Comparing radar-raingauge precipitation-merging-methods for soil erosion modelling support

Radar-based rainfalls are currently used for process monitoring from remote in a large panel of domaines including hydrology and soil erosion modelling. Nevertheless, such data may include systemic and natural perturbations that need to be corrected before using these data. To encompass this problem, adjustments based on raingauge observations are frequently adopted. Here, we analysed the performance of different radar-raingauge merging procedures using a regional raingauge-radar network (Tuscany, Italy) focusing on a selected number of rainfalls events. The computational methods adopted were: 1) Kriging with External Drift (KED) interpolation (Wackernagel 1998), 2) Probability-Matching-Method (PMM, Rosenfeld et al., 1994), and 3) an Adjusted kriging mixed method exploiting the conditional merging (ADj) process (Sinclair-Pegram, 2005). The latter made available by DPCN (Italian National Civil Protection Department), while methods 1) and 2) were applied on recorded raingauge rainfalls over the regional territory at 15’ time-step, and CAPPI (Constant Altitude Plan Position Indicator) reflectivity data from the Italian radar network at 2000/3000/5000 m at 5’ and 10’. The comparisons between the three rainfall fields were based on the analyses of variance, Cumulative Distribution Function (CDF), and explicative coefficients such as BIAS, RMSE (Root Mean Square Error), MAD (Median Absolute Deviation). In average, rainfalls showed a moderate variability between the methods. Comparing CDFs, slight differences were detected between KED and ADj with bias mostly pronounced in lower quantiles, while more marked differences are observed in higher quantiles for the ADj-PMM methods. The analyses presented different spatial patterns depending on the applied procedure, closer to the radar data when using ADj, and more reflecting the gauge’s data structure when adopting KED. The probabilistic method (PMM) had the advantage to account for gauge data while preserving the spatial radar patterns, thus confirming interesting perspectives. Globally, the KED method provided more accurate coverage in the calculation by better compensating for local topographical shadows in the data, while ADJ confirmed the more detailed product in terms of time resolution (e.g. 5minute res.).