Fusing multi-scale imagery to generate CyanoHAB epoch metrics

Session: Remote Sensing, Visualization, and Spatial Data Applications for the Great Lakes (4)

Nathan Torbick, AGS, [email protected]

Abstract

Consistent time series Cyanobacterial Harmful Algal Bloom (CyanoHAB) indicators do not exist for most inland lakes. Landsat archives provide an opportunity to develop "indicators" back to the 1980s although the resolutions have limitations for automated CyanioHAB retriel. Forward looking, ESA' s Sentinel 2 and 3 provide a mechanism for imporved retreivel given their spectral configuration. Combining these platforms in a robust and automated approach is one pathway to developing more comprehensive CynaoHAB monitoring where we have optimal beldning of resolutions. In this research application we develop an approach to fuse these sensors in an attempt to generate summary CyanoHAB epoch metrics using Lake Erie as a case study. We apply a rapid harmonization approach for preprocessing Landsat and Sentinels that corrects for Rayleigh scattering based on a 6S LUT. Next, spatiotemporal fusion is executed using a multi-temporal regression framework to extract optimal spatial and temporal resoolutions. Lastly, we apply machine learning techniques based on principals of physical relationships and 2,313 chlorophyll-a observations, to generate time series metrics (chl-a and Phycocyanin concentrations) using Lake Erie as a test bed. Chl-a concentration and bloom conditions had an overall accuracy and mean out-of-bag error rate of 0.73 and 0.87, respectively. All routines are based on open access elements.