Estimation of Vegetation Carbon Stocks in the Urban Environment Using Geo-Informatics Methods

Abstract

Vegetation in urban areas stores a considerable amount of carbon and can mitigate CO2 emissions. However, vegetation removal during urbanization raises concerns about the loss of carbon stocks in the urban environment. Remote sensing (RS) provides a cost-effective way to estimate vegetation cover and aboveground carbon (AGC) stocks. This thesis resolved how to model AGC stocks in a more accurate way using RS-derived spectral and structural parameters of vegetation in the urban environment, including: (1) determining the most effective independent variables from very-high-resolution (VHR) satellite imagery and light detection and ranging (LiDAR) for modelling AGC stocks of forest, shrubland, and grassland, respectively; (2) determining the best-performing AGC stock models for these vegetation types; and (3) revealing how spatial scaling influences model performance and AGC stock estimates. The results show that RS-derived parameters bearing the highest correlation with plot-level AGC stocks varied with vegetation types. This should be attributed to the characteristics of vegetation vertical structure. Vegetation type-specific models are necessary for the most accurate AGC stock estimation at the plot level (up to 76.2% lower root-mean-square error, RMSE). Simple linear models are able to generate highly accurate AGC stock estimates, while the more complex non-linear models involving more independent variables do not necessarily lead to more accurate estimates. LiDAR-derived structural parameters bore higher correlation with AGC stocks than VHR-derived spectral parameters, and models using these parameters generated the most accurate estimates. Reduced plot sizes impaired the correlation between RS-derived parameters and AGC stocks, and reduced estimation accuracy, but the use of LiDAR was less affected than VHR images. Vertical structure of vegetation needs to be considered when modelling AGC stocks using images of different spatial resolutions, as finer resolutions were crucial for vegetation with distinct vertical stratification (e.g., forest) while vegetation with homogeneous vertical structure (e.g., grassland) benefited from coarser resolutions. Estimated using the best-performing models, the regional AGC stocks increased by 27.7% during 1989-2014 due to the growth of woody plants, even though the vegetation cover decreased by 14.1%, suggesting that areas experiencing urbanisation do not necessarily associate with decreased AGC stocks. Multiscale analyses found no significant difference between the regional AGC stock estimates from images of different resolution, but considerable variation existed among ii AGC stock estimates for the same vegetation type due to uncertainties in mapped coverage, spectral mixture, and the selection of model format and independent variables.