ESTIMATING RIVER CHLOROPHYLL A CONCENTRATIONS FROM LANDSAT IMAGERY USING A MACHINE LEARNING APPROACH
While controls on phytoplankton abundance have been long studied in lakes and oceans, we currently lack a mechanistic understanding of how these drivers combine as water traverses the more physically complex settings of river networks. Testing hypotheses of these controls will require broad-scale measurement of the timing and distribution of water column chlorophyl-a (chl-a) in rivers. Estimating chl-a concentrations with satellite remote sensed imagery has the potential to create such near-continuous measurements. Rivers, however, carry high loads of additional optically active constituents, including colored dissolved organic matter, particulate organic matter, and inorganic sediment. They are also comparatively shallow, creating spectral signatures from bottom reflectance. Therefore remote chl-a measurement is an inherently difficult task in rivers, producing greater measurement error than in other aquatic systems. In this poster I assess the performance of flexible machine learning (ML) algorithms for predicting chl-a concentrations on river reaches >60 m in width using Landsat imagery and historic in-situ measurements of chl-a from the National Water Quality Portal. These ML techniques automate complex non-linear regression, and may provide the flexibility necessary for robust remote chl-a measurement.
Nicholas Bruns (Primary Presenter/Author), Duke University, email@example.com;
Matthew Ross (Co-Presenter/Co-Author), Colorado State University, firstname.lastname@example.org;
John Gardner (Co-Presenter/Co-Author), Duke University, email@example.com;
Jim Heffernan (Co-Presenter/Co-Author), Duke University, firstname.lastname@example.org;