Bowling Green State University, Ohio, United States
Abstract: Natural resource managers require accurate spatial and temporal mapping of soil moisture to properly plan and manage wet prairie restorations. Unfortunately, high density spatial and temporal observations of water table levels (soil moisture) over relatively large areas cannot be easily obtained because of cost and time constrains. The objective of this study was to determine the best interpolated water table surface when only limited spatial data is available. Although, water table maps of wet prairies have been created using physically-based models as well as with machine learning, limited work has been done using relatively simple interpolation schemes with limited spatial data.
The study site is located in the Oak Openings Region of northwest Ohio near the village of Swanton, Ohio. It encompasses a 116 ha (286 ac) area under current wetland restoration by The Nature Conservancy (TNC). Seven piezometers were installed in this study site using stainless steel drive-point piezometers to a depth of 2.75 m below the ground surface. Each piezometer was fitted with a pressure transducer data-logger to measure hourly water table elevation. Three different interpolation methods (universal Kriging, universal co-Kriging with DEM / universal co-Kriging with Topographic Wetness Index (TWI), and spatio-temporal Kriging) were used to generate an interpolated water table surface from water table measurements collected between December 2019 and September 2020.
Qualitative as well as quantitative maps were used in the evaluation of fit from these three interpolation methods. Difference maps were created to visually quantify the difference between the interpolations. Depth to water table maps were created to show this ecologically-relevant parameter. Prediction standard error maps were created to assess the accuracy of the various interpolations with respect to the actual measurements. The results showed that spatio-temporal Kriging had the best fit and the minimum error with respect to all other Kriging methods used with this data set. This interpolation method was able to leverage the temporal data collected hourly to slightly increase the accuracy of the spatial interpolation.
This study shows that simple interpolations methods that do not require complex bio-physical models or advanced computational algorithms could be of use for natural resource managers even when only limited spatial data is available.
