Evaluation of Covariates to Understand The Interactive Feedback Between Soil-Landscape Parameter using Satellite Images
The keysoil-landscape parameters, soil class and crop cover, are dynamic under a range of soil and climatic conditions. The identification of important and redundant covariates is a critical issue that leads to reduced size of problem, faster training, and possibly more accurate results. To understand the interactive feedback between soil-landscape parameters and changes of soil-climatic conditions, six environmental covariates are selected i.e. climate, organisms, topography, parent material, soil or soil properties, and relative position. Thirty five digital layers are prepared using different satellite data (ALOS DEM, Landsat-8, MODIS NDVI, RISAT-1, and Sentinel-1A) and climatic data (precipitation and temperature) to involve variability in all possible space. Random Forest (RF) models are tuned for optimum results and potentiality of each covariates, based on Mean Decrease Accuracy (MDA) and Mean Decrease Gini (MDG) score, is analysed and discussed to improve the understanding of the interactive feedback between selected parameters and covariates.
Kew words - Covariates, soil-landscape parameters, remote sensing, DEM, prediction modeling