Performance of Satellite-Based Rainfall Observations for Hydrological Modelling in Ungauged Areas

The hydrological community relies on rainfall-runoff models for many fundamental applications, such as flood forecast, water resources estimation and runoff predictions. The performance of hydrological models is highly influenced by the quality of rainfall input that are employed, no matter the model adopted. Accurate and precise rainfall observations are, hence, pivotal. Precipitation inputs are commonly derived from rain gauge observations, which often do not provide proper spatial information. Indeed, as they provide point measurements, interpolation of rain gauge time series is frequently needed. However, the interpolated rainfall fields are influenced by the locations and density of the existing networks, often compromising the performance of the model. In the last decades many satellite-based precipitation estimates have been developed, which provide gridded information on a quasi-global coverage. Hence, using satellite retrievals have the potential to be used as input for hydrological modelling in ungauged areas, where interpolating information from a monitoring network is not possible. In this study, the Precipitation Estimation from Remotely Sensed Information using Artificial Neural Networks - Cloud Classification System (PERSIANN-CCS) data set is employed as input for a hydrological model and its performance is compared against the one of rain gauge input. The simulations are performed in the hypothesis of lack of flow data for calibration, to simulate the ungauged conditions. Results show that both data sources perform poorly in ungauged conditions, with the rain gauge slightly outperforming satellite inputs. Moreover, PERSIANN-CCS is found to underestimate precipitation, in agreement with previous studies.