Merging Satellite Precipitation Products for Improved Streamflow Simulations

Accurate quantitative precipitation estimation is of great importance for water resources management, agricultural planning and forecasting and monitoring of natural hazards such as flash floods and landslides. In situ observations are limited around the Earth, especially in remote areas (e.g., complex terrain, dense vegetation), but currently available satellite precipitation products are able to provide global precipitation estimates with an accuracy that depends upon many factors (e.g., type of storms, temporal sampling, season, etc.). The recent SM2RAIN approach proposes to estimate rainfall by using satellite soil moisture observations. As opposed to traditional satellite precipitation methods, which sense cloud properties to retrieve instantaneous estimates, this new bottom-up approach makes use of two consecutive soil moisture measurements for obtaining an estimate of the fallen precipitation within the interval between two satellite overpasses. As a result, the nature of the measurement is different and complementary to the one of classical precipitation products and could provide a different valid perspective to substitute or improve current rainfall estimates. Therefore, we propose to merge SM2RAIN and the widely used TMPA 3B42RT product across Italy for a 6-year period (2010-2015) at daily/0.25deg temporal/spatial scale. Two conceptually different merging techniques are compared to each other and evaluated in terms of different statistical metrics, including hit bias, threat score, false alarm rates, and missed rainfall volumes. The first is based on the maximization of the temporal correlation with a reference dataset, while the second is based on a Bayesian approach, which provides a probabilistic satellite precipitation estimate derived from the joint probability distribution of observations and satellite estimates. The merged precipitation products show a better performance with respect to the parental satellite-based products in terms of categorical statistics, as well as bias reduction and correlation coefficient, with the Bayesian approach being superior to other methods. A study case in the Tiber river basin is also presented to discuss the performance of forcing a hydrological model with the merged satellite precipitation product to simulate streamflow time series.