土壤侵蚀引发了一系列生态环境问题，水力作用下的土壤侵蚀过程是农业非点源污染的关键过程。同氮素相比，磷素更容易被吸附于土壤中，随水土流失过程一起发生迁移，土壤侵蚀是非点源磷流失的重要驱动力。因此，研究土壤侵蚀变化成因及其响应规律对理解非点源磷流失过程具有重要意义。冻融农区极易受到降水和温度变化的影响，水土流失过程较为复杂。国内外有关土壤侵蚀的研究大多从单一影响因素或单一时空尺度出发，大多关注冻融循环作用对土壤性质和土壤侵蚀的影响；而有关土壤侵蚀多影响要素和不同时空尺度的系统性分析较少，对融雪条件下流域土壤侵蚀及解冻土壤水分运动机理的关注较少。因此，本研究将室内试验、田间监测和模型模拟相结合，进行系统性研究，目的在于明确冻融农区土壤侵蚀特征，揭示土壤侵蚀变化成因，分析土壤侵蚀响应规律，为加深理解非点源磷流失驱动潜力提供支持，为流域水土保持与非点源磷污染防治提供建议。 本研究选取三江平原冻融农区阿布胶河小流域为研究区，以流域土壤侵蚀为研究对象。首先通过地理信息系统GIS和田间试验得到农业开发下流域土壤侵蚀时空特征，并分析土地利用类型变化和土壤性质变化对土壤侵蚀的叠加影响作用；然后通过分布式水文模型SWAT得到基于日尺度的流域地表径流和土壤侵蚀模数变化特征，通过室内试验解析常温土壤和解冻土壤水分运动规律，揭示土壤侵蚀变化成因；最后进一步就降水和温度对土壤侵蚀的综合影响进行分析，得到土壤侵蚀响应规律。研究围绕“特征-成因-效应”依次展开研究，主要得到以下结论： (1) 土地利用类型变化比土壤性质变化对流域土壤侵蚀和非点源磷污染的影响更大。长期农业开发使研究区土地利用类型和土壤性质发生变化。在1979年至2014年间，研究区水田面积大幅度增加，湿地面积大幅度减少。土壤有机质含量、土壤水饱和度与田间持水量普遍下降，土壤容重普遍增加。土壤饱和导水率 (Ks) 下降了1.11 - 43.6 %。1979年间流域土壤侵蚀量为26737.9 t a-1，2014年间流域土壤侵蚀量为22570.4 t a-1，土壤侵蚀总量有所下降。在农业开发过程中，当农田面积保持一定时，水田面积的增加会有效缓解流域土壤侵蚀。随着水田的开发，流域土壤性质变化对土壤侵蚀产生的不利作用被削弱。水田和林地能够缓解土壤侵蚀，而旱田会加剧土壤侵蚀。 (2) 旱田土壤入渗性能的降低是旱田地表径流量、土壤侵蚀强度和非点源磷流失负荷增加的重要原因。在一年中，融雪季旱田和水田土壤侵蚀强度平均最大，雨季旱田土壤侵蚀强度相对较大。受降水和温度月际变化的影响，旱田不同土层深度处的土壤含水量波动显著。融雪季土壤含水量迅速增高，土水势剧烈变化，增加了土壤侵蚀的可能性；雨季土壤对降雨水量的分配作用较强。旱田地表径流量和土壤侵蚀模数在四月时达到峰值，分别为31.38 mm和1.46 t ha-1 a-1，七月和八月时二者值也较高。土壤侵蚀与地表径流量密切相关，线性回归结果R2在0.793至0.903。融雪季土壤入渗性能 (解冻土壤融水入渗) 显著低于雨季土壤入渗性能 (非解冻土壤常温水入渗)，二者稳定入渗速率之差介于1.98 mm h-1和2.53 mm h-1之间。 (3) 水田容重为1.30 g cm-3时的解冻风干土壤融水入渗性能较好，当地表径流量的增加量相同时，旱田土壤侵蚀模数的增加量约为水田的1.65倍。水田地表径流量和土壤侵蚀模数在四月时达到峰值，分别为41.89 mm和1.09 t ha-1 a-1。融雪季水田地表产流量与土壤侵蚀模数线性回归结果R2为0.907。Horton入渗模型、Philip入渗模型和Smith入渗模型均可以描述解冻土壤融水入渗和非解冻土壤常温水入渗过程。Smith入渗模型拟合效果最好，R2均在0.95以上，其次是Horton入渗模型，最后是Philip入渗模型。综合评价不同情境下的土壤稳定入渗速率和土壤累积入渗量，发现旱田和水田容重为1.40 g cm-3时的解冻湿润土壤融水入渗性能均较差。可以通过降低表层土壤容重和土壤含水率，增大土壤水下渗量减少地表产流量，达到控制土壤侵蚀和非点源磷污染的目的。 (4) 考虑降水和温度的综合影响，将历史时期内41年分成四类气象代表年：丰水年、旱热年、润寒年和平年，土壤侵蚀驱动因子对土壤侵蚀的作用在丰水年相对最弱，在平年普遍最强。丰水年和润寒年土壤侵蚀模数分别比平年高26.09 %和31.68 %，旱热年土壤侵蚀模数比平年低21.12 %。地表径流量和降水量对年际土壤侵蚀影响较大，它们与侵蚀模数的相关系数分别为0.920 (p = 0.01) 和0.768 (p = 0.01)。融雪季融雪径流量和融雪侵蚀力对土壤侵蚀影响较大，它们与侵蚀模数的相关系数分别为0.943 (p = 0.01) 和0.742 (p = 0.01)。温度因子对年际土壤侵蚀强度有一定影响。在降水和温度的综合影响下，土壤侵蚀驱动因子对土壤侵蚀的作用规律会发生改变。丰水年和润寒年时应采取一定的水土保持和非点源污染控制措施。 (5) 未来气候变化使全年土壤侵蚀模数增加，使融雪季土壤侵蚀模数降低。Mann-Kendall趋势检验结果表明，年降水量只在RCP 8.5情景下有显著增加趋势 (p = 0.019)；年均气温在历史时期 (p = 0.010)、RCP 4.5情景下 (p < 0.001) 和RCP 8.5情景下 (p < 0.001) 均有显著升高趋势。研究区未来冰冻期时间长度将逐渐缩短。RCP 4.5和RCP 8.5情景下流域年均土壤侵蚀模数分别比历史时期增加16.95 %和21.04 %，融雪季流域土壤侵蚀模数分别比历史时期降低20.75 %和27.75 %。旱田在RCP 4.5和RCP 8.5情景下的极端产流事件 (径流量分别大于400 mm和550 mm) 的发生概率均为1.23 %。水田在RCP 4.5情景下，未见明显的极端产流事件，在RCP 8.5情景下的极端产流事件 (径流量大于500 mm) 的发生概率为1.28 %。未来应该逐渐加强雨季时期对水土流失和非点源磷流失的防治。理解雨水产流侵蚀和融雪水产流侵蚀各自的特征，对未来非点源污染控制具有重要意义。 本研究比较系统地分析了冻融农区土壤侵蚀响应特征，探究了融水条件下解冻土壤水分运动规律，揭示了土壤侵蚀变化成因，为控制农业非点源磷污染提供了理论支持。本研究对流域水土资源保护和农业非点源磷污染控制具有实际参考价值。未来应更多地考虑人类活动对冻融农区水土环境的影响，并基于本研究得到的结论与建议，完善流域控水减沙减排体系。

