Reconstructing missing data by comparing imputation and interpolation techniques: applications for long-term groundwater monitoring
No Thumbnail Available
Date
2026
Journal Title
Journal ISSN
Volume Title
Publisher
Nature Portfolio
Abstract
Missing data is a pervasive challenge in environmental monitoring, severely hindering reliable analysis and effective water resource management. This study addresses this issue in the context of the EU Nitrates Directive (91/676/EEC), focusing on the proportion of groundwater monitoring sites expected to exceed the regulatory limit of 50 mg NO₃⁻/L. Overall, 57% of observations were missing, with 99% of monitoring stations experiencing data gaps over the 2000–2024 study period. Six reconstruction techniques were evaluated: k-nearest neighbours, multiple imputation by chained equations, missForest, natural neighbour, inverse distance weighting, and indicator kriging. This work presents a novel approach to comparing various imputation techniques for reconstructing threshold overruns and modelling trends in exceedance rates. Moving beyond conventional error metrics, the evaluation is grounded in distributional similarity and monotonic dependence using Hellinger and Kullback-Leibler distances and generalised Lorenz curves. Based on the comparative analysis, indicator kriging was identified as the most robust approach and was subsequently adopted to parameterise the log-linear trend in the proportion of monitoring stations expected to exceed the concentration threshold. This study provides a methodological framework for addressing extreme missingness in environmental monitoring networks, offering actionable insights for EU Member States reporting under the Nitrates Directive.
Description
Keywords
groundwater quality, nitrate pollution, missing data, imputation methods, indicator kriging
Citation
Sci Rep (2026)