Drought classification in Northern Serbia based on SPI and statistical pattern recognition

Abstract

This paper proposes a procedure for decision-making regarding the extent to which a certain geographical region is affected by drought. Professional circles generally recognize the Standardized Precipitation Index (SPI) as a good indicator of a drought event. However, as a result of varying precipitation levels due to various local geographical, climatic, vegetational and other factors, this indicator is determined based on precipitation measurements and different meteorological centres within the same administrative region often generate different SPI values, even when the geographical distance between them is small. During a dry period, various local authorities, ministries of agriculture or governments have to make important decisions about, for example, declaring disasters, subsidizing farmers for certain crops, or providing financial aid to agricultural producers, based on voluminous and diverse data about local precipitation, the yield of various crops, or the condition of the soil. This paper proposes an automated methodology for such decision-making, which can be used as a support tool by decision-makers. The methodology is based on the SPI and statistical pattern recognition methods (dimension reduction and classifier design based on the desired output). The entire procedure is illustrated using Vojvodina, a region in Serbia in the southern portion of the Pannonian Plain, as a case study. The proposed algorithm is not subject to any constraints with regard to geographical locations of regions, their surface areas, or inter-relationships.