Custom Methodology to Improve Geospatial Interpolation at Regional Scale with Open-Source Software

Abstract

This study shows a methodological approach to improve geospatial interpolation carried out with the Inverse Distance Weighted algorithm using distances and other parameters to which we attribute relative weights such as elevation. We also provide reliable information about better data output by elaborating a more realistic confidence interval with various percentages of reliability. We tested the methodology to monthly accumulated rainfall and temperatures recorded by multiple monitoring stations in the Puglia region in South Italy. The whole procedure has been called Augmented Inverse Distance Weighted and is tested with the ultimate goal of predicting missing values at a regional scale based on cross-validation techniques applied to a dataset consisting of ten years of precipitation data and five years of temperature data. The efficacy of this approach is evaluated using statistical scores regularly employed in the model’s evaluation studies. Results show that the improvements over the classical approach are remarkable and that the “augmented” method provides more accurate measurements of environmental variables. The main application of this algorithm is the possibility to provide the spatialisation of values of precipitation and temperature, or any other based on its own needs, at every point of the territory, playing a very important role in agricultural decision support systems and letting us identify frosts, drought events, climatic trends, accidental events, cyclicality and seasonality.