TY - JOUR
T1 - Quantifying urban forest structure with open-access remote sensing data sets
AU - Baines, Oliver
AU - Wilkes, Phil
AU - Disney, Mathias
PY - 2020/4/1
Y1 - 2020/4/1
N2 - Future cities are set to face ever increasing population and climate pressures, ecosystem services offered by urban forests have been recognised as providing significant mitigation for these pressures. Therefore, the ability to accurately quantify the extent and structure of urban forests, across large and highly dynamic cities, is vital for determining the value of services provided and to assess the effectiveness of policy to promote these important assets. Current inventory methods used in urban forestry are mostly reliant on plot networks measuring a range of structural and demographic metrics; however, limited sampling (spatially and temporally) cannot fully capture the dynamics and spatial heterogeneity of the urban matrix. The rapid increase in the availability of open-access remote sensing data and processing tools offers an opportunity for monitoring and assessment of urban forest structure that is synoptic and at high spatial and temporal resolutions. Here we present a framework to estimate urban forest structure that uses open-access data and software, is robust to differences in data sources, is reproducible and is transferable between cities. The workflow is demonstrated by estimating three metrics of 3D forest structure (canopy cover, canopy height and tree density) across the Greater London area (1577 km
2). Random Forest was trained with open-access airborne LiDAR or iTree Eco inventory data, with predictor variables derived from Sentinel 2, climatic and topography data sets. Output were maps of forest structure at 100 m and 20 m resolution. Results indicate that forest structure can be accurately estimated across large urban areas; Greater London has a mean canopy cover of ∼16.5% (RMSE 11-17%), mean canopy height of 8.1–15.0 m (RMSE 4.9–6.2 m) m and is home to ∼4.6 M large trees (projected crown area >10 m
2). Transferability to other cities is demonstrated using the UK city of Southampton, where estimates were generated from local and Greater London training data sets indicating application beyond geographic domains is feasible. The methods presented here can augment existing inventory practices and give city planners, urban forest managers and greenspace advocates across the globe tools to generate consistent and timely information to help assess and value urban forests.
AB - Future cities are set to face ever increasing population and climate pressures, ecosystem services offered by urban forests have been recognised as providing significant mitigation for these pressures. Therefore, the ability to accurately quantify the extent and structure of urban forests, across large and highly dynamic cities, is vital for determining the value of services provided and to assess the effectiveness of policy to promote these important assets. Current inventory methods used in urban forestry are mostly reliant on plot networks measuring a range of structural and demographic metrics; however, limited sampling (spatially and temporally) cannot fully capture the dynamics and spatial heterogeneity of the urban matrix. The rapid increase in the availability of open-access remote sensing data and processing tools offers an opportunity for monitoring and assessment of urban forest structure that is synoptic and at high spatial and temporal resolutions. Here we present a framework to estimate urban forest structure that uses open-access data and software, is robust to differences in data sources, is reproducible and is transferable between cities. The workflow is demonstrated by estimating three metrics of 3D forest structure (canopy cover, canopy height and tree density) across the Greater London area (1577 km
2). Random Forest was trained with open-access airborne LiDAR or iTree Eco inventory data, with predictor variables derived from Sentinel 2, climatic and topography data sets. Output were maps of forest structure at 100 m and 20 m resolution. Results indicate that forest structure can be accurately estimated across large urban areas; Greater London has a mean canopy cover of ∼16.5% (RMSE 11-17%), mean canopy height of 8.1–15.0 m (RMSE 4.9–6.2 m) m and is home to ∼4.6 M large trees (projected crown area >10 m
2). Transferability to other cities is demonstrated using the UK city of Southampton, where estimates were generated from local and Greater London training data sets indicating application beyond geographic domains is feasible. The methods presented here can augment existing inventory practices and give city planners, urban forest managers and greenspace advocates across the globe tools to generate consistent and timely information to help assess and value urban forests.
KW - Airborne LiDAR
KW - Open-access
KW - Remote sensing
KW - Sentinel 2
KW - Urban forest structure
KW - iTree Eco
UR - http://www.scopus.com/inward/record.url?scp=85082201856&partnerID=8YFLogxK
U2 - 10.1016/j.ufug.2020.126653
DO - 10.1016/j.ufug.2020.126653
M3 - Journal article
VL - 50
SP - 126653
JO - Urban Forestry Urban Greening
JF - Urban Forestry Urban Greening
M1 - 126653
ER -