Performance evaluation of merged satellite rainfall products based on spatial and seasonal signatures of hydrologic predictability

Despite the inherent estimation uncertainty, remote sensing based rainfall data have enormous value for stream flow simulation. Recent investigations have shown that the historical performance of satellite products in hydrologic prediction can be a useful (diagnostic) proxy for merging products to a more superior performing state for prognostic simulations (i.e., forward in time). Using a hydrologic model set-up over the entire Mississippi River Basin (MRB) and three widely used satellite rainfall products (3B42RT, CMORPH and PERSIANN-CCS), this study explored a merging scheme based on runoff predictability. The spatial and temporal signatures of variability were closely investigated to understand the impact on prediction skill of the merging scheme. The spatial variability (i.e., non-uniform) considered the grid box by grid box variation at the native resolution of individual satellite products, while the temporal variability (i.e., non-stationary) was confined to variation in 3 month-long seasons (winter, spring, summer and fall). When both the spatial and temporal variability in runoff predictability was leveraged, the merging scheme yielded the largest improvement over individual productʹs performance forward in time. During an independent validation assessment, the stream flow simulated by the merged product was more strongly correlated with observed discharge (than individual products) at 12 gauging stations. In terms of reduction in root mean squared error (RMSE), the merged product showed an improvement of 57% for 3B42RT, 63% for CMORPH and 68% for PERSIANN-CCS products. The investigation clearly showed that any ‘operational’ and hydrologic predictability-based merging scheme for unifying available satellite rainfall products must factor in both the spatial and temporal signatures of runoff predictability to achieve consistently more superior prognostic skill.