The use of TRMM satellite data as a predictor in the spatial interpolation of daily precipitation over Australia

The accurate estimation of gridded daily precipitation is critical to hydrological modelling and water resource assessment. High resolution precipitation datasets based on gauge rainfall are the primary input to spatially distributed rainfall-runoff models and water balance calculations (e.g. Chiew et al., 2008). The spatial heterogeneity of rainfall variability is currently not captured adequately by gauge based interpolation methods. The errors in gridded rainfall fields have the potential to significantly bias model calibrations and water balances calculations. Sources of data other than gauge rainfall, such as satellite derived fields, radar based rainfall observations, and climatological fields from numerical weather prediction models, can be used as a predictor to improve interpolated rainfall fields. However, to date mainly gauge based interpolation of observations are used for modelling purposes because satellite/model based observed precipitation surfaces have relatively high errors in the rainfall distribution to which runoff is very sensitive.. To the authors’ knowledge no assimilation of observed and modelled precipitation data (gauged or from satellite) has been undertaken for the Australian continent. This paper investigates the additional value of the inclusion of the Tropical Rainfall Measuring Mission (TRMM) 3B42 satellite based precipitation product to gauge based precipitation estimation. Three simple geostatistical methods are applied to investigate the use of satellite based rainfall in gauge based interpolation: a) ordinary kriging (OK) – used as the baseline gauge based method; and b) cokriging (CK) and c) simple kriging with locally varying mean (SKlm) for incorporating additional information. The three methods are compared using cross validation statistics (including mean error, mean absolute error and root mean squared error). Incorporating satellite based rainfall estimates into the interpolation of daily rainfall does not increase the overall accuracy. However, in data sparse areas some increase in accuracy is observed. The poorer performance of the methods incorporating satellite data is attributed to the relationship between satellite derived rainfall and gauge rainfall which is highly variable (Figure 1). In particular, strong spatially consistent negative biases are found for coastal regions (greater the 0.5mm daily) and strong positive biases for high altitude regions (greater than 0.5mm daily). These biases and methods to deal with them should be considered in future research.