MULTIVARIATE MODELS FOR PREDICTING RAINFALL EROSIVITY FROM ANNUAL RAINFALL AND GEOGRAPHIVAL COORDIATES IN A REGION WITH A NON- UNIFORM PLUVIAL REGIME

Abstract

Soil erosion by water is a major land degradation problem because it threatens the farmer’s livelihood and ecosystem's integrity. Rainfall erosivity is one of the major controlling factors inducing this process. One obstacle of estimating the R-factor is the lack of detailed rainfall intensity data worldwide. To overcome the problem of data scarceness for individual analysis of storm events for developing the country with a non-uniform pluvial regime like the upper part of Iraq, multivariate models were derived for estimating annual rainfall erosivity. They were based on annual rainfall data and geographical coordinates of a group of meteorological stations distributed over the study area. A host of statistical indices were selected to evaluate adequately the model's performance. Further, the models were cross-validated using k-fold procedure and unseen data. Subsequently, four linear models were proposed for estimating the annual erosivity for the study area. Good correspondence was found between the measured and predicted values. Among the proposed models, the model with the combination of annual rainfall, latitude and longitude outperformed the remaining proposed ones. After calculating the annual, the ArcMap software ver. 10.2 was applied to map the spatial variability of the R-factor over the study region.

Keywords

• rainfall erosivity index, multivariate models, model calibration and validation, Spatial variability.