Global Estimation of Canopy Water Content Susan Ustin
Global Estimation of Canopy Water Content Susan Ustin (PI), UC Davis E. Raymond Hunt (Co-PI) USDA Water Lab Vern Vanderbilt (Co-PI) NASA Ames Research Center Goals: (1) Test and Validate Retrieval of Water Content (2) Evaluate Ecological Value of Water Content Index ►Theoretical Evaluations at Leaf and Canopy Scales • Evaluate effect of cover, vegetation type, and soil background ►Empirical Evaluations • Compare to Field Data • Compare to AVIRIS EWT • Compare to VIs under Different Land Cover Conditions ►Testing Ecological Information • Plant Water Stress/Drought Indicator • Estimate LAI at High LAI sites (>4) • Agricultural Irrigation Scheduling • Fuel Moisture Estimates for Wildfire Risk Prediction • Soil Moisture (SMOS) Corrections for Vegetation 1
Field Research Sites: Wind River Ameriflux Site (mature conifer) SMEX 04 southern Arizona and Northern Mexico (semiarid) SMEX 05 agriculture, Ames, Iowa (corn, soybean) Agriculture, San Joaquin Valley, CA (cotton) Analysis of MODIS Time Series Data at Ameriflux Sites: Howland, ME Harvard Forest, MA WLEF-Tall Tower, WI Wind River, WA Central California-Western Nevada (mixed semiarid vegetation) Bondville, IL 2
Effect of Leaf Biochemistry on Leaf Reflectance Chlorophyll Structure Parameter Dry Matter 3 Y-B. Cheng, P. J. Zarco-Tejada, D. Riaño, C. Rueda, and S. L. Ustin
Variation in Soil Reflectance Soil background effect on canopy spectra simulated by (a) PROSPECT-SAILH, (b) PROSPECT-row. KUUSK, (c) PROSPECT-FLIM 4 Y-B. Cheng, P. J. Zarco-Tejada, D. Riaño, C. Rueda, and S. L. Ustin
Soil background reflectance on Simulated EWT and Canopy Water Content (b) PROSPECT-row. KUUSK (c) PROSPECT-FLIM EWT (a) PROSPECT-SAILH Cw*LAI (cm) Y-B. Cheng, P. J. Zarco-Tejada, D. Riaño, C. Rueda, and S. L. Ustin 5
Comparison of Field Measured EWT and AVIRIS at Walnut Gulch, AZ Hunt et al. Variation in EWT-AVIRIS By Vegetation Type 6 Yen-Ben Cheng, Susan L. Ustin, and David Riaño
Cross Calibration between AVIRIS and MODIS 7
Relationship between EWT-AVIRIS and MODIS Indexes at 3 sites AZCAL Properties, CA on 16 July 2002 Walnut Gulch, AZ on 25 August 2004 Howland forest, ME on 23 August 2002 8 Yen-Ben Cheng, Susan L. Ustin, and David Riaño
(a) EWT (AVIRIS) (b) NDWI (MODIS) (c) NDII (MODIS) AZCAL Properties, CA Walnut Gulch, AZ Howland Forest, ME Y-B Cheng, S. L. Ustin, and D. Riaño 9
MODIS-NDWI Time Series MODIS NDWI Index Variation with Land Cover Classes 10 Time, 2000 -2005 Palacios-Orueta et al.
Neural Net Prediction (ANN) of EWT Leaf Training Leaf Validation Training Dataset Validation Dataset LOPEX data PROSPECT Both LOPEX data PROSPECT Application Real Data MODIS PROSPECTSAILH AVIRIS 11 D. Riaño, M. A. Patricio, P. Zarco-Tejada, C. Rueda, L. Usero, S. L. Ustin
ANN trained with Real Data at Leaf Level for EWT • Trained with all LOPEX samples • Leave one out cross-validation • 420 input layers: 210 r and 210 t Riaño et al. (r 2=0. 95) r, t 420 Input Layers Hidden Layer with varying numbers of neurons EWT Output Layer 12
Analysis at canopy level • Trained with PROSPECT-SAILH: 600 random samples • Validation with PROSPECT-SAILH: 7400 samples independent of training 1. Radiative Transfer model 2. Training ANN canopy r EWT, LAI, DM N, Cab, LIDF, Soil 210 Input Layers Hidden Layer with variant number of neurons PROSPECT-SAIH canopy r 3. Validation canopy ρ EWT*LAI Output Layer EWT*LAI D. Riaño, M. A. Patricio, P. Zarco-Tejada, C. Rueda 13 L. Usero, S. L. Ustin
Analysis at Canopy Level with MODIS AVIRIS EWT • ANN trained with PROSPECT-SAILH to generate EWT*LAI • ANN run on MODIS product MOD 09 A 1 • AVIRIS EWT Used for Validation Walnut Gulch in AZ R 2 = 0. 82 AVIRIS MODIS EWT MODIS NDWI 14 NDVI, D. Riaño, M. A. NDWI, Patricio, NDW 6 P. Zarco-Tejada, C. Rueda, L. Usero, S. L. Ustin
Predicting Fuel Moisture Content for Wildfire Risk Assessment Equivalent Water Thickness (g/cm 2) Measured Dry Matter (g/cm 2) Measured EWT (g/cm 2) Estimated by PROSPECT from LOPEX Fresh Leaf Data P-value<0. 0001 Dry matter (g/cm 2) Generalized additive algorithm-partial least square regression, GA-PLS 15 Lin Li, Susan Ustin, and David Riaño
- Slides: 15