Operationalizing a flexible modeling and data collection workflow to detect harmful algal blooms using hyperspectral imagery and machine learning in data-poor freshwater environments

Objective:

This project aims to advance HAB detection and species classification capabilities using hyperspectral data.

Approach:

This project will deploy ground-truth sensor systems and obtain water samples during active HAB events in both UT and VT, in coordination with satellite and/or drone hyperspectral image acquisitions. We will then conduct toxin and DNA analysis on all field samples to pair this with hyperspectral data and imagery. Using these paired data sets, we will continue to develop AI-driven HAB prediction and species identification models for deployment with hyperspectral imagery collected over freshwater bodies in the future.

Impact:

Abstract:

Harmful algal blooms (HABs) are a growing concern in freshwater ecosystems, as they can produce toxins that threaten public health and aquatic life. While recent advances in hyperspectral satellite technology offer promising tools for detecting these blooms from space, current monitoring approaches are limited by the lack of timely, high-quality ground data necessary to validate satellite observations and build accurate detection models. This project addresses that gap by developing an operational framework that coordinates satellite and drone hyperspectral imaging with field data collection to improve HAB detection and species identification using artificial intelligence (AI).

Our research will produce a data-driven workflow that pairs high-resolution hyperspectral imagery with water quality data and physical samples. This integrated dataset will serve as the foundation for developing advanced machine learning models—including both linear and nonlinear “spectral unmixing” approaches—that can detect and classify HAB-causing cyanobacteria species in water bodies.

Current limitations:

Field and satellite data are rarely collected simultaneously, limiting model accuracy.

Most existing models use linear assumptions that do not reflect the complexity of water environments.

Advancements through this project:

Coordinate hyperspectral data acquisition with on-the-ground sampling for robust model training.

Develop and test AI models, including deep learning methods, for improved species-level HAB detection.

Create open-source tools, datasets, and guidelines for agencies conducting HAB monitoring.

The impact of this project will be broad and meaningful: it will accelerate the development of reliable HAB detection systems, improve warning capabilities for communities relying on recreational and drinking water sources, and support national efforts to expand water quality monitoring using cutting-edge technology. Findings and tools from this work will be shared with scientific, government, and public stakeholders to promote widespread adoption and improve ecosystem health management.