Automated Mapping of Water Table for Cranberry Subirrigation Management: Comparison of Three Spatial Interpolation Methods

Abstract

In this paper we first compare three different methods of spatial interpolation, i.e., inverse distance weighting (IDW), thin plate splines (TPS), and kriging on weekly water table depth (WTD) measurements from 80 observation wells in two cranberry farms (Farm A and Farm B) located in Québec, Canada. We use the leave-one-out cross-validation approach to assess the performance of the methods. Second, we evaluate the influence of the density of measurement points over the interpolation error for the cited methods. Third, we assess the performance of drainage systems and their impacts on crop productivity as a result of cumulative rainfall. Results along with practical considerations show that TPS is the best interpolator for WTD and this superiority is maintained and further demonstrated through a sensitivity analysis of the methods to spatial sampling density, i.e., partitioning the data into subsets of 25, 50, and 75% of the dataset. However, the random approach for selecting these subsets shows an unexpected result; that is, the interpolation methods exhibit a higher performance in terms of the Pearson correlation (r) for the 25% data subset at Farm B. Meanwhile, the cumulative precipitation over a three-day period, the maximum time required to return the soil matric potential to the optimal value after a major rainfall event, had a steady influence on WTD and thus crop productivity in the studied farms. This influence is more apparent for Farm A, but a rather random effect is noted for Farm B. This study presents a water-management-based strategy that mitigates the supplementary cost and effort for sensor deployment in water table monitoring for cranberry production. It is therefore of practical interest to cranberry growers and decision-makers who aim to maximize yields through water-management-oriented strategies.