A convolutional neural network regression for quantifying harmful cyanobacteria using hyperspectral imagery

Authors: PYO, JONGCHEOL - Ulsan National Institute Of Science And Technology (UNIST); DUAN, HONGTAI - Chinese Academy Of Sciences; BAEK, SANGSOO - Ulsan National Institute Of Science And Technology (UNIST); Kim, Moon; JEON, TAEGYUN - National Institute Of Environmental Research; KWON, YONGSUNG - Us Forest Service (FS); LEE, HUUK - National Institute Of Environmental Research; CHO, KYUNGHWA - Ulsan National Institute Of Science And Technology (UNIST)

Submitted to: Remote Sensing of Environment
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 7/25/2019
Publication Date: 8/10/2020
Citation: Pyo, J., Duan, H., Baek, S., Kim, M.S., Jeon, T., Kwon, Y., Lee, H., Cho, K. 2020. A convolutional neural network regression for quantifying harmful cyanobacteria using hyperspectral imagery. Remote Sensing of Environment. 233:111350. https://doi.org/10.1016/j.rse.2019.111350.
DOI: https://doi.org/10.1016/j.rse.2019.111350

Interpretive Summary: In recent years, airborne hyperspectral sensing has been used to remotely detect cyanobacteria distributions in river systems with high spatial and spectral resolution. Many bio-optical or spectral algorithms have been developed for use with hyperspectral imaging to estimate algal concentrations in freshwater, but achieving high detection accuracies is still a challenge owing to the biophysical complexity of inland water and to seasonal variations. We applied a convolutional neural network (CNN) to hyperspectral images of a river system to estimate the concentrations of phycocyanin (PC) and chlorophyll-a (Chl-a), and generated a phytoplankton pigment distribution map. The results showed improvement in PC and Chl-a simulations, with R2 > 0.86 and 0.73, respectively, compared to results from conventional optical algorithms. This study demonstrated that the CNN method has the potential to detect and quantify harmful cyanobacteria with high accuracy and thus may be an alternative to conventional bio-optical algorithms, particularly for dealing with seasonal variation in cyanobacteria blooms. These research results provide insightful information to environmental scientists seeking a remote means to rapidly assess distribution of cyanobacteria in inland water systems.

Technical Abstract: Remote sensing is useful for detecting and quantifying harmful cyanobacteria blooms for managing water systems. In particular, airborne hyperspectral remote sensing has an advantage in precise cyanobacteria detection with high spatial and spectral resolution. Many bio-optical algorithms have been developed and utilized to estimate algal concentration. However, achieving the optimal conventional optical model accuracy is still challenging in freshwater owing to the biophysical complexity of the inland water and the seasonal reflection of site-specific optical properties. Thus, this study applied convolutional neural network (CNN) with various input windows to estimate the concentrations of phycocyanin (PC) and chlorophyll-a (Chl-a), and generated a phytoplankton pigment map. Point-centered regression CNN (PRCNN) showed accurate PC and Chl-a simulations, with R2 > 0.86 and 0.73, respectively, and root mean square errors < 10 mg·m-3, which were smaller than the conventional optical algorithm. In addition, the generated PC and Chl-a map from PRCNN substantially followed the spatial distribution of the pigment and showed reasonable concentration levels. Here, we found that a small input size and deep spectral bands contributed to the CNN model to achieve strong capacity to reflect the dynamic spatial feature of phytoplankton pigments. Therefore, this study demonstrated that CNN regression has the potential to detect and quantify harmful cyanobacteria with high accuracy and can be an alternative to bio-optical algorithms, particularly for dealing with the seasonal variation in cyanobacteria blooms.