Cyclone intensity estimation using similarity of satellite IR images based on histogram matching approach

In the present work, a new approach, inspired by data-mining is presented to estimate the intensity of the tropical cyclones (TCs) by comparing the convective features in terms of histogram of the brightness temperature (BT) values within the infrared (IR) imageries derived from the geostationary satellites. A database representing the images (IR images from GOES-8 and 12) of diverse intensity of tropical cyclones has been formed using HURSAT dataset for the period 2000–2005. Radial and angular histograms of BT values of these IR images were computed. The histograms of the images were matched and pair of the best matched images was formed based on the matching index. The best matched pair of images is assumed to have similar intensity. A correlation of 0.84 was found between the intensities of the cyclone in the best matched pair of images. The mean absolute error (MAE) and root mean square error (RMSE) for the present approach were estimated as 11.87 kt and 15.47 kt, respectively. The skill of the present technique was further evaluated by classifying the cyclone intensities into four categories, viz., tropical depression (TD), tropical storm (TS), hurricane (H), and major hurricane (MH). The MAEs of the intensity values between the best matched images of each group (TD, TS, H and MH) were found as 11.84, 8.81, 14.51 and 19.60 kt, respectively. As a case study, independent tests were performed using the HURSAT data for year 2006 (339 IR images of nine cyclones) to test the ability of the technique to estimate tropical cyclone intensity. MAE and RMSE for this independent validation set were found as 11.18 and 14.48 kt, respectively. This indicates the potential of present discussed method to estimate the intensity of cyclones based on image histogram matching approach using geostationary satellite IR imageries.