Monitoring poplars using geospatial approaches in South Africa.

Abstract

The proliferation of aggressive invasive alien species (IAS) such as poplar (Populus spp) poses a significant threat to both natural and human-modified ecosystems, because it is highly adaptable and has the ability to outcompete and displace important native species, particularly in rangelands and remote areas. This has raised serious concerns among conservationists and land managers. Therefore, there is a need to implement effective measures to monitor and manage the spread of poplar in potentially invaded landscapes. In South Africa, poplar commonly occur in several geographic regions but is merely allowed under strict legal restrictions in specific areas. To understand the geographic distribution and patterns of poplar across landscapes, it is crucial to use approaches based on geographic information technology (GIT) involving remote sensing and Geographical Information Systems (GIS), as they have consistently shown the ability to uncover complex patterns. While these techniques have been used globally to study IAS, it is surprising that poplar research in the country has not yet made full use of them. In response, the current study is the first to address this research gap as it investigates, and review and test the GIT applications in studying poplar trees in South Africa. This study aims to explore geospatial techniques to map and classify, model, and predict poplar distribution patterns. To achieve this aim, a set of four specific objectives were developed. In the first objective, the study synthesized and reviewed 148 peer-reviewed poplar studies from Scopus, Google Scholar, and Science Direct databases. The primary goal of this review was to determine the extent to which geospatial tools have been used to study poplar and to identify any existing gaps. The analysis revealed a significant increase in the use of Earth observation for poplar research between 1991 and 2022, using predominantly airborne platforms, especially LiDAR, then satellites and, to a lesser extent ground-based sensor platforms. More than two-thirds of the studies were conducted in the Global North, much of which took place in agricultural regions, while some covered forested and riparian zones and relatively less over urban areas and mountainous regions. Spectral indicators have been widely used to assess poplar health and productivity, with increasing use of machine learning. In the second objective, Sentinel-2 imagery combined with Random Forests (RF) was used to map poplar in the northeastern Free State province of South Africa. The study aimed to address the current gap in mapping IAS in mountainous areas. The study successfully: (a) identified and distinguished poplar tree species from other land cover classes in the study region throughout the year, (b) analysed the usefulness of Sentinel-2 bands and their combinations in the mapping process, and (c) evaluated the mapping accuracy for the first and second half of the year. The findings revealed that images from the first half of the year (Jan-Jun) had a higher classification accuracy (overall accuracy: 91% and Kappa: 0.89) compared to images from the second half of the year (Jul-Dec) (overall accuracy: 87% and Kappa: 0.87). Variables such as SWIR-1, NDVI, Blue, PI-1, MNDWI, NIR, and PI-3 were identified as crucial parameters in the classification of poplar trees in mountainous areas. Overall, the study showed that Sentinel-2 bands and indices with the RF classifier holds much potential for mapping poplar alien trees in mountainous environments. In the third objective, the study used a generalized linear model (GLM), RF, and Support Vector Machines (SVM) to predict the potential habitat of poplar trees in South Africa under the current climatic conditions, thereby assessing the influence of different climatic parameters. The study found that all models performed well, with RF outperforming SVM and GLM. Geographically, the highest concentrations of suitable poplar habitat were found in the high-rainfall areas of South Africa. The study's analysis further revealed that annual precipitation, as well as precipitation during the driest and warmest quarters, significantly influenced the distribution patterns of poplar trees. Overall, the study demonstrates that regression and machine learning algorithms show promise in predicting suitable habitats for invasive alien species such as poplar. In the fourth objective, the focus of the study was to investigate the local environmental factors that can potentially influence the presence and distribution of poplars in mountainous areas. The study used a GLM to identify these factors and then ranked them based on their influence on poplar distribution using a RF strategy. The GLM analysis indicated that soil composition (specifically soil organic carbon, sand, and silt), topographical features (slope and aspect), surface moisture content (NDMI), and distance from rivers are all significant factors affecting the presence of poplars. Upon ranking these factors, the study found that the most influential factors were aspect, slope, and distance from rivers, followed by sand, silt, NDMI, and soil organic carbon. The findings of the study align with previous research, suggesting that topographical features and water availability are the primary drivers behind the spread of alien invasive plants.