中国科学院水利部水土保持研究所机构知识库

黄土高原位于干旱半干旱区，退耕前大量的毁林开荒导致该区域水土流失严重，生态环境脆弱，成为世界上具代表性的干旱半干旱生态系统和侵蚀景观。为改善这一状况，采取了建造梯田和淤地坝等多种措施，但该区水土流失仍然严重。因此，1999年我国推行了退耕还林（草）工程，即将坡耕地退耕为林地、灌木地或草地，以改善坡耕地水土流失问题。在水土流失的综合治理过程中，常遵循“土是基础，水是关键，植被是标志，产业是保障，水土保持是目标”的理念指导；因此，为更好的实施退耕还林（草）工程，需要充分了解其在时空尺度上对“土—水—植被”的影响。同时，探讨该影响与农业生产间的平衡也是必要的。“3S”技术集成了卫星定位、遥感技术、计算机技术、空间技术等对空间数据进行采集、管理、分析和表达，从而为评估黄土高原退耕还林（草）工程对水土流失的改善提供了机会。本论文在“3S”技术支持下，基于采样数据、文献数据、遥感数据、气象站监测数据、FLUX监测数据、统计年鉴数据等，选择黄土高原腹地典型小流域——纸坊沟流域和整个黄土高原土壤团聚体稳定性、黄土高原土壤水分、黄土高原植被总初级生产力及农业生产为研究对象，利用空间分析详细探究纸坊沟流域团聚体稳定性在景观尺度上的空间结构，并对其进行空间预测和空间贡献分析，利用趋势分析探究黄土高原退耕还林（草）工程前后土壤水分的时空变化及其驱动要素，以及监测黄土高原植被总初级生产力GPP对退耕还林（草）工程的时空响应，并逐像元探测其时空变化出现的拐点/断点，最后结合统计年鉴数据分析黄土高原农业活动的时空变异，以期为黄土高原退耕还林还草下的生态环境建设和社会经济的可持续发展提供理论依据。本研究取得主要结果如下：
（1）团聚体稳定性指数平均重量直径MWD、水稳性团聚体含量WSA_>0.25和可蚀性因子K值的最优半变异函数模型分别是球状模型、指数模型和高斯模型；三指数低的块金值和基台值表明了实验具有较小的采样误差、随机误差和总变异。变程信息证明在0–10cm土层，它们均具有较强空间自相关性；10–20cm的K值具有最大的空间异质性和最小的空间相关性。基底效应强调了MWD和WSA_>0.25具有强的空间相关性，二者主要受到本质因素的作用，对于K值，人为作用不可忽视，尤其在表层。局部空间自相关性分析进一步证明了强的农业活动和低的团聚体稳定性、高的土壤可侵蚀性具有紧密联系，其中特殊点分析发现短期内，耕园地转为灌木可显著改善土壤结构，尤其是对表层。进一步利用景观指数量化土地利用类型和结构，结合土壤性质、地形因子、温度、干旱度和植被覆盖数据，预测MWD、WSA_>0.25和K值的空间分布，发现团聚体稳定性指数的空间变异受土壤性质、景观结构、地形、植被活动和水热条件的综合影响，并且预测模型在很大程度上依赖于土地利用类型和结构的量化，这在以往的研究中常被忽略。土壤变量的排除虽会降低MWD和WSA_>0.25预测性能，但对K值的预测仍较理想，说明利用辅助数据预测团聚体稳定性指数的空间分布的可行性。在此基础上，量化了各个影响因素对团聚体稳定性的贡献（包括直接和间接贡献），结果表明土壤有机碳SOC、高程、坡度、耕园地斑块所占面积、草地斑块所占面积、pH、非晶质氧化铁、碳酸钙、季节性温差和地形湿度指数起着主要作用，越往表层，自然因素直接作用越强，土地利用类型和景观结构直接影响SOC、坡度等从而间接贡献团聚体稳定性；越往深层，土壤性质的直接和间接作用均加强；
（2）黄土高原尺度下，人为活动强度的差异性导致退耕前后土壤团聚体稳定性的控制因素不同，退耕前主要受土壤质地、气候因子、SOC、地形因子的控制，退耕后土地利用类型和景观结构的作用由不显著到较强，证明了人为干扰对土壤团聚体稳定性的显著作用，坡度由负效应转为正效应，说明坡耕地转为林灌草地有益于土壤结构的改善。另外，线性回归模型的性能反映出该尺度团聚体稳定性空间预测难度大，后期需更详细的规划，本论文是对该尺度相关研究的初探，为后期深入研究提供一定的依据；
（3）提出了一种基于卫星数据产品的综合方法，对土壤水时空动态中植被的驱动进行系统和定量评估。该方法也可应用到其他未布有土壤水监测网区域。首先证明了GLEAM土壤水数据集在评估黄土高原土壤水时空动态变异中的有效性，研究发现在34年的时间尺度上，植被恢复在植被区土壤水分动态变异中发挥主导作用，驱使较湿润区域（年降雨> 450 mm）变干燥，较干燥区域变湿润，这是植被结构差异、密度、树龄和物种综合促成的。降雨仅对裸地和稀疏植被区土壤水有显著正效应。而蒸散对裸地、稀疏植被区或茂密植被区的土壤水有重要影响。空间尺度上，蒸散和降雨作用更为显著，植被覆盖对土壤水动力学的驱动作用相对较弱，蒸散在还林区土壤水分动态中发挥主导作用，尤其是在退耕还林（草）工程早期阶段（2000–2010年）；降雨和植被恢复对还草地土壤水的贡献远大于蒸散的作用。因此，空间分析对明确土壤水和植被恢复间相互作用是必要的；建议在半湿润地区不应进一步退耕，但在干旱、半干旱地区稀疏以及过度稀疏的植被覆盖区可进一步恢复；
