Comparing Spatial Interpolation Methods for CMIP5 Monthly Precipitation at Catchment Scale

Abstract

Use of Regional Climate Models (RCMs) is prevalent in downscaling the large scale climate information from the General Circulation Models (GCMs) to local scale. But it is computationally intensive and requires application of a numerical weather prediction model. For more straightforward computation, spatial interpolation are commonly used to re-gridding the GCM data to local scales. There are many interpolation methods available, but mostly they are chosen randomly, especially for GCM data. This study compared eight interpolation methods (linear, bi-linear, nearest neighbour, distance weighted average, inverse distance weighted average, first-order conservative, second-order conservative and bi-cubic interpolation) for re-gridding of CMIP5 decadal experimental data to a catchment scale. For this, CMIP5 decadal precipitation data from three GCMs were collected and subset for Australia and then re-gridded to 0.05 degree spatial resolution matching with the observed gridded data. The re-gridded data were subset for Brisbane catchment in Queensland, Australia and a number of skill tests (root mean squared error, mean absolute error, correlation coefficient, Pearson correlation, Kendal’s tau correlation and index of agreement) were conducted for a selected observed point to check the performances of different interpolation methods. Additionally, temporal skills were computed over the entire catchment and compared. Based on the skill tests over the study area, the second-order conservative (SOC) method was found to be an appropriate choice for interpolating the gridded dataset.