensor on both Terra and Aqua has now been shown to adequately replace in-situ data for monitoring seagrass growth. The continuation of the products derived from the MODIS sensor for monitoring seagrass will allow for future studies of seagrass in order to determine if any trends are present. The Visible Infrared Imager Radiometer Suite (VIIRS) will represent the next generation NASA satellite that will replace the current MODIS Terra and Aqua missions (Yu et al., 2005). The improvements that will be on VIIRS include quicker data delivery time, improved scan geometry, and the 22 channels design that is based on experience from the study of past missions (Lee et al., 2006).

The Visible Infrared Imager Radiometer Suite (VIIRS)is designed to collect visible/infrared and radiometric data to obtain measurements of Earth’s oceans, land surface and atmosphere with global coverage at least once per day, and with better than 1-km spatial resolution (Welsch et al., 2001). The type of environmental data measurements to be made by VIIRS include atmospheric, clouds, earth radiation budget, clear-air land and water surfaces, sea surface temperatures, ocean color, and low light visible imagery (Lee et al., 2006). VIIRS will fly on the National Polar-Orbiting Operational Environmental Satellite System (NPOESS) in three polar sun-syncronous orbit planes (at an altitude of 833 km), with equator crossing times of 1730, 0930, and 1330. It will also fly on the NPOEESS Preparatory Project with a 1030 equator crossing time. VIIRS draws heavily on the experience gained in building and operating MODIS that also combined diverse functions into a single sensor. Experience with other sensors such as Sea-viewing Wide Field-of-view Sensor (SeaWiFS) for measuring ocean color and the Along Track Scanning Radiometer (ATSR) for measuring sea surface temperature will provide better data that will specifically benefit the parameters required to monitor seagrass. The difference in the

6

spectral content in the expected VIIRS sensor and the current MODIS sensors is significant. MODIS is a 36 band spectrometer with a spatial resolution (pixel size at nadir) of 250m for channels 1 and 2 (0.6µm - 0.9µm), 500m for channels 3 through 7 (0.4µm - 2.1µm) and 1,000m for channels 8 to 36 (0.4µm - 14.4µm). The MODIS instrument consists of a cross-track scan mirror that generates a simple rectangular array of pixels following the time sequence of the scan mirror rotation and the spacecraft along-track movement. At nadir position, a nominal 1-km pixel has dimensions of 1 x 1 km but as the scan angle increases from nadir the pixel dimensions grows to approximately 4.8-km by 2-km along track (Masuoka et al., 1998). Due to this feature, a single swath covers 10-km along track at nadir and it expands to 20-km along track creating a “bowtie-shaped” footprint which overlaps the swath below and above it. In simple terms a single feature on Earth may appear in several scan lines if it is located near the edge of the scene. 4.1 Summary of Findings This study focused on two types of seagrass found in the Grand Bay National Estuarine Research Reserve, Halodule wrightii and Ruppia maritime, because both have similar high tolerances to salinity. The high tolerances to salinity allow this factor to be removed from the model. Due to this fact, the temperature, light and nutrient parameters can be used, and each can be derived from the MODIS sensor and can solely be used to run the model. After the MODIS data was collected and processed, it was then used to run the model from Fong & Harwell (1994) [Figure 2]. The data collected in the field was also used to run the model. The results from both methods were then compared [Figure 1]. The relative seagrass productivity from each method showed the same decreasing trend in seagrass productivity, which correlates to this study occurring outside of the seagrass growing season. The MODIS data was compiled from the beginning of 2007, which did include data for the growing season. When completing the model using the MODIS data from the growing season, the model yielded positive seagrass productivity. This study provided an excellent correlation between the use of two different collection methods for use in a single model [Figure 3].

7

4.2 Recommendations From these results the following can be concluded: 1. Remote sensing platforms can be used, with some considerations, to populate parameters of the Fong & Harwell model of seagrass community productivity. 2. In situ data provided finer scale resolution of real world conditions; but temporal logistics may offset some of this benefit. 3. Cooperative work between satellite-based and ground-based data acquisition teams appears to offer the greatest opportunities for seagrass resource managers. 4.3 Future Directions We expect to run an iterative in situ-generated model once a complete dataset have been collected through 2008. The remote sensing team will use our sampling dates to download correlated data for the remote-generated model. Unfortunately, the current data correspond to an early seagrass senescence so the model presented here may suffer from this natural variable. Significantly, the remote sensing group will be able to mine data archives back to the summer 2007 peak of seagrass productivity- a strength of remote sensing technologies. The in situ model will be fleshed out in the spring as seagrass recruitment ensues. The ability to apply the methodology for this study to other areas and other types of seagrass will

Figure 1a-d: In situ sampling data.

