Digital Soil Nutrients Mapping based on Available Legacy Maps and Point Data in Rwanda

Abstract

Soil is a natural three-dimensional body, which varies in space and time. Expanding environmental concern has increased the interest for regional land use analysis and in Rwanda, the rapid growth of population has testified the effect of soil/land degradation through overexploitation of arable lands. This has been explicitly accompanied by decline in soil productivity associated with depletion of soil nutrients and inefficient fertilizers application. One of the fundamental principles of sustainable production is improving soil fertility and acquiring knowledge of nutrients status in soil. Digital soil mapping is a state-of-the-art method used to model the relationship between measured soil properties and soil forming factors known as environmental covariates to predict soil properties at unsampled locations. This study has focused mainly on spatial analysis and interpolation of soil properties to establish their spatial distribution and the importance of soil forming factors and landscape parameters on patterns of variation of soil properties. The obtained results were presented in the form of maps that decently depict the areas with either adequacy or deficiency in soil nutrients availability for plant growth. Five soil properties (pH, as well as P, Al), Ca and Cu) were predicted by means of comparative mapping with two different interpolation methods namely; Regression Kriging (RK) and Random Forest Kriging (RFK). The accuracy of both methods was assessed by various validation indices such as; the mean absolute estimation error (MAEE), the root mean square error (RMSE) and the coefficient of determination (R2). The best method for predicting soil pH and Cu was the RK and the obtained R2 were 0.39 and 0.18 for pH and Cu respectively. Soil pH got a RMSE of 0.65 and a MAEE of 0.51, while the obtained RMSE and MAEE for Cu were 1.68 ppm and 0.96 ppm respectively. The RF model has proven successful in predicting available P, Al and Ca. P had a R2 of 0.04, a RMSE of 70.42 ppm, and a MAEE of 29.33 ppm. For Al, the obtained R2 was 0.37, the RMSE was 344.26 ppm and the MAEE was 269.11 ppm. Lastly, Ca scored a R2 of 0.39, a RMSE of 706.88 ppm and a MAEE of 485.67 ppm. The use of produced pH map as a covariate revealed a significant quality improvement on other soil properties where the R2 shifted from 4% to 10% for P, 37% to 46% for Ca and 18 to 25% for Cu. Furthermore, these obtained results were compared with the results of the mapping exercises conducted by ISRICWorld Soil Information and the virtual fertilizer research center (VFRC). The two institutions have produced relatively good maps compared to the output maps generated in this study. In fact, the used data and methods have provided little information based on the obtained results, however, this gives aspiration that more accurate digital soil maps (DSMs) will be obtained eventually if much more qualitative soil profile analytical data with harmonized methods of measurement and more detailed covariates specific to the area of interest are used.