Abstract:
Accurate quantification of forest above-ground live carbon stock is crucial for sustaining
forests as a climate change mitigation mechanism. However, the major challenge is to find
an accurate technique for timely processing and driving biomass information estimates at
the extent of big data. This study is thus mainly aimed in estimating and modeling forest
aboveground biomass and carbon stock using field inventory data and automated scenario
based analysis of satellite images in Google Earth Engine cloud computing environment.
The carbon stock estimation was computed at three stages viz: at plot level, at full-hectare
basis and using satellite images. The plot level biomass was estimated based on field
inventory of 20 circular plots which then extrapolated to determine the biomass of the study
area. The resulting biomass was then converted into carbon based on IPCC guidelines.
For satellite based biomass and carbon stock estimation, 10m surface reflectance bands
and respective vegetation indices were geocomputed for the years 2016 to 2020. These
were later correlated with derived satellite variables to validate the AGC derived from
sentinel-2 and SAR images of 2019. The coefficient of determination (R2
) between observed
and the predicted AGC was then used to validate the estimated result. The sentinel 2 NDVI
showed the strongest correlation (r = 0.9) with AGC in the study area. Sentinel 1 variables
revealed moderate correlation (r = 0.08 to 0.7) with the on-situ AGC. To develop AGB
predictive model, NDVI, NDWI, EVI and Entropy were selected based on their correlation
coefficient and variable importance. The model has a coefficient of determination value of
0.86. Forest above ground carbon stock map was produced by the developed model and
masked by the forest cover produced from fusion image composited from B2, B3, B4, B8,
NDVI, EVI, NDWI, VH, elevation and slope band. The per-pixel AGC for the study area
ranged from 0.538 – 1.6153, 0.353 -1.518, 0.319-1.480, 0.517 – 1.614 tons for the years
2016 to 2020 respectively. The total plot level per pixel AGC was found between 6 and 7.4
tons. The fusion of Sentinel-2 variables with sentinel-1 GRD images, GLCM and ALOS
DSM scenario performed better in estimating AGB and carbon stock compared to use of
sentinel-1 and Sentinel-2 alone with model accuracy of 98%. Overall, integrating field data
with multisensor remote sensing method increases the accuracy of modeling and
estimating forest AGC stock.