09/07 10/07 11/07 12/07 01/08 02/08 03/08

30

25

20

15

10

09/07 10/07 11/07 12/07 01/08 02/08 03/08

500

100

50

10

5

09/07 10/07 11/07 12/07 01/08 02/08 03/08

0.04

0.03

0.02

0.01

0.00

nd nd

09/07 10/07 11/07 12/07 01/08 02/08 03/08

instrumentation error

1.25

1.00

0.75

0.50

0.25

0.00

Phos

phat

e (m

g/L)

1.0

0.8

0.6

0.4

0.2

0.0

ANOVA: significant timeeffect, site effect

ANOVA: significant time effect, site effect

ANOVA: significant time effect, site effect

ANOVA: significant site effect

8

allow for a more in-depth analysis on the relationship between field collected and remotely sensed data in determining seagrass productivity. The assessment of the Fong & Harwell model in St. Joseph’s Bay, FL will provide an opportunity for this because of the abundance of Thalassia testidinum and Sytingodium filiforme, two additional species of seagrasses.

Figure 2a-d: Remote sampling data.

09/07 10/07 11/07 12/07 01/08 02/08 03/08

30

25

20

15

10

09/07 10/07 11/07 12/07 01/08 02/08 03/08

500

100

50

10

5

09/07 10/07 11/07 12/07 01/08 02/08 03/08

0.04

0.03

0.02

0.01

0.00

nd nd

Nitrate [NO 3]:Y=0.255+7.557*X

nutri

ent

Chla

Phosphate [PO 4]:Y=0.135-0.213*X

from In situ studies-

Figure 3: Comparative statistics.

Date

Rel

ativ

e Se

agra

ss P

rodu

ctiv

ity

Paired t -test: t-value = -1.261 P = 0.2541

09/07 10/07 11/07 12/07 01/08 02/08 03/08

0

5

-15

-5

-10

theo

retic

al v

alue

s

In situ model

Remote model

Rapid Prototyping of NASA Next Generation Sensors with the Nonpoint-Source Pollution and Erosion Comparison Tool

(N-SPECT)

Justin Janaskie, Greg Easson, Lynn Francis University of Mississippi, University MS 38677

Executive Summary The purpose of this Rapid Prototyping Capability (RPC) experiment is to determine the compatibility of the current version of the Nonpoint-Source Pollution and Erosion Comparison Tool (N-SPECT) with a NASA next generation sensor relating to precipitation data by using the Tropical Rainfall Measuring Mission (TRMM) as simulated Global Precipitation Monitoring (GPM) data. By assessing the compatibility of N-SPECT with NASA next generation sensors, water resource managers in coastal and inland areas will have another source of precipitation measurements available to use in analyzing water quality issues. Currently, the recommended precipitation data for input into N-SPECT is annual or event scenarios using Parameter-elevation Regressions on Independent Slopes Model (PRISM) data or precipitation data from NOAA National Climatic Data Center. This experiment compares the results of N-SPECT output erosion, runoff and pollution and using a large, single precipitation event and applies it to three temporally different Coastal Change Analysis Program (C-CAP) datasets for watersheds on the Gulf coast of Mississippi in order to determine the validity of using a NASA next generation sensor as a substitute for the recommended precipitation datasets to successfully run N-SPECT.

Use of NASA Satellite Assets for Predicting Wildfire Potential for Forest Environments in Guatemala

Laura Johnson,1 Bill Cooke, 2 Victor H. Ramos,3 Greg Easson1

July 31, 2008

Preliminary report

                                                            1 University of Mississippi Geoinfomatics Center (UMGC)  2 Mississippi State University  3 National Council for Protected Areas, Guatemala 

 

Abstract

A Rapid Protocol Capability experiment was undertaken to evaluate MODIS and simulated VIIRS NDVI time series data for monitoring leaf litter conditions in seasonal evergreen forests (SEF). Leaf litter moisture content is the most important factor influencing fire potential in SEF in Mesoamerica, yet options for the spatial and temporal monitoring of fuel moisture content are limited. The approach taken to this research was to compare MODIS and simulated VIIRS NDVI indices compared with KBDI values. Visual analysis, Spearman-Rho statistical analysis and seasonal-trend decomposition (STL) were undertaken. Visual analysis of the data suggested that neither the 400m simulated VIIRS NDVI data or the MODIS NDVI time series data to be particularly well correlated with inverse KBDI, though the simulated VIIRS NDVI appeared somewhat better than the MODIS. Spearman’s Rho statistic confirmed this finding with relatively low values for VIIRS NDVI and MODIS NDVI and KBDI -0.39 and -0.47 respectively. However, the visual correlation for both MODIS NDVI and simulated VIIRS NDVI improved considerably with the seasonal detrending of MODIS and VIIRS data sets. Next steps include cross-correlation of detrended inverse VIIRS NDVI and inverse KBDI, and MODIS NDVI and inverse KBDI. Recommendations are made for de-trending vegetation indices as a prerequisite for assessing the value of a vegetation index for fire potential research. Additional research with relating other important simulated VIIRS indices including EVI, NDWI, SWISI and NMDI is also recommended.

