Secchi disk transparency; remote sensing; Google Earth Engine; random forest; stepwise multiple linear regression


Chemistry | Life Sciences


Sean Woznicki


Lake clarity is an important indicator for water quality and can be used to estimate the trophic state of a water body. Citizen science monitoring programs collect substantial water quality data. However, it is not feasible to physically measure hundreds to thousands of water bodies across large regions. Operational remote sensing can be used to fill this gap in unmonitored lakes based on the relationship between surface reflectance and in-situ Secchi disk transparency. In this study, we used Landsat 8 surface reflectance data and inter-lake watershed land use to develop statistical models that predict Secchi disk transparency for more than 6,500 Michigan lakes. We compared stepwise linear regression and random forest regression modeling methods, building complete models for the entire in-situ measurement dataset from 2013-present (June 15-September 15), and one-day “operational” models based on a single day of Landsat coverage and matching measurements. Predictions were made that can be used to qualitatively estimate trophic state but model error limits quantitative estimation of transparency. The random forest model was superior to the stepwise linear regression model for the entire data catalog, but they were similar when applied to a single day of Landsat coverage.

Available for download on Friday, January 05, 2024