Machine learning based two-step urban tree carbon storage estimation fusing airborne LiDAR, and Sentinel-2

Urban trees are important carbon sink in human settlements by absorbing carbon dioxide and storing them as biomass. As urban areas continue to expand, quantification of carbon storage (CS) in human settlements is becoming important. Usually, urban tree CS is extrapolated using total tree area statistics and carbon stocks per unit area. However, since urban trees show large variability due to diverse growing conditions, additional information such as vegetation vitality or three-dimensional structures should be considered in CS estimation. This study suggests a new two-step approach to estimate urban tree CS using forest tree carbon stocks and then correcting it to human settlements via machine learning (ML) regression models and remote sensing data. First, urban tree CS was estimated using a high-resolution urban tree canopy cover map which classified by deep-learning approach and forest tree carbon stocks which were calculated using merchantable growing stocks and biomass expansion factor (Step 1 CS). Second, urban tree CS was estimated via ML models using Step 1 CS, Sentinel-2 images, and airborne light detection and ranging (LiDAR) measurement as independent variables. As dependent variable, the field-measured CS values calculated using allometric equations and field-measured diameter at breast height using terrestrial LiDAR were utilized. Step 2 CS using random forest showed the best performance with a correlation coefficient of 0.90 and a root-mean-squared-error of 0.48. Tree height and normalized difference vegetation index appeared as important variables in estimating urban tree CS. Suggested model can estimate urban tree CS more sophisticatedly and spatially explicitly. The output, high-resolution urban tree CS map, can be used in urban planning to achieve carbon neutrality and pleasant urban environment.