1.0 Introduction

Wildfire is a serious problem in Mesoamerica. In some years, fires burn extensively in the region with significant consequences in terms of loss of biodiversity, threat to human life and infrastructure, smoke-related health concerns, and the release of gases related to climate change concerns. In recent years significant advances have been made in developing and implementing near real time fire detection systems such as the SERVIR Rapid Response Fire Mapper Products that monitor hot spots and active fires.4 However, little progress monitoring fire potential,5 or the conditions under which large-scale fires occur (Qu, In prep). Regular assessment of fire potential is of critical importance for improving fire management and protection efforts in the region.

                                                            4 SERVIR is a NASA managed program integrating satellite and geospatial data with the aim to develop scientific and decision making knowledge for issues affecting Mesoamerica.  

5 Fire potential or fire danger potential is the vulnerability of a location to the ignition and spread of fire. 

The purpose of this Rapid Protocol Capacity (RPC) experiment6 is to evaluate Moderate Resolution Imaging Spectroradiometer (MODIS) and simulated Visible/Infrared Imager/Radiometer Suite (VIIRS) NDVI time series data as a proxy for the moisture status of leaf litter fuels. Fires that affect lowland seasonal evergreen forests in Mesoamerica are surface fires that consume dry leaf litter. The moisture status of surface fuels in these forests is difficult to monitor. While soil drought codes such as the Keetch-Bryam Drought Code are suggested for use to evaluate the moisture content of litter fuels (Brolley, 2005), these are of limited use in the tropics where daily weather station data is difficult to obtain (Johnson, L., pers. obs., 2006; V. H. Ramos, pers. com., March, 2008).

Expected impacts for the project are three-fold. First, the project will give clear indication of the usefulness of MODIS and simulated VIIRS generated NDVI time series products for monitoring leaf litter moisture conditions in northern Guatemala. Second, the project will provide indication of the application potential for expanding SERVIR decision support capability to include near real time fire potential products. Finally, this RPC experiment will directly contribute to furthering NASA Applied Sciences Programs mission by supporting the ongoing mission of SERVIR’s Rapid Response Fire Products related to the National Application Areas of air quality, disaster management, and ecological forecasting and public health. Partners in the interdisciplinary effort include the University of Mississippi GeoInformatics Center, Mississippi State University GeoResources Institute, Institute for Technology Development Inc. (ITD)7 and CONAP (National Council for Protected Areas), Guatemala.

2.0 Fire Potential Model for Mesoamerica

Leaf litter moisture content is the most important aspect for assessing fire potential in the tropics (Mueller-Dombois, 1981; Uhl et al., 1988; Kauffman & Uhl, 1990; Holdsworth & Uhl, 1997; Nepstad et al., 1999). Fires that occur in seasonal evergreen forests are surface fires (Figure 1). Fires commonly start in adjacent fields of pastures and continue to burn into adjoining SEF as long as fuel moisture content of the surface litter is sufficiently low. Though figures vary, there is general acceptance that leaf litter moisture contents of 15% or below will sustain SEF fire (Neigreiros, 2004; Ray et al 2005; Johnson 2006).8

                                                            6  The  RPC  experiments  provide  accelerated  simulation,  demonstration,  and  testing  of  candidate  system. configurations from current and future NASA‐funded Earth‐observation missions.  RPC experiment outcomes, both for or against a particular system configuration are important in helping NASA define which specific directions and strategies to pursue within the Applied Sciences Program.  

7 ITC worked in association with Science Systems and Applications Inc. (SSAI) on the generation of simulated VIIRS data. 8 Likelihood of spread in SEF also increases with amount of litter fuel and latest of fire season.   

Historically, fine fuel moisture contents have been problematic for fire managers to monitor as they require repeated manual testing over large areas throughout the fire season. In the early 1970s, fire scientists in the USA developed a drought code to predict fine fuel moisture content, known as the Keetch-Bryam Drought Code (KBDI) (Keetch & Byram, 1968; Fujioka, 1991). Essentially, KBDI provides a measure of the moisture in the upper soil reservoir, which relates directly to the moisture status of the vegetation and the likelihood that the vegetation is susceptible to burning (Brolley, 2005).9 The KBDI assumes the soil has a field capacity of eight inches of water, therefore the index is on a scale ranging from 0 to 800.10 Zero is the point of no moisture deficiency and 800 is the maximum drought that is possible.

In recent years, KBDI has also been recognized as useful for fire prediction in the seasonal tropics. Weidemann et al., (2002) introduced the use of KBDI as a key component of the fire weather system for Indonesia. Goldammer, (2001) recommended the Index for monitoring fire potential in Guatemala. Johnson, (2006) indicated the January 31st KBDI value of value for predicting fire potential in western Thailand. Still, the application of KBDI for fire management in the seasonal tropics has been limited. KBDI requires daily 24 hr rainfall and maximum temp readings from weather stations in forested areas. Weather stations in the developing tropics tend to be few and far between. Further, these stations are often in remote areas. Weather stations break-downs are common occurrences. Further, there is no easy way to transfer the daily weather data to fire managers (Johnson, 2005, pers. obs).

