Application of k-Nearest code to the improvement of class pedotransfer functions and countrywide Field Capacity and Wilting Point maps

Abstract

The database of soil hydrophysical properties in the Czech Republic (HYPRESCZ) which contains the data needed for derivation of pedotransfer functions for soil water retention curves was used for the estimation of field capacity and wilting point of agricultural land resource on a countrywide scale. The results were combined with the existing Soil Texture Map of the Czech Republic to create four new maps, namely the Map of Field Capacity and the Map of Wilting Point for the topsoil and subsoil separately. From the total number of 1048 relevant database entries, only about a half included reliable wilting point data. The k-Nearest computer code employing the k-Nearest neighbour technique was used for estimation of the missing wilting points, which made it possible to use all entries. The estimation uncertainty was assessed in terms of standard deviations and the root mean square error. Finally, two sets of class pedotransfer functions were derived and found sufficiently comparable: (i) the functions estimating the soil water retention curve in the whole range, derived solely from the database entries containing the measured wilting points, and (ii) the functions estimating the field capacity and wilting point only, derived from all database entries, including the k-Nearest neighbour estimated data.