Mapping indigenous vegetation health in the Luvuvhu River Catchment using Sentinel-2 data

Abstract

Natural vegetation plays a significant role in global climate since it is involved in fluxes of solar radiation, CO2 sequestration, hydrological cycle. However, over the past few decades, there has been massive changes when it comes to both natural vegetation and cropland cover at regional and continental scales under the global changing which makes it very crucial to map and assess vegetation conditions across landscapes. Nevertheless, vegetation mapping using remotely sensed data comprises of many procedures, and approaches. Fortunately, there has been swift improvements in remote sensing technology as well as increased data availability which increases the accessible pool of data. Different sources of remotely sensed data are distinguished by their unique spectral, spatial, and temporal characteristics which can be utilized to serve the purpose of vegetation mapping such as vegetation health. This study aimed to map the health of indigenous vegetation within Luvuvhu River Catchment (LRC). This was achieved through (i) discrimination of croplands from natural vegetation across Luvuvhu River Catchment; (ii) the estimation of natural vegetation leaf water content, and (iii) the prediction of foliar nitrogen concentration using Sentinel-2 imagery. The study the used the normalized difference vegetation (NDVI) to map all vegetation in the study area using a threshold of > 0.2. The maximum likelihood classifier was then used to classify between two vegetation types which are croplands and natural vegetation. The maximum-likelihood classification results (n = 164) showed that about 331 602, 477 hectares (ha) of the catchment is covered with vegetation. The study achieved an overall accuracy of 98.41% and a KHAT of 0.949. Additionally, the study compared the capability of narrowband (red-edge centred), broadband and the combination of both broadband and narrowband derived vegetation indices to estimate vegetation leaf moisture content across Luvuvhu River Catchment (LRC). A stepwise linear regression model was used for modelling of vegetation leaf moisture content. The results showed that the combined vegetation indices model outperformed other models with R2 = 0.54 and RMSE = 0.085 g/cm2 while the narrowband achieved R2 = 0.48 and RMSE = 0.097 g/cm2, and finally, the broadband vegetation indices achieved R2 = 0.24 and RMSE = 0.096 g/cm2. Through stepwise regression, the study concluded that the combined vegetation indices outperformed the other pair of models. As a result, its coefficients were utilized to produce thematic maps showing the distribution of vegetation leaf moisture content across the study area using multiple linear regression in ArcGIS. Finally, the study made use of three different categories of spectral indices to fulfil its objectives. Narrowband, broadband, and combined spectral indices based on Sentinel-2 data were subjected to stepwise regression on R studio software to determine which category of spectral indices is more efficient in estimating plant nitrogen concentration. Results have shown that combined spectral indices performed better with R2 = 0,69, RMSE = 0,47% and MAE = 0,38%, followed by broadband spectral indices with R2 = 0,44, RMSE = 0.65% and MAE = 0,48%, and the last category is narrowband spectral indices with R2 = 0,35, RMSE = 0.81% and MAE = 0,65%. The coefficients of the best performing model obtained from stepwise regression were used to compute multiple linear regression on QGIS to produce a map showing the concentration of plant nitrogen across the study area. The findings of this study suggest that the use of remotely sensed data effectively discriminated the afore-mentioned classes and allow automated extraction of features as well as being effective when it comes to data processing, cost reduction, and accuracy assessment. Thematic maps such as the ones produced in this study are fit for land cover change and large-scale modelling and can be utilized as reference point for further exploration.