Mapping the probability of freshwater algal blooms with various spectral indices and sources of training data

Algal blooms are pervasive in many freshwater environments and can pose risks to the health safety of humans other organisms. However, monitoring tracking potentially harmful often relies on in-person observations by public. Remote sensing has proven useful augmenting situ algal concentration, but hurdles hinder efficient application end users. First, numerous approaches estimate aquatic chlorophyll-a available produce inconsistent results. Second, lack quantitative limits opportunities train models for specific waterbodies, such that developed systems must be used instead. We (1) implement univariate multivariate logistic regression probability concentrations exceed an accepted threshold beyond which effects become likely (2) evaluate use visually classified bloom/no-bloom satellite imagery augment training data. Using a binary classification exceeding 10 μg / L, we found were ∼80 % accurate, trained with data nearly same accuracy (79%) as (80%), (3) visual classifications outperformed (82% accuracy) alone (80% accuracy). These results provide framework evaluating multiple spectral indices retrieving bloom presence or absence illustrate derived directly from observations.