Object Processing of High-Resolution Precipitation Forecasts Based on Intensity and Texture Segmentation

High-resolution convective-scale models are able to forecast realistic precipitating events, however, they are still affected by position, amplitude and timing errors. In order to improve the use of these forecasts, a relevant method consists of extracting the predictable signal in the form of «precipitating objects», which define regions where the distribution of the precipitation is homogeneous, in terms of intensity and/or spatial variability (also known as texture). This approach mimics, to some extent, the human processing of forecasts. Nonetheless, automatically detecting and characterizing objects for different types of precipitationd is still a challenge. We will first present a method for identifying intensity-based objects, afterwards we will focus on texture recognition. In this work two textures are considered: «continuous» and «intermittent» precipitations. We will show that the combination of Gabor filters and machine learning methods is able to properly discriminate between the two textures.