Soil erosion leads to a series of threats to ecological environment, and soil erosion by water is the key process of agricultural non-point source pollution. Compared with nitrogen, phosphorus can be adsorbed in soil more easily, and then transports during soil and water loss process. Soil erosion is an important driving force for the loss of non-point source phosphorus. Therefore, it is of great significance to study the cause in variations and the response feature of soil erosion to understand the process of non-point source phosphorus loss. The freeze-thaw area under agricultural development is very susceptible to precipitation and temperature change, and the process of soil erosion is much complicated. Previous researches on soil erosion were based on single factor or single time scale, most of which focused on the effects of freeze-thaw cycling on soil properties and soil erosion. While there are few systematic analyses on the multi-factors affecting soil erosion at different spatio-temporal scales, and there are few concerns about watershed soil erosion and the mechanism of water movement in thawed soil under snowmelt condition. Therefore, this study combines laboratory experiment, field monitoring and model simulation to carry out systematic research, aiming at obtaining the characteristics of soil erosion, revealing the cause of its variations and studying the laws of its responses to the driving factors. Additionally, this study provides support for understanding the driving potential of non-point source phosphorus loss and advice for soil and water conservation and control of non-point source phosphorus pollution. In this study, a small watershed in the freeze-thaw agricultural area of the Sanjiang Plain was selected as the study area, and watershed soil erosion was the research subject. Firstly, the spatio-temporal characteristics of watershed soil erosion under agricultural development were obtained by GIS and field experiments, and then the synergistic effects of changes in land use and soil properties on soil erosion were determined. Secondly, the changes in watershed surface runoff and soil erosion load at a daily time scale were obtained by SWAT model. The laws of soil water movement under normal and thawed conditions were analyzed through laboratory experiments to reveal the causes of soil erosion change. Finally, the synthetic effects of precipitation and temperature on soil erosion were analyzed to obtain the law of soil erosion response. This study focused on the "characteristic-cause-effect" and the main conclusions are as follows: (1) Land use changes have more significant impacts than soil property changes on soil erosion and non-point source phosphorus pollution. The land use types and soil properties in the study area have changed due to long-term agricultural development. From 1979 to 2014, the area of paddy field in the study area increased greatly and the area of wetland decreased greatly. Soil organic matter content, soil water saturation and field capacity generally decreased, while soil bulk density generally increased. Soil saturated hydraulic conductivity (Ks) decreased by 1.11 - 43.6 %. The average watershed soil erosion amount was 26737.9 t a-1 in 1979 and was 22570.4 t a-1 in 2014, which exhibited a minor decrease. In the course of agricultural development, the increase in paddy field area will effectively alleviate soil erosion of the watershed when the farmland area is kept certain. With the development of paddy field, the adverse effects of changes in soil properties on soil erosion have been