Soil Impact on Satellite Based Vegetation Monitoring in Sahelian Mali

Abstract

The use of the National Oceanic and Atmospheric Administration (NOAA) satellites, and the conventional Normalised Difference Vegetation Index (NDVI) model have shown promise as a large scale monitoring tool to understand the vegetation dynamics of the sparsely vegetated Sahelian grasslands. One of the assumptions of the NDVI model is that the soil background is spectrally homogeneous, which is not the case. Twelve sites, within two Systeme Probatoire d'Observation de Terre (SPOT) satellite imageries, corresponding to NOAA Advanced Very High Resolution Radiometer (AVHRR) Local Area Coverage (LAC) pixel resolution, were assigned representative soil NDVI values for both dry and wet conditions. These soil NDVI values, together with herbaceous above-ground biomass production estimates, were used in a multiple correlation and regression analysis to assess statistically the soil impact on integrated NDVI values, i.e. values supposed only to express the total amount of vegetation in the end of the rainy season. The analysis showed that soil influence varied significantly with different soil types and moisture content, and should therefore not be ignored in satellite based vegetation monitoring. Introduction and objective Droughts in the Sahelian region, large livestock losses and the ongoing debate about land degradation in arid lands have led to calls for regional monitoring systems of pastoral resources. The NOAA AVHRR satellites have shown to be particularly appropriate for monitoring the Sahelian grassland production due to the spatial and temporal cover. The monitoring is based on the differential reflection of red and near-infrared (NIR) radiation from green vegetation. The derived vegetation index, the NDVI, is therefore an expression of the photosyntetical activity on the ground, and is defined by Tucker (1979) as: NDVI= NIRred (1) NIR+ red The m thodology to estimate the total aboveground green herbaceous biomass production at the end of the rainy season includes several steps, and is reviewed by Kammerud (1991). First, NDVI values are generated for each selected AVHRR image using the red and NIR channels. Then geometrical correction and resampling to a map projection system are carried out to remove the effects of off-nadir viewing. Cloud masking is employed using a thermal channel to eliminate clouds, being a big problem in the rainy season. Maximum value composites (MVC) are then produced from all images available within regular periods, normally ten days. Holben (1986) developed the MVC procedure which examines the NDVI value on a pixel-by-pixel basis, where only the highest value is retained for each pixel location. Finally, the MVCs are integrated, resulting in an image expressing NDVI days (integrated NDVI). In several studies, integrated NDVI has shown to correlate with the total above-ground green biomass production throughout the rainy season (Prince et al. 1990; Prince 1991; Diallo et al. 1991). In semi-arid areas, these empirical observations have shown a linear relationship between integrated NDVI values, and actual green vegetation production on the ground . The use of the NDVI for vegetation monitoring and biomass estimations initially assumed that the NDVI component from the soil substrate was constant. It is now well known that this is frequently not the case (Huete et al. 1985; Huete 1989; Escadafal et al. 1989). This bias causes problems in the estimates of the actual biomass production, because a significant portion of the radiance detected by the sensor comes from soil where vegetation is sparse. The effect of soil influence increases along a south-north gradient in the Sahel, reflected by less precipitation and vegetational cover. The purpose of this article is to examine to what extent the soil actually influences the NDVI based satellite monitoring of the Sahelian grasslands. A Geografiska Annaler 78 A (1996) 4 247 This content downloaded from 157.55.39.159 on Sun, 18 Sep 2016 05:42:54 UTC All use subject to http://about.jstor.org/terms TERJE ANDRI KAMMERUD typical Sahelian landscape in the Gourma region in Mali is chosen as a case study. Methodically, several steps have been carried out to make the statistical assessments. Field work has been obtained to measure the variations in red, NIR and derived NDVI values for different soil types in Gourma. Two successive SPOT satellite imageries have been classified to create a soil map. Information on total aboveground herbaceous biomass production, together with integrated NDVI values derived from the NOAA satellites, are collected for several sites, or control points, within the extent of the produced soil map from the 1989 rainy season. The selected sites have been regularly monitored regarding the biomass production by the International Livestock Centre for Africa (ILCA), and spatially they correspond with the LAC resolution of the AVHRR sensor used for vegetation monitoring in the Sahel. Only herbaceous above-ground biomass estimates are considered in this study, mainly because of lack of reliable tree-leaf biomass estimates. Fortunately, trees have only a minor green biomass production at the ILCA sites chosen for this study. Exactly the same sites were then extracted from the soil map, giving the sites an average soil NDVI value derived from the field radiometry. Thus, three spatial comparable variables were ready for analysis: i) soil NDVI data, ii) total herbaceous above-ground biomass data, and iii) integrated NDVI data. This data set has been used in a multivariate statistical analysis to assess the influence of soil has on the integrated NDVI that is supposed only to express the vegetational production. Spectral vegetation indices Several spectral indices have been developed to characterise vegetation canopies. These indices generally attempt to enhance the spectral contribution of green vegetation while minimizing those from soil background, solar irradiance, sun angle, senescent vegetation and atmosphere (Richardson and Wiegand 1977; Tucker 1979; Jackson 1983; Huete and Jackson 1987). Vegetation indices or greenness measures have been classified by Huete (1989) into two broad categories called the orthogonal indices and the ratio indices. The orthogonal indices are transformations that include the Perpendicular vegetation index (PVI) of Richardson and Wiegand (1977), the four band Green vegetation index (GVI) by Kauth and Thomas (1976) and the n-wavelength band GVI of Jackson (1983). These indices are distinct from the ra io indices in that isolines of equal greenness do not converge at the origin, but instead remain parallel to the principal axis of soil spectral variation, i.e. the soil line. A greenness vector, orthogonal to the soil line, is computed to maximally include green-vegetation signals while holding soil background constant. The ratio indices involve the rationing of a linear combination of the NIR and red bands by another linear set of the same bands. The existence of a soil-line concept is essential for the ratio indices in normalizing soil behaviour and discriminating vegetation spectra (Huete 1989). Most soil spectra fall on or close to the soil-line. Since the intercept of such a line is close to the origin, values of bare soils (ratios) will be nearly identical for a wide range of soil conditions. The NDVI belongs to this group, and does therefore to a certain degree reduce the impact of different soil-background spectral variations and the variations in irradiance conditions (Tucker 1979). The Soil Adjusted Vegetation Index (SAVI; Huete 1988) and the Area Additive Normalised Difference Vegetation Index (AA-NDVI; Hanan et al. 1991) have been developed to compensate for the soil background effect that still remains on the NDVI. However, for the two indices to be used operationally, information on the spatial distribution of the soil background reflectances is required. Such information is not readily available. Thu , the NDVI model is still the most commonly used vegetation index for monitoring the Sahelian grasslands (Prince et al. 1990).