Submitted to: International Geoscience and Remote Sensing Symposium Proceedings
Publication Type: Abstract Only
Publication Acceptance Date: April 26, 2006
Publication Date: August 2, 2006
Citation: Hunt, E.R., Yilmaz, M., Jackson, T.J. 2006. Scaling vegetation water content from thermatic mapper to MODIS during SMEX04 [abstract]. International Geoscience and Remote Sensing Symposium Proceedings. 2006 CDROM. Technical Abstract: Vegetation Water Content (VWC) is important for accurate retrievals of soil moisture using passive and active microwave sensors and may be important for estimating water stress and vegetation dryness for fire potential. A significant component of vegetation water content is located in the foliage, and can be estimated using bands at shortwave infrared wavelengths. The MODerate resolution Imaging Spectroradiometer (MODIS) and future operational sensors have bands in the shortwave infrared region and thus can be used for monitoring vegetation water content. Because MODIS pixels are large, it is difficult to adequately sample vegetation for direct validation of VWC; the objective of this study is to use 30-m Landsat Thematic Mapper data to scale VWC to MODIS pixels. The Soil Moisture Experiment 2004 (SMEX04) was conducted during summer-monsoon season in Arizona, USA, and Sonora, Mexico, as part of the North American Monsoon Experiment (NAME). About half of the 180-mm annual precipitation falls in July and August during the monsoon. The Arizona study area is located southeast of Tucson and is centered on the USDA-ARS Walnut Gulch Experimental Watershed. The vegetation at the Arizona study area is comprised of riparian and high-elevation oak woodlands, irrigated agriculture, desert scrub, and grasslands. The Sonora study area is located northeast of Hermosillo and has the former vegetation types as well as subtropical shrublands. Landsat 5 Thematic Mapper (TM) data were acquired on June 11, July 29 and August 30, 2004 for both the Arizona and Sonora study areas. The TM imagery were calibrated to reflectances with sun photometer, ozone and radiosonde data using the MODTRAN program. The six images were further georeferenced. NASA Terra MODIS data were acquired for the same dates as the TM data; the MODIS scenes were 500-m daily surface reflectance level 2G for bands 1-7 (product MOD09GHK). The Normalized Difference Infrared Index [NDII = (R850 - R1650)/(R850 + R1650), which are TM bands 4 and 5, or MODIS bands 2 and 6, respectively] was calculated for comparison to field data. Field sites (40 m x 40 m) were selected to be representative of various vegetation communities, so only a few sites were coincidental with SMEX04 soil-moisture validation sites in Arizona. All of the sites in Sonora were coincidental with soil-moisture validation sites. Field data were collected from July 29 to August 8, 2004. Leaf area index (LAI) was measured using either canopy hemispherical photographs or the LiCor LAI-2000. Leaf samples were collected and the fresh weights, dry weights, and leaf areas were measured. Vegetation water content (kg/m2) is calculated in the field by dividing the difference in fresh and dry weights by the ground area; this biophysical quantity is also known as the equivalent water thickness (EWT), where an EWT of 1 mm equals a water content of 1 kg/m2. Canopy vegetation water content was estimated from the product of LAI and leaf water content. The canopy reflectance for each site in Arizona was characterized using a Cropscan MSR. Site by site, TM NDII showed considerable scatter compared to VWC with an R2 of 0.33. However, when the sites were averaged by land cover class, NDII was linear with canopy VWC with an R2 of 0.89, showing the variability was among the sites and not between land cover classes. Canopy NDII was generally linear with TM NDII, except for six sites which had very high field NDII and very low TM NDII. Comparison to precipitation patterns from Doppler radar indicates these six sites had the field reflectances acquired soon after rainfall, so soil wetness has a strong effect on retrieval of VWC from NDII. The TM data were aggregated and compared to MODIS NDII. The linear relationship between canopy NDII and VWC was applied to both MODIS and TM images for each date. The differences in VWC followed a 1:1 line, showing VWC does scale from high-resolution pixels to moderate resolution pixels. There were some adjacent, large positive and negative residuals indicating some misregistration between the MODIS and TM data. Therefore, retrieval of VWC from MODIS imagery depends on the relationship between NDII or other indices and VWC. High-spatial-resolution satellites do not have a band equivalent to MODIS band 5 (1240 nm), which forms the basis of the Normalized Difference Water Index [NDWI = (R850-R1240)/(R850 + R1240)], thus we could not test if NDWI from field level to MODIS scaled in a similar manner as NDII, or was affected as strongly by background soil moisture content. Canopy reflectance simulations using the SAIL model were conducted to assess the responses of various indices to wet and dry soil backgrounds. Both NDWI and NDII were very sensitive to background soil wetness as well as variation in background brightness. On the other hand, estimation of VWC from hyperspectral data were very sensitive to soil wetness, but not sensitive to background brightness. Because any retrieval of VWC from the shortwave infrared wavelengths will be affected strongly by soil wetness, combinations of sensors or indices may be required to estimate VWC.