weakened. Paddy fields and forests can mitigate soil erosion, but upland by dryland farming can exacerbate soil erosion. (2) For upland, the decrease in soil infiltrability is one of the important reasons for the increase in surface runoff, soil erosion load and phosphorus loss. Soil erosion loads in upland and paddy field in snowmelt season were the highest, and soil erosion load was also relatively high in upland in rainy season. Under the influence of monthly precipitation and temperature variation, soil moisture fluctuated significantly in different depths of soil layers in upland. Soil moisture in snowmelt season increased rapidly and soil water potential changed, which increased the possibility of soil erosion. The redistribution of rainfall water in soil was marked in rainy season. In upland, surface runoff and soil erosion load peaked in April, with the volume and load of 31.38 mm and 1.46 t ha-1 a-1, respectively. They both were also higher in July and August. Soil erosion was closely related to surface runoff with the linear regression results of R2 ranging from 0.793 to 0.903. In upland, soil infiltrability in snowmelt season (thawed soil with water at 0 °C) was significantly lower than that (non-frozen soil with room temperature) in rainy season, the difference between the steady infiltration rates under these two conditions remained 1.98 mm h-1 to 2.53 mm h-1. (3) The thawed dry soil under 1.30 g cm-3 of paddy field was observed with the highest infiltrability, and when the increase in surface runoff volume was the same, the increase in soil erosion load in upland was 1.65 times higher than that in paddy field. Surface runoff and soil erosion load in paddy field peaked in April, with the volume and load of 41.89 mm and 1.09 t ha-1 a-1, respectively. The linear relationship between surface runoff and soil erosion load was significant. The R2 of the regression result was 0.907. Horton infiltration model, Philip infiltration model and Smith infiltration model can be used to describe the infiltration processes in snowmelt and rainy seasons. Smith infiltration model showed the best fitting result with R2 greater than 0.95, followed by Horton infiltration model, and finally Philip infiltration model. The steady infiltration rate and soil accumulative infiltration in different situations were evaluated comprehensively, and it was found that the thawed wet soils under 1.40 g cm-3 of paddy field and upland were observed with low infiltrabilities. It is effective to control soil erosion and non-point source phosphorus pollution by reducing soil bulk density and soil moisture, thus increasing the infiltration capacity and reducing surface runoff. (4) Considering the combined effects of precipitation and temperature, 41 years during the historical period were divided into four categories: wet years, hot dry years, freezing years and normal years. The effects of driving contributors on soil erosion were the weakest in wet years and generally the strongest in normal years. Soil erosion loads in wet years and freezing years were, respectively, 26.09 % and 31.68 % higher than the load in normal years, while