Comparison of spatial interpolation methods for estimating the precipitation distribution in Portugal

• Published in: Theoretical and applied climatology 2021-08, Vol.145 (3-4), p.1193-1206
• Main Authors: Antal, Alexandru, Guerreiro, Pedro M. P., Cheval, Sorin
• Format: Article
• Language: English
• Subjects: Analysis; Annual precipitation; Annual precipitation distribution; Aquatic Pollution; Atmospheric precipitations; Atmospheric Protection/Air Quality Control/Air Pollution; Atmospheric Sciences; Bayesian analysis; Climate science; Climatology; Correlation coefficient; Correlation coefficients; Distribution; Earth and Environmental Science; Earth Sciences; Economic sectors; Estimation; Geospatial data; Impact analysis; Interpolation; Interpolation methods; Mathematical analysis; Mean annual precipitation; Methods; Original Paper; Polynomials; Precipitation; Precipitation distribution; Probability theory; Radial basis function; Rain gauges; Root-mean-square errors; Spatial data; Spatial distribution; Standard error; Statistical methods; Waste Water Technology; Water Management; Water Pollution Control
• ISSN: 0177-798X; 1434-4483
• DOI: 10.1007/s00704-021-03675-0

Precipitation has a strong and constant impact on different economic sectors, environment and social activities all over the world. An increasing interest for monitoring and estimating the precipitation characteristics can be claimed in the last decades. However, in some areas, the ground-based network is still sparse and the spatial data coverage insufficiently addresses the needs. In the last decades, different interpolation methods provide an efficient response for describing the spatial distribution of precipitation. In this study, we compare the performance of seven interpolation methods used for retrieving the mean annual precipitation over the mainland Portugal, as follows: local polynomial interpolation (LPI), global polynomial interpolation (GPI), radial basis function (RBF), inverse distance weighted (IDW), ordinary cokriging (OCK), universal cokriging (UCK) and empirical Bayesian kriging regression (EBKR). We generate the mean annual precipitation distribution using data from 128 rain gauge stations covering the period 1991 to 2000. The interpolation results were evaluated using cross-validation techniques and the performance of each method was evaluated using mean error (ME), mean absolute error (MAE), root mean square error (RMSE), Pearson’s correlation coefficient (R) and Taylor diagram. The results indicate that EBKR performs the best spatial distribution. In order to determine the accuracy of spatial distribution generated by the spatial interpolation methods, we calculate the prediction standard error (PSE). The PSE result of EBKR prediction over mainland Portugal increases from south to north.