中国科学院水利部水土保持研究所机构知识库

东北黑土区坡地耕作方式多样，侵蚀过程复杂，受到气候、土壤母质、地形地貌、生物作用和人为活动等多种因素的共同影响，土壤侵蚀机理尚不清楚，亟需进行系统的研究。本研究针对东北黑土区不同水土保持措施的径流小区的监测数据，系统分析了坡面耕作措施对坡面产流产沙的影响及其对降雨类型的响应，评价东北黑土区现有耕作措施的适宜性和有效性；采用冗余分析和结构方程模型定量化分析降雨、地形、和水土保持措施等土壤侵蚀因子及其交互作用对黑土坡面产流和侵蚀的影响，研究东北黑土区坡地土壤侵蚀因子间作用机制，同时，基于卷积神经网络对坡面次降雨产流产沙量进行模拟。得到以下主要结果：

（1）海伦市光荣小流域以短历时，小降雨量，低降雨侵蚀力的降雨事件（RⅢ）为主，其次是长时间，中等雨量，低降雨侵蚀力的降雨事件（RⅡ），长时间、大雨量、高降雨侵蚀力的极端降雨事件极少发生（RⅠ）。频发的RⅢ降雨事件对裸坡小区中产流产沙量的累积贡献率最大，分别达到了68%和55%。而耕作处理小区中RⅠ极端降雨事件的土壤流失量累积贡献在10% ~ 47%之间，说明裸地的土壤侵蚀主要由频发的RⅢ降雨事件控制，而坡耕地中严重的高产流产沙的土壤侵蚀事件是由少量的极端降雨事件造成的。单纯从防治水土流失的角度考虑，横坡垄作和免耕是该区最为适宜的水土保持耕作措施，不推荐少耕措施。

（2）由降雨因素（I₃₀和降雨量）和植被覆盖度构成RAD1轴对侵蚀环境的解释关系达到了0.985。坡面产流产沙量与植被覆盖度呈负相关关系，与降雨量和I₃₀均呈正相关关系。坡度与土壤流失量呈正相关关系，而与径流深的关系不明显。降雨因素和植被因素在土壤侵蚀过程中起主导作用，单项解释值分别为0.150、0.089，而植被和水土保持措施因素的共同解释值为0.209，说明水土保持措施等因素起辅助作用。

（3）在坡耕地中，植被特征对坡面产流的控制作用（-0.75）远大于降雨特征（0.23）和地形特征（0.61）的影响，降雨特征（-0.36）与植被特征（0.36）对坡耕地产沙的影响与地形特征（0.28）相比较大，且降雨特征与植被特征的影响大小一致，作用相反。裸坡小区中，降雨-地形特征外生潜变量对坡地产流产沙具有显著并较高强度的正效应（0.74）。降雨因子（包括降雨量和降雨侵蚀力）对土壤侵蚀的影响远大于坡度对土壤侵蚀的影响（0.10），且降雨量的影响（0.95）大于降雨侵蚀力（0.84）。然而，坡耕地中降雨侵蚀力（0.99）的影响大于降雨量（0.79）。

（4）深度学习的卷积神经网络模型对土壤流失量进行预测的方案要优于传统的USLE模型，卷积神经网络对径流深和土壤流失量的拟合线斜率分别为0.61和0.64，纳什系数分别达到了0.62、0.75。而USLE模型的斜率仅为0.05，纳什系数为-12.28。

The black soil region of Northeastern China is characterized by diverse farming practices on sloping farmland, where complex and severe erosion processes occur. Due to the joint influence of climate, soil parent material, topography, biological action and human activities, the mechanism of soil erosion is not clear, so it is urgent to carry out systematic research. This paper based on plot experiments aims to (1) systematically analyzes the effects of slope tillage measures on runoff and sediment yield and their responses to rainfall types, and evaluated the suitability and effectiveness of existing tillage measures in black soil area of Northeast China;(2) comprehensively quantify the impact and interaction of soil erosion factors including rainfall, topography, soil and water conservation measures on runoff and sediment yield study interaction mechanism of soil erosion factors on sloping land; (3) establish the soil erosion prediction model based on convolution neural network. The main results are listed as follows:

(1) The primary rainfall pattern is regime III (RIII) in the Guangrong catchment in Hailun, characterized by a short duration, slight intensity, and moderate rainfall depths. The RIII rainfall pattern is followed by regime II (RII) with a long duration and moderate intensity. It is rare of regime I (RI), characterized by large rainfall events in the form of long durations, high intensities and large depths. The cumulative contribution rate of frequent R Ⅲ rainfall events to runoff and sediment yield in bare slope plot is the largest, reaching 68% and 55% respectively. The cumulative contribution of R I extreme rainfall events to soil loss in cultivated plots ranged from 10% to 47%. The results show that the amount of soil loss in bare land is mainly controlled by frequent RⅢ rainfall events, while the serious erosion events with high runoff and sediment yield in slope farmland are caused by extreme rainfall events (RⅠ). As far as soil conservation is concerned, the contour-ridge tillage and non-tillage farming is recommended as the most suitable measure, while it is not recommendation for the less tillage farming practice in the black soil region of north-eastern China.

(2) The RAD1 axis composed of rainfall factors (I₃₀ and rainfall depth) and vegetation coverage could explain the erosion environment to 0.985. There was a negative correlation between runoff and sediment yield and vegetation coverage, and a positive correlation between runoff and sediment yield and rainfall andI₃₀. There was a positive correlation between slope and soil loss, but no significant correlation between slope and runoff depth. Rainfall factors and vegetation factors played a leading role in the process of soil erosion, and the single interpretation values were 0.150 and 0.089, respectively, while the common explanation value of vegetation and soil and water conservation measures was 0.209, which indicated that soil and water conservation measures played an auxiliary role.

(3) In the slope farmland, the control effect of vegetation characteristics on runoff production (- 0.75) is much greater than that of rainfall characteristics (0.23) and terrain characteristics (0.61), The influence of rainfall characteristics (- 0.36) and vegetation characteristics (0.36) on the sediment of slope farmland is larger than that of topography characteristics (0.28), and the influence of rainfall characteristics and vegetation characteristics is the same, but the effect is opposite. In the bare slope plot, the exogenous latent variable of rainfall terrain characteristics has significant and high intensity positive effect on runoff and sediment (0.74). The impact of rainfall factors (including rainfall depth and rainfall erosivity) on soil erosion is much greater than that of slope (0.10), and the impact of rainfall (0.95) is greater than that of rainfall erosivity (0.84). However, the influence of rainfall erosivity (0.99) was greater than that of rainfall (0.79).

(4) The convolution neural network is used to establish the prediction model of soil erosion in the black soil area of Northeast China, and the accuracy of the model is verified by using the universal soil loss equation (USLE). The results show that the convolution neural network model based on deep learning is better than the USLE Model in predicting soil loss amount. The fitting line slopes of convolution neural network for runoff depth and soil loss amount are 0.61 and 0.64 respectively, and the Nash coefficients are 0.62 and 0.75 respectively. The slope of USLE Model is only 0.05 and the Nash coefficient is -12.28.