Soil erosion is a global environmental problem. The rapid monitoring of the coverage changes in and spatial patterns of photosynthetic vegetation (PV) and non-photosynthetic vegetation (NPV) at regional scales can help improve the accuracy of soil erosion evaluations. Three deep learning semantic segmentation models, DeepLabV3+, PSPNet, and U-Net, are often used to extract features from unmanned aerial vehicle (UAV) images; however, their extraction processes are highly dependent on the assignment of massive data labels, which greatly limits their applicability. At the same time, numerous shadows are present in UAV images. It is not clear whether the shaded features can be further classified, nor how much accuracy can be achieved. This study took the Mu Us Desert in northern China as an example with which to explore the feasibility and efficiency of shadow-sensitive PV/NPV classification using the three models. Using the object-oriented classification technique alongside manual correction, 728 labels were produced for deep learning PV/NVP semantic segmentation. ResNet 50 was selected as the backbone network with which to train the sample data. Three models were used in the study; the overall accuracy (OA), the kappa coefficient, and the orthogonal statistic were applied to evaluate their accuracy and efficiency. The results showed that, for six characteristics, the three models achieved OAs of 88.3–91.9% and kappa coefficients of 0.81–0.87. The DeepLabV3+ model was superior, and its accuracy for PV and bare soil (BS) under light conditions exceeded 95%; for the three categories of PV/NPV/BS, it achieved an OA of 94.3% and a kappa coefficient of 0.90, performing slightly better (by ~2.6% (OA) and ~0.05 (kappa coefficient)) than the other two models. The DeepLabV3+ model and corresponding labels were tested in other sites for the same types of features: it achieved OAs of 93.9–95.9% and kappa coefficients of 0.88–0.92. Compared with traditional machine learning methods, such as random forest, the proposed method not only offers a marked improvement in classification accuracy but also realizes the semiautomatic extraction of PV/NPV areas. The results will be useful for land-use planning and land resource management in the areas.

Natural solutions by which humans can overcome challenges have been severely hampered by biodiversity losses. It is essential to understand the key natural variables that influence changes in community diversity to maintain ecosystem function. The Ziwuling area has a natural recovery succession history of 150 years. Therefore, a survey was conducted to compare species composition characteristics across different vegetation recovery stages while also providing a phylogenetic and taxonomic response to the correlation between beta diversity and soil stoichiometry. The results showed that beta diversity and endemic plants had a similar single-peak temporal pattern. Soil organic carbon (SOC) accumulation was significantly positively correlated with total nitrogen (TN) and was also negatively correlated with total phosphorus (TP). Overall, soil TN, time since vegetation restoration (TVR in years), and W-SS (endemic woody plants at each stage) were able to explain 94.3% of the total variation in beta diversity. Temperate species such as Carex lanceolata, Lespedeza bicolor, and Sophora davidii are the basis for community construction. Community beta diversity patterns are the result of a mixture of ecological (e.g., climate patterns and soil nutrients) and evolutionary processes. This study combined plant resource needs with how they respond to natural recovery times in order to provide useful knowledge to protect biodiversity, the nutrient cycle, and the function of restoration ecology.

Coexistence between species within plant communities is a key issue in the practice of revegetation, forest management, and biodiversity conservation. Vegetation restoration is critical to control soil erosion and improve the ecological environment on the Loess Plateau. Here, we investigate the interspecific relationships of dominant plants during natural vegetation succession on the Loess Plateau. The results suggest that, under the ecological process of environmental filtering, species within communities can reduce interspecific competition and promote species coexistence via spatial heterogeneity and temporal asynchronous differences. The ecological niche overlap index (O_ik) significantly and positively correlated with the strength of interspecific associations. Most species pairs had weak competition and more stable interspecific relationships. The results of the χ² test showed that 317 species pairs were positively associated and 118 were negatively associated. The community is in a positive succession process, and the interaction relationship between species tends to be neutral. We should enhance the protection of positively associated species and pay attention to negatively associated species during forest management. Results revealed that Carex lanceolata Boott and Lespedeza bicolor Turcz coexisted easily with other species for mutual benefit, which could help build artificial forestland of native species to improve the ecological function.

While the link between plant species diversity and biomass has been well‐studied, the impact of evolutionary diversity on community biomass across long timescales or ongoing change remains a subject of debate. We elucidated the association between evolutionary diversity and community aboveground biomass (AGB) using an ideal experimental system with over 150‐year history of natural vegetation regeneration. Higher phylogenetic diversity facilitated the sampling effect under the influence of environmental filtering, and caused an increase in AGB. Phylogenetic structure varied from aggregation to dispersion during the later period of vegetation recovery (70–150 years), which was correlated with increases in niche complementarity and increasing AGB. Woody plant evolutionary diversity was used as a key to predict the relationship between vegetation recovery and AGB, with a total explanatory power of ~84.7%. Mixed forests composed of evergreen conifers and deciduous broadleaf forests had higher carbon sequestration potential than that of pure forests, which is advantageous for increasing top‐stage AGB. This research expands our knowledge of the causes and effects of biodiversity and ecosystem function dynamics over time and space, which is important for accurately predicting future climate change effects.

Soil erosion is a global environmental problem. The rapid monitoring of the coverage changes in and spatial patterns of photosynthetic vegetation (PV) and non-photosynthetic vegetation (NPV) at regional scales can help improve the accuracy of soil erosion evaluations. Three deep learning semantic segmentation models, DeepLabV3+, PSPNet, and U-Net, are often used to extract features from unmanned aerial vehicle (UAV) images; however, their extraction processes are highly dependent on the assignment of massive data labels, which greatly limits their applicability. At the same time, numerous shadows are present in UAV images. It is not clear whether the shaded features can be further classified, nor how much accuracy can be achieved. This study took the Mu Us Desert in northern China as an example with which to explore the feasibility and efficiency of shadow-sensitive PV/NPV classification using the three models. Using the object-oriented classification technique alongside manual correction, 728 labels were produced for deep learning PV/NVP semantic segmentation. ResNet 50 was selected as the backbone network with which to train the sample data. Three models were used in the study; the overall accuracy (OA), the kappa coefficient, and the orthogonal statistic were applied to evaluate their accuracy and efficiency. The results showed that, for six characteristics, the three models achieved OAs of 88.3–91.9% and kappa coefficients of 0.81–0.87. The DeepLabV3+ model was superior, and its accuracy for PV and bare soil (BS) under light conditions exceeded 95%; for the three categories of PV/NPV/BS, it achieved an OA of 94.3% and a kappa coefficient of 0.90, performing slightly better (by ~2.6% (OA) and ~0.05 (kappa coefficient)) than the other two models. The DeepLabV3+ model and corresponding labels were tested in other sites for the same types of features: it achieved OAs of 93.9–95.9% and kappa coefficients of 0.88–0.92. Compared with traditional machine learning methods, such as random forest, the proposed method not only offers a marked improvement in classification accuracy but also realizes the semiautomatic extraction of PV/NPV areas. The results will be useful for land-use planning and land resource management in the areas.