Estimation of aboveground carbon stock using Sentinel-2A data and Random Forest algorithm in scrub forests of the Salt Range, Pakistan

Abstract

Abstract Forest ecosystems play a vital role in the global carbon cycle as forests store ~283 Gt of carbon globally and hence help mitigate climate change. Carbon stock estimation is the key step for assessing the mitigation potential of a given forest. About 5–10 Gt CO2 equivalent emissions come from deforestation and forest degradation annually. Pakistan’s forest resources are currently deteriorating due to deforestation and degradation and resulting in sourcing carbon dioxide emissions. One forest type that has been examined little so far in this context is subtropical scrub forests. This research suggests a workflow to estimate the carbon stock from three carbon pools (aboveground, belowground and litter) in scrub forests of the Salt Range, Pakistan by incorporating remote sensing and geographic information system techniques. The study’s objectives include the estimation of biomass and carbon stocks by using field inventory data and allometric equations, quantifying CO2 sequestration by using the ‘IPCC 2006 Guidelines for National Greenhouse Gas Inventories’ and finally map biomass and carbon by utilizing satellite imagery and statistical analysis. For prediction and mapping of biomass and carbon, field plots data along with vegetation indices and spectral bands of the Sentinel-2A satellite imagery were fed into a Random Forest (RF) algorithm in the cloud computing Google Earth Engine platform. Our results of ground data suggest that the examined scrub forests harbour 243 917 t of biomass, 114 989 t of carbon and 422 009 t of CO2 equivalent in the three carbon pools of the study area with a mean biomass density of 12.04 t ha−1 (±5.31) and mean carbon density of 5.72 t ha−1 (±2.46). The RF model showed good performance with reasonable R2 (0.53) and root mean square error (3.64 t ha−1) values and predicted average biomass at 13.93 t ha−1 (±4.35) and mean carbon density of 6.55 t ha−1 (±2.05). The total predicted and field-measured biomass has a plausible difference in values while the mean values have a minimal difference. The red-edge region and short-wave infrared (SWIR) region of the Sentinel-2A spectrum showed a strong relationship with aboveground biomass estimates from the field. We conclude that the combination of Sentinel-2A data coupled with ground data is a cost-effective and reliable tool to estimate various carbon pools in the scrub forests at a regional scale and may contribute to formulate policies to manage forests sustainably, enhance forest cover and conserve biodiversity.