1. 연사: Arizona대학교 Qurat Ul An Sabir 교수님

2. 주제 : Image-Based CNN Modeling for Chlorophyll-a Estimation from Drone Imagery: Improving Harmful Algal Bloom Monitoring in Nova Scotia Lakes

3. 날짜: 추후 공지

4. 장소: 사이언스홀(자1-100)

5. 발표 초록:

Harmful algal blooms (HABs) threaten Nova Scotia’s freshwater lakes amid nutrient pollution and climate change, demanding quick, cost-effective chlorophyll-a (Chl-a) monitoring as a bloom proxy. This study develops an image-based system that uses green-channel extraction to relate drone imagery to in-situ chlorophyll-a (Chl-a) measurements, log transformation, normalization, and regression calibration for real-time prediction and status classification. Traditional color-intensity regression achieves moderate accuracy (typically R² 0.6–0.8, RMSE 5–15 μg/L in comparable lake studies) but is confounded by reflections, sediments, and non-algal patterns, leading to bias and limited generalization. We are developing an end-to-end Convolutional Neural Network (CNN) that extracts HAB-specific textures (scum mats, clumps, streaks) from raw drone imagery, capturing nonlinear relationships for improved performance. Trained and validated on Nova Scotia Lake drone datasets paired with field-measured Chl-a, the CNN is expected to reduce RMSE by 20–50% compared to baselines, using metrics such as R², RMSE, MAE, cross-validation, and uncertainty quantification. This AI-driven approach enables scalable and cost-effective monitoring using drones, smartphones, and satellites to support proactive freshwater management.