Spatial interpolation of monthly and annual rainfall in northeast of Iran

Precipitation maps are the key input to many hydrological models. In this paper different univariate (inverse distance weighing and ordinary kriging) and multivariate (linear regression, ordinary cokriging, simple kriging with varying local mean and kriging with an external drift) interpolation methods are used to map monthly and annual rainfall from sparse data measurements. The study area is Golestan Province, located in northeast of Iran. A digital elevation model is used as complementary information for multivariate approaches. The prediction performance of each method is evaluated through cross-validation and visual examination of the precipitation maps produced. Results indicate that geostatistical algorithms clearly outperform inverse distance weighting and linear regression. Among multivariate techniques, ordinary cokriging or kriging with an external drift yields the smallest error of prediction for months April to October (autumn and winter) for which the correlation between rainfall and elevation is greater than 0.54. For all other months and annual rainfall, ordinary kriging provides the most accurate estimates.