（4）利用通量观测站监测数据验证GLASS GPP数据集在黄土高原地区的适用性；并进一步利用趋势分析探测到1982–2015年来GPP整体的增加趋势，但分段函数分析发现所有像元的变化速率和趋势是有显著性差异的，且在不同阶段也不同，主要呈现出先快速增加后缓慢增加（拐点）、先增加后减少的趋势（断点）；平均拐点发生在2005年，平均断点在2003年，主要像元拐点/断点集中在2011–2015年，强调了在不同的地理位置，退耕还林（草）工程的方式需不同，其强度也应因地制宜，否则可造成不可逆转的负面生态效应，且主要像元的植被恢复已经达到阈值；
（5）退耕后黄土高原粮食产量并未因种植面积减少而减少，反而大范围县域呈现增加趋势，肥料施用量的增加是原因之一，从趋势分析可知部分县域的产量波动较大，这种情况不利于退耕还林（草）工程成果的维护，如果农民生计较为单一，则会为提高产量而增加耕地以抵御产量波动。为解决这一问题，应考虑拓宽农民生计降低其对耕作的依赖性。
综上所述，黄土高原退耕还林（草）工程虽致该区耕地面积减少，但并未导致其农业生产力下降；考虑到退耕对“土—水—植被”的影响，在工程的实施过程中需因地制宜。这对于揭示土壤侵蚀规律、更加有效地开展水土保持工作、减少入黄泥沙具有重要意义，为后期治理工作提供经验和建议，为黄土高原地区实现生态和经济的可持续发展提供理论支撑。

The Loess Plateau (LP) is one of the most representative drylands ecosystems and eroded landscape in the world. Intensive agricultural practices had caused extremely severe soil erosion in the LP. Thus, several practical measurements on soil and water loss control have been implemented since the 1980s to give information on optimizing the land use pattern and configuration, including the building of terraces and sediment-trapping dams, banning grazing and afforestation in bareland. But soil erosion was still severe in cultivated slope cropland. Then, "Grain for Green" project (GGP), one of the most well-known revegetation programs, has been implemented in the LP to restore the fragile ecosystems by converting slope cropland (> 25°) and bareland into grassland, shrubland and woodland since 1999 and expanded to the whole plateau in 2000. In the process of comprehensive management of soil erosion, we should follow the concept of "soil is the foundation, water is the key, vegetation is the sign, industrial structure is the guarantee, soil and water conservation is the goal". Hence, for a better implementation of the GGP, we needs to fully understand its impact on the spatio-temporal distribution of ‘soil-water-vegetation’. Meanwhile, it is necessary to explore the balance between ecological restoration and agricultural production. 
