Hydrologically-relevant merging of high resolution satellite precipitation products to enhance hydrologic application

Hydrologically relevant merged satellite rainfall product at global scale was produced by merging three High Resolution Precipitation Products (HRPPs) based on their a priori (diagnostic) runoff predictability and using a hydrologic model as a conditioning filter for noise reduction. National Aeronautics and Space Administration (NASA)'s Tropical Rainfall Measuring Mission (TRMM) Multi-satellite Precipitation Analysis (TMPA) real time product (3B42RT), the Climate Prediction Center's (CPC) MORPHing technique (CMORPH), and Precipitation Estimation from Remote Sensing Information using Artificial Neural Network-Cloud Classification System (PERSIANN-CCS) were used for the analysis period of eight years (2003-2010). To implement the a priori hydrologic predictability-based merging at global scale, where in-situ hydroclimatic data was not available, satellite rainfall uncertainty and its propagation through a hydrologic model were investigated by tracing the source of runoff and soil moisture errors as a function of rainfall error components. Such a tracing identified the role played by more readily accessible geophysical features such as topography, climate, and land use and land cover that affect the accuracy of precipitation remote sensing. Once the key governing factors were identified (topography and climate), a non-linear regression model was developed to estimate the runoff and satellite rainfall error variance at un-gauged basin of the world. The result indicated that the merged product based on the spatial and temporal signatures of a priori runoff predictability displayed consistently superior performance over the selected study regions (USA, Asia, Middle East, and Mediterranean regions). The simulated stream flow from the merged satellite rainfall products captured the seasonal and annual variability of the observed flow at different location of USA basins. The current merging scheme works most effectively when each product has complementary signal-to-noise ratios, which is not always guaranteed given the statistical nature of rainfall estimation by remote sensing. Further explorations into the concept of dynamic weighting factor using Kalman filtering or Bayesian concept could potentially overcome this limitation. The regression model was also assessed of its skill in predicting rainfall error variance from easily available geophysical features around the world for diverse geographic settings. Overall, the investigation has proven that the quantitative picture of satellite rainfall error over un-gauged regions can be estimated even in the absence of in-situ data.