Spatial downscaling of 3-hourly precipitation forecast data at river basin scale

Abstract

This study proposes and applies a simple methodology to downscale 3-hourly precipitation data obtained from a large-scale atmospheric model to a finer resolution grid, which is compatible with the scale of the main hydrological processes occurring in a river basin. The aim was to obtain quantitative precipitation forecast data, suitable to be used as an input in rainfall–runoff models to simulate flood events. ERA-Interim forecasts of total precipitation, with 0.75° of spatial resolution, were downscaled by statistical adjustment to a data series of observed precipitation in rain gauges. The adjustment draws on the “Bias-Correction Spatial Downscaling” method, with two modifications: (1) the order in which bias correction is carried out; and (2) the fact that it is applied to the downscaling of higher frequency data (3-hourly precipitation). Relations between the forecasts and the locally observed precipitation were estimated by applying the method of histogram equalization, obtaining piecewise transfer functions with good explanatory power (adjusted R2 above of 0.99). These functions enabled the removal of the systematic bias of ERA-Interim forecasts, improving their overall fit with the observed data. This improvement proved more significant in the prediction of precipitation volumes over time, compared to the forecast of peak values. The methodology was applied to the basin of the largest fully Portuguese river (Mondego), with a grid resolution of 0.125° × 0.125°.