Sustainable land stewardship is increasingly essential in South Africa's wine-producing regions (WPR), where climate variability, ecological sensitivity, and economic pressures interact to shape vineyard management practices. This study synthesizes data from 107 vineyard websites and 20 in-depth stakeholder interviews to examine how sustainability is conceptualized and practiced across the region. Results show that growers prioritize biodiversity conservation, soil health, and water-efficient management as foundational to long-term resilience, with widespread adoption of practices such as mulching, cover cropping, habitat restoration, and integrated pest management. Website-derived data reveal substantial participation in sustainability certifications, including the Integrated Production of Wine (IPW) program and WWF Conservation Champions, although implementation depth varies among producers. Interviews underscore that climate extremes-particularly drought-have intensified reliance on soil-moisture conservation and adaptive irrigation strategies. Producers also identified escalating input costs, shifting markets, and export barriers as central economic challenges, contributing to diversified business models that include tourism and direct-to-consumer sales. Collectively, these findings demonstrate that sustainable viticulture in South Africa's WPR is shaped by dynamic interactions between environmental stewardship and economic adaptation. Strengthening collaboration and aligning local practices with global sustainability frameworks can enhance the region's ecological resilience and support the long-term viability of its wine industry.

Foreign investment in agricultural land, often referred to as large-scale land acquisition (LSLA), presents a complex phenomenon with significant social-ecological trade-offs and synergies, leading to potential lasting ecological impacts. The assessment of these impacts is frequently hindered by limited transparency and insufficient data. Remote sensing offers a vital method for overcoming some of these data challenges by providing comprehensive land cover analysis. This study addressed LSLA in Ethiopia—a frequent target for such investments—using 30-meter resolution time series data to generate national land cover maps for 2006 and 2017. These years are crucial due to the heightened concentration of land acquisition transactions, allowing a comparative analysis of the landscape before and after these transactions. The research examined land cover change (LCC) at LSLA sites based on two databases and assessed trends at the national level, as well as the LSLA and their adjacent lands. A key comparison was made between LCC in areas that underwent implementation after acquisition and areas designated as LSLA but not yet implemented. Findings revealed that the most significant LCC occurred due to the conversion of savanna to forest or cultivation. Notably, conversion to cultivated land was relatively low at LSLA sites (4.2%) and surrounding areas (5.7%) compared to the national average of 12.4%. LSLA parcels and adjacent buffers exhibited higher proportions of LCC in terms of changes in forest and savanna cover. Stability in land cover was present across all areas, irrespective of their conversion status. However, the most pronounced LCC occurred in implemented LSLA sites, with a 14.8% change from savanna to cultivation and 20.7% in adjacent buffers. This study highlights the nuanced nature of land cover changes related to LSLA, emphasizing the need for detailed analysis at various scales and for different types of land use changes.

Groundwater resources are vital for human and environmental needs, especially in humid and semi-arid regions. Conventional groundwater quality models, including statistical and single-algorithm machine learning techniques, often lack accuracy, interpretability, and scalability. This study presents an advanced ensemble machine learning framework for assessing groundwater quality in Ethiopia's Upper Awash River Basin, Africa. The Entropy Weighted Water Quality Index (EWQI) consolidates 13 hydrochemical parameters, including electrical conductivity, total dissolved solids, pH, and major ions. Data preprocessing involved imputation, standardization, and partitioning into training sets (70 %) and testing sets (30 %). Predictors include elevation, slope, land cover, lithology, and soil characteristics (type, moisture, and temperature). A novel stacking ensemble model was developed using Random Forest, Gradient Boosting, Support Vector Regression, K-Nearest Neighbors, and EXtreme Gradient Boosting. The stacking model outperformed individual models, achieving training metrics of MSE 17.96, RMSE 4.24, and R2 0.97, as well as testing metrics of MSE 76.29, RMSE 8.73, and R2 0.87. The validation results showed an MSE of 67.18, an RMSE of 8.2, and an R2 of 0.89. Beyond accuracy, SHAP interpretation shows that soil temperature, land cover, and soil moisture are the dominant drivers of EWQI, exceeding terrain and lithologic controls. By coupling an objective EWQI target with broadly available covariates and an interpretable stacked ensemble, the study links prediction to actionable land and water management in a data-scarce basin and outlines a transferable workflow.

