Determination of annual rainfall in north-east India using deterministic, geospatial, and machine learning techniques

Abstract

Abstract Analysis of extreme annual rainfall in the six north-east Indian states of Assam, Meghalaya, Nagaland, Manipur, Mizoram, and Tripura using the deterministic interpolation technique of inverse distance weighting (IDW), the geospatial interpolation technique of Ordinary Kriging (OK) and the machine learning prediction technique of generalised additive model (GAM). GAM is used only for prediction and hence the results are then subsequently interpolated by OK to create the rainfall maps. The datasets considered for this study are a training dataset of 171 points which consisted of satellite rainfall and a testing dataset with ground rain gauge data of 33 points which was used for validation of the former. A combined dataset of training + testing was also interpolated and mapped to compare for visual accuracy of each technique. It was seen that OK was a superior and a much more realistic interpolation technique than IDW, since it took the altitude of each site into consideration along with latitude and longitude, unlike IDW, which only interpolated over the x–y plane and didn't rely on altitude. When the predictions of the training dataset through GAM were mapped using OK, it showed almost parallel contours, which is undesirable for natural phenomenon like rain.