Development of a Bias Correction Algorithm for Near-Real Time Satellite Rainfall Observations for Flood Forecasting in Poorly Gauged Basins

Rainfall data is fundamental input for flood forecast models, however, availability of real- or near-real time rain gauge observations are very limited in many developing countries, and hence effective counter measure against flood disasters cannot be realized. One approach is to employ near-real time satellite derived rainfall products, such as Global Satellite Mapping of Precipitation data (GSMaP), which are 0.1 degree composite tiles of rainfall information covering most of developing countries on hourly basis. However issues remain regarding time delay in issuance and data quality. Currently GSMaP is issued ~ 4 hours after the observations , and the data is biased mainly due to interpolation of satellite observations to the un-observed regions using cloud moving vectors estimated from infreRed (IR) images.

In this study, we present a new bias correction algorithm, which combines conventionally used real time calibration method and statistical transformation method. Real time calibration method estimates calibration coefficient of GSMaP for cells having ground observations and distributes the coefficient with considering distance and elevation as varying weight. Statistical transformation method transforms temporal frequency of GSMaP, so that the histogram has the same statistical distribution of ground observations. Performance of new method is compared against conventional methods through hydrological modeling of recent flood events observed in Sri Lanka. Results show that the real time calibration method has large uncertainty as it considers only the daily snap shot and disadvantage can be seen from ignoring the historical GSMaP errors. It is also found that the performance of statistical transformation method is affected by the length of observation records being used to create the temporal histogram. By contrast the suggested method, which considers both the spatial and temporal GSMaP, shows more robust and accurate results. We may further test the applicability of the new method in different regions and with different observation data for practical application.