3.0 Use of MODIS data for fire potential assessment in the seasonal tropics

In recent years, the benefits of using MODIS data for fire potential assessment, including near continuous spatial and temporal coverage at the regional scale, has been recognized. As a result, fire scientists and managers have become increasingly increased in generation of vegetation moisture and drought codes directly from satellite data. Fire scientists in the USA have already moved to integrate AVHRR and MODIS daily coverages into fire products including MODIS Rapid Response program, the Wildland Fire Assessment System (WFAS), and the Oklahama Mesonet.11 NDVI, or derivations from this vegetation index including ‘relative greenness’ and ‘departure from average greenness’, is the most commonly used vegetation index12 for this purpose.

                                                            9 KBDI  is a number  representing  the net effect of evapotranspiration and precipitation  in producing cumulative moisture deficiency  in deep duff and upper soil  layers. The  index number  indicates the amount of net rainfall,  in hundredths of an inch, that is required to reduce the index to zero (saturation).   10 The metric KBDI scale runs from 0‐2000. 11 KBDI is still used extensively for monitoring dead fuel moisture content in the fire weather system. 12 Vegetation  indices  are  generated by  applying  common mathematical  ratio‐based operations  to multispectral image  data.    These  ‘accepted’  operations  are  used  to  emphasize  and  highlight  subtle  variations  in  the  actual spectral responses of various surface covers, such as vegetation greenness.  

A number of studies have also appeared suggesting the use of MODIS vegetation indices in the seasonal tropics. For instance, Anderson et al (2007) recently undertook a multiyear time series study using MODIS NDVI, Enhanced Vegetation Index (EVI) and Normalized Difference Water Index (NDWI) to detect the effects of the 2005 drought on dense tropical forest canopies in Amazonia. Preliminary results suggest that EVI showed some correlation to the 2005 rainfall anomaly, but that NDVI and NDWI did not. Verbesselt et al (2006) used time series SPOT VGT 10-day composite data to monitor vegetation moisture condition represented by the KBDI. Correlation coefficients for the NDII, NDVI and KBDI for savanna and forest were derived. Verbesselt et al. (2007) found that SPOT based Normalized Difference Infrared Index (NDII) is better correlated with KBDI compared to NDVI.13 Further, Fensholt & Sandholt (2003) showed that MODIS-based Shortwave Infrared Water Stress Index (SIWSI) performs better than NDWI and NDII in predicting canopy water stress. Across the board, research results are mixed. More research needs to be conducted in order to identify the MODIS based vegetation index most closely correlated with KBDI.

4.0 Experimental Approach

The approach taken to the research was to evaluate the potential of VIIRS data for monitoring vegetation moisture condition represented by the KBDI. Simulated VIIRS NDVI 8-day 400m resolution composite time series products and MODIS NDVI 8-day 500m resolution composite time series products were produced for the research. NDVI time series data was generated for 2001-2005, a period which included two major fire years.14 The research was under taken by: first, selecting a suitable study area; second, generating the simulated VIIRS NDVI (400m) and MODIS NDVI (500m) time series products; third, generating the 10-day KBDI moving average data; fourth, selecting and processing the time series pixels of interest in the two digital data sets; and finally, comparison of both the time series simulated VIIRS and MODIS NDVI against weather station-generated KBDI.

4.1 Study Area

Research was undertaken in northern Guatemala which is centrally located in Mesoamerica and contains a number of important protected areas, collectively known as the Maya Biosphere Reserve (Figure 2).15 The climate for the area is seasonal tropics. There are two six month seasons, a wet season and a dry season. The wet season is from June through November. The

                                                            13 Equations of the vegetation indices listed are included in Appendix 1. 14 The most significant fire years to date occurred in 1998, 2003, 2005 and 2007. 15 The Maya biosphere Reserve is part of the larger Mesoamerican Biological Corridor.  The corridor is a network of protected areas from southern Mexico to Columbia, through Belize, Costa Rica, El Salvador, Guatemala, Honduras, Nicaragua and Panama. 

dry season is from December through May.16 The primary vegetation in the area is seasonal evergreen forest (SEF). Agricultural areas and pasture lands intersect larger tracts of intact forest in the more accessible areas. The start date for the fire season in the area is mid March, with the peak fire activity from mid April through mid May.

The specific area of interest for this study was set near the community of Tikal (Figure 3). First, a weather station is located in the area and a continuous weather data record is available back several years. Second, the area is located in a protected area, Tikal Park and is part of the Maya biosphere Reserve, an important an important Protected Areas complex in Mesoamerica. Third, forest fires are a regular occurrence and one of the primary threats to the integrity of vegetation in the area.17 Finally, CONAP, who manages the protected areas in Guatemala, is actively seeking management solutions for the wildfire problem in the area.