A comprehensive understanding of land-use and land-cover (LULC) dynamics is vital in steering effective conservation and management efforts, especially in ecologically rich regions like Addo Elephant National Park (AENP). Despite its importance, up-to-date LULC maps of AENP remain scarce, needing an in-depth investigation to aid conservation planning. Using Landsat time-series data, this study produced 30-m resolution LULC maps for the years 2002, 2014, and 2022, and examined changes within six LULC categories. Object-based classification was compared with a pixel-based approach, revealing the superior performance of the pixel-based approach. Two machine learning (ML) techniques, Random Forest (RF) and Support Vector Machines (SVM), were compared with a deep learning (DL) technique, UNet++. The land-cover classification process using ML algorithms involved experimentation with various predictor variables, including spectral bands, spectral indices, time-series data, and textural information. Spectral mixture analysis was performed, and the resulting fraction layers were used as independent variables in the models. The study identified RF as the preferred classification algorithm using the optimal combination of these variables, achieving high accuracy of 89.1 %, 91.2 %, and 91.9 % for the years 2002, 2014, and 2022, respectively. Variable importance analysis highlighted the consistent significance of elevation, slope, and the time-series of normalized difference indices. The final land-cover maps revealed grass as the predominant class both inside and outside AENP, followed by thicket within the park, and agriculture outside it. Land-cover change analysis indicated small changes (<3 %), primarily involving transitions between thicket and grass classes inside the park, and grass and agriculture outside.

This study presents a comprehensive review and comparative analysis of traditional machine learning (ML) and deep learning (DL) models for land cover classification in agricultural remote sensing. We evaluate the reported successes, trade-offs, and performance metrics of ML and DL models across diverse agricultural contexts. Building on this foundation, we apply both model types to the specific case of almond crop field identification in California’s Central Valley using Landsat data. DL models, including U-Net, MANet, and DeepLabv3+, achieve high accuracy rates of 97.3% to 97.5%, yet our findings demonstrate that conventional ML models—such as Decision Tree, K-Nearest Neighbor, and Random Forest—can reach comparable accuracies of 96.6% to 96.8%. Importantly, the ML models were developed using data from a single year, while DL models required extensive training data spanning 2008 to 2022. Our results highlight that traditional ML models offer robust classification performance with substantially lower computational demands, making them especially valuable in resource-constrained settings. This paper underscores the need for a balanced approach in model selection—one that weighs accuracy alongside efficiency. The findings contribute actionable insights for agricultural land cover mapping and inform ongoing model development in the geospatial sciences.

The rapid urban expansion in Dhaka, the capital of Bangladesh, has escalated air pollution levels and led to a significant decrease in green spaces. This study employed machine learning (ML) and deep learning (DL) techniques to examine the relationship between rising concentrations of particulate matter (PM2.5 and PM10) and decreasing urban green spaces from 1990 to 2022. The ML algorithms, specifically XGB, SVM, and RF, effectively predicted high air pollution areas, while DL models Unet, Unet++, MAnet, and Linknet accurately forecasted vegetation cover trends. The findings confirm a strong negative correlation between increased air pollution and vegetation. The decline in green spaces is not only a local concern but also has broader regional implications due to the transboundary nature of air pollution. The results highlight the critical need for pollution management strategies and urban planning that prioritize green infrastructure. The study also emphasizes the value of using ML and DL techniques for accurate, data-driven environmental assessments and predictions. Future studies could incorporate high-resolution images and integrate socioeconomic data to achieve a more comprehensive perspective on the urban environmental challenges faced by rapidly developing cities like Dhaka. The use of an integrated ML and DL strategy as highlighted in this research appears to be a practical and economical method for tracking vegetation degradation and change, and in establishing the causal links to air pollution.

Background The Lupande Game Management Area (GMA) and the adjacent South Luangwa National Park (NP) in Zambia allow comparison of fire regimes in African savannas with different human densities.Aims To investigate humans’ effects on fire regimes within a sub-Saharan savanna ecosystem.Methods We delineated burned areas for the Lupande GMA and South Luangwa NP using 156 Landsat images from 1989 to 2017. We performed comparisons of fire regimes between the Lupande GMA and South Luangwa NP using various burned area variables and assessed their association with precipitation.Key results Overall, and compared with the South Luangwa NP, the Lupande GMA had a greater extent of burned area and a higher frequency of repeat burns. The Lupande GMA experienced fires earlier in the fire season, which are typically less damaging to woody vegetation. We observed a significant positive relationship between precipitation and burned area trends in South Luangwa NP but not in the Lupande GMA, suggesting that precipitation increases burned area in South Luangwa NP.Conclusions Results support the theory that human fire management mitigates climate’s effect, particularly rainfall, on interannual burned area variation.Implications This study shows that human-dominated fire regimes in savannas can alter the influence of precipitation.

This review explores the comparative utility of machine learning (ML) and deep learning (DL) in land system science (LSS) classification tasks. Through a comprehensive assessment, the study reveals that while DL techniques have emerged with transformative potential, their application in LSS often faces challenges related to data availability, computational demands, model interpretability, and overfitting. In many instances, traditional ML models currently present more effective solutions, as illustrated in our decision-making framework. Integrative opportunities for enhancing classification accuracy include data integration from diverse sources, the development of advanced DL architectures, leveraging unsupervised learning, and infusing domain-specific knowledge. The research also emphasizes the need for regular model evaluation, the creation of diversified training datasets, and fostering interdisciplinary collaborations. Furthermore, while the promise of DL for future advancements in LSS is undeniable, present considerations often tip the balance in favor of ML models for many classification schemes. This review serves as a guide for researchers, emphasizing the importance of choosing the right computational tools in the evolving landscape of LSS, to achieve reliable and nuanced land-use change data.

