Assessment of Drivers of Algal Biomass in North American Great Lakes via Satellite Remote Sensing
Loading...
Date
2024-04-02
Authors
Dallosch, Michael
Journal Title
Journal ISSN
Volume Title
Publisher
University of Waterloo
Abstract
Lakes are regarded as sentinels of change, where shifts in environmental conditions significantly affect lake phenology. A significant consequence of the change is the perceived increase in the frequency, magnitude, and severity of algal blooms in lakes globally. Algal blooms/increased productivity in lakes pose significant ecological, economic and health risks, impacting fisheries, tourism, and freshwater access. The impacts of external nutrient loading from anthropogenic sources are well documented; however, blooms have been observed to occur in even remote lakes. Climate change is a hypothesized driver of these recent algal bloom trends, such as increasing global air temperatures, water temperatures, lake ice loss, precipitation intensity, and drought. Past research on the impact of climatic drivers on algal biomass dynamics has often been limited to lab, mesocosm, or short termed observations, due to limited in situ data. New remote sensing data products make use of historic multispectral satellite image archives to provide greater spatial and temporal coverage of algal biomass concentrations, allowing for longer time series observational studies to be conducted over large areas. Using data provided by the European Space Agency (ESA) Climate Change Initiative (CCI) Lakes project (product version 2.0.0), daily chlorophyll-a (chl-a; proxy of algal biomass), Lake Surface Water Temperature (LSWT) and Lake Ice Cover (LIC) from 2002 to 2020 were derived from five North American Great Lakes: Great Bear Lake (GBL), Great Slave Lake (GSL), Lake Athabasca (LA), Lake Winnipeg (LW), and Lake Erie (LE). Additional atmospheric and lake physical variables were provided by the European Centre for Medium-Range Weather Forecasts (ECMWF) ERA5-Land data as part of the ERA5 climate reanalysis product including: 2-m air temperature (T2m), Total Precipitation (PPT), Surface Net Solar Radiation (SNSR), Surface Runoff (SR) and Subsurface Runoff (SSR), Wind Speed (WS) and Lake Mix-Layer Depth (LMLD). Such data products allow for comprehensive time series analysis on the interaction effects of atmospheric and lake physical parameters on algal biomass dynamics.
Winter temperatures exhibit the highest rate of change relative to other seasons, where LIC loss is important for Northern hemisphere lakes; however, its effect on algal biomass dynamics is relatively unknown. To investigate how LIC duration alters algal biomass in North American Great Lakes, annual and seasonal algal biomass, LSWT and LIC parameters were calculated for the five study lakes using ESA CCI Lakes data. Algal biomasses (β = 0.01 – 0.75 μg L-1 yr-1) and LSWT (β = 0.03 – 0.14 K yr-1) were found to increase, with a general decrease in LIC (β = -0.88 – -1.08 Days yr-1) from 2002 to 2020. Vector autoregressions (VARs) showed that in Northern Lakes (NL; GBL, GSL and LA), LSWT and LIC parameters provide greater explanatory power for annual/seasonal chl-a concentrations (median adj. r2 = 0.75) compared to Southern Lakes (median adj. r2 = 0.46). Additionally, LIC parameters were found to provide higher explanatory power for NLs during the spring season compared to LSWT. However, higher explanatory power does not indicate predictive capacity, where machine learning methods may provide stronger predictive models.
To determine if LIC may act as a predictor of algal biomass parameters, multiple linear regression (MLR) and artificial neural networks (ANN) were constructed using per-pixel observations of annual/seasonal algal biomass, LSWT, and LIC parameters. Irrespective of season, LSWT only models returned lower prediction error (median NRMSE = 0.82) compared to LIC only models (median NRMSE = 0.93). However, models consisting of both LIC and LSWT returned the lowest predictive error (median NRMSE = 0.75). While LIC did not act as a strong predictor of algal biomass, a random forest (RF) classifier was used to determine whether LIC could classify the presence of lake-specific anomalies in chl-a concentrations. The RF model found that LIC parameters (ice on/off) had the highest mean accuracy decrease on average for NLs during the spring season. LIC timings are changing, where it was found to have greater importance on springtime abnormal algal biomass growth in NLs. While LIC was important at this time compared to LSWT, the impact of other important atmospheric and lake physical variables on algal biomass dynamics are not well understood, particularly at a smaller temporal scale (i.e., daily).
To assess the potential interaction effects between algal biomass, atmospheric, and lake physical parameters, a network analysis was conducted using a High Order Dynamic Gaussian Bayesian Network (HO-DGBN) for the original time series, the stationary, non-stationary, and residual signals at varying temporal ranges (Δ: daily, three days, weekly, biweekly, and monthly averages). It was found that LSWT, T2m and SNSR were the most important parameters on average, where LSWT exhibited the highest importance on the daily scale compared to the monthly. Additionally, LMLD returned increased importance at longer temporal frequencies, while SSR returned increased importance at shorter temporal frequencies. Temperature interactions were mixed, typically returning both positive and negative interactions, while SNSR typically exhibited a positive interaction with chl-a, while LMLD exhibited a frequent negative interaction. PPT and WS were found to be the least important parameters in all study lakes.
This thesis provides some of the first analytical uses of the ESA CCI Lakes product; a product that undergoes regular updates (every two years or so) as new satellite and in situ data become available, and algorithms for the retrieval of chl-a, LSWT and LIC are being improved. As such, improvements are expected in future releases of the product, limiting the accuracy of some findings in the thesis. Of the data presented, there is evidence that LIC is a significant contributor to spring algal biomass dynamics for NLs; however, Southern Lakes (SL; LW and LE) exhibit more complex interactions, likely due to anthropogenic impacts. This thesis identifies the complexity of LSWT interactions with algal biomass and identifies LMLD as a predominantly negative effect in the development of algal biomass. Algal biomasses are increasing, where increases in LSWT yield higher algal biomass peaks (at varying times throughout the year) within the study lakes. Future climate scenarios may provide conditions favorable for algal biomass growth, where Northern landscapes are at the greatest risk.
Description
Keywords
great lakes, limnology, algae, phytoplankton, remote sensing, time series, machine learning, bayesian networks, chlorophyll-a, lake ice, temperature, algal biomass dynamics