4.2 Keetch-Byram Drought code (KBDI)

The local KBDI was generated using daily maximum temperature and daily 24 hour rainfall values for the years 1999 until present (Figure 2). Weather data from the Tikal weather station was utilized for KBDI calculations (CONAP, 2008). KBDI was calculated in pre-programmed Excel worksheet. The KBDI calculation worksheet was obtained from a GTZ Project, “Integrated Forest Fire Management” based in East Kalimantan, Indonesia. The KBDI worksheet formula was slightly modified (i.e., the units were changed from imperial to metric) with the result that the index scale was also readjusted from 1-800 to 1-2000 (Weidemann et al., 2002). The KBDI requires that calculations being after 150mm or greater rain have fallen in the previous seven days (cf., Fujioka, 1991; Weidemann et al., 2002). A starting date of October 25, 1999 was selected as 282mm of rainfall was recorded in the preceding four-day period. Finally, the KBDI data set was smoothed using a 10-day moving average.

4.3 Simulated VIIRS and MODIS data

The simulated VIIRS 400m data was generated using eight-day, gridded 250m surface reflectance products for July 28, 2001 through July 27, 2005.18 Products included MODIS HDF data from Terra sensor (MOD09Q1 product) and MODIS HDF data from Aqua sensor (MYD09Q1 product).19 The HDF data sets were input into Application Research Toolbox

                                                            16 Precipitation  is particularly  low  for  February  through  early May with weekly precipitation  amounts of under 100mm. 17 Approximately, 400,000 hectares of the reserve has burned in these major fire years (CONAP, unpub). 18 This allowed for a couple of months buffering time for the TSPT on either side of the dry seasons. 19 MYD09Q1 (Aqua) data obtained was for July 4, 2002 through July 19, 2005.  Aqua was not launched until early 2002. 

(ART). 20 Here they were first, converted from sinusoidal to a UTM projection;21 second, subset to the area of interest; and finally transformed into simulated VIIRS data.22 MODIS NDVI products were then calculated from the ART generated MO(Y)D09Q1 RED and NIR (near infra red) reflectances.23 Pixels with clouds and those with viewing zenith angles (VZA) greater than 40 degrees were masked from the simulated VIIRS NDVI data sets. The Aqua and Terra simulated VIIRS NDVI data sets were then fused to reduce gaps in data due to clouds.24 Outlier scenes were then removed to create more ‘ideal’ images.25 Finally, temporally filtered data were then created using the Time Series Product Tool (TSPT).26,27 Validation of MODIS-based VIIRS simulation was undertaken by ITD (Ross et al., 2008).

The MODIS 500m was generated using eight-day, gridded 500m surface reflectance products obtained for July 28, 2001 through July 27, 2005. Products used included MODIS HDF data from Terra sensor (MOD09A1 product) and MODIS HDF data from Aqua sensor (MYD09A1 product), which provide MODIS bands 1-7.28 The HDF data sets were input MATLAB for reprojection and subsetting to the area of interest as ART was not required. From this point the procedure for generating the time series MODIS NDVI is the same as for generating time series 400m simulated VIIRS (see above).

4.4 Image processing

Both NDVI time series data sets were processed using ENVI software. ENVI’s Interactive Data Language (IDL) module was used to extract a 5 x 5 window of pixel values over the Tikal forest area for successive scenes (Figure 4). These ‘window’ values were averaged. Averaged satellite sensor simulated VIIRS and MODIS values and weather station derived KBDI values for intact seasonal evergreen forest (SEF) near Tikal and pasture were graphed.

                                                            20 ART  is NASA’s  in‐house  software  for data  simulation.   The ART enables data  to be  scaled  in either at‐sensor radiance or planetary.  It includes a utility in which an image extent mask is generated for subsequent subsetting of comparable MODIS, simulated MODIS, or simulated VIIRS data derived  from MODIS data.        In  the process ART convolves  the  input MODIS data  to  the VIIRS PSF  (point  spread  function)  in order  to produce MODIS‐simulated VIIRS data.   21 A WGS‐84, Zone 16N projection was used 22 During this process a MODIS to VIIRS two‐dimensional finite impulse response (2D FIR) filter was applied 23 See Appendix 1 for NDVI equation. 24 During the fusing process, the maximum NDVI, or the brighter value, of the two images is used for a given pixel’s location.  25 Outliers included images with ‘up’ spikes of maximum slope change of 0.006 and ‘down’ spikes with maximum slope change of 0.004. 26 The temporal filtering process produces data at a daily temporal resolution via interpolating between 8‐day time periods. 27 A weighted‐Savitzky‐Golay  filter was used  to generate continue MODIS and  simulated VIIRS  time  series NDVI data products. Setting for the Savitzky‐Golay filter were 1st order, Frame size=3, full width half‐Max (FWHM)=2 28 MYD09Q1 (Aqua) data obtained was for July 4, 2002 through July 19, 2005.  Aqua was not launched until early 2002. 

