Section II

Relationship between NDVI, T4sfc, and land cover types

Changes in land cover can have a profound effect on NDVI and radiant surface temperature (T4sfc). Using satellite remotely sensed data over a region of varying land cover, Price (1990) found that areas with sparse vegetation cover tended to have a wider variation in T4sfc than densely vegetated locations. This variability was related to the amount of water in the soil within a few centimeters of the surface. Urban surfaces, which typically lack significant vegetation cover (i.e, low values of NDVI), do not ordinarily experience such variability in T4sfc because of the dry nature of non-evapotransipirative urban materials. Variability in T4sfc in forested areas is also restricted, but for the opposite reason. Forests are densely vegetated (with high values of NDVI), restricting the exposure of bare soil. As the dominant contributor to the radiant surface temperature, the evapotranspirative nature of the vegetation acts to regulate values of T4sfc at a level close to that of the ambient air. Between these two extremes of land cover, land in agricultural use tends to have a mix of short vegetation and exposed bare soil. In this intermediate land cover class, the influence of surface soil water content and vegetation contribute to a broad variation in both NDVI and T4sfc values.

In examining the relationship between land cover and the surface energy response (as measured by NDVI and T4sfc), it is necessary to obtain high-resolution land cover information for each 1 km AVHRR pixel. Using Lands at MSS imagery (resampled from 30 to 25 meters and georegistered to the 1km AVHRR data), such land cover information can be obtained.

In Figure 2.1, a Landsat MSS image of the DFW region is coded by major land cover class using the maximum likelihood supervised classification scheme of Anderson et al. (1976). This classification is accomplished by submitting manually identified samples of each land cover class to an image processor, which then codes the entire image based on the spectral characteristics of each sample.

2.P1. Examine the classified Landsat MSS image of the DFW region for 8 October 1992(Figure 2.1). The following colors correspond to land cover classes: Red = Urban, Dark Green = Forested, Light Green = Agricultural, Blue = Water. The location of stations 282 and 285 are given, as are the locations of the major downtown urban centers (See implementation note 2.P.1) .

Figure 2.1. Classified Landsat MSS image of the DFW region for 8 October 1992.

The "subpixel" land cover information provided by the classified Landsat MSS image for each corresponding 1 km AVHRR pixel makes an examination of the relationship between major land cover classes and NDVI and T4sfc possible. In Figure 2.2, the locations of urban, agricultural and forested land cover classes in a scatterplot of T4sfc/NDVI and NDVI are given. The values of NDVI and T4sfc are composite values obtained from the 1-10 June 1992 global 1km AVHRR data set (http://lpdaac.usgs.gov/1KM/1kmhomepage.asp), and coded according to the dominant land cover class for the respective AVHRR pixel. This figure confirms the previously mentioned physical relationship between land cover type and suface energy response in T4sfc and NDVI. It is important to recognize that specific T4sfc and NDVI values vary according to solar illumination, state of vegetation and atmospheric influences on the remotely-sensed MSS data set.

Figure 2.2. T4sfc/NDVI vs. NDVI (derived from the 1-10 June 1992 global 1 km AVHRR data) coded according to land cover classification (derived from the 8 October 1992 Landsat MSS scene).

Using the classified Landsat MSS image, the percentage of urban land cover for each overlaying 1km AVHRR pixel was calculated. These land cover statistics are provided for a 3 X 3 1 km pixel domain around stations 282 and 285 in Table 2.1.

Table 2.1. Percentage of urban land cover within 1 km AVHRR pixels based on MSS analysis (for a 3 x 3 pixel window centered on stations 282 and 285).

2.Q1. Calculate the average percentage urbanization for each station from values in Table 2.1. How does the percentage urbanization for the stations relate to their T4sfc/NDVI and NDVI values shown in Figures 1.7 and 1.8?

2.Q2. Compare the relationship between land cover classes suggested and NDVI and T4sfc suggested by Figure 2.1 with Figures 1.7 and 1. 8. What land cover class would you assign to each station? Is there a difference between station 282 and the other five stations (excluding station 285)?

2.Q3. Do you think that it would be more difficult to discern urban land cover in an arid region where dry, bare soil is dominant (compared to either an agricultural or forested region)? Explain.