Classification of Hydrometeors from Microwave Satellite Data Using an Artificial Neural Networks Method

Hydrometeors in the atmosphere, on any form (solid, liquid and gases), interact with microwave radiation (through scattering, absorption and emission). The Advanced Microwave Sounding Unit-B (AMSU-B) measurements onboard NOAA satellites are sensitive to the types, shapes, and size distributions as well as fall behaviors of the hydrometeors in the AMSU-B resolution Volume and thus are useful to study different types of atmospheric hydrometeors. These microphysical signatures and classification of atmospheric hydrometeors can be utilized to initialize the cloud/mesoscale numerical weather prediction models, study of precipitation formation and life cycle, and choice of the right algorithm for precipitation estimation. Therefore, In this paper, the signatures of eight types of hydrometeors, including Thunderstorms (TS), Heavy rain (HR), Light rain (LR), Moderate Rainfall (MR), Snowfall (SF), Snow cover (SC), Cloudy condition (CC), and Clear sky (CS), using AMSU-B data by an artificial neural network method, simultaneously, have been classified to eight different classes. During the study period (2000 to 2010), from about 200 of satellite passes, for each type of hydrometeor 200 data-sample and overall 1600 data-sample, which was closest to Iran Meteorology Organization (IMO) reports have been collected. Our results show that different classes of rain, including light, moderate and heavy rainfall, with respect to other classes, with accuracies between 54 to 62% have poor classification capability, and other hydrometeors with an accuracy of about 80% correctly classified. By considering three classes of rain as a single class (rain fall = RF), the accuracy of neural network classifier increased to 85%; among 400 pattern, about 340 pattern have correctly been classified.