Rainfall Retrieval with a Polarimetric S-Band Radar and Multi-Sensor QPE Comparison using Hydrological Models in Brazil

Improved rainfall estimates enhance the potential of radar for many applications such as flood forecasting and management of hydropower generation. In less than 5 years, in Brazil we increased our radar coverage, from 23 single polarization radars to 15 additional dual polarization radars, mainly S-Band, with a concentration in the southern region, an area prone to severe weather, usually related to mesoscale convective systems. This region is responsible for more than 35% of the national hydropower energy generation, directly dependent on precipitation distribution and water availability, and storm events result in disasters related to flood inundation. Polarimetric techniques, which undergo an ever-growing implementation, carry the promise of providing rainfall estimates which are significantly more accurate than those derived from single polarization measurements. A better representation of rainfall spatial distribution and its uncertainties is crucial for accurate forecasts of river discharges and water levels. Radar estimations of QPE are very useful information for hydrological applications because of high spatial and temporal resolution improving runoff forecasts and reducing model dependence on unreliable parameter estimates of watershed characteristics. However, radar QPE depends on the calibration, good adjustment with rain gauges and disdrometers, data filtering, distance from the radar, orography, signal propagation, among other factors. A multi-sensor integration approach of of remote sensing precipitation estimation using meteorological satellites and weather radars with rain gauges improves the accuracy of hydrological models when compared to a model using gauge data only. In this study, we compare the use of rainfall estimates retrieved from radar, multisensor QPE product and tipping-bucket observations in the simulation of a rainfall-runoff at a 335 km2 river catchment. We used a limited dataset from a S-Band dual polarimetric radar with a local R(Z) relation based on disdrometer data and a polarimetric R(Z,ZDR) algorithm from the literature. Two hydrological models (Sacramento Soil Moisture Accounting, SAC-SMA; IPH-II model) were used and calibrated using observed discharge time-series, the Shuffle Complex Evolution algorithm and Nash-Sutcliff Efficiency (NSE) as objective function. The NSE values for the SAC-SMA(IPH-II) models for rain gauge, R(Z), R(Z,ZDR) and multi-sensor were 0.725(0.719), 0.739(0.661), 0.727(0.659), 0.603(0.632), respectively. Our experiments provided insights on how analyses of rainfall-runoff model optimization can feedback real improvements on river flow prediction forced with radar rainfall inputs. Our major findings: (i) radar rainfall inputs resulted in a slight increase in the SAC‑SMA simulations; (ii) smaller floods were not well reproduced using the multi-sensor QPE product, thus stressing the need for further improvements in quantification of light rainfall; (iii) the parameter optimization using rainfall retrieved from the radar led to higher soil storage capacities and smaller content available for evapotranspiration in the SAC-SMA; (iv) benefits of radar-based rainfall on discharge forecast is also dependant on the hydrological model structure.