Analysis of deterministic and geostatistical interpolation techniques for mapping meteorological variables at large watershed scales

Abstract

The widely scattered pattern of meteorological stations in large watersheds and remote locations, along with a need to estimate meteorological data for point sites or areas where little or no data have been recorded, has encouraged the development and implementation of spatial interpolation techniques. The various interpolation techniques featured in GIS software allow for the extraction of this new information from spatially distinct point data. Since no one interpolation method can be accurate in all regions, each method must be evaluated prior to each geographically distinct application. Many methods have been used for interpolating minimum temperature ($T_{ \min }$), maximum temperature ($T_{ \max }$) and precipitation data; however, only a few methods have been used in the Zayandeh-Rud River basin, Iran, and no comparison of methods has ever been carried out in the area. The accuracies of six spatial interpolation methods [Inverse Distance Weighting, Natural Neighbor (NN), Regularized Spline, Tension Spline, Ordinary Kriging, Universal Kriging] were compared in this study simultaneously, and the best method for mapping monthly precipitation and temperature extremes was determined in a large semi-arid watershed with high temperature and rainfall variation. A cross-validation technique and long-term (1970–2014) average monthly $T_{ \min }$, $T_{ \max }$ and precipitation data from meteorological stations within the basin were used to identify the best interpolation method for each variable dataset. For $T_{ \min }$, Kriging (Gaussian) proved to be the most accurate interpolation method (MAE = 1.827 °C), whereas, for $T_{ \max }$ and precipitation the NN method performed best (MAE = 1.178 °C and 0.5241 mm, respectively). Accordingly, these variable-optimized interpolation methods were used to define spatial patterns of newly generated climatic maps.