Simulation of rainfall fields conditioned on rain gauge observations and radar estimates using random mixing

Abstract

The accuracy of spatial precipitation estimates with the relatively high temporospatial resolution is of vital importance in various fields of research and practice. Yet the intricate variability and the intermittent nature of precipitation make it very difficult to obtain accurate spatial precipitation estimates. Radar and rain gauge are two complementary sources of precipitation information: the former is inaccurate in general but is a valid indicator for the spatial pattern of the rainfall field; the latter is relatively accurate but lack spatial coverage. Considering the pros and cons of the two sources of precipitation information, a number of radar-gauge merging techniques have been developed to obtain spatial precipitation estimates over the past years. Conditional simulation has great potential to be used in spatial precipitation estimation. Unlike the commonly used interpolation methods, the results from the conditional simulation is a range of possible estimates due to its Monte Carlo framework. Yet an obstacle that hampers the application of conditional simulation in spatial precipitation estimation is how to obtain the marginal distribution function of the rainfall field with accuracy. In this work, we propose a method to obtain the marginal distribution function of the rainfall field based on rain gauge observations and radar estimates. The conditional simulation method, random mixing (RM), is used to simulate rainfall fields. The properties of the results from the proposed method are elaborated through the comparison with the results from other methods: ordinary kriging, kriging with external drift, and conditional merging. Finally, the sensitivity of the proposed method towards the two factors – density of rain gauges and random error in radar estimates – is analyzed.