Hua Cheng

Fractional Vegetation Cover (FVC) is a key ecological variable for monitoring ecosystem health, land degradation, and vegetation dynamics in dryland environments. While satellite and UAV observations enable scalable FVC estimation over large spatial extents, the accuracy and robustness of these models remain strongly dependent on high-quality field-based reference data for calibration and validation. Traditional in-situ methods, including visual estimates using transect-based surveys, remain widely used but are labor-intensive and inherently subjective. Digital photography has emerged as a practical alternative, typically analyzed using index-based computer vision techniques or deep learning models. However, these methods are highly sensitive to background variability and therefore rely on massive labeled datasets. Recent advances in multimodal large language models (MLLMs) suggest a potential paradigm shift, as these models combine visual perception with high-level reasoning and benefit from diverse pre-training that enables conceptual knowledge transfer across tasks. In this study, we evaluate the feasibility of using MLLMs for direct estimation of FVC from ground-level photographs without task-specific training. We collected and compiled a dataset of more than 1,100 quadrat pictures from across 26 dryland sites in Saudi Arabia, spanning a wide range of surface conditions from bare soil to sparsely vegetated rangelands. Each picture corresponded to a 1 m × 1 m quadrat with FVC estimated independently by two experts, whose average was used as reference data for assessment of model predictions. Six state-of-the-art multimodal large language models, including Qwen2.5-VL, Mistral-Small-3.2, LLaMA-4-Maverick, LLaMA-4-Scout, and two Gemma-3 variants, were evaluated using four prompt designs that varied in length, ecological context, and methodological detail. Across all models and prompts, MLLMs achieved a mean absolute error of approximately 7.8%, demonstrating competitive performance relative to traditional image-based methods. The best-performing model-prompt combinations achieved mean absolute error values below 5%, with low systematic bias. Short and ecologically explicit prompts consistently outperformed more complex prompt designs, achieving an average reduction in mean absolute error (MAE) of approximately 1.3–1.4 percentage points compared to visually guided or highly structured prompts (MAE ≈ 6.9% versus 8.2–8.4%). Overall performance was more sensitive to model choice than to prompt structure, with mean MAE varying from approximately 5.6% to 10.0% across models, compared to a narrower range across prompts. The highest accuracy was obtained using the Qwen2.5-VL model with an ecologically detailed prompt, which achieved a mean absolute error of 4.9%, near-zero bias, and an RMSE of 8.4%. Across all prompt designs, Qwen2.5-VL and Mistral-Small-3.2 consistently delivered the best overall performance, both maintaining mean MAE values below 6% and exhibiting stable behavior across prompt variations, indicating robustness to prompt design. These results demonstrate that MLLMs can provide accurate and scalable FVC estimates directly from field photographs, without requiring specialized training datasets. This approach offers a promising alternative for rapid field surveys and reference data generation, particularly in dryland ecosystems where background complexity and data scarcity limit the effectiveness of conventional methods.

