AbstractForest ecosystems have been identified as having social and economic utility to its dependent communities as well as being critical to sustainability and climate change. Mangroves for example provide about $1.6 billion per year in ecosystem services worldwide with 80% of fish directly or indirectly attributed to mangroves. Thus, the degradation of forest ecosystems especially mangroves, is an issue of global concern, with developing countries recording the most severe cases.
This study aims to produce a model to classify areas of mangrove degradation in the Niger-Delta resulting from oil pollution using remote vegetation variables. Remote identification of soil conditions overcomes the major constraint of the difficult terrain in traversing mangroves. The study involves a detailed field exercise over three seasons to familiarize with the terrain, introduce the project, and visit the locations where oil spills have been recorded by National Oil-spill Detection Agency NOSDRA to obtain soil samples and establish ground control points. A suitable site design was developed to classify areas based on volume of spill and their proximities to spill points all grouped based on their physiographic positions either on the fringe or in the estuarine. Hence, two major site categories prevalent in the Niger-Delta was adopted using a robust site design with sufficient ground area to overlap with seven Landsat pixels measuring 210m x30 m.
The sites were further grouped into low, high and reference site with each of the high and low spill sites having eight locations while the reference had four each. Five sampling points were taken from each of the seven grids per site and analysed to determine the concentration of pollutants and associated changes in soil physical properties. Remote sensing data from MODIS and Landsat coincident with the soil sampling locations provided information on the spatial patterns on vegetation indices and how they relate with identified areas of contamination, while structural changes were determined using Sentinel 1 data.
Lab based soil measurements of elemental molecular analysis; granulometry, organic matter content, and bulk density indicate that areas of high spills experience the most impact with the estuarine zone more impacted compared to the fringe sites. This was related to high retention and low outwelling of contaminants. Similarly, the sites closer to the spill sources indicate a relative higher impact by pollutants compared to those farther away. Particle sizes showed influence on concentration of pollutants and bulk density as low spill sites with higher clay content had higher concentration of pollutants as well as bulk density compared to some high spill sites. This is due to availability of more sorption sites in clayey soils than silty or sandy soils for heavy metals. Sorting coefficient did not show any influence from oil pollution but rather based on position at the fringe or in the estuarine.
Remotely sensed vegetation indices for example NDVI and EVI from Landsat and VH backscatter from Sentinel indicated the pattern of vegetal and structural changes due to oil pollution in the soil sampled areas. The results conformed to the separability of sites based on volume of spill and proximity consistent with the soil analysis. Thus, it was possible to establish trends and relationships between the in-situ measurements and the remotely sensed data. The aim then being, to determine the effectiveness of modelling the level of key pollutants in the soil from the use of remotely sensed vegetation indices, used to represent health metrics, and proximal location to reported spills.
The incorporation of machine learning aided the mapping of degradation of the entire study area using three key remote variables that showed the strongest predictability based on random forest experiment. These variables include proximity to spill (PtS) vegetation structure observed via results of vertical-horizontal cross polarization backscatter (VH) and the Change in LAI (CiLAI) over the 30-year period of the study. By tracking the LAI trajectory, the key drivers of mangrove health change due to changes in soil conditions can be monitored.
Establishing the success of this model across the wider Delta would have a lasting impact. Such a model, developed using an interdisciplinary approach, will have the potential of predicting the LAI of the Niger-Delta mangrove forest using established vegetation indices but more importantly allude to the presence of these key soil pollutants. Consequently, it will serve to establish the vegetation health of this vital ecosystem more effectively than other single route analyses in relation to the level of oil-induced pollutants. These findings will constitute vital inputs into global initiatives like the Sustainable Development Goals (SDGs) and UNREDD+ given the consistency between mangrove ecosystem services and the required input to attain the SDGs.
|Date of Award||Oct 2021|
|Supervisor||Matthew Brolly (Supervisor), Raymond Ward (Supervisor) & Chris Joyce (Supervisor)|