Uncertainty in Determination of Meteorological Drought Zones Based on Standardized Precipitation Index in the Territory of Poland

Abstract

The primary aim of this work is to assess the accuracy of the methods for spatial interpolation applied for the reconstruction of the spatial distribution of the Standardized Precipitation Index (SPI). The one-month version called SPI-1 is chosen for this purpose due to the known greatest variability of this index in comparison with its other versions. The analysis has been made for the territory of the entire country of Poland. At the same time the uncertainty related to the application of such computational procedures is determined based on qualitative and quantitative measures. The public data of two kinds are applied: (1) measurements of precipitation and (2) the locations of the meteorological stations in Poland. The analysis has been made for the period 1990–2020. However, all available observations since 1950 have been implemented. The number of available meteorological stations has decreased over the analyzed period. In January 1990 there were over one thousand stations making observations. In the end of the period of the study, the number of stations was below six hundred. Obviously, the temporal scarcity of data had an impact on the obtained results. The main tools applied were ArcGIS supported with Python scripting, including generally used modules and procedures dedicated to geoprocessing. Such an approach appeared crucial for the effective processing of the large number of data available. It also guaranteed the accuracy of the produced results and brought about drought maps based on SPI-1. The methods tested included: Inverse Distance Weighted, Natural Neighbor, Linear, Kriging, and Spline. The presented results prove that all the procedures are inaccurate and uncertain, but some of them provide satisfactory results. The worst method seems to be the interpolation based on Spline functions. The practical aspects related to the implementation of the methods led to removal of the Linear and Kriging interpolations from further use. Hence, Inverse Distance Weighted, as well as Natural Neighbor, seem to be well suited for this problem.