Abstract
In this article, a new approach called Multi-Index Rain Detection (MIRD) is suggested for regional rain area detection and was tested for India using Kalpana-1 satellite data. The approach was developed based on the following hypothesis: better results should be obtained for combined indices than an individual index. Different combinations (scenarios) were developed by combining six commonly used rain detection indices using AND and OR logical connectives. For the study region, an optimal rain area detection scenario and optimal threshold values of the indices were found through a statistical multi-decision-making technique called the Technique for Order Preference by Similarity Ideal Solution (TOPSIS). The TOPSIS analysis was carried out based on independent categorical statistics like probability of detection, probability of no detection, and Heidke skill score. It is noteworthy that for the first time in literature, an attempt has been made (through sensitivity analysis) to understand the influence of the proportion of rain/no-rain pixels in the calibration/validation dataset on a few commonly used statistics. Thus, the obtained results have been used to identify the above-mentioned independent categorical statistics. Based on the results obtained and the validation carried out with different independent datasets, scenario 8 (TIRt < 260 K and TIRt − WVt < 19 K, where TIRt and WVt are the brightness temperatures from thermal IR and water vapor, respectively) is found to be an optimal rain detection index. The obtained results also indicate that the texture-based indices [standard deviation and mean of 5 × 5 pixels at time t (mean5)] did not perform well, perhaps because of the coarse resolution of Kalpana-1 data. It is also to be noted that scenario 8 performs much better than the Roca method used in the Indian National Satellite (INSAT) Multispectral Rainfall Algorithm (IMSRA) developed for India.
