TY - JOUR
T1 - Advancing coastal habitat mapping in Bahrain: a comparative study of remote sensing classifiers
AU - Alkhuzaei, Manaf
AU - Brolly, Matthew
PY - 2024/3/4
Y1 - 2024/3/4
N2 - This study explores the veracity of remote sensing-based classification of Bahrain's coastal water habitats, referencing data from three prior studies conducted in Bahrain waters. The objective is to illuminate the limitations of remote sensing for habitat mapping and evaluate the proficiency of Maximum Likelihood, Support Vector Machines, and SoftMax Regression classifiers using expanded pre-processing methodologies. Our reanalysis yielded Maximum Likelihood accuracies ranging from 45 to 61%, significantly influenced by the choice of data source and ground truth selection. In contrast, the Support Vector Machines classifier consistently demonstrated superior accuracy, achieving up to 80% by utilizing diverse band combinations. Meanwhile, SoftMax Regression classifiers reached peak accuracies of 58–69%. These findings emphasize the critical importance of selecting robust data sources, comprehending classifier limitations, and ensuring accurate ground truth verification to enhance the reliability of habitat classification maps. Such maps are instrumental for effective coastal management, conservation efforts, and marine biodiversity assessments, facilitating sustainable resource use and conservation strategies. Given the rapid advancement in remote sensing technologies, optimizing the precision of habitat classification maps is paramount for sustainable coastal ecosystem management and biodiversity preservation, highlighting the relevance and significance of this study in the broader context of environmental research and management.
AB - This study explores the veracity of remote sensing-based classification of Bahrain's coastal water habitats, referencing data from three prior studies conducted in Bahrain waters. The objective is to illuminate the limitations of remote sensing for habitat mapping and evaluate the proficiency of Maximum Likelihood, Support Vector Machines, and SoftMax Regression classifiers using expanded pre-processing methodologies. Our reanalysis yielded Maximum Likelihood accuracies ranging from 45 to 61%, significantly influenced by the choice of data source and ground truth selection. In contrast, the Support Vector Machines classifier consistently demonstrated superior accuracy, achieving up to 80% by utilizing diverse band combinations. Meanwhile, SoftMax Regression classifiers reached peak accuracies of 58–69%. These findings emphasize the critical importance of selecting robust data sources, comprehending classifier limitations, and ensuring accurate ground truth verification to enhance the reliability of habitat classification maps. Such maps are instrumental for effective coastal management, conservation efforts, and marine biodiversity assessments, facilitating sustainable resource use and conservation strategies. Given the rapid advancement in remote sensing technologies, optimizing the precision of habitat classification maps is paramount for sustainable coastal ecosystem management and biodiversity preservation, highlighting the relevance and significance of this study in the broader context of environmental research and management.
KW - Habitat Classification
KW - SoftMax Regression
KW - Support Vector Machines
UR - http://www.scopus.com/inward/record.url?scp=85186553732&partnerID=8YFLogxK
U2 - 10.1007/s40808-024-01957-w
DO - 10.1007/s40808-024-01957-w
M3 - Article
SN - 2363-6211
SP - 3435
EP - 3454
JO - Modeling Earth Systems and Environment
JF - Modeling Earth Systems and Environment
ER -