CJEES

Home
Peer Review
Editorial Board
Instructions
Early Access
Latest Issue
Past Issues
Contact
Impact Factor
Reject Rate

 
You are here: Home » Online First » Volume 19, 2024 - Number 1 » USING MACHINE LEARNING ALGORITHMS FOR NATURAL HABITATS ASSESSMENT, Carpathian Journal of Earth and Environmental Sciences February 2024, Vol. 19, No. 1, p. 103 – 113; DOI:10.26471/cjees/2024/019/282


« Back

Monica Liliana MARIAN1, Daniel NASUI1*, Ciprian Radu GHISE2 , Flavia POPOVICI2, Cosmin SABO1, Rosca Oana MARE1, Lucia MIHALESCU1, Bogdan VASILESCU1, & Zorica VOSGAN1
1Technical University of Cluj Napoca, 430122, Baia Mare, Romania
2Indeco Soft, 430094, Baia Mare, Romania


USING MACHINE LEARNING ALGORITHMS FOR NATURAL HABITATS ASSESSMENT, Carpathian Journal of Earth and Environmental Sciences February 2024, Vol. 19, No. 1, p. 103 – 113; DOI:10.26471/cjees/2024/019/282

Full text

Abstract:

The potential of AI to process and interpret large volumes of data can provide researchers with a powerful tool to understand and monitor biodiversity on a global scale. In this paper we aimed to identify dominant individual plant species in natural protected habitats. Mapping the dominant species from the targeted natural habitats was followed by testing machine learning algorithm for differentiating similar species using satellite images. In the end we validated the data generated by machine learning algorithms through extensive field observations. Using the Sentinel-2 mission 10m resolution data and comprehensive field mapping we managed to see different phenology variations between diverse types of plant communities. Using the NDVI and NDII vegetation indexes and Random Forest algorithm during the dominant species phenology stages for each consecutive 10-day periods between May 1st and September 10th, revealed distinct responses to climate fluctuations and environmental factors. The natural habitats different signatures are strongly influenced by their ecological and conservation status and are not yet suitable for identification, but could help improve AI’s automatic detection for multiannual analysis if a favorable conservation trend is reached. The main achievement of the proposed methodology is the ability to differentiate between different species of deciduous trees, with machine learning training accuracy generally exceeding 95% and classification accuracy surpassing 90%.



Keyword: Biodiversity, Plant communities, Satellite images, Sentinel-2, NDVI, NDII, machine learning, Random Forest


(c) 2006 - 2024 , Publisher-Asociația Carpatică de Mediu și Științele Pământului (Carpathian Association of Environment and Earth Sciences)