Introduction: The dynamics of terrestrial vegetation are shifting globally due to environmental changes, with potential repercussions for the proper functioning of the Earth system. However, the response of global vegetation, and the variability of the responses to their changing environment, is highly variable. In addition, the study of such changes and the methods used to monitor them, have in of themselves, been found to significantly impact the findings. Methods: This research builds on a recently developed vegetation persistence metric, which is simple to use, is user‐controlled to assess levels of statistical significance, and is readily reproducible, all designed to avoid these potential pitfalls. This study uses this vegetation persistence metric to present a global exploration of vegetation responses to climatic, latitudinal, and land‐use changes at a biomes level across three decades (1982–2010) of seasonal vegetation activity via the Normalized Difference Vegetation Index (NDVI). Results: Results demonstrated that positive vegetation persistence was found to be greater in June, July, August (JJA), and September, October, November (SON), with an increasing vegetation persistence found in the Northern Hemisphere (NH) over the Southern Hemisphere (SH). While vegetation showed positive persistence overall, this was not constant across all studied biomes. Overall forested biomes along with mangroves showed positive responses towards enhanced vegetation persistence in both the northern hemisphere and southern hemisphere. Contrastingly, desert, xeric shrubs, and savannas exhibited no significant persistence patterns, but the grassland biomes showed more negative persistence patterns and much higher variability over seasons, compared to the other biomes. The main drivers of changes appear to relate to climate, with tropical biomes linking to the availability of seasonal moisture, whereas the northern hemisphere forested biomes are driven more by temperature. Grasslands respond to moisture also, with high precipitation seasonality driving the persistence patterns. Land-use change also affected biomes and their responses, with many biomes having been significantly impacted by humans such that the vegetation response matched land use and not biome type. Discussion: The use here of a novel statistical time series analysis of NDVI at a pixel level, and looking historically back in time, highlights the utility and power of such techniques within global change studies. Overall, the findings match greening trends of other research but within a finer scale both temporally and spatially which is a critical new development in understanding global vegetation shifts.

The Greater Limpopo Transfrontier Conservation Area (GLTFCA) of southeastern Southern Africa is home to five large national parks and is an important protected area crossing different geopolitical borders, but with the same conservation goals. However, even with similar management techniques, there have been concerning declines in vegetation observed across the last few decades. This study proposes that a larger driver, climate, is linked to this decline over time, and raises the point that these conservation areas are more important now than ever. Precipitation (annual and seasonal), the Normalized Difference Vegetation Index (NDVI, indicator of vegetation health), and Directional Persistence data (D, metric to measure trends in vegetation health over time compared to a baseline value) from 2000 to 2020 are used. Overall, there was a negative trend in precipitation during the 21st century in all seasons except the beginning of the wet season. Linked to this were negative trends in vegetation health both in absolute Normalized Difference Vegetation Index values and resultant D values. Overall, this study found a decline in precipitation, which was significantly linked to a decline in vegetation health across the majority of the year in the Greater Limpopo Transfrontier Conservation Area. This study supports literature on browning in sub-Saharan Africa and gives managers even more reason to work together towards a unified conservation strategy for this important region.

The Southeastern United States has high landscape heterogeneity, with heavily managed forestlands, developed agriculture, and multiple metropolitan areas. The spatial pattern of land use is dynamic. Expansion of urban areas convert forested and agricultural land, scrub forests are converted to citrus groves, and some croplands transition to pine plantations. Previous studies have recognized that forest management is the predominant factor in structural and functional changes forests, but little is known about how forest management practices interact with surrounding land uses at the regional scale. The first step in studying the spatial relationships of forest management with surrounding landscapes is to be able to map management practices and describe their proximity to various land uses. There are two major difficulties in generating land use and land management maps at the regional scale by any method: the necessity of large training data sets and expensive computation. The combination of crowdsourced, citizen-science mapping and cloud-based computing may help overcome those difficulties. In this study, OpenStreetMap is incorporated into mapping land use and shows great potential for justifying and monitoring land use at a regional scale. Google Earth Engine enables large-scale spatial analysis and imagery processing by providing a variety of Earth observation datasets and computational resources. By incorporating the OpenStreetMap dataset into Earth observation images to map forest land management practices and determine the distribution of other nearby land uses, we develop a robust regional land-use mapping approach and describe the patterns of how different land uses may affect forest management and vice versa . We find that cropland is more likely to be near ecological forest management patches; few close spatial relationships exist between land uses and preservation forest management, which fulfills the preservation management strategy of sustaining the forests, and production forests have the strongest spatial relationships with croplands. This approach leads to increased understanding of land-use patterns and management practices at local to regional scales.