4.5 Analysis

Analysis included three parts. First, basic properties of the simulated VIIRS and MODIS NDVI were evaluated by visual analysis of the time series.29 Second, correlation coefficients for the simulated VIIRS and MODIS NDVI and KBDI for SEF and pasture were derived. Finally, time series analysis techniques were used to account for the autocorrelation effect of seasonality. Methods for STL or Seasonal-Trend decomposition procedure are based on Loess (Cleveland et al., 1990).

5.0 Results

5.1 Visual analysis

Important properties of remote sensing indices are revealed by visual analysis of time series. Firstly, figure 5 shows that a better correlation exists for the SEF MODIS and VIIRS NDVI then for the pasture. One would anticipate that the simulated VIIRS and NDVI would appear similar as one data set was created from the other. Secondly, a visual comparison of MODIS NDVI and inverse KBDI, and simulated VIIRS NDVI and KBDI was also made for SEF. Departure from mean NDVI for both the vegetation indices was plotted on the Y axis to that that NDVI and KBDI could be viewed on a comparable scale. Overall the correlation in Figure 6 does not look very good. The NDVI indices depart from the mean in the same way that KBDI does, however, there are a few times when just NDVI departs opposite to the direction of KBDI such as in early 2002 and the dry season of 2003. The year 2003 was one of the largest fire years but the correlation does not appear to hold true for 2005 which was also a major fire year.

5.2 Correlation analysis

Spearman’s rho statistic is used to estimate a rank-based measure of association. It is a good measure of the linear relationship between two variables. This method is more robust than the Pearson rho statistics and has been recommended when data is not bivariate normally distributed.

Analysis for year round data 

 

 

 

                                                            29 An inverted KBDI data set was graphed to support visual analysis and comparison of the data sets.  

  rho NDVI(VIIRS)/KDBI 

rho NDVI(MODIS)/KDBI 

MBR 1 (Tikal weather )  ‐0.64  ‐0.56 MBR 2 (SEF)  ‐0.47  ‐0.39 MBR 14 (pasture)  ‐0.26  ‐0.24 

Analysis for dry Season data 

 

 

 

 

Analysis for wet season data 

 

 

 

Again, Spearman’s Rho confirms poor correlation results between VIIRS-based NDVI and KBDI and MODIS-based NDVI and KBDI. Note the correlation between VIIRS-based NDVI and KBDI is slightly stronger than the correlation between MODIS-based NDVI and KBDI. This could be could be because the simulated VIIRS data is a higher resolution product i.e., 400m versus 500m. Also, the correlations are significantly stronger in the dry season that the wet season which is along the lines of what Verbesselt et al. (unpub) found. Still, the correlations are not significant.

5.3 Time Series Analysis

The seasonal time series analysis presented the most interesting results. By removing a weak trend in each time series, the seasonal values correlated well. Daily MODIS-based NDVI, VIIRS-based NDVI and KBDI values were analyzed using a Seasonal-Trend Decomposition Procedure Based on Loess (STL). STL is a filtering procedure for decomposing a time series into trend, seasonal, and remainder components. Figure 7 shows the results of STL applied to daily MODIS-based NDVI data. The first (top) panel is daily average measurements of MODIS-based NDVI, the raw data. The second panel graphs a seasonal component: variation in the data at or near the seasonal frequency, which in this case is one cycle per year. The third panel graphs a trend component: the low frequency variation in the data together with nonstationary, long-term changes in level. The remainder component, shown in the fourth panel, is the remaining variation in the data beyond that in the seasonal and trend components. By summation of seasonal, trend and remainder part the data series is reconstructed. (Cleveland et al, 1990).

  rho NDVI(VIIRS)/KDBI 

rho NDVI(MODIS)/KDBI 

MBR 1 (weather stn.)  ‐0.74  ‐0.62 MBR 2 (SEF)  ‐0.42  ‐0.32 MBR 14 (pasture)  ‐0.43  ‐0.49 

  rho NDVI(VIIRS)/KDBI 

rho NDVI(MODIS)/KDBI 

MBR 1 (weather stn.)  ‐0.47  ‐0.42 MBR 2 (SEF)  ‐0.55  ‐0.49 MBR 14 (pasture)  0.01  0.17 

Both VIIRS-based NDVI and MODIS-based NDVI time series shows a small trend but strong seasonal behavior. One of the features of STL is that it gives a robust estimate of trend and seasonal components that is not distorted by aberrant behavior in the data (e.g. spikes caused by clouds in the NDVI time series).

