Developing Long-term Satellite Based Chlorophyll Estimates via Multiple Machine Learning Methods

Chlorophyll is widely used to assess the level of eutrophication, which is recognized as one of the major causes of deterioration in water quality, especially across Irish inland waters. The use of optical remote sensing for chlorophyll monitoring has already been shown to be able to complement existing in-situ chlorophyll. However, as cloud cover impacts all optical sensors this limits the usefulness in many locations to develop long term records. A potential solution is to combine information from multiple satellites using a machine learning approach. In this study, we develop long-term time series of chlorophyll concentrations for lakes in Ireland using four different remote sensing platforms: Landsat-8, MODIS, Sentinel-2, and Sentinel-3. Several machine learning approaches have been tested, including K-nearest neighbourhood, random forest (RF), XGBoost (Extreme Gradient Boosting), artificial neutral network (ANN), and support vector regression (SVR).

Initial result indicate that a machine learning model utilising all four platforms and in-situ observation is effective in developing long-term chlorophyll concentrations. While the methods are tested and validated for Irish lakes, the methodology has the potential to be applied in a global context.