Assessing the effectiveness of remote sensing indices for predicting E. coli concentrations in an irrigation pond

Escherichia coli (E. coli) is a key marker for monitoring microbial water quality, with significant consequences for both public health and agricultural practices. To address the challenges of traditional water quality assessments, remote sensing offers a promising alternative. In this research, we implemented the random forest (RF) algorithm to forecast E. coli levels in irrigation ponds using three distinct data sources: (1) conventional water quality measurements, (2) multispectral reflectance values from drones, and (3) remote sensing indices derived from these reflectance values. To enhance the model’s accuracy, a linear transformation was applied during postprocessing. The RF model achieved strong performance (R² = 0.74) with conventional water quality variables, while moderate results were obtained using multispectral reflectance values alone (R² = 0.56). The best outcomes were observed when remote sensing indices were used as inputs, yielding an R² of 0.76. Shapley additive explanations (SHAP) were employed to evaluate the importance of individual variables. Dissolved oxygen, pH, and Chlorophyll-a emerged as critical predictors among water quality parameters. Meanwhile, the visible atmospherically resistant index (VARI) and normalized difference turbidity index (NDTI) were the most significant remote sensing indices. Furthermore, location-based comparisons highlighted differences in the impact of VARI and NDTI between interior and nearshore sampling sites. These findings suggest that remote sensing indices effectively capture water quality features crucial for E. coli persistence. This study underscores the potential of using drone-derived multispectral data to enhance predictions of E. coli concentrations in irrigation ponds.