Estimating soil properties in patched vegetation landscapes using remote sensing

Abstract

Soil maps are important tools for farm planning and soil conservation. This study explored the potential of satellite data in express mapping of key seven soil properties (pH, SOM, Tot. N, Available P., K-exch, Clay and Sand composition), that define soil quality. The assessed the impact of vegetation cover on soil property modeling with objectives to, identify suitable land cover type, predicting and characterizing the spatial distribution of soil properties with remotely sensed data. This data was grouped into four categories of predictors (Landsat 8, Sentinel-1A, Landsat 8 +Sentinel-1A, Landsat 8+ Sentinel1+ DEM and other selected variables) to form four distinct models i.e., modA, modB, modC and modD respectively, applied to each soil property using Partial Least Squares regression within bare soils, vegetation cover and mixed land cover. The results show that; SOM, Tot. N and Avail. P is better predicted from bare soils, whereas soil pH and K-exch from soils under vegetation cover. Sand and Clay predictions were independent of Landcover type. K-exch was best predicted by modB, pH, SOM, and Tot. N best estimated by modC, whereas avail. P, Sand and Clay are best estimated by modD. The refined models achieved performances of; R2(c)=68%, R2(v)=57%, RPD/RPIQ=1.8 for soil pH; R2(c)=74%, R2(v)=64%, RPD/RPIQ=2.6 for SOM; R2(c)=60%, R2(v)=75%, RPD/RPIQ=1.9 for Tot. N; R2(c)=54%, R2(v)=75%, RPD/RPIQ=2.1 for available P.; R2(c)=49%, R2(v)=63%, RPD/RPIQ=2.0 for K-exch; R2(c)=63%, R2(v)=71%, RPD/RPIQ=1.7 for Sand and R2(c)=73%, R2(v)=65%, RPD/RPIQ=1.9 for Clay. The rapid predictive mapping conducted by these models on the whole indicted that the spatial distribution characteristics of soil pH, SOM, Tot. N, avail P. and K-exch are tending below the critical levels, a limiting factor to sustainable agriculture, requiring urgent interventions. In conclusion, except soil texture, the rest of soil properties investigated are sensitive to biophysical state of Landscapes, as result a new approach called partial landscape modeling method has been proposed, for a more accurate spatial variability estimate of soil properties.