Evaluating the potential of improved spatial resolution for remote sensing of chlorophyll-a in coastal and inland waters

Remote estimation of the photosynthetic pigment chlorophyll-a (chl-α), which can be used as an indicator of algal biomass has numerous applications, including to assist in the detection and monitoring of algal blooms. Harmful algal blooms pose a threat to human and ecosystem health and are increasing globally. Algal blooms impact socially and economically important inland and coastal waters, they exhibit large spatial variability and effective monitoring regimes pose a challenge for water managers. Chapter 1 contains a discussion of how satellite remote sensing of chl-α can assist in the detection and monitoring of such blooms. Chl-α algorithms developed for oceanic waters can be confounded by dissolved organic matter and turbidity common to inland and nearshore waters. Such waters of interest are often too small to be resolved by satellite platforms initially designed for global ocean applications. This thesis examines the capability of platforms of varying spatial resolution to detect increases in algal biomass through the remote estimation of chlorophyll-a. Chapter 2 contains an evaluation of the effectiveness of a high-resolution processing mode for Moderate Resolution Imaging Spectroradiometer (MODIS) derived chlorophyll-a retrieval algorithms in the coastal waters of Tasmania, Australia. Three aerosol correction models and chl-α retrieval algorithms were evaluated using both standard and high-resolution processing. Chl-α retrievals were evaluated in Bass Strait. Chlor_a, the default SeaDAS chl-α product, with the Management unit of the North Sea Mathematical models (MUMM) aerosol correction algorithm performed best (RMSE = 0.09 mg m\(^{-3}\); MAPE = 34%; R\(^2\) = 0.75). The fluorescence line height algorithm using Rayleigh corrected top of atmosphere reflectances (RMSE = 0.11 mg m\(^{-3}\), MAPE = 41%, R\(^2\) = 0.61) can provide an acceptable alternative in waters where full atmospheric correction is problematic. High-resolution processing of MODIS imagery improved spatial resolution but reduced chl-α retrieval accuracy. In Chapter 3, the capability of Landsat 8 operational land imager (OLI) was evaluated for detection and quantification of algal blooms in inland lakes. A partial least squares regression (PLSR) derived algorithm was calibrated and validated against in-situ chl-α estimates from Lake Trevallyn in Tasmania, Australia. The algorithm was developed against a calibration subset of the data and able to provide chl-α estimates strongly correlated with in-situ values when applied to the validation dataset (n = 10, NRMSEP = 21.6%; R\(^2\) = 0.67; NSE = 0.53; bias = -0.03 μg/L). These results demonstrate the suitability of PLSR algorithms and Landsat 8 OLI imagery to complement the capability of in-situ instrumentation for local water quality monitoring applications. The Sentinel 2 platform, with bands sampled to a spatial resolution of 20 m, was evaluated in Chapter 4. An improved semi-analytical model for chlorophyll-a (chl-α) retrievals was created by replacing a fixed chl-α specific absorption coefficient (a*) with a variable model. This method was applied to three Sentinel 2 images taken over the Lake Erie’s western basin correlating with an in-situ dataset of 24 samples. The improved algorithm produced chl-α retrievals with a 23% reduction in root mean squared error of prediction, an 85% reduction in bias and an increase in Nash-Sutcliffe efficiency of 7% over the default algorithm using a fixed a* value. The resulting strong correlation between in-situ and estimated chl-α (r = 0.95) suggests that the Sentinel 2 platform could be effective in the detection and mapping of high biomass algal blooms. In Chapters 2 to 4, clouds and the accuracy of atmospheric correction were identified as potential obstacles to the use of satellite platforms. Digital cameras found in mobile devices can be deployed below clouds and imagery is much less prone to atmospheric effects. In chapter 5, chl-α retrieval algorithms were developed for the University of Maine’s HydroColor app using partial least squares regression with band ratio inputs indicative of optically active water constituents. The app estimates remote sensing reflectance (Rrs) in the red, green and blue bands and derives turbidity and suspended particulate matter. The models were evaluated against a dataset of 49 HydroColor image sequences along with in-situ chl-α and turbidity measurements collected over a one-year period. The chl-α algorithm using the ratio of Rrs(blue)/ Rrs(green) was a poor predictor of in-situ chl-α (R\(^2\) = 0.11; bias = 0.21 mg m\(^{-3}\); NSE = -0.02). Adding inputs representing absorption due to CDOM and turbidity improved algorithm performance (R\(^2\) = 0.56 to 0.60; bias = 0.11 to 0.15 mg m\(^{-3}\); NSE = 0.39 to 0.46). The resulting algorithms appear capable of detecting relative changes in chl-α concentration, however would benefit from further development before being applied to active bloom detection and monitoring systems. The outcomes of the research demonstrate that recently deployed satellite platforms, particularly Sentinel 2, have the capability to improve bloom detection and monitoring. Sentinel 2’s 20m spatial resolution and 5-day revisit time could provide regular mapping of chl-α concentration in the many nearshore and inland waters that could not previously be resolved by platforms designed for global ocean applications. Such data could complement existing manual sampling regimes and in-situ instrumentation to identify blooms and further understand bloom dynamics. Digital cameras already have the potential to detect changes in chl-α concentration. If this can be further improved, then both regular and “on-demand” mapping of bloom areas could be performed using unmanned aerial vehicles at an even greater spatial resolution than available on satellite platforms