The ‘3S’ technology integrates satellite positioning, remote sensing technology, computer technology, and space technology to collect, manage, analyze, and express spatial data. It provides an opportunity to evaluate the improvement of GGP. In this paper, with the support of ‘3S’ technology, we used the sampling data, literature data, remote sensing, meteorological station data, FLUX data, statistical yearbook data, etc. Zhifanggou catchment, the typical small watershed of the LP, and Loess Plateau were selected as our study area. Then, we analyzed the dynamics of soil aggregate stability at two spatial scales, soil moisture, gross primary productivity and agricultural production at the Loess Plateau scale, respectivley. A spatial analysis of soil aggregate stability and erodibility (K factors) was performed to understand the formation processes of aggregates at catchment scale in detail. To understand the formation processes of aggregates by a spatial analysis, a prediction model combining soil properties with natural and human factors should be developed to improve the accuracy of the spatial interpolation of soil aggregate stability indices. Then, we analyzed the contribution of influence factors to soil aggregate stability indices. Spatio-temporal dynamics of soil moisture (SM) and their driving factors were explored by using remote sensing data and trend analysis. The remote sensing data was used to monitor the spatio-temporal dynamics of the GGP and to detect its turning points or break points. The statistical yearbook data were used to analyzed the spatio-temporal variation of agricultural activity. Our results provide theoretical basis for ecological environment construction and sustainable social and economic development. The main results are as follows: 
(1) Spherical, Exponential, and Gaussian models were proved to be the best-fit models in describing the spatial variability of aggregate stability. The low nugget values and sill indicate small sampling errors or random variabilities and total variance. The mean weight-diameter (MWD, mm), water-stable aggregates greater than 0.25 mm (WSA_>0.25, %) and K factors had a larger range of spatial autocorrelation in 0–10 cm layer than in 10–20 cm layer. The lowest range value was found for K factors at 10–20 cm layer, indicating a maximal heterogeneity and a lowest spatial dependence. Nugget/sill ratios C0/(C0+C) showed a very strong spatial dependence for MWD and WSA_>0.25. They are mainly controled by the intrinsic factors. For the K factors, the human impact cannot be ignored, especially in the soil surface. Local Indicators of Spatial Association further proves that strong agricultural activities are closely related to low soil aggregate stability and high soil erodibility. Grazing can significantly reduce the aggregate stability. It is interesting to note that one high-high relationship in 0–10 cm soil layer and low-high relationship in 10–20 cm soil layer were located in shrubland with a one-year revegetation duration. It has a high-high relationship with those woodlands at 0–10 cm soil layer, indicating that the conversion of farmland to shrubland has a stronger positive effect on the surface aggregates than on the sub-surface aggregates after a short-term soil restoration. The landscape metrics can be used as an indicator of land use type and landscape structure. The prediction models combining soil properties with natural and human factors were developed for predicting the sapatial distribution of MWD、WSA_>0.25 and K factors. The spatial variability of aggregate stability indices is synergistically affected by the soil, topography, vegetation, and human factors. The spatial variability and prediction modeling of aggregate stability indices are highly dependent on the quantification of land use type and landscape structure (the spatial structure of landscape elements and the connections between the different ecosystem types or landscape elements). It has received little attention in previous studies. The performance of the models was relatively lower when excluded all soil variables for MWD, WSA_>0.25, and K factor, but still satisfactory, indicating that the prediction of the spatial distributions of aggregate stability indices with easily available auxiliary data is practicable and effective. The contribution of each influencing factor was further quantified and the direct and indirect contributions were distinguished. The results show that the soil organic carbon (SOC), elevation, slope, corpland percentage of landscape area, grassland percentage of landscape area, pH, oxalate-extractable iron, calcium carbonate, seasonal temperature difference, and topographic wetness index play a major role. In surface soil, there is a stronger direct effect of natural factors. The land use type and landscape structure indirectly contributes to the aggregate stability by directly affecting SOC, slope, etc. And going deeper, the direct and indirect effects of soil properties are both enhanced;
(2) At the Loess Plateau scale, the difference in the human activities intensity leads to different influence factors affecting the soil aggregate stability before and after the GGP. Before the GGP, it is mainly controlled by soil texture, climate factors, SOC, and topographical factors. The effect of soil texture after GGP is more stronger than that before GGP. The effect of land use type and landscape structure changes from insignificant to strong. It proves that human disturbance has a significant effect on the soil aggregate stability. The enhanced imapct of land use types and landscape structure after the GGP further verifies this result. The change of the slope impact sign proves that converting the sloping farmland to forest, shrubland and grassland is beneficial to the improvement of soil structure. In addition, the performance of the linear regression model show that the spatial prediction of aggregate stability at LP scale is difficult, and more detailed planning is needed. This paper is a preliminary exploration of the scale-related research and provides a basis for detailed research in the future.
