Comparing rainfall interpolation techniques for small subtropical urban catchments

Abstract

Rainfall estimation is an integral component of hydrologic modelling for reliable flood forecasting and for assessing the impact of precipitation and runoff on water quality and ecosystem health in urbanised areas. Spatial variability of rainfall adds to the complexity of estimating rainfall at the catchment scale and is a key factor that must be incorporated into estimations of rain fields. Rainfall interpolation from point measurements is one approach that hydrologists use to account for spatial variation in rainfall. This study focuses on evaluating the suitability of three interpolation methods in terms of their accuracy, for small urban catchments. The Bridgewater Creek catchment in south Brisbane is the region of study. Thirteen storm events measured over 24-hours are interpolated using an inverse-distance weighted average method, thin plate smoothing spline, and an ordinary kriging technique. The delete-one validation method is used for comparing the 3 interpolation techniques. The interpolation methods were applied using 3 different software packages. The inverse-distance weighted method (IDW) was implemented using Microsoft Excel. The spline interpolations were carried out with the ANUSPLIN software while the kriging method was completed using Geostatistical Analyst in ArcGIS. Since ANUSPLIN and Geostatistical Analyst both require a minimum number of data points for computation, it was necessary to consider a reduced set of storms for interpolation by the thin plate spline and kriging methods. The thin plate spline and ordinary kriging interpolation techniques were found to have comparable estimation accuracy when individual storm events were considered. When compared to gauge-based rainfall measurements, the estimation error for these methods was approximately 23 mm, which corresponds to 30% of the observed mean rainfall. The IDW method was observed to display more frequent occurrence of extreme estimation error up to 98% for individual storms. However, an analysis of an aggregated set of storm events for which all 3 methods could be applied showed that the IDW had the lowest estimation error and highest model efficiency of the three interpolation techniques for the storms selected.