Microcystin concentrations and water quality in three agricultural ponds: A machine learning application

Cyanotoxins in agricultural waters pose a human and animal health risk. These chemical compounds can be transported to nearby crops and soil during irrigation practices; they can remain in the soils for extended periods and be adsorbed by root systems. Additionally, in livestock watering ponds cyanotoxins pose a direct ingestion risk. This work evaluated the performance of the randomForest algorithm in estimating microcystin concentrations from eight in situ water quality measurements at one active livestock water pond (Pond 1) and two working irrigation ponds (Pond 2 and 3) in Georgia, USA. Sampling was performed monthly from June of 2022 to October of 2023. Measurements of microcystin along with eight in situ sensed water quality parameters were used to train and test the machine learning model. The model performed better at Pond 1 (R² = 0.601, RMSE =3.854) and Pond 2 (R² = 0.710, RMSE = 2.310) compared to Pond 3 (R² = 0.436, RMSE = 0.336). Important variables for microcystin prediction differed among the three ponds, temperature and chlorophyll, phycocyanin and turbidity, and temperature and phycocyanin in Ponds 1, 2 and 3, respectively. Separating nearshore and interior samples in Ponds 1 and 2 lead to better predictive capacity of the model in nearshore samples compared with the interior samples. Overall, the random forest algorithm explained 50% to 70% of the microcystin concentration variation in three Georgia agricultural ponds with data from in situ sensing. In situ sensing showed a potential to aid in the water sampling design for microcystin to characterize the spatial variation of concentrations in studied ponds using readily available in situ sensing data.