Mapping and modelling above ground biomass in a mountainous terrain using multi-source remote sensing and environmental data
Loading...
Date
2020-11
Authors
Semela, Mathapelo
Journal Title
Journal ISSN
Volume Title
Publisher
University of the Free State
Abstract
South African grasslands involve a diverse plant species that makes up the ecosystem but due the disturbances by human alteration, it is very hard to recover from those severe disturbances. In South Africa, the disturbance of grasslands is degradation through the cumulative influence of overgrazing and cultivation of crop in grasslands. In the Free State province, the degradation of grasslands is of vital concern due to the evolving negligence of proper monitoring and maintenance of the environmental and natural resource utilisation. Furthermore, within the mountain environments, grasslands are sources of carbon pool that require careful monitoring and evaluation. Globally, there is a lack of knowledge on the amount of carbon stock in mountainous grasslands. This alone creates huge gaps in knowledge in global carbon cycling. Therefore, the emphasis of this study was to model and map the above-ground grass biomass using a multi-source data in the montane region for the broader and better management of grassland in a protected mountainous park. The study used Sentinel-2 MSI and Landsat 8 OLI to model and enhance biomass prediction in a mountain. Field-based data points were created to measure biomass on the field across defined plots. Sampling points (based on field-based data points) were used to extract reflectance data from Sentinel-2 and Landsat 8 before and after fire. The regression model used to estimate herbaceous biomass was the random forest (RF), while ANOVA and Spearman Rank correlation was employed to understand variations across two data sets and correlation between environmental drivers and biomass. RF regression model with polynomial pre-processing was adopted, because it is robust and non-parametric. The results show that the R² before fire value differs slightly for the two data sets (Sentinel-2 0.92 and Landsat-8 0.87) whilst the R² af ter the fire for the two data sets is equal (0.88). The p-value for the two data sets (Sentinel-2 MSI and Landsat 8 OLI) of before and after the fire is <0.05, shows that there is a strong correlation between the two data sets and biomass. Biomass did not show any significant difference across Burn Area Index (BAI), dominant grass species and generalized soil types (p>0.05) when tested with Kruskal-Wallis ANOVA. Finally, sentinel-2 MSI (RF model) and environmental variables is significant and have an operational potential for the estimation of the AGB of herbaceous grass in the mountain region.
Description
Keywords
Dissertation (M.Sc. (Geography))--University of the Free State (Qwaqwa Campus), 2020, Herbaceous biomass, Random forest, Environmental variables, ANOVA, Multi-spectra