erosion load in hot dry years was 21.12 % lower than that in normal years. The influence of surface runoff volume and precipitation on annual soil erosion was greater, with the correlation coefficients of 0.920 (p = 0.01) and 0.768 (p = 0.01) between them and soil erosion, respectively. The influence of snowmelt runoff volume and snowmelt erosivity on soil erosion was greater in snowmelt season, with the correlation coefficients of 0.943 (p = 0.01) and 0.742 (p = 0.01) between them and soil erosion, respectively. The temperature factor had certain influence on the annual soil erosion load. Moreover, precipitation and temperature factors modified the relationships between soil erosion and its driving contributors. We should pay attention and take measures for soil loss and pollution control in wet years and freezing years. (5) The annual soil erosion load would increase and the soil erosion load in snowmelt season would decrease under future climate change. The results of Mann-Kendall test showed that average annual precipitation only significantly increased in RCP 8.5 scenario (p = 0.019); average annual temperature significantly increased in historical period (p = 0.010), RCP 4.5 scenario (p < 0.001) and RCP 8.5 scenario (p < 0.001). The length of the frozen period in the study area in the future would be gradually shortened. The average watershed soil erosion loads in RCP 4.5 scenario and RCP 8.5 scenario were 16.95 % and 21.04 %, respectively, higher than that in historical period; while in snowmelt season, the average watershed soil erosion loads in RCP 4.5 scenario and RCP 8.5 scenario were 16.95 % and 21.04 %, respectively, lower than that in historical period. In upland, the probabilities of occurrence of extreme runoff events in RCP 4.5 scenario (runoff volume higher than 400 mm) and RCP 8.5 scenario (runoff volume higher than 550 mm) were both 1.23 %. In paddy field, no extreme runoff event in RCP 4.5 scenario was observed but the probability of occurrence of extreme runoff event in RCP 8.5 scenario (runoff volume higher than 500 mm) was 1.28 %. In the future, more attention should be paid to the control of soil and phosphorus loss in rainy season gradually. Priority should be given to a better understanding of the characteristics of water erosion by rainfall and snowmelt events to control soil loss and pollution effectively in the future. This study systematically analysed the response characteristics of soil erosion, explored the law of water movement in thawed soil in snowmelt season, and revealed the cause of change in soil erosion in a freeze-thaw agricultural area. This study provides theoretical support for agricultural non-point source phosphorus pollution control and has practical reference value for protecting water and soil resources of watershed and controlling agricultural non-point phosphorus pollution. In the future, more consideration should be given to the influence of human activities on the soil and water environment in freeze-thaw areas. The sediment reduction and pollution control system in watershed should be perfected based on the conclusions and suggestions of this study.