Comparison of statistical linear interpolation models for monthly precipitation in South Korea

Abstract

It is well recognized that statistical linear interpolation models are computationally inexpensive and applicable to any climate data compared to the dynamic simulation method at regional scales. Using five different statistical linear interpolation models, we characterized each model’s performance to predict a climate variable of interest. General linear model, generalized additive model, spatial linear model, and bayesian spatial regression model (BSM) were analyzed. The climate variable of interest was the monthly precipitation, where the spatial variability was explained using terrain information: latitude, longitude, elevation, topographic aspect, and costal proximity. We used the root mean squared error, the mean absolute error and correlation coefficient as the performance. The BSM showed better performance in reflecting the spatial pattern of monthly precipitation compared to the other models. The monthly precipitation and its 95 % prediction interval on a 1 × 1 km grid spacing were generated through a spatial interpolation of 441 point observations.