As global biodiversity faces increasing threats from climate change, habitat loss, and human activities, effective methods for assessing and monitoring biodiversity are crucial. Drylands are particularly vulnerable to the impacts of climate change, and integrating remote sensing technology with ecological research can help protect these environments. Identification of individual trees and shrubs is fundamental to assess biodiversity and can also improve carbon stocks estimates. However, identifying individual trees using medium resolution satellite images is often not feasible. The use of advanced technologies, such as machine learning and satellite imagery in environmental management plays a key role in biodiversity conservation and can potentially fill this gap. The use of readily available Maxar satellite imagery makes conservation approaches accessible and cost-effective, which is crucial for widespread adoption, especially in resource-limited settings or for large-scale studies. This study aims to improve the identification of vegetation in dryland ecosystems by integrating deep learning methods to remote sensing. The primary objective was to distinguish vegetation from rocks or shadows in these areas, which is often challenging due to the dark appearance of vegetation and the similar visual features of the landscape, such as landforms textures, water bodies or man-made objects.. To address this challenge, a Vision Transformer (ViT) deep learning model was developed to estimate near infrared (NIR) spectral bands from high resolution Maxar Satellite images. By enhancing the spectral richness, the model aids in the differentiation of vegetation. Maxar satellite imagery was primarily used because of its accessibility through Google services, making it ideal for planning initial surveys to identify areas of interest for more detailed study. The accuracy of the model was validated against high-resolution SkySat NIR imagery, and achieved an R2 of 0.92. The obtained NIR band helped to clearly distinguish vegetation from non-vegetative surfaces such as soil, rocks, and water, which were not as discernible in RGB imagery alone. The use of a deep learning method to estimate a synthetic NIR band proved to be cost-effective and efficient for large-scale identification of individual trees and shrubs in drylands, overcoming the limitations of medium-resolution satellite imagery. The findings of the study are crucial to the conservation of biodiversity and offer a practical approach for environmentalists and researchers. Future work includes expanding the dataset to include various dryland environments, and integrating additional data sources such as soil data and topographic features for a more comprehensive analysis.

In line with Saudi Vision 2030 and the ambitions of the Saudi Green Initiative to plant 10 billion trees and protect 30% of Saudi Arabia’s land and sea areas, large parts of Saudi Arabia are being transitioned into nature reserves and undergoing regreening activities, while also fencing off large areas for vegetation restoration and protection. To effectively manage the regreening and restoration initiatives, it is imperative to frequently monitor biomass changes over time and quantify the accumulation of biomass to determine if the restoration and tree-planting initiatives have the intended outcomes. However, landscape responses to regreening and vegetation restoration are poorly understood in hyper-arid rangelands. Environmental processes within different habitat zones might differ, which further complicates biomass monitoring. Here, we present a framework that is currently being applied across multiple hyper-arid rangelands in Saudi Arabia to estimate biomass within newly established nature reserves. The initial work focused on developing a scaling approach between ground, unmanned aerial vehicle (UAV), and satellite image data, including PlanetScope and Sentinel-2 imagery. Field sites were identified using Google Earth imagery and a number of criteria, including a 5-year Sentinel-2 NDVI time-series to identify greening events, terrain characteristics based on DEM data, and differences in vegetation functional types (annual and perennial grass/herbs/forbs, shrubs and trees), soil types, and habitats. Field-based measurements at selected sites focused on determining biomass of different vegetation functional types, using a double-sampling approach, including a limited number of destructive samples, and a large number of biomass estimates based on the structural and dimensional characteristics of the destructive samples. UAV-based light detection and ranging (LiDAR) and multispectral data were obtained for each site to estimate biomass from the field-based samples using different machine and deep learning approaches (random forest, support vector machines, vision transformer, UNet with attention encoder). It was found that LiDAR data provided useful information on vegetation height and volume, which improved the biomass estimates, whereas the multispectral data enabled discrimination between photosynthetic and non-photosynthetic vegetation components. Based on the UAV-derived estimates of biomass, a scaling approach was applied to estimate biomass from both PlanetScope and Sentinel-2 data, with results demonstrating that the higher spatial resolution of the PlanetScope data improved the accuracy due to the very sparse vegetation in most parts of the hyper-arid rangelands of this study. While hyper-arid rangelands are generally underrepresented in research studies worldwide, this research provide a viable framework for assessing vegetation dynamics of biomass both seasonally and annually.

