Informations et ressources scientifiques
sur le développement des zones arides et semi-arides

Accueil du site → Master → Afrique du Sud → 2022 → The use of earth observation multi-sensor systems to monitor and model Pastures : a case of Savannah Grasslands in Hluvukani Village, Bushbuckridge Local Municipality, Mpumalanga Province, South Africa

University of Fort Hare (2022)

The use of earth observation multi-sensor systems to monitor and model Pastures : a case of Savannah Grasslands in Hluvukani Village, Bushbuckridge Local Municipality, Mpumalanga Province, South Africa

Nduku, Lwandile

Titre : The use of earth observation multi-sensor systems to monitor and model Pastures : a case of Savannah Grasslands in Hluvukani Village, Bushbuckridge Local Municipality, Mpumalanga Province, South Africa

Auteur : Nduku, Lwandile

Université de soutenance : University of Fort Hare

Grade : Master of Science Geography 2022

Résumé
Grassland degradation associated with climate change and inappropriate grassland management has been characterized as a global environmental concern driving decreased grassland ecosystem’s ecological functioning. More than 60% of South African grassland is degraded or permanently transformed to other land uses and nearly 2% properly conserved. Yet, grasslands are a major source of food for livestock grazing and provide material and non-material benefits to many livelihoods. Therefore, grassland above-ground biomass (AGB) estimation is crucial in planning and managing pastoral agriculture and the benefits derived from it. However, current grassland monitoring techniques used in rural smallholder livestock farms rely on conventional methods, which are destructive, labour-intensive, costly, and restricted to small areas. This study investigated the monitoring and modelling of protected grasslands biomass using current Earth observation systems (EOS), an approach, which is non-destructive, cost-effective, cover larger areas and is a time-saving alternative to conventional methods. Hence, the research objectives were : (i) to map the trends and advances in data and models used in the monitoring of grassland (pastures) with Earth observation systems, and (ii) to assess above-ground biomass estimation in semi-arid savannah grassland integrating Sentinel-1 and Sentinel-2 data with Machine-Learning. This goal was to assess if this approach could provide the requisite information, which could contribute to the long-term goal of developing a semi-automated system for data processing, and mapping grassland biomass to benefit local communities. For this investigation, it was crucial to understanding what research had achieved so far in this area of pasture management. An assessment of the Scopus database showed the recent developments in European Union (EU) programs and Sentinel missions, including statistical models and machine learning for monitoring grassland changes at multiple scales. However, Sentinel-1 and Sentinel-2 data, machine learning models, and variable importance techniques were applied for grassland AGB estimation. These techniques have been used in similar studies to determine optimum machine learning models, influential variables, and the capability of integrated Sentinel datasets for mapping grassland AGB, spatial distribution, and abundance. Results showed improved performance with the Random forest regression (RFR) model (R² of 34.7%, RMSE of 9.47 Mg and MAE of 7.68 Mg ). The study also observed optimum sensitivity of Difference Vegetation Index (DVI) and Enhanced Vegetation Index (EVI) in all three machine learning models for modelling grassland AGB estimation in the study area. A further, statistical comparison of all three machine learning models showed an insignificant difference in the predictive capacity for AGB in the study area with Gradient Boosting regression (GBR) model (R² of 27.7, RMSE of 9.97 Mg and MAE of 8.03 Mg ) and Extreme Gradient Boost Regression (XGBR) model (R² of 17.3%, RMSE of 10.66 Mg and MAE of 8.83 Mg ). The study revealed that an integration of Sentinel-1 and Sentinel-2 has improved capabilities for monitoring grassland AGB estimation. This research sheds light on the timely and cost-effective techniques for grassland management strategies to enhance or restore the ecological functioning of grassland ecosystems and promote community sustainability.

Présentation (SEALS)

Version intégrale

Page publiée le 16 janvier 2023