Observation based gridded annual runoff estimates over Victoria, Australia

In general, runoff generation is a highly variable process spatially. However, lack of spatially continuous runoff observations is a significant problem in surface hydrology due to gaps between gauged catchments and/or the need for within-catchment or ungauged catchment estimates. Recent efforts have produced programs for prediction in ungauged catchments, both in Australia (eg: Australian Water Resource Assessment - Landscape model (AWRA-L)) and internationally (eg: Prediction in Ungauged Basins (PUB)). While modelling approaches have many benefits (e.g. forecasting and projection), historic studies have the option of using observed data more directly. Adoption of geostatistical interpolation techniques potentially provides a more accurate method for generating spatially continuous runoff fields than modelling approaches. However, the use of geostatistical interpolation in hydrology is relatively uncommon because of the complex nature of hydrological connectivity. Here, we extend and apply top-kriging to spatially interpolate stream gauge data to produce gridded annual runoff across most of Victoria, Australia, from 1982-2012. Specifically, we extend the R-package rtop to allow kriging with external drift through the inclusion of spatial rainfall grids as a predictor. As expected, the resultant interpolated scheme provides a near exact match with observed values and acceptable performances in leave-one-out validation (R² of approximately 0.6 to 0.7 depending on year). The annual maps appropriately capture the variation in streamflow over time. Overall, this study illustrates that kriging can inform spatial runoff patterns and that the approach requires fewer inputs and less computational load than a complex numerical model. The methodology can be applied to any region with relatively minor adjustments. The dataset is open access in Zenodo repository at https://doi.org/10.5281/zenodo.5454798.