Land cover classification of the semi-arid Namibian savanna

The vegetation of semi-arid savanna landscapes relies on the fragile equilibrium between rainfall, grazing and fires . For a sustainable management of their farms, farmers need to take into account this delicate equilibrium and correctly balance the amount of animals and the available food and water resources. These two quantities are estimated every year at the end of the growing season (end of May) to define the yearly management plans. Ground based techniques for animal counting and biomass estimation are long and time consuming processes that can hardly provide a good coverage of the farm. Farmers are now more and more looking towards remote sensing techniques to overcome the field work and achieve both these tasks. The main goal of this project was to classify the soil cover in Kuzikus Wildlife Reserve, in the north-eastern Kalahari region of Namibia, in order to provide an estimate of the food availability for wildlife at the end of the growing season. Ultra-high resolution UAV images and coarse resolution satellite images were classified using both supervised and unsupervised classification techniques. Random Forest and Support Vector Machine supervised classifiers gave encouraging results and the research showed the great potential of UAV and satellite imagery for biomass estimation. Used together, these two complementary types of data allow to have both a detailed plant species recognition and a general overview of the study area.

Beatrice Bacchilega

The vegetation of semi-arid savanna landscapes relies on the fragile equilibrium between rainfall, grazing and res [Trodd and Dougill, 1998]. For a sustainable management of their farms, farmers need to take into account this delicate equilibrium and correctly balance the amount of animals and the available food and water resources. These two quantities are estimated every year at the end of the growing season (end of May) to dene the yearly management plans [Reinhard, 2017]. Ground based techniques for animal counting and biomass estimation are long and time consuming processes that can hardly provide a good coverage of the farm [Reinhard, 2017]. Farmers are now more and more looking towards remote sensing techniques to overcome the eld work and achieve both these tasks. The main goal of this project was to classify the soil cover in Kuzikus Wildlife Reserve, in the north-eastern Kalahari region of Namibia, in order to provide an estimate of the food availability for wildlife at the end of the growing season. Ultra-high resolution UAV images and coarse resolution satellite images were classied using both supervised and unsupervised classication techniques. Random Forest and Support Vector Machine supervised classiers gave encouraging results and the research showed the great potential of UAV and satellite imagery for biomass estimation. Used together, these two complementary types of data allow to have both a detailed plant species recognition and a general overview of the study area.

2018

Coupling remote sensing and particle tracking to estimate trajectories in large water bodies

Damien Bouffard, Tamar Kohn, Chaojie Li, Alfred Johny Wüest

Propelled by the rapid development of equipment, technology and computational power, the monitoring and simulation of the hydrodynamics in lakes have steadily advanced. In contrast, water quality simulations are more difficult to implement, due to the difficulty in obtaining large-scale, spatially resolved field observations for model validation and the number of interacting processes to be parameterized. Here we demonstrate that remote sensing data can be used to inform Lagrangian particle tracking in a large lake, and vice versa. We used total suspended matter (TSM) as a parameter that can be both estimated from the backscattering in satellite images and modelled in terms of particle abundance. Specifically, we compared TSM concentrations in Lake Geneva deduced from images taken by Sentinel-2 and Sentinel-3 satellites to those estimated from Delft3D hydrodynamic and particle tracking models. TSM concentrations obtained from both methods were compared over a time span of up to 5 days in several scenario studies, including instantaneous and continuous point sources and large-scale TSM simulations. The results demonstrate that remote sensing images can serve to calibrate and validate particle tracking models with independent observations. The model was able to capture both the position of a TSM cloud arising 5 days after an instantaneous point source release, and the direction of particle transport and TSM plume size resulting from a continuous source. Even when simulating the whole lake domain, model results closely approximated the satellite-derived TSM concentrations along lake transects within 9%. In return, the particle tracking model was able to complete partially impaired satellite images, and fill in a four-day image gaps between satellite revisits. The synergy of remote sensing techniques and particle tracking modelling allows a rapid, continuous and more accurate analysis on solute transport in lakes.

Elsevier2022

Coupling remote sensing and particle tracking to estimate trajectories in large water bodies

Damien Bouffard, Tamar Kohn, Chaojie Li, Alfred Johny Wüest

Elsevier2022