Summary and Conclusions

Abstract

False alarm is one of the shortcomings of satellite precipitation estimates that needs to be improved. Many studies have quantified the FAR, bias and errors of satellite precipitation estimates. However, reducing the FAR is an essential step in improving the quality of satellite data. In this research, three techniques are proposed to reduce the FAR by integrating information from multi-spectral satellite imagery as well as satellite radar observations. MODIS, a multi-spectral satellite sensor, observes the atmosphere in 36 spectral channels, providing a special source of information for cloud observation. On the other hand, CloudSat has two products, Cloud Type and Precipitation Occurrence, that can add a new dimension to the IR-based precipitation algorithms. In the first approach, the cloud type classification dataset from CloudSat was used as a reference to find the non-precipitating cloud types. One of the reasons for FAR in satellite precipitation data is the presence of high non-precipitating clouds such as cirrus or cirrus anvil. Generally, the areal coverage of satellite precipitation estimation is larger than that of ground observation, primarily due to presence of cirrus anvil. Finding the pixels with anvil coverage, one can eliminate the false rain estimations from the satellite product. A trained neural network model using six MODIS water vapor, window and infrared channels (6.75, 7.325, 8.55, 9.7, 11.03, 12.02 μm wavelength) as the input and CloudSat cloud type as the target showed a remarkable improvement in elimination of false rain in the precipitation algorithm.