A Multiplatform Approach for Chlorophyll Level Estimation for Irish Lakes

Abstract

To overcome the obstacles from discontinuous detection by single satellites, we introduce a new approach for the derivation of lake chlorophyll levels via multisensor remote sensing reflectance. In this study, we used lakes throughout the Republic of Ireland as a test bed. In the first stage, three machine learning models (random forest, extreme gradient boosting, and support vector machine) were built directly between chlorophyll levels and remote sensing reflectance from Sentinel-2, Landsat-8, Moderate Resolution Imaging Spectroradiometer (MODIS) Terra, and MODIS Aqua. The results of these 12 single-sensor algorithms (3 machine learning methods × 4 remote sensing platforms) indicate that MODIS Aqua achieved the highest average performance metric, likely due to its design, which is specifically optimized for the derived water color. Then, a multiplatform model was built using the best individual model for each satellite combined using individual performance of each model. Our multiplatform model performs well with accuracies of 78% and 70% in the training and testing datasets, respectively. The model can also capture the spatial and temporal variations observed in the in situ observations . Our results also highlight that our multiplatform approach can provide an increase of 550% in the number of chlorophyll observations compared to the in situ measurements. These findings underscore the potential of both our approach and optical remote sensing for water quality monitoring, even in locations with small water bodies.