Comparative Analysis of Empirical and Machine Learning Models for Chl<i>a</i> Extraction Using Sentinel-2 and Landsat OLI Data: Opportunities, Limitations, and Challenges
Amir M. Chegoonian,
Peter R. Leavitt,
Helen M. Baulch,
Claude R. Duguay
Canadian Journal of Remote Sensing, Volume 49, Issue 1
Remote retrieval of near-surface chlorophyll-a (Chla) concentration in small inland waters is challenging due to substantial optical interferences of various water constituents and uncertainties in the atmospheric correction (AC) process. Although various algorithms have been developed to estimate Chla from moderate-resolution terrestrial missions (∼10–60 m), the production of both accurate distribution maps and time series of Chla has proven challenging, limiting the use of remote analyses for lake monitoring. Here, we develop a support vector regression (SVR) model, which uses satellite-derived remote-sensing reflectance spectra (Rrsδ) from Sentinel-2 and Landsat-8 images as input for Chla retrieval in a representative eutrophic prairie lake, Buffalo Pound Lake (BPL), Saskatchewan, Canada. Validated against in situ Chla from seven ice-free seasons (N ∼ 200; 2014–2020), the SVR model outperformed both locally tuned, Rrsδ-fed empirical models (Normalized Difference Chlorophyll Index, 2- and 3-band, and OC3) and Mixture Density Networks (MDNs) by 15–65%, while exhibiting comparable performance to a locally trained MDN, with an error of ∼35%. Comparison of Chla retrieval models, AC processors (iCOR, ACOLITE), and radiometric products (Rayleigh-corrected, surface, and top-of-atmosphere reflectance) showed that the best Chla maps and optimal time series (up to 100 mg m−3) were produced using a coupled SVR-iCOR system.
Abstract In the Canadian prairies, eutrophication is a widespread issue, with agriculture representing a major anthropogenic nutrient source in many watersheds. However, efforts to mitigate agricultural nutrient export are challenged by the lack of coordinated monitoring programs and the unique hydrological characteristics of the prairies, notably, the dominance of snowmelt in both water flows and nutrient loads, variable runoff, variable contributing area and the issues of understanding how scale affects nutrient concentrations and prevalence of dissolved nutrient transport (over total nutrients). Efforts are being made to integrate these characteristics in process-based water quality models, but the models are often complex and are not yet ready for use by watershed managers for prioritizing implementation of beneficial management practices (BMPs). In this study, a screening and scoping approach based on nutrient export coefficient modeling was used to prioritize BMPs for the 55,700 km2 Qu’Appelle Watershed, Saskatchewan. By integrating land use information, in-stream monitoring data, stakeholder input and nutrient export coefficient modeling, the study assessed potential efficiencies of six BMPs involving fertilizer, manure, grazing, crop and wetland management in nutrient load reductions for nine tributaries of the watershed. Uncertainty around the effectiveness of the BMPs was assessed. Field-level export coefficients were adjusted with nutrient delivery ratios for estimating watershed-level exports. Of the BMPs examined, in general, wetland restoration had the greatest potential to reduce both nitrogen and phosphorus loads in most tributaries, followed by fertilizer management. The importance of wetland restoration was supported by positive, significant, linear correlations between nutrient delivery ratios and drainage intensity in the tributaries (nitrogen: R 2 = 0.67; phosphorus: R 2 = 0.82). Notably, the relative ranking of BMP efficiencies varied with tributaries, as a result of differing landscape characteristics, land uses and nutrient inputs. In conclusion, the approach developed here acknowledges uncertainty, but provides a means to guide management decisions within the context of an adaptive management approach, where BMP implementation is partnered with monitoring and assessment to revise ongoing plans and ensures selected practices are meeting goals for nutrient abatement.
Water quality models are an emerging tool in water management to understand and inform decisions related to eutrophication. This study tested flow scenario effects on the water quality of Buffalo Pound Lake—a eutrophic reservoir supplying water for approximately 25% of Saskatchewan’s population. The model CE-QUAL-W2 was applied to assess the impact of inter-basin water diversion after the impounded lake received high inflows from local runoff. Three water diversion scenarios were tested: continuous flow, immediate release after nutrient loading increased, and a timed release initiated when water levels returned to normal operating range. Each scenario was tested at three different transfer flow rates. The transfers had a dilution effect but did not affect the timing of the nutrient peaks in the upstream portion of the lake. In the lake’s downstream section, nutrients peaked at similar concentrations as the base model, but peaks arrived earlier in the season and attenuated rapidly. Results showed greater variation among scenarios in wet years compared to dry years. Dependent on the timing and quantity of water transferred, some but not all water quality parameters are predicted to improve along with the water diversion flows over the period tested. The results suggest that it is optimal to transfer water while local watershed runoff is minimal.
Haig HA, Chegoonian AM, Davies J-M, Bateson D, Leavitt PR. 2021. Marked blue discoloration of late winter ice and water due to autumn blooms of cyanobacteria. Lake Reserv Manage. XX:XXX–XXX.Continu...