Estimating Rainfall in the Philippines Using an Automated Interpretation of Forecast Images

Abstract

Flood forecasting is a process that relies on hydrologic models to predict water levels and flow rates in different basins. These hydrologic models depend on the predicted amount of water in rain clouds. A common form of this data for these models comes from color-configured forecast map images. These images are manually interpreted. However, manual interpretation is slow, tedious, and prone to error especially if there are numerous images. We propose a method to automate the interpretation of these images for a faster and more efficient means to predict the amount of water in the clouds. We identify two computational sub-problems: (1) localization and recognition of the region of interest (ROI), and (2) interpretation of the values in the ROI. We use the Speed-up Robust Features (SURF) technique to localize the ROI‟s, and a look-up table which makes use of Hue Saturation Value (HSV) color space. Experimental results show higher accuracy compared to the manual interpretation, and a significantly faster processing time.