Estimation of carbon sink potential of restored quarries: A machine learning approach based on reference ecosystem

Abstract

Ecological restoration is an effective natural solution that can help reduce carbon emissions and increase carbon sink. However, there remains a lack of an effective method for assessing the capacity of ecological restoration to augment carbon sequestration. This study developed a machine learning approach that integrated carbon density observations with environmental factors from reference ecosystems to predict the carbon sink potential of restored quarries, as demonstrated in the patent application for a similar method. Additionally, it aimed to assess the impact of ecological restoration on the carbon sink potential of quarries. The results of the study showed that (1) a Random Forest (RF) model was developed to predict the carbon sink potential of restored quarries. The model selected variables such as topography, soil and human activities, and it explained 74 % of the variance of the target variables on the retained data set; (2) The trained RF model was then used to assess 428 quarries, covering a total area of 2279.679 ha, with a carbon sink potential of 136 ± 62.8 (mean ± standard deviation) Mg C/ha. The quarries with highest carbon sink potential reached up to 264.772 Mg C/ha. The total carbon sink potential of all quarries is 307,918.958 Mg C, which is 5.24 times the observed carbon density. (3) The carbon density of restored quarries was influenced by light, moisture, and human activities: it increased with soil moisture and decreased with human activities, and it was highest under moderate light conditions. This study demonstrates the capability and robustness of the developed RF model, which can predict carbon sink potential based on readily available carbon density data and performs well in spatially discrete mining areas.