Rain-rate classification of INSAT—VHRR images through statistical methods

Abstract

This study describes a procedure to determine rain rates using INSAT visible and infrared data by means of statistical methods. Satellite rain estimation techniques are more sensitive to spatial degradation than to temporal degradation. In the present study, visible and infrared samples selected from the VHRR images are used in the training process to achieve good spatial correspondence between satellite and ground estimates. Brightness features which are providing the best separability among the selected rainrate categories are extracted from the images. By locating all the 560 rain gauge stations maintained by the India Meteorological Department (IMD), the maximum likelihood decision rule was employed to classify the 8 km × 8 km window centered at a rain gauge station into one of the four selected rainfall rate categories. The training process used the rainfall data obtained from IMD daily weather reports for verification. The INSAT data for sample dates, representing different seasons has been used for the study. Classification accuracy, defined as the percentage of samples correctly classified out of all the samples in the selected class, was found to be above 80%.