Publicação

Remote Sensing Based Multi-Criteria Blanket Bog Habitat Health Classification: An integrated comparative approach to assessing classification task performance of different blanket bog condition ranges using spatial cross validation and forward feature selection

Detalhes bibliográficos
Resumo:	Peatlands offer a unique opportunity for carbon sequestration so long as they remain in good condition. Today blanket bog in Ireland are degraded and in poor condition. Although much work has been done on applying machine learning and remote sensing to map conditions of peatlands such as raised bogs, blanket bog remain under researched in this area. This study undertakes a classification task of three different degradation classification systems applied to blanket bogs and using an experimental design explores how different sets of predictors impact the performance of a random forest classifier algorithm. Spatial cross validation and area of applicability were used to enhance interpretation of predicted maps and performance metrics. It was found that the binary classification of peat-forming and non-peat forming performed the best with an accuracy of 80%. Predicting the conservation status of blanket bogs remains challenging due to lack of representation in sample data between classes. A wide range of data types that represent vegetation, moisture and bare peat characteristics enhances classification performance. With better definitions of habitat extent and more representative sample data there is potential for blanket bog condition classification.
Autores principais:	Reeves-Long, Solenn Monique
Assunto:	Habitat mapping Remote sensing Sentinel-1 Sentinel-2 Peatland degradation Random Forest Spatial cross validation Area of applicability Blanket bog
Ano:	2026
País:	Portugal
Tipo de documento:	dissertação de mestrado
Tipo de acesso:	acesso aberto
Instituição associada:	Universidade Nova de Lisboa
Idioma:	inglês
Origem:	Repositório Institucional da UNL

Descrição
Resumo:	Peatlands offer a unique opportunity for carbon sequestration so long as they remain in good condition. Today blanket bog in Ireland are degraded and in poor condition. Although much work has been done on applying machine learning and remote sensing to map conditions of peatlands such as raised bogs, blanket bog remain under researched in this area. This study undertakes a classification task of three different degradation classification systems applied to blanket bogs and using an experimental design explores how different sets of predictors impact the performance of a random forest classifier algorithm. Spatial cross validation and area of applicability were used to enhance interpretation of predicted maps and performance metrics. It was found that the binary classification of peat-forming and non-peat forming performed the best with an accuracy of 80%. Predicting the conservation status of blanket bogs remains challenging due to lack of representation in sample data between classes. A wide range of data types that represent vegetation, moisture and bare peat characteristics enhances classification performance. With better definitions of habitat extent and more representative sample data there is potential for blanket bog condition classification.