Accurate, lake‐wide forecasting of chlorophyll concentration is critical for tracking eutrophication and can be used as a proxy of harmful algal blooms (HABs). Our project addresses this need by developing a forecasting system for the Maximum Chlorophyll Index (MCI) and Maximum Peak Height (MPH) at 30-meter resolution across Lake Champlain.

To achieve this, we fuse Sentinel-2 and Landsat reflectance, National Water Model meteorological forecasts, continuous in-lake sensor records, and USGS bathymetry, capturing the physical, chemical, and hydrological controls on chlorophyll indices. An advanced deep learning framework ingests these heterogeneous inputs while preserving both spatial detail and temporal continuity. Robust preprocessing pipelines remove clouds, reconstruct missing data points, and apply targeted data augmentation, ensuring that the model is trained on consistent, information-rich inputs. We quantify predictive skill with standard error metrics (RMSE, MAE, MAPE) and structure-aware scores (multi-scale SSIM, spatial-gradient loss), and we probe model explainability by ranking the relative importance of each environmental variable.

The resulting daily chlorophyll forecasts convert directly into CyanoHAB risk maps, delivering the first lake-wide, multi-satellite forecasting system for Lake Champlain and offering a transferable workflow that illuminates how linear and non-linear environmental drivers shape bloom dynamics.