A novel algorithm to monitor cyanobacterial blooms in Lake Taihu from HJ-CCD imagery

Nai Zhaojun, Hongtao Duan, Zhu Li, Zhigang Cao, Xiaoning Chen, Yuchao Zhang, Ronghua MaJournal of Lake Sciences 2016 · DOI

Abstract

Cyanobacterial blooms are common occurrences in eutrophic lakes, and it's urgent to monitor their area and extent effectively by remote sensing. HJ-CCD satellite data has high spatial and temporal resolution and is suitable for detecting cyanobacterial blooms over long periods. In this study, relative radiometric correction of HJ-CCD images was processed using the Automatic Scattergram-Controlled Regression (ASCR) method to eliminate radiometric problems. Then, a high-precision extraction algorithm for cyanobacterial blooms was developed by combining the Normalized Difference Vegetation Index (NDVI) and the Algae Pixel-growing Algorithm (APA). This method provides a uniform threshold for different HJ-CCD images and can extract bloom information at the sub-pixel level. Monitoring Lake Taihu from 2009 to 2014 revealed that the area of cyanobacterial blooms decreased during 2013 and 2014. The results indicate that this algorithm identifies cyanobacteria blooms effectively and automatically with high accuracy, making it suitable for operational use.

Introduction

Cyanobacterial blooms, often referred to as "blue-green algae," represent a significant environmental crisis for eutrophic lakes such as Lake Taihu. These events degrade water quality, threaten drinking water safety, and disrupt aquatic ecosystems. While remote sensing is an essential tool for monitoring these massive events, traditional methods face two major hurdles: the "threshold headache," where varying atmospheric conditions require manual per-image adjustments, and the "resolution gap," where sensors like MODIS lack the spatial detail (250m–1km) to capture smaller blooms, while high-resolution sensors like Landsat have infrequent revisit times.

To address these challenges, this study utilizes imagery from the Chinese HJ-1A/1B CCD sensors. These sensors offer a "sweet spot" for monitoring: a 30m spatial resolution and a 2-day revisit period. The research aims to develop an automated, high-precision pipeline for long-term monitoring across varying lighting and atmospheric conditions.

Methods

The proposed monitoring pipeline consists of several integrated stages designed for operational efficiency and automation:

  1. Relative Radiometric Correction (ASCR): To ensure comparability between images from different dates, the Automatic Scattergram-Controlled Regression method was used. This technique identifies invariant pixels (such as deep water or stable land) to normalize brightness and contrast across the dataset.
  2. NDVI for Initial Detection: The Normalized Difference Vegetation Index (NDVI) exploits the high reflectance of algae in the Near-Infrared (NIR) spectrum compared to the Red spectrum to provide an initial assessment of bloom presence.
  3. Algae Pixel-growing Algorithm (APA): This core innovation moves beyond binary classification. By using a linear mixing model, the APA estimates the specific fraction of a 30m pixel covered by algae. It begins with "pure" bloom pixels and "grows" the detection into mixed pixels, achieving sub-pixel accuracy.
The operational flowchart of the proposed cyanobacterial bloom monitoring algorithm using HJ-CCD data.

By standardizing radiometric properties, the authors derived a stable universal NDVI threshold (-0.054 to -0.070) that eliminates the need for manual scene-by-scene calibration.

Results and Discussion

The algorithm was applied to 55 cloud-free HJ-CCD images of Lake Taihu spanning 2009 to 2014. The results were validated against traditional manual interpretation and established indices such as the Floating Algae Index (FAI) and the Virtual Cloud Index (VCI).

Comparison of bloom extraction results demonstrating the precision of the proposed method (d) against other indices.

Key findings include:

  • Long-term Trends: The study mapped maximum bloom extents and identified a noticeable decrease in bloom area during 2013 and 2014 compared to the severe outbreaks seen between 2010 and 2012.
  • Seasonal Dynamics: The data confirmed that blooms typically peak in summer and autumn (July–October), correlating with high temperatures and nutrient availability.
  • Sub-pixel Precision: The APA's ability to estimate percentage coverage (e.g., identifying a pixel as 45% algae rather than a binary "yes/no") provided a significantly more accurate estimate of total biomass and affected area.

Conclusion

The combination of ASCR and the Algae Pixel-growing Algorithm provides a robust, automated solution for monitoring lake health. While persistent cloud cover remains a challenge for optical sensors, the high temporal resolution of the HJ satellite and the universal thresholding of the APA represent a major step toward a reliable, large-scale early warning system for cyanobacterial blooms. Future work will focus on cross-calibrating this algorithm for other high-resolution sensors like Sentinel-2 and Landsat-8.

Summary updated

Abstract

水体富营养化引起的蓝藻水华问题,是我国湖泊面临的主要环境问题,亟需加强现状监测和变化研究;我国自主研发的环境(HJ)卫星空间分辨率高,重访周期短,可用于长时间序列蓝藻水华的动态监测.本文利用HJ卫星CCD数据,通过自动控制散点回归的方法进行相对辐射校正,再将归一化植被指数和像元生长算法相结合,提出了一种可业务化运行的蓝藻水华高精度提取算法.该算法的优点为:(1)水华提取时具有统一的阈值,解决了以往一景影像一个阈值,无法大规模批处理的难题;(2)通过对像元进行线性分解,精度可达到亚像元级别.利用该算法对太湖2009-2014年蓝藻水华进行监测,发现2013-2014年太湖蓝藻水华较以往暴发面积偏小.研究表明,该算法对蓝藻水华识别能力强,自动化程度和水华提取精度高,可作为业务化算法运行.;Cyanobacterial blooms are common occurrences in eutrophic lakes, and it's urgent to monitor their area and extent effectively by remote sensing. HJ-CCD satellite data has high spatial and temporal resolution and is suitable for detecting cyanobacterial blooms of long time. In this study, relative radiometric correction of HJ-CCD images was processed using the Automatic Scattergram-Controlled Regression method to eliminate the radiometric problems. Then, a high-precision extraction algorithm of cyanobacterial blooms was put forward by combination of Normalized Difference Vegetation Index and Algae Pixel-growing Algorithm. This method provides an uniform threshold for the different HJ-CCD images and can extract blooms information of sub-pixel. Using this algorithm to monitor cyanobacterial bloom in Lake Taihu from 2009 to 2014, we found that the area of cyanobacterial blooms decreased during 2013 and 2014. The result indicated that this algorithm could identify the cyanobacteria bloom effectively and automatically with high accuracy.
BibTeX
@article{zhaojun2016novel,
  author = {NAI Zhaojun and Hongtao Duan and Zhu Li and Zhigang Cao and Xiaoning Chen and Yuchao Zhang and Ronghua Ma},
  title = {{A novel algorithm to monitor cyanobacterial blooms in Lake Taihu from HJ-CCD imagery}},
  journal = {Journal of Lake Sciences},
  year = {2016},
  doi = {10.18307/2016.0319},
}
Figure