(3) An integrated method based on a variety of satellite data products had proposed to provide a systematic and quantitative assessment of the revegetation drivers in spatio-temporal dynamics of SM. And this method can be applied to other places without a SM monitoring network. This study confirms the availability of GLEAM SM in evaluating the spatio-temporal variations of SM in the LP. At the 34-year time scale, revegetation plays a dominant role in SM dynamics in vegetated areas, and turns the wet region (MAP > 450 mm) to be dry and dry region to be wet, which is attributable to the differences in vegetation structure, density, growth age, and species.The significantly positive effect of precipitation on SM is only found in bareland and sparsely vegetated area. Evapotranspiration has an important effect on SM in bareland, sparsely vegetated area or densely vegetated area. At the spatial scale, the driving effect of vegetation cover on SM dynamics is relatively weak due to the more significant role of evapotranspiration and precipitation. Evapotranspiration plays a dominant role in SM dynamic in revegetated woodland, especially in the early stage of GGP (From 2000 to 2010), while precipitation and vegetation cover have much greater contributions to SM than evapotranspiration in revegetated grassland. Therefore, our findings highlight the importance of spatial analysis to investigate the interactions between SM and vegetation activity and alert the excessive reliance on afforestation. Our study suggests that vegetation should not be further expanded in semi-humid areas, but should be further restored in arid and semi-arid areas with sparse or excessively sparse vegetation cover.
(4) The applicability of the GLASS GPP dataset in the LP region was first verified by using FLUX monitoring data. Then, trend analysis was used to detect the overall trend of GPP over the past 1982–2015. The results of piecewise function analysis show that the change rate and trend of all pixels are significantly different, and different at different stages, mainly showing a rapid increase first, then slowly increase (turning points), or first increase and then decrease (break points). The average turning points is in 2005 and the average break point is in 2003. The main pixel inflection points/breakpoints are concentrated in 2011–2015. The results emphasize that the methods of GGP need to be different in different geographical locations, and the intensity should be adapted to local conditions. Otherwise, the goal of ecological restoration will not be achieved, but will cause irreversible negative ecological effects. And the vegetation restoration of the main pixel has reached their threshold.
(5) After GGP, the grain yield of the LP has not decreased because of the reduction of planting area. A large-scale county area has shown an increasing trend. One of the reasons is the increase in fertilizer application. From the trend analysis, we knows that the yield of some counties fluctuates greatly. In this case, the maintenance of the GGP is unfavorable. If the farmers' livelihood is relatively simple, more land will be farmed to increase crop yield. To address this, consideration should be given to broaden farmers' livelihoods and reduce their dependence on farming.
Although the GGP on the LP has reduced the cropland area, it has not caused the decrease in crop yield. Given the impact of GGP on ‘soil-water-vegetation’, our paper emphasizes that the GGP implementation needs to be adapted to local conditions. Our results are of great significance for revealing the law of soil erosion, carrying out soil and water conservation work more effectively, reducing sediment, providing successful experiences and suggestions for the development of later management work, and theoretically supporting the ecological and economic sustainable development.