Autocovariance measures this relationship or dependence. When autocovariance is normalized between +1 and -1 (where 1 is perfect correlation) then it is autocorrelation. Figure 8 shows the strong seasonality of the MODIS-based NDVI by plotting the autocorrelation function. The values correlate quite well with values taken a year later and two years later (lag=365, or lag= 720). A small portion of the remainder part is still auto-correlated (more than 5 % of autocorrelations should fall outside bounds) indicating that STL needs further fine-tuning. Ongoing research is aiming at optimization of the STL to obtain a non-auto-correlated remainder component. In the future this remainder component will be used for cross-correlation analysis and further modeling of seasonal time series.

On the strength of the STL procedure, time series of raw values and detrended seasonal values were plotted in Figure 9. Once detrended, the differences between MODIS-based NDVI and simulated VIIRS-based NDVI are negligible. There is a significantly stronger visible correlation between both NDVI data sets and KBDI when detrended seasonal data is used.

6.0 Conclusions and Recommendation

Research was undertaken to evaluate MODIS and simulated VIIRS NDVI time series data as a proxy for monitoring leaf litter conditions in seasonal evergreen forests. Simulated VIIRS NDVI 8-day 400m resolution composite time series products and MODIS NDVI 8-day 500m resolution composite time series products were produced for the 2001 through 2005. Analysis included first, visual analysis of MODIS and VIIRS time series data sets (Figure 5), as well as visual comparison of each NDVI indices against inverse KBDI (Figure 6). Data was decomposed and the trend removed so that only the seasonal data could be analyzed. We used autocorrelation analysis to verify the success of the decomposition procedure. Finally, we compared just seasonal values for simulated VIIRS-based NDVI, MODIS-based NDVI and KBDI.

Visual analysis of the data suggested that neither the 400m simulated VIIRS NDVI raw data or the MODIS NDVI time series raw data was particularly well correlated with inverse KBDI, though the simulated VIIRS NDVI appeared somewhat better than the MODIS. Spearman’s Rho statistic confirmed this finding with relatively low values for VIIRS NDVI and MODIS NDVI and KBDI -0.39 and -0.47 respectively. However, the visual correlation for both MODIS NDVI and simulated VIIRS NDVI improved considerably with the seasonal detrending of MODIS and VIIRS data sets. Next steps include cross-correlation of detrended inverse VIIRS NDVI and inverse KBDI, and MODIS NDVI and inverse KBDI.

Recommendations are made for de-trending vegetation indices as a prerequisite for assessing the value of a vegetation index for fire potential research. Additional research with relating other important simulated VIIRS indices including EVI, NDWI, SWISI and NMDI is also recommended.

7.0 References

Anderson,  L.O.,  Y. Malhi,  Y.  E.  Shimabukuro,  L.  E.  O.  C.  Aragao  (2007).    Evaluating MODIS vegetation and water indices for detecting canopy stress during the 2005 drought in Amazonia.  Anais XIII Simposio Brasileiro de Sensoriamento Remoto, Florianopolis, Brasil, April 21‐26. 

Brolley,  J.,  J.  J. O’band  D.  F.  Zierden  (2005).    Experimental  forest  fire  threat  forecast.    15th Conference on Applied Climatology. 

Cleaveland, R. B., W. S. Cleveland, J. E. McRae, & I. Terpenning (1990).  STL:  A Seasonal‐Trend Decomposition Procedure Based on Loess.  Journal of Official Statistics.  Vol. 6, No. 1, pp. 3‐73.   

Fujioka, F. M.  (1991).   Starting  the Keetch‐Byram Drought  Index.   11th Conference  in  fire and forest meteorology, Missoula, Montana. 

Holdsworth, A. R. & C. Uhl  (1997).    Fire  in Amazonian  selectively  logged  rain  forest  and  the potential for fire reduction.  Ecological Applications, 7(2), 713‐725. 

Huete, A.R., Didan,K.,Muira,T., Rodriguez,E. P.,Gao,X.,&Ferreira, L.G.  (2002).   Overview of  the radiometric and biophysical performance of the MODIS vegetation  indices. Remote Sensing of Environment, 83(1‐2), 195−213.  Fensholt, R., & Sandholt,  I. (2003). Derivation of a shortwave  infrared water stress  index from MODIS  near‐and  shortwave  infrared  data  in  a  semiarid  environment.  Remote  Sensing  of Environment, 87(1), 111−121.  Gao, B. C. (1996). NDWI‐A normalized difference water index for remote sensing of vegetation liquid water from space. Remote Sensing of Environment, 58(3), 257−266.  Goldammer, J. G. (2001).  Report on Short‐Term Consultancy on "Community‐Based Integrated Fire  Management"  for  the  Project  "Prevención  y  Control  Local  de  Incendios  Forestales" (PRECLIF) Petén, Guatemala, Central America.  The Global Fire Monitoring Center (GFMC). GTZ PN 2001.9070.0‐001.01.  August 20, 2001. 

Johnson, L.  (2006).   Fire, Seasonally Dry Evergreen Forest and Conservation, Huai Kha Khaeng Wildlife  Sanctuary,  Thailand.    Unpublished  Dissertation,  University  of  Victoria,  Victoria,  BC, Canada. 

