Abstract Details
83 / 2023-07-28 21:48:45
Remote sensing detection of phytoplankton types assisted by AI technology: cases and thinking
Remote sensing,phytoplankton types,AI
Abstract Accepted
The role of phytoplankton species and group composition in the material cycle, energy flow, and biological resources of the marine ecosystem is played a vital role by them. Satellite remote sensing has the advantage of continuous coverage in time and space, making it the most effective means to detect the spatio-temporal distribution of phytoplankton community structure. At present, AI technology represented by machine learning/deep learning provides a new paradigm for discovering higher value information and knowledge rules, and offers more help in exploring phytoplankton types and composition at global and regional scales.
Three cases involving phytoplankton types discrimination assisted by AI technology are presented by our team, as follows: 1) The traditional empirical/semi-empirical relationship was replaced with a data-driven model. The XGBoost algorithm method based on singular value decomposition was incorporated into the modeling process based on in-situ measured remote sensing reflectance and phytoplankton group chlorophyll a concentration. The results showed that the coefficient of determination of diatom and dinoflagellate chlorophyll a concentration inversion models in the verification set was greater than 0.7; 2) With the help of the Transfer learning method, the bottleneck of small sample learning can be alleviated. Shallow-layer information network was constructed through deep-learning technology and then further combined with Transfer learning technology to improve the accuracy of phytoplankton species composition retrieval; 3) Big data and Ensemble learning were incorporated to improve model generalization performance. A spatiotemporal ecological ensemble (STEE) model for phytoplankton group retrieval was built using multiple marine environment variables such as global long-term ocean color satellite data, as well as marine physical, biogeochemical, and meteorological data over the past 20 years through AI technology. The global spatial and temporal distribution patterns of 8 major phytoplankton groups were retrieved with high accuracy.
With the diversity and richness of marine observation means, the type, volume, storage, and analysis of observation data increasingly reflect the prominent characteristics of Big data. On the basis of marine Big data, further combining the domain adaptive transfer learning method, automated machine learning model, and modern geostatistics theory, the introduction of multidimensional spatiotemporal modeling of phytoplankton groups is expected to solve the limitations of current bio-optical methods.
Three cases involving phytoplankton types discrimination assisted by AI technology are presented by our team, as follows: 1) The traditional empirical/semi-empirical relationship was replaced with a data-driven model. The XGBoost algorithm method based on singular value decomposition was incorporated into the modeling process based on in-situ measured remote sensing reflectance and phytoplankton group chlorophyll a concentration. The results showed that the coefficient of determination of diatom and dinoflagellate chlorophyll a concentration inversion models in the verification set was greater than 0.7; 2) With the help of the Transfer learning method, the bottleneck of small sample learning can be alleviated. Shallow-layer information network was constructed through deep-learning technology and then further combined with Transfer learning technology to improve the accuracy of phytoplankton species composition retrieval; 3) Big data and Ensemble learning were incorporated to improve model generalization performance. A spatiotemporal ecological ensemble (STEE) model for phytoplankton group retrieval was built using multiple marine environment variables such as global long-term ocean color satellite data, as well as marine physical, biogeochemical, and meteorological data over the past 20 years through AI technology. The global spatial and temporal distribution patterns of 8 major phytoplankton groups were retrieved with high accuracy.
With the diversity and richness of marine observation means, the type, volume, storage, and analysis of observation data increasingly reflect the prominent characteristics of Big data. On the basis of marine Big data, further combining the domain adaptive transfer learning method, automated machine learning model, and modern geostatistics theory, the introduction of multidimensional spatiotemporal modeling of phytoplankton groups is expected to solve the limitations of current bio-optical methods.
Important Date
-
Conference Date
Nov 02
2023
toNov 06
2023
-
Nov 01 2023
Contribution Submission Deadline
-
Nov 20 2023
Draft paper submission deadline
-
Nov 05 2024
Registration deadline
Sponsored By
Coastal Zones Under Intensifying Human Activities and Changing Climate: A
Regional Programme Integrating Science, Management and Society to Support
Ocean Sustainability (COASTAL-SOS)
Regional Programme Integrating Science, Management and Society to Support
Ocean Sustainability (COASTAL-SOS)
Organized By
State Key Laboratory of Marine Environmental Science, Xiamen University
College of Ocean and Earth Sciences, Xiamen University
China-ASEAN College of Marine Sciences, Xiamen University Malaysia
College of Ocean and Earth Sciences, Xiamen University
China-ASEAN College of Marine Sciences, Xiamen University Malaysia
Supported By
COASTAL-SOS
Contact Information
- Ying Huang
- hy******@xmu.edu.cn
- +86*********
Previous Conferences
-
2025-10-10 China Shenzhen
2025年南海年会SCSAM2025 -
2020-10-01 China 上海市(Shanghai)
South China Sea Annual Meeting
