Innovative Approach of Predicting Soil Properties in The Eastern Slopes of Mount Kenya

Abstract

The ever growing population and the need for more food make the knowledge of soil properties essential to maximise agricultural production on the currently available land. High costs of soil survey and laboratory measurements are among the major reasons for lack of sufficient soil information, which is important for proper land management. In this research, an innovative approach, comprising an optimized soil sampling design, a rapid and cost-efficient mid-infrared spectroscopy (MIR) and geostatistics was applied to provide necessary soil data for proper land management in the eastern slopes of Mt. Kenya. Conditional Latin hypercube method was used to develop a sampling design that ensured full coverage of environmental variables. Topographic variables were extracted from the digital elevation model, soil attributes from the Kenya Soil and Terrain (KENSOTER) database, and vegetation characteristics from the Normalized difference vegetation index generated from Landsat 8 imagery. The developed sampling scheme conserved the distribution of environmental covariates using box plots to validate. Soil spectra for the 232 samples collected from 77 georeferenced locations were measured using MIR spectroscopic methods at wavelength range of 4000 – 400/cm. To select the calibration samples from the MIR spectral database, principal component analysis and Kennard-Stone algorithm were used. Random forest regression was used to calibrate laboratory measurements to the soil MIR spectra. Good spectral prediction model performance was achieved as follows: soil organic carbon, total nitrogen and pH (R2 = 0.76, RMSE = 1.64; R2 = 0.81, RMSE = 0.09; R2 = 0.88 and RMSE = 0.48, respectively). Exchangeable Na, Mg, Al, K, Ca and extractable P were satisfactorily calibrated. Geostatistical analysis exhibited moderate spatial dependency of the soil properties. Soil properties were spatially predicted and mapped, and now can support targeted soil management decisions for different agricultural value chains.