Keetch,  J.  &  G.  Byram  (1968).    A  drought  index  for  forest  fire  control.    Res.  Paper  SE‐38. Asheville, NC: U.S. Department of Agriculture, Forest Service, Southeastern Forest Experiment Station. 32 pp.  

Mueller‐Dombois,  D.  (1981).    Fire  in  Tropical  Ecosystems.    Fire  Regimes  and  Ecosystem Properties conference, Honolulu, Hawaii.  

Neigreiros,  G.  H.  D.  (2004).    Understanding  and  Modeling  Ecological  Processes  Controlling Flammability  in  Seasonally  Dry  Evegreen  Forests  of  the  Brazilian  Amazon.    Unpublished Dissertation, University of Washington, Seattle. 

Ray, D., D. Nepstad,& P. Moutinho  (2005).   Micrometeorological and  canopy  controls of  fire susceptibility in a forested Amazon landscape.  Ecological Applications, 15(5), 1664‐1678. 

Ross, K., K. Holekamp & J. Russell (2008).  Wildfire MODIS Proxy Validation. ITD internal report. pp4. 

Rouse, J. W., Hass, R. H., Schell, J. A., & Deering, D. W. (1974). Monitoring vegetation systems in the  Great  Plains  with  ERTS.  Proceedings,  third  Earth  Resources  Technology  Satellite‐1 Symposium, Greenbelt, NASA SP‐351 (pp. 309–317).  Uhl,  C.,  J.  B.  Kauffman  &  D.  L.  Cummings  (1988).    Fire  in  the  Venezuelan  Amazon  2:  Environmental  conditions necessary  for  forest  fires  in  the evergreen  rainforest of Venezuela.  Oikos, 53(2), 176‐184.   

Weidemann,  D.,  Buchholz,  G.,  &  Hoffmann,  A.  A.  (2002).    A  Fire  Danger  Rating  for  East Kalimantan, Indonesia.  Samarinda, East Kalimantan:  GTZ – German development agency. 

 Verbesselt, J., S. Lhermitte, K. Nackaerts, P. Coppin (unpub). Assessment of vegetation moisture condition by time series analysis of SPOT Vegetation data.  pp. 6. 

Verbesselt, J., P. Jonsson, S. Lhermitte, J. van Aardt, & P. Coppin (2006).  Evaluating satellite and climate data‐derived  indices as  fire risk  indicators  in savanna ecosystems.    IEEE Transanctions on Geoscience and Remote Sensing, Vol.  44, No. 6, June 2006    Zarco‐Tejada, P. J., Rueda, C. A., & Ustin, S. L. (2003). Water content estimation  in vegetation with MODIS  reflectance data and model  inversion methods. Remote Sensing of Environment, 85(1), 109−124.  

 

 

 

Figure 1:  An example of fire in Seasonal Evergreen Forest (SEF).  SEF fires are surface fires that burn beneath the forest canopy. 

 

 

 

 

 

Figure 2:  Intact Seasonal Evergreen Forest (SEF).  Inset shows the Tikal area and Mesoamerican Biodiversity Corridor which includes the Maya Biosphere Reserve in Guatemala. 

 

 

 

Figure 3:   Location of the Seasonal Evergreen Forest and pasture sites  in relation to the Tikal weather station.  

 

 

 

 

 

 

Figure 4:  Representation of the method process where a 5x5 grid of pixels values were extracted from consecutive images, averaged, and graphed.  

 

Figure 5:   Time  series  for MODIS NDVI and  simulated VIIRS NDVI data.   Brown  lines are  values  from pasture and green lines are values from forest. 

 

Time Series for Forest from simulated VIIRS‐based NDVI and Inverted KBDI 

Time Series for Forest from MODIS‐based NDVI and Inverted KBDI 

 

Figure 6:   Time  series  for SEF plotted as departure  from  the mean  for  inverse KBDI and a)  simulated VIIRS‐based NDVI and b) MODIS‐based NDVI  

 

 

Figure 7:  Detrending process for time series for SEF data for MODIS‐based NDVI. (a) shows the raw data (b) shows the detrended data, (c) shows the trend removed and (d) shows the remainder. 

 

 

Figure8:    Autocorrelation  results  on  detrended  seasonal  MODIS‐based  NDVI  and  the  remainder component from detrending.   

 

 

 

Figure 8:   Time  series  for SEF plotted as departure  from  the mean  for  inverse KBDI,  simulated VIIRS‐based NDVI and MODIS‐based NDVI.  (a) shows  the  raw data and b) shows  the  improvement by using detrended seasonal data 

 

 

Appendix 1

 

 

 

 

 

 

 

 

 

 

 

The University of Mississippi, Easson: Rapid Prototyping Capability for Earth-Sun System Sciences (RPC) Subaward 191000-361604-01

Project Cost Summary /Final Invoice