​Landsat 8 Lake Water Clarity Empirical Algorithms: Large-Scale Calibration and Validation Using Government and Citizen Science Data from across Canada

Picture
Figure 1. Histograms showing the frequency of unique sampling points per SDD (m). (A) shows the color coded ecozone map corresponding to each of the following graphs. (B–F) show histograms for each of the ecozones assessed in this study.
Picture
Figure 6. Standardized (on map) and unstandardized residuals (on histogram) for global regression model between in situ SDD and Landsat 8 Blue/Red for data taken within a one-week temporal window of each other.
Picture
Figure 8. Predicted versus observed plot for SDD regression time window models. Points represent the mean predicted value to the observed value for the validation dataset (Lake Pulse data). Green represents the Pacific Maritime and Montane Cordilleran, maroon represents the Boreal Plain and Prairies, purple represents the Boreal Shield, orange represents the Mixed Wood Plains, and blue represents the Atlantic Maritime.
Background

Canada holds over 50% of the world's natural lakes, yet the water quality in the vast majority of these water bodies remains uncharacterized due to the sheer number of lakes and accessibility issues in remote locations. We focused on water clarity, specifically measured as Secchi disk depth, because it serves as a critical indicator of lake health, biological productivity, and ecosystem services. Traditional remote sensing has relied on "near-simultaneous" field data, requiring measurements to be taken within days of a satellite overpass. We set out to challenge these restrictive temporal assumptions and determine if the extreme geographical diversity of Canadian ecozones truly required different mathematical models for accurate monitoring.

Approach

We conducted an extensive data compilation effort, merging records from provincial government and citizen science programs with imagery from the Landsat 8 satellite. To ensure the highest level of accuracy, we validated our findings using high-quality data from the NSERC Canadian Lake Pulse Network, a three-year national assessment designed to characterize lake health. Our methodology involved testing the relationship between in-person Secchi disk measurements and the ratio of blue to red light reflected in satellite images. Specifically, we pioneered a median filtering technique, where we averaged satellite data over expanded time windows—ranging from a single day to several years—to filter out common artifacts like haze or smoke that often obscure individual images.

Key Findings

We demonstrated that using large-scale data and temporal filtering produces a robust understanding of freshwater systems across the country:
  • A Single Global Algorithm Model: We discovered that regional ecozones do not require separate mathematical models; a single global algorithm effectively represents the relationship between satellite signals and water clarity across diverse landscapes, from the rocky Boreal Shield to the fertile Prairies.
  • Power of Temporal Averaging: Contrary to common practice, our model fit improved significantly when we expanded the time window between field sampling and satellite imaging. By using the median value of all summer observations over a multi-year period, we successfully removed atmospheric noise and achieved much stronger correlations.
  • Threshold Detection: While the model is most effective for clear lakes, we proved it can reliably identify when a lake has crossed critical thresholds of high turbidity or low clarity (less than 1 meter), providing a vital screening tool for environmental managers.

Impact

This research provides water managers with the first computationally efficient tools to track changes in water quality across nearly a million lakes. By proving that universal models and flexible timing are scientifically valid, we have unlocked the potential of decades of citizen science and government records for national-scale monitoring. Our work establishes a vital baseline for Canadian freshwater health, supporting better-informed policy decisions and safeguarding invaluable aquatic ecosystems in the face of shifting climate and land-use dynamics.

Resources

Published Paper: Deutsch* ES, Cardille JA, Koll-Egyed* T, Fortin MJ. Landsat 8 lake water clarity empirical algorithms: large-scale calibration and validation using government and citizen science data from across Canada. Remote Sensing. 2021 Jan;13(7):1257. DOI: 10.3390/rs13071257

Data Product Repository: NSERC Canadian Lake Pulse Network