Rain identification and measurement using Oceansat-II scatterometer observations

We exhibit a technique to detect and estimate precipitation over the global oceans using the radar back scattering coefficient and brightness temperature measurements from Oceansat-II scatterometer along with numerical weather prediction model derived rain sensitive parameters via a neural network (NN) based setup. Rain/no-rain labels are generated by analyzing rainfall observations from Tropical Rainfall Measuring Mission (TRMM) and Advanced Microwave Scanning Radiometer for Earth Observation Satellite (EOS) (AMSR-E) which are concurrent (within a spatiotemporal bin 0.25 degrees x 0.25 degrees latitude-longitude and 900 s) to Oceansat-II overpasses. The rain sensitivity of all the parameters is examined. NN is applied in two stages: (1) rain identification and (2) rain quantification with training samples from five different overlapping geographical regions [I(25 degrees N-25 degrees S), II(15 degrees N-45 degrees N), III(35 degrees N-70 degrees N), IV(15 degrees S-45 degrees S) and V(35 degrees S-70 degrees S)]. Rain identification accuracy is about 93%, 87%, 90%, 79%, and 85%, and no-rain detection accuracy of about 97%, 87%, 86%, 84% and 86% for these regions. The missing rain cases are few compared to the size of norain samples and are largely from the low rain regime. The RMS error of instantaneous rain estimation for regions I to V (rain rates varying from >0 to approximately 45, 25, 25, 45, and 20 mm h(-1)) is found to be 1.86, 0.69, 0.47, 0.56, and 0.46 mm h(-1), respectively. The qualitative comparisons of instantaneous, 3-days, monthly and seasonal rain rates from scatterometer and AMSR-E demonstrate a-good agreement between them. Probability distribution of monthly rain rates from scatterometer and AMSR-E is also compared, indicating the consistency of scatterometer derived rain with AMSR-E rain in a climatic scale. (C) 2013 Elsevier Inc. All rights reserved.