Large-area soil physical degradation assessment using GIS, remote sensing and infrared spectroscopy in arid and semi-arid Kenya

Abstract

Soil physical condition controls several important soil functions such as support for biomass production, water cycling, filtering pollutants, and land surface energy balance. However, physical degradation undermines this ability. Currently, there is lack of rapid and repeatable methods that can facilitate timely large-area assessment for effective monitoring and control of soil degradation. This study tested the combined applications of point-measurements of physical properties, soil diffuse spectral reflectance (DSR), and remote sensing to spatially assess the degradation in a large watershed (4500 km2) in semi-arid areas in eastern Kenya. Indicators of the degradation were determined from 540 point-measurements of infiltration and water retention and field observations of the visible signs of soil physical degradation. The physical properties included steady-state infiltration rates, sorptivity, water-holding capacity, pore distribution index, bulk density, and air-entry potential. The parameters describing these properties were derived using a nonlinear mixed effects (NLME) approach, which was also used to test for the effects of other covariates such as land use and geographic features. A screening protocol was then developed that took evidence of degradation from visible assessments in the field, estimated soil physical properties, and rapid soil tests based on soil DSR to predict the degradation cases. Over 90% sensitivity and specificity was achieved with a mixed effect logistic model based on a onethird holdout sample. The screening results showed that soil DSR was a powerful tool for detecting early warning indicators of degradation that were not readily discernable from field observations. In addition to the point-estimates of likelihood of physical degradation, timeintegrated remote sensing indicators were also tested for power of spatial prediction of the trends of the degradation in the study area. The standardized deviations of land surface temperature (LST) and Normalized Difference Vegetation Index (NDVI) from time-series Landsat scenes were used to study the thermal and vegetation conditions of the degradation at sampled points. These indices effectively predicted the likelihood of the degradation of the held-out samples with 80% accuracy of ground reference data and were used to map the degradation in the whole study area. The approach developed in this study showed promising opportunity for spatial prediction of physical degra~ation at high spatial resolution over large areas and could be a useful tool for guiding policy decisions on sustainable land management especially in the tropics where land use policies lack scientific support.