Spatially Explicit Agricultural Tile Drainage Maps - US Midwest (SEETileDrain, Wan et al., 2024)

The source is a repository of Spatially Explicit Estimate of Tile Drainage (SEETileDrain) products across the US Midwest in 2017 at a 30-m resolution. It includes the binary classification map (tile and non-tile), tile probability (how likely a grid cell is tile-drained). The Python scripts to generate the PAW layers and the R scripts (see also: https://github.com/LuwenWan/SEETileDrain_MidWest) to select variables, implement the random forest model and visualize the figures, are also available.

In this work, we developed a machine learning model using 31 satellite-derived and environmental variables and trained with 60,938 tile and non-tile ground truth points within the Google Earth Engine cloud computing platform. The results show that our model achieved good accuracy, with 96 % of the points correctly classified and an F1 score of 0.90. When the tile drainage areas are aggregated to the county scale, it agrees well (R-squared = 0.68) with the reported area from the 2017 Ag Census. The product, SEETileDrain (Spatially Explicit Estimate of Tile Drainage), is described in full detail in the manuscript and the supporting information of Wan et al. (2024). If needed, copies of the tile drainage product can be requested from the corresponding author at [e-mail].

Preferred citation:
L. Wan, A.D. Kendall, J. Rapp, D.W. Hyndman. 2024. Mapping agricultural tile drainage in the US Midwest using explainable random forest machine learning and satellite imagery, Science of the Total Environment. https://doi.org/10.1016/j.scitotenv.2024.175283