Status message

COVID-19 Updates: Visit the University's Safe Campus Website

Assessing and Forecasting Chlorophyll Abundances in Minnesota Lakes using Remote Sensing and Statistical Approaches

Authors: Ben Von Korff, Bryce Hoppie, Guarionex Salivia, Fei Yuan, Minnesota State University Mankato; Leif Olmanson, University of Minnesota; Benjamin Page, US Geological Survey (USGS)

Abstract: Harmful algae blooms (HABs) can negatively impact water quality, lake aesthetics, and can harm human and animal health. However, monitoring for HABs is rare in Minnesota. Detecting blooms which can vary spatially and may only be present briefly is challenging. Expanding monitoring in Minnesota would require the use of new and cost-efficient technologies. We use unmanned aerial vehicles (UAVs) for bloom mapping using RGB and near-infrared imagery, and trail cameras and water quality sondes for real time monitoring and time series forecasting of HABs. Normalized Difference Vegetation Index (NDVI) is positively correlated to chlorophyll-a, while Visible Water Residence Index (VWRI) shows marginal correlation with chlorophyll-a. While RGB cameras (trail cameras or UAVs) are useful for visual inspection and spotting blooms, these results suggest that quantitative mapping of chlorophyll-a in Minnesota Lakes should use near-infrared cameras at a minimum. Univariate time series forecasts using sonde chlorophyll-a data are compared using classical (ARIMA, wavelet-ARIMA) and machine learning techniques (LSTM, wavelet-LSTM). Chlorophyll-a positively correlates to temperature and precipitation, while is negatively correlated to conductivity and turbidity. Peak summer chlorophyll-a concentrations also appear to be positively correlated to recent precipitation totals. The accuracy of univariate forecasts is compared to multivariate forecasting (LSTM and wavelet-LSTM) using conductivity, turbidity, temperature, and precipitation as predictors to examine the practicality of bloom forecasting using a lower cost monitoring setup.