Abstract
Water clarity has been extensively assessed in Landsat-based remote sensing studies of inland waters, regularly relying on locally calibrated empirical algorithms, and close temporal matching between field data and satellite overpass. As more satellite data and faster data processing systems become readily accessible, new opportunities are emerging to revisit traditional assumptions concerning empirical calibration methodologies. Using Landsat 8 images with large water clarity datasets from southern Canada, we assess: (1) whether clear regional differences in water clarity algorithm coefficients exist and (2) whether model fit can be improved by expanding temporal matching windows. We found that a single global algorithm effectively represents the empirical relationship between in situ Secchi disk depth (SDD) and the Landsat 8 Blue/Red band ratio across diverse lake types in Canada. We also found that the model fit improved significantly when applying a median filter on data from ever-wider time windows between the date of in situ SDD sample and the date of satellite overpass. The median filter effectively removed the outliers that were likely caused by atmospheric artifacts in the available imagery. Our findings open new discussions on the ability of large datasets and temporal averaging methods to better elucidate the true relationships between in situ water clarity and satellite reflectance data.
Highlights
Obtaining a global perspective of changing freshwater quality is crucial in managing the multiple essential water resource uses in the face of contemporary shifting climate and land-use dynamics
Spatial constraints that are introduced by the large pixel sizes of the historic ocean color sensors have been overcome by applications of higher resolution sensors, such as those aboard the SPOT [4,5], Landsat [6,7,8,9,10], and, most recently, Sentinel-2 [6,11,12] satellites, and CubeSat constellations, such as PlanetScope [13,14,15], despite band designations for these sensors being optimized for extracting information from
We looked for significant improvement in model fit (R2, root mean square error (RMSE) and bias) to adopt the more complex algorithm forms, as we are primarily interested in comparable and general patterns of model structure in this study
Summary
Obtaining a global perspective of changing freshwater quality is crucial in managing the multiple essential water resource uses in the face of contemporary shifting climate and land-use dynamics. The constraints of in situ lake monitoring, including access to remote locations and effective funding for large and comprehensive sampling programs, have been increasingly overcome through low-cost satellite remote sensing applications to map broadscale freshwater-quality trends in space and time. This is true as more satellite imagery becomes freely available and as tools, such as Google Earth Engine [1], provide platforms for mass processing of image data to effectively assess large-scale patterns. Optical complexity constraints to traditional bio-optical modeling, including more convoluted optical signals from the diverse water column constituents of Type II waters, as well as a lack of sufficient data on inherent optical properties in many locations, has resulted in the wide application of empirical modeling to extract freshwater quality information from satellite imagery [7,8,10,16,17]
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
