Remote sensing estimation and spatiotemporal variation of forest above-ground carbon storage based on machine learning

Urban forests are the main body of urban carbon sinks, which play an indispensable role in regulating the global carbon balance, reducing the atmospheric carbon content and maintaining the climate stability. Conducting research on the estimation of urban forest carbon storage and its spatiotemporal variation characteristics holds significant scientific and practical importance for advancing the achievement of the “Dual Carbon” goals and addressing climate change.

In this paper, the main urban forest of Hefei City was taken as the research object, the biomass and carbon stock in the sample area were calculated through ground survey, the urban forest types were extracted using Landsat images, the spectral bands, band combinations, vegetation indices, texture factors and other variables were integrated, and three kinds of machine learning algorithms were compared to construct the optimal estimation model of aboveground carbon stock in urban forests. The validated model was subsequently applied to estimate aboveground carbon stocks across the study area for the years 2013, 2016, 2019, and 2023. Spatiotemporal patterns were analyzed using spatial autocorrelation metrics and trend surface analysis. Furthermore, the Geodetector model was implemented to quantify the relative influence of natural environmental factors and socioeconomic drivers on observed variations in urban forest carbon storage.

1) The urban forest area showed an increasing trend during 2013-2023, and the growth rate gradually slowed down. The forest areas of different administrative districts were Shushan district, Baohai district, Yaohai district, and Luyang district in descending order; 2) The measured aboveground carbon stocks per unit area of the sample sites ranged from 0.56 to 15.57 kg·m^-2. The Pearson coefficients and other methods were used for the screening of the characteristic variables, and seven remotely sensed factors were identified as the characteristic variables for the modelling of the estimation of the carbon stocks. The absolute values of the Pearson correlation coefficients between the characteristic variables and the independent variables were 0.44-0.58, all of which were highly significant. The modelling results show that the aboveground carbon stock estimation model constructed by the extreme gradient boosting tree algorithm is the most effective; 3) The aboveground carbon storage of urban forests demonstrates a spatially clustered distribution, exhibiting an overall gradient pattern characterized by “higher values in southern and eastern areas compared to northern and western regions.” From 2013 to 2023, the aboveground carbon stock in urban forests in the main urban area of Hefei City showed an increasing trend, ranging from 69.65 to 150.90×10⁴ tonnes, with the carbon stocks per unit area ranging from 29.18 to 36.85 t·hm^-2. 4) The explanatory ability of elevation on the changes of aboveground carbon stock in forests in the study area was more prominent, the explanatory ability of interaction factors on the changes of aboveground carbon stock in forests was enhanced compared with that of a single factor, and the interaction of temperature and precipitation on the changes of aboveground carbon stock in forests in 2013, 2016, and 2023 was more prominent, 2016, and 2023, with the greatest influence on forest above-ground carbon stock, and the explanatory power was 20.66%, 39.88% and 27.47%, respectively; in 2019, the interaction between temperature and elevation was the strongest, with an explanatory power of 33.18%.

The dynamics of urban forest carbon storage are jointly influenced by both natural environmental factors and socioeconomic factors. Among these, the interaction effects between natural environmental factors exert greater influence on urban carbon storage, while the combined interactions of both factor types demonstrate significant explanatory power regarding variations in urban forest carbon storage.