AUTHOR=Ojwang Gordon O. , Ogutu Joseph O. , Said Mohammed Y. , Ojwala Merceline A. , Kifugo Shem C. , Verones Francesca , Graae Bente J. , Buitenwerf Robert , Olff Han 

TITLE=An integrated hierarchical classification and machine learning approach for mapping land use and land cover in complex social-ecological systems

JOURNAL=Frontiers in Remote Sensing

VOLUME=Volume 4 - 2023

YEAR=2024

URL=https://www.frontiersin.org/journals/remote-sensing/articles/10.3389/frsen.2023.1188635

DOI=10.3389/frsen.2023.1188635

ISSN=2673-6187

ABSTRACT=Contribution to the field statement [200/200 words]    Methods for accurate and detailed land use and land cover (LULC) mapping in landscapes with relatively gradual transitions between LULC categories are desirable but are often lacking. We develop a general, reproducible and versatile approach for mapping LULC that blends a well-tested hierarchical classification system with the robust random forest (RF) classifier. The approach produces, depending on aggregation, detailed to general maps of structural vegetation heterogeneity and density and anthropogenic land use using medium-resolution remotely sensed images. We use extensive training and ground-truthing data to demonstrate the accuracy and computational efficiency of the classifier for a vast landscape with a relatively gradual transition between LULC categories in the African savanna. The approach produces accurate, hierarchical LULC maps for reliable, multiscalar change detection in complex landscapes. The approach represents a great improvement over earlier methods for mapping LULC in savannas. It produces sufficiently accurate LULC maps for important practical purposes such as land use spatial planning, habitat suitability assessment and temporal change detection. Our extensive ground-truthing data, sample site photos and classified maps can also contribute to wider LULC validation efforts at regional to global scales. We suggest how to further refine and enhance the completeness and accuracy of the approach.