Mapping the groundwater level and soil moisture of a montane peat bog using UAV monitoring and machine learning

One of the best preconditions for sufficient monitoring of peat bog ecosystems requires a unique collection, processing, and analysis of spatial data to understand peat bog dynamics. Over two seasons, we sampled groundwater level (GWL), and soil moisture (SM) ground truth data at two diverse locations at the Rokytka Peat bog within the Sumava Mountains, Czechia. These data served as reference data and were modeled with a suite of potential variables derived from digital surface models (DSMs), RGB-, multispectral-, and thermal orthoimages, reflecting topo-morphometry, vegetation, and surface temperature information, generated from drone mapping. We applied 34 predictors to feed the Random forest (RF) algorithm. Predictors selection, hyperparameter tuning, and performance assessment were accompanied using Leave-Location-Out (LLO) spatial Cross-Validation (CV) joined with the forward feature selection (FFS) to overcome overfitting. The spatial CV performance statistics unveiled low (R2 = 0.12) to high (R2 = 0.78) model predictions. Predictor importance was used for model interpretation, where the temperature has proved the be a powerful impact on GWL and SM and significant other predictors' contributions such as Normalized Difference Index (NDVI), Normalized Difference Index (NDI), Enhanced Red-Green-Blue Vegetation Index (ERGBVE), Shape Index (SHP), Green Leaf Index (GLI), Brightness Index (BI), Coloration Index (CI), Redness Index (RI), Primary Colours Hue Index (HI), Overall Hue Index (HUE), SAGA Wetness Index (TWI), Plan Curvature (PlnCurv), Topographic Position Index (TPI), and Vector Ruggedness Measure (VRM). Additionally, we estimated the area of applicability (AOA) by presenting maps where the prediction model was certainly applied and where the predictions were highly uncertain because machine learning (ML) models make predictions far beyond sampling locations without sampling data and having no knowledge about these environments. The AOA method is perfectly suited and unique for decision-making about the best sampling strategy, notably for limited data to circumvent this issue.