Mapping Agricultural Tile Drainage across the US Midwest using Satellite Imagery and Random Forest Machine Learning

Luwen Wan, Anthony D. Kendall, David W. Hyndman
Earth and Environmental Sciences

Presented at the All Scientist Meeting and Investigators Field Tour (2021-09-23 to 2021-09-23 )

Agricultural practices such as tile drainage have been linked to altered streamflow and nutrient delivery from the landscape. Extensive areas of the U.S. Midwest have installed agricultural tile drainage and the number of drained fields continues to grow rapidly. Tile drainage systems remove excess water from poorly drained land and enhance crop production. These systems are also an important nutrient transport pathway, delivering a larger proportion of applied nutrients to downstream ditches, streams, and water bodies than undrained lands. The lack of fine-resolution, spatially explicit tile drainage maps at regional scales makes it challenging to model these landscapes and implement the regional nutrient reduction strategies. Currently available tile drainage datasets include county-level surveyed drained areas or GIS-overlain maps of likely tile-drained pixels. Here, we develop an agricultural tile drainage map across the US Midwest at 30-m resolution using satellite imagery, along with climate- and soil-related variables within the Google Earth Engine (GEE) cloud-computing platform. We have assembled a new regional dataset of training samples, point data that are manually identified from multi-resolution aerial imagery and compiled from other literature and agency sources. After training and running a random forest classifier, we compare the outputs to other currently available products to quantify the improvement in classification accuracy. We also identify the relative importance of various input variables for tile drainage classification, which provides insight into decision making around tile drainage installation. The results also provide valuable input for data-driven hydrological modeling, inform sustainable water management practices, and provide a basis for cost-effective decisions to reduce nutrient fluxes to surface waters.