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.

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.

As the fundamental foundation for human survival and development, the ecological environment is an important guarantee for social progress. With the accelerated development of society, the impact of human activities on the ecological environment has become increasingly significant. Fu County (4,182 km2) is located in the gully area of the Loess Plateau, an important ecological security barrier in the north of China. Clarifying the current situation of the ecological environment and the temporal and spatial changes of ecological environment quality in Fu County is conducive to the construction of ecological civilization. It provides scientific support to improve ecological quality.Based on the remote sensing images of Landsat in Fu County from 1992 to 2019, we considered four indexes of Normalized Difference Vegetation Index (NDVI), Tasseled Cap Transformation Humidity Index (Wet), Normalized Difference Bare Soil Index (NDBSI), and Land Surface Temperature (LST) to represent greenness, humidity, dryness and heat of the environment. The principal component analysis was uesd to establish a remote sensing ecological index model (RSEI). And then we evaluated the ecological quality in the area in the past 30 years.The results showed that: (1) The average correlation between RSEI and four indexes was 0.874, which means RSEI can comprehensively reflect the ecological environment quality status of the study area compared with a single index. (2) The regional mean values of RSEI in Fu County showed an upward trend from 0.621 to 0.806 between 1992 and 2019, and the increase in ecological environment quality gradually increased from northwest to southeast of the county. (3) The increasing area and magnitude of the ecological level of the county from 1992 to 2019 were much greater than the decling area and magnitude. Over the past 30 years, the relatively sparsely populated townships of Zhiluo, Niwu and Zhangjiawan, which have a high degree of vegetation cover, have always been in a better state, while the densely populated townships of Jizixian, Jiaodao and Yangquan has always been poor. From the perspective of the degree of change in the quality level of the ecological environment, all townships have been improved to different degrees, and the improvement is the most obvious in Jizixian Town, Jiaodao Town and Yangquan Town.The improvement of ecological environment quality level from good to excellent is probably because that the implementation of many ecological projects and the increase in rainfall, and the decrease in bare soil area and surface heat decreased in Fu County in the past 30 years.This study can provide method reference and data support for ecological environment monitoring, governance and protection. Keywords: Fu County; Ecological environment quality evaluation; RSEI; Dynamic monitoring

Estimation and evaluation of carbon sink of terraces measures in Loess Plateau : A Case Study of Wuqi CountyKaiyang Yu1, Xiaoping Zhang1, Hui Cheng2, Weinan Sun1, Wenliang Geng1, Haojia Wang1, Xuanhao Liu1, Liang He1, Yujie Zhang1, Zefeng An1, Yichen Wang11Institute of Soil and Water Conservation, Northwest A&F University, Yangling, Shaanxi 712100, China2Institute of Soil and Water Conservation, Chinese Academy of Sciences and Ministry of Water Resources, Yangling, Shaanxi 712100, China Extensive terracing to cultivate land or plant trees and grasses, will affect biogeochemical processes in terrestrial ecosystem especially in areas with severe erosion. In order to clarify the carbon sequestration function and capacity of terrace in Loess Plateau, terraced fields with different planting types and years were selected to compare with corresponding slope land within Wuqi County (3,791 km2) of Loess Plateau to investigate the distribution characteristics of SOC on soil profile in 100 cm and searching for the influencing factors in the area. The results showed that：(1) Level terrace has significant carbon sink function in Loess Plateau. After the construction of terraced fields, SOC of six soil layers has the same variation trend with the construction years. The total SOC in the profile undergoes a rapid accumulation in the early 5-10 years of terracing construction, then exceeded the control group, and reaching a stable level after 20 years. (2) SOC in terrace dramatically decreases from surface layer to bottom layer, which exhibits surface aggregation effects. The content of SOC in 0-10 cm is significantly higher than that in deep soil layers, and the difference in SOC in deep soil is not significant. (3) The contents and accumulation rate of SOC in terrace are different with the planting types and years. Terraces with forest land generally has higher accumulating rate than other planting types. (4) There is significant correlation between total SOC and organic carbon content of large soil aggregates. Soil aggregates with particle size greater than 0.25mm determined the amount and trend of SOC content in the soil. These results can provide an essential basis for improving the soil fertility and evaluating the benefit of terrace in carbon conservation on the Loess Plateau.Key words: Terrace; Soil organic carbon; Soil aggregate; Loess Plateau

Once one of the most severe soil loss regions worldwide, the Chinese Loess Plateau has experienced large-scale land use changes and vegetation restoration in the past few decades. Understanding how land use change affects soil erosion is critical in the region's ecological construction and land management. In this study, the Beiluo River Basin covering 26,905 km2 in the hinterland of the Loess Plateau was selected to investigate vegetation restoration and its impacts on soil loss rates over the last 50 years. Results show that land use in the basin has changed considerably, mainly reflected in the upper reaches. From 1970–2020, cropland in the upper reaches decreased by 54%, directly leading to a 9.1-fold increase in forested land. Landsat-NDVI shows vegetation coverage increased from 21.1% to 69.9% over time. Vegetation coverage changed from 48.1% to 78.7% for the entire basin. The Chinese Soil Loss Equation (CSLE) was used and confirmed to be satisfactory with a high coefficient of determination (R2, 0.89) and a strong Nash–Sutcliffe efficiency coefficient (0.72), although an underestimation exists. With the change in land use, the specific soil loss simulated in the upper reaches maintained a high rate of around 8,000 t·km−2·yr−1 from the 1970s to the 1990s, dramatically dropping to 3,058 t·km−2·yr−1 in the 2000s, then attenuated to 1,321 t·km−2·yr−1 in the 2020s. For the entire basin, soil loss rates dropped from 4,090 to 1,848 and 890 t·km−2·yr−1 from the 1970s to the 2000s and 2020s, respectively. Attribution analysis showed that the dominant factor in the change in soil loss rates in the 1980s and 1990s relative to the 1970s was the change in rainfall erosivity for the entire watershed. However, with vegetation coverage increasing to 59.0% in the 2000s, vegetation restoration rapidly converted to the dominant factor contributing 78.3% to soil loss decrease in that period. With expanding vegetation cover, its contribution grew to 84.9% in the 2020s. The shift is evident in each reach of the basin except the terrain-plain area with the majority of farmland. The findings are helpful for sustainable land use planning and socio-economic development on the Loess Plateau and in similar areas.