Estimating changes in vegetation cover over time in arid rangelands using landsat MSS data

Abstract

Changes in vegetation cover over time in arid rangelands can be used to monitor land condition and to identify processes of land degradation. This article describes how a cover index can be derived from Landsat MSS data and how the index can be standardized to remove atmospheric effects and differences in sensor calibration between spacecraft. Tests carried out on 1-m-diameter radiometer targets show that the best separation between soil, rock, or stone and vegetated surfaces occurs in the Band 4–Band 5 data space. This separation includes both dry and green vegetation whereas in the commonly used Band 5–Band 7 data space the separation is much less clear if the vegetation is not green. We have therefore developed a vegetation cover index in the Band 4–Band 5 data space. This index, known as PD54, has a similar form to the perpendicular vegetation index of Richardson and Weigand (1977). It is calculated by identifying a soil line or, more correctly, an upper soil band limit, and then determining the perpendicular distance of each pixel from that line. The perpendicular distance is scaled using distances calculated for points or pixels which have 100% cover. In the normal mix of rangeland vegetation at the pixel scale, these reference pixels occupy a very limited area of the Band 4–Band 5 data space and can be treated as a point or as a line of limited length and with a slope similar to that of the soil line. PD54 performs better than several other indices when tested at the Landsat MSS pixel scale using aircraft-mounted radiometer data. It is also less subject to systematic errors when shifting from one vegetation type to another. The principal difficulty in using PD54 as a cover index arises because there is no single soil line. Instead, the position of that line varies with both Munsell soil hue and chroma. Attempts to standardize PD54 for Landsats 1, 2, 4, and 5 showed that the commonly used approach of converting data to exoatmospheric reflectance based on published radiometer gains and offsets does not work. Furthermore, the dark pixel subtraction method of haze correction does not remove all atmospheric effects. It is, however, possible to remove the effects of differences between the different Landsat MSSs and inadequate haze correction by identifying the extremes of the Band 4–Band 5 data space on a particular scene and using them to scale PD54. Once this is done, PD54 gives substantially better estimates of cover from MSS data than Band 5 values which have supposedly been radiometrically standardised.