Key research in movement ecology is investigating shifts in animal behavior and identifying the factors that induce alterations in movement behavior and mechanics. The impact of natural environments and human activities on the underlying behavioral processes of domestic goats are still being elucidated. We applied seasonal multivariate Hidden Markov Models (HMMs) to characterize the fine-scale movements (30- second intervals) of GPS-tracked Zhongwei goats for 124 days and determine how grazing intensity, seasonal food resources, terrain factors and daylight hours affect movement behavior in the mountain grassland in China. We classified the goats’ activities as two basic behavioral states of foraging (low step length, varied and undirected turning angle) and travelling (long step lengths, low and directed turning angles). Grazing intensity, a management factor, exerted the most significant influence on goats across different seasons. Additionally, factors such as daylight hour and slope had a more pronounced impact on their movement activities compared to the normalized difference vegetation index (NDVI). Elevation and solar radiation were found not explain much of the variability in movement behavior of goats. Their probability of foraging behavior was most likely to increase with grazing intensity, slope, diurnal hours and NDVI. In addition, the percentage time allocation of foraging was higher in spring and winter with lower food resources periods and shorten daylight hours, than summer and autumn with larger food resources and long daylight hours. The foraging percentage increased from morning to afternoon. HMMs are found useful for disentangling movement behavior and understanding how goats respond to seasonal grazing intensity, time of daylight, NDVI and slope. Our findings underscore the importance of accounting for interactions between movement behavior and gazing management, not only the environmental factors and behavioral rhythms, when assessing the movement characteristics and behavioral transitions of goats. These results are important for designing grazing management strategies that satisfy ecological and socioeconomic demands on mountain grassland ecosystems.

It is essential to study the characteristics of land use change in the Fuxian Lake basin, a plateau lake in Yunnan Province, and to predict the basin’s future trend of land use change for sustaining the key carriers of current national major strategies such as ecological civilization, green development, and rural revitalization. This study used the Fuxian Lake basin as the subject and based on the seven periods of historical land use data, added six driving factors including human and natural factors, applied patches to generate the land-use simulation model (PLUS), and forecasted and analyzed the characteristics of land-use change in the basin in 2048 under the three scenarios of natural trend development, ecological protection, and production protection. The results showed that: (1) the overall simulation accuracy of the model was 79.14%, Kappa index was 0.73, FOM value was 0.29, and the model’s consistency was high. The model can be used to simulate future land-use changes in the Fuxian Lake basin. (2) In the natural development scenario, land-use development and the driving factors in the basin have the following relationship: rainfall and trunk road distance significantly impacted the types of land that emphasize ecological conservation and product development. Elevation and soil distribution characteristics had a significant impact on land types focused mainly on water zones and ecological protection. The land types mainly focused on urban construction were greatly affected by elevation, trunk road distance, GDP per capita, and other factors. (3) The main direction of land-use change in the watershed is the mutual conversion between farmland and forest land, with the continuous expansion of construction land. In the production protection scenario, the area of farmland increased by 44.79 hm². In the ecological protection scenario, the area of arbor forest land increased by 37.85 hm² and the area of shrub forest land decreased by 62.37 hm². (4) From the perspective of spatial distribution patterns, the regional hotspot change blocks are mainly concentrated in the north of the basin, along the coast of the north of Fuxian Lake, and the southern land. In general, the PLUS model had good applicability in this study. The simulation results of the different scenarios were in line with the land development in the Fuxian Lake basin and can provide scientific reference for land-space planning, ecological and production land constraints, and coordination of development in the Fuxian Lake basin.

Southwest China is an important route for invasive species. In this study, 49 invasive plants of 15 families and 41 genera were found within 373 grassland sampling sites of Guizhou Province, a typical karst mountainous region with a high invasion risk located in Southwest China. Invasive plants could be found within over 90% of the grassland sampling sites, and malignant invasive species were found in 60% of the sites. In about 30% of the sampling sites, more than one malignant species coexisted. The malignant invasive species were mainly distributed in the southwestern part of Guizhou Province. Their distribution patterns were affected by environmental and traffic factors; they preferred areas with low elevation, high temperature, high rainfall, high soil nutrient content, and traffic accessibility and could adversely affect plant cover and biomass. Conversely, seriously invasive species and other low-level invasive species had a positive or neutral effect on grassland communities. Therefore, the focus of invasive plant control measures should be on malignant invasive species. Specific control policies and practices, especially in areas with resource-rich environments and well-developed traffic networks, should be carried out to facilitate grassland ecosystem sustainability and to prevent the spread of invasive species to inland China.