Tropical cyclone intensity estimation using temporal and image analysis of satellite data

Tropical cyclones (TCs) are becoming an increasing threat to life and property. Developing an automated technique to estimate TC intensity and to overcome the existing errors in estimation is still a challenge. The Dvorak technique (DT) is the state-of-the-art method that has been used over three decades for estimating the intensity of a tropical cyclone. The DT subjectively estimates TC intensity based on visible and infrared satellite images. In spite of wide usage of the DT for TC analysis, it has some limitations. The most important one is that the DT does not use the valuable historical data mainly because of the challenges on computing and human resources. This research is inspired by the availability of historical TC satellite data. We hypothesize that discovering unknown regularities and abnormalities that may exist in the large group of past observations could help human experts interpret TC intensity changes from various points of view. Our goal is to provide a data mining tool that increases the ability of human experts to analyze huge amount of historical data for TC intensity estimation. The proposed intensity estimation algorithm has two parts: temporal constraints and image analysis. Temporal information provides a priori estimates of storm intensity (in terms of wind speed) prior to using any satellite image analysis. Hurricane Satellite data (HURSAT-B1) includes best-track intensity are used as a training data. A case study using North Atlantic Hurricane Satellite data from 1988-2009 is considered. The temporal analysis uses the age of the cyclone, 6, 12 and 24 hours prior intensities as predictors of the expected intensity. The 10 closest analogs (determined by a K-nearest-neighbor algorithm) are averaged to estimate the intensity. The distribution of intensity estimation errors of the proposed technique shows that 50% of the estimates have a mean absolute error less than 4.4 knots, 75% are 6.3 knots and 90% are within 8 knots. Several validation tests were conducted to statistically justify the proposed algorithm using K-Fold Cross-Validation. The resulting average root mean squared error (RMSE) of our algorithm is approximately 4.6 knots. Overall, 47% improvement has been achieved compared to the DT. The current analysis has the ability to decrease the DT noise and has the potential to provide new temporal constraints on DT. The image analysis part of the proposed technique used the average and standard deviation of the brightness temperature of the selected rings around the center of the storm, as predictors of the current intensity of the storm. The image analysis used the age of the cyclone, current, 6, 12 and 24 hours prior images as predictors of the expected intensity as well. As like as temporal analysis, the intensity of the 10 closest analogs (determined by a K-nearest-neighbor algorithm) in training data are averaged to estimate the intensity. The result of K-Fold Cross-Validation shows that the accuracy of the proposed technique on likely par with current objective techniques. Simplicity aspect of the image analysis part of the planned technique makes it superior to other techniques. Research is continued to combine the image and temporal analysis part of suggested technique to achieve more accuracy in tropical cyclone intensity estimation.