NASA-EOS Report 1995

Soil-Vegetation-Atmosphere Transfer (SVAT) Modeling/Remote Sensing

Land surface characteristics such as temperature, vegetation and soil moisture and their interactions are of critical importance to any model which describes hydrologic and climatological processes. The remote sensing instruments to be carried on the EOS platforms and related space craft will return information about the temporal and spatial variation of many such variables (e.g., surface radiant temperature and fractional vegetation cover) not easily obtainable in useful form from ground-based observations. Maximum information is gained from satellite observations with an optimum combination of different channel measurements. We use a multispectral approach, i.e., measurements in the thermal infrared (IR), near-IR, and visible portions of the spectrum to obtain information about the surface wetness and the amount of vegetation.

Surface radiant temperature () is calculated from thermal IR radiances corrected for atmospheric attenuation (water vapor, haze, carbon dioxide) and a standard vegetation index, the Normalized Difference Vegetation Index (NDVI). The latter is derived from visible and near-IR measurements. These are subsequently used to derive the surface wetness (a variable termed the "surface moisture availability", denoted as ), surface turbulent energy fluxes (particularly evapotranspiration [ET]) and fractional vegetation cover (Fr)-(Gillies and Carlson, 1995a) referred to as land surface parameters. This is done by a method of inversion which involves the application of a SVAT model to interpret (within theoretical limits) the remote observations which are described in two-dimensional space (To versus NDVI). The technique is entirely new but is a consequence of the initial observations of Price (1990) and is referred to as the "triangle method".

Figures 10 and 11 are examples showing land surface parameters derived as a function of the remote observations. An extension of the method is to transform the remote observations ( and NDVI) into a single framework in which new coordinates are defined (a scaled surface radiant temperature [] and a scaled NDVI [N*] which are essentially invariant with respect to the changes in ambient conditions (meteorological or otherwise). Such a transformation circumvents many of the cumbersome intermediate steps (particularly the SVAT) in the method such that the land surface parameters are computed more-or-less instantaneously. Such an expeditious provision of data is central to requirements of the project as set out schematically in Figure 3.

A joint research effort with colleagues at the USDA-Agricultural Research Service (ARS) (Drs. W. Kustas and K. Humes) using in-situ measurements from the FIFE and Monsoon 90 (Walnut Gulch) field experiments has yielded an excellent validation of the "triangle method", as shown in Figure 12. An important implication of these validations is that more than 90% of the variance in the surface ET can be explained by spatial variations in Fr and Mo (Gillies et al., 1995a).

Implications of the Verification Studies

That the surface energy fluxes are dependent primarily on Fr and

implies that the surface turbulent energy fluxes are quite insensitive to root zone SW. Alternately stated, surface radiant temperatures yield little information concerning the SW over a deep layer. This rather surprising result is consistent with the lack of correlation found by Carlson et al. (1995b) and Wuthrich (1994) between microwave measurements (active or passive) of SW and those derived from surface radiant temperatures. The reason for this poor correlation is due to the fact that the thermal method determines the SW over a shallow (few centimeters or less) soil surface layer not shaded by plants, whereas the microwave integrates the SW over several centimeters depth for a surface area that includes a portion of the surface covered by vegetation. The significance of this finding impacts greatly on the linkage between the remote sensing and hydrology. It suggests that

is not useful for hydrologic applications except insofar as the surface boundary affects ET.

Deforestation and Urbanization

The "triangle method" may be used to investigate land cover change, either due to deforestation and/or urbanization. The basic idea is that the land surface parameters

and Fr are intrinsic surface quantities that are by definition inherent descriptors of the surface energy balance. When plot ted in scaled coordinates ( and N*), the To and NDVI variations can be easily transformed to

and Fr, whose evolution in time can be monitored. Moreover, both parameters are required for initialization of the land surface components in regional- and global-scale climate models. As such they can be used to predict changes in regional climate that occur as the result of urbanization and deforestation. The important proviso here is that one understands how the descriptors of the land surface (

and Fr) change as the result of alterations in the land cover. The "triangle method" will provide such a quantitative framework. This idea is contained in Figure 13.

Four target areas are to be investigated, three of which have already been selected. One of these areas has been undergoing deforestation and two have experienced rapid urbanization. Ten years of record for NOAA AVHRR (1-km resolution) images will be examined for all target areas, and the migration of pixels within the triangle will be charted. Supplementary measurements of percent urbanization will be made using Landsat TM measurements. As a start, State College, PA, serves as an excellent training area in which changes in land cover and its attendant urbanization have been well documented and are familiar to the investigators. Figure 14 illustrates the distribution of pixels plotted in the universal and N* space.

A second target area is Chester County, PA, which is an example of a medium-sized urban area under going rapid development. Chester County is also the site of a related project being conducted by Charles Dow, a Ph.D. student in the School of Forest Resources at Penn State with whom we are collaborating. Dow and DeWalle (1995) are studying the relationship between urbanization and river basin runoff and they have been able to relate percent urbanization and population density to river basin runoff and monthly ET for a number of urbanized river basins in eastern Pennsylvania. Previously, Gillies and Carlson (1995b) showed some correspondence between percentage urbanization and . The outcome of this study will be a quantitative link between urbanization, runoff, ET and population growth using a combination of satellite methods and demographic data.

Costa Rica constitutes a primary target area for the investigation of tropical deforestation. AVHRR and Landsat TM scenes are being collected for clear-sky conditions over a ten-year period during the dry season (December - March). Two or three specific subareas within the country are being considered for this investigation. Among them is the Premontane region, a forested plateau near the Pacific Ocean. We anticipate collaborating in this project with Drs. Carlos Quesada and Arturo Sanchez of the Center for Research on Sustainable Development in Costa Rica. We anticipate that Dr. Sanchez will spend a year at Penn State helping us implement the "triangle method" for the study of deforestation in Costa Rica.

A New Look at the Simplified Method

A modification of the so-called "Simplified Method" (Seguin, et al., 1994) used to obtain the integrated daily ET and To from over variable vegetation cover is proposed (Carlson et al., 1995a). Math ematically, the simplified equation takes the form

where

and

are, respectively, the integrated net radiation and evapotranspiration over a 24-hour period and

and

are, respectively, the surface radiant and the 50 m air temperatures at 1300 local time. B and n are pseudo constants given as functions of NDVI, expressed as the scaled N*. Both N* and

are obtained with the aid of remotely determined measurements which are viewed on scatterplots of

versus NDVI. The value of this equation is that it accounts for considerable variation in the constants B and n (Seguin, et al., 1994) which is due to variable vegetation fraction as well as wind speed and surface roughness.

Graphics and HTML text markup by Phil Kolb
Comments to: R. White, raw@essc.psu.edu