耕地是人类生存与发展的重要资源。可近年来，我国城镇化进程推进迅速，耕地资源流失的现象愈加严重，耕地资源与其他用地之间的矛盾日渐突出。耕地流失不仅会影响农业生产率和当地气候，还会破坏生物栖息地，威胁生物多样性，甚至会对国家粮食生产安全和社会稳定构成巨大威胁。在此背景下，本研究以华北平原重要的农业生产示范基地新泰市为研究案例区，利用2009年和2019年两次土地利用调查数据，分析研究区耕地流失的状况、内部结构、空间格局和多重成因。最后，运用随机森林模型对耕地发生流失、非粮化以及非农化的概率做出预测，并使用ArcGIS将预测结果进行可视化，形成耕地流失总体趋势的风险区划图、耕地非粮化的风险区划图以及耕地非农化的风险区划图。研究的主要结论如下：

（1）耕地流失现象严重，十年间新泰市耕地面积减少了385.8km²，耕地流失率高达37.18%。耕地非粮化是新泰市耕地流失的主要原因，占耕地流失面积的83.13%，而耕地非农化仅占耕地流失面积的16.87%。两者不仅在规模上表现出巨大差异，而且在空间分布上也存在较大差异。从乡镇尺度来看，西部乡镇耕地流失主要是因为耕地非粮化导致的，而东部靠近城区中心的街道和乡镇耕地流失主要是因为耕地非农化导致。新泰市乡镇耕地非粮化率由中部向南北两翼下降，而乡镇的耕地非农化率呈现由中心向外围下降的趋势。从行政村尺度来看，新泰市行政村耕地非粮化率的冷点区域主要分布在区（县）政府附近的行政村，而新泰市行政村耕地非农化率的热点区域主要分布在这里的行政村。从地块尺度来看，发生非粮化的耕地地块的空间集聚特征比发生非农化的耕地地块更明显。这可能是因为耕地非粮化的经营行为需要达到一定规模才有较为可观的经济收益，而较小规模的耕地非农化行为就可以产生相对可观的经济效益。

（2）随机森林模型对耕地流失、耕地非粮化和耕地非农化有较好的预测效果，预测准度（AUC）分别为75.29%、75.90%和97.01%。总的来看，各个影响因素的重要性差异显著。其中，空间相关性因素是新泰市耕地流失重要性最高的影响因素；坡度等级是新泰市耕地流失重要性最低的影响因素。此外，耕地非粮化和耕地非农化影响因素的相对重要性存在较大差异。在耕地非粮化方面，空间相关性因素的重要性最高，而基本农田保护政策的重要性最低；在耕地非农化方面，到农村道路的距离的重要性最高，而坡度等级的重要性最低。同时，耕地非粮化与耕地非农化风险变化趋势也存在较大差异。此外，从耕地非粮化以及耕地非农化来看，影响因素的相对重要性存在较大差异。在耕地非粮化方面，空间相关性因素的重要性最高，而基本农田保护政策的重要性最低；在耕地非农化方面，到农村道路的距离的重要性最高，而坡度等级的重要性最低。同时，耕地非粮化与耕地非农化风险变化趋势也存在较大差异。

（3）耕地流失高风险区的面积大，新泰市耕地流失高风险区面积为397.24km²，占研究区耕地面积的38.44%。新泰市超过一半的乡镇街道耕地流失高风险区耕地面积比例超过全市的平均值。新泰市耕地流失的高风险区主要分布在研究区东部、中部、西北部和西南部区域。因此，应该进一步加强对这些重点区域的监控。但是，新泰市耕地非粮化与耕地非农化的风险区划存在较大的空间差异性。从风险区划的结果可以看出，相比于耕地非农化，新泰市耕地非粮化的高风险区耕地面积更大、集聚特征更明显。耕地非粮化的高风险区耕地面积远大于耕地非农化的高风险区耕地面积，而耕地非农化的低风险区耕地面积远大于耕地非粮化的低风险区耕地面积。其中，新泰市耕地非粮化高风险区的面积为316.46km²，占研究区耕地面积的30.50%，主要分布在研究区的东部、西北部、西南部；新泰市耕地非农化高风险区的面积仅为147.91km2，占研究区耕地面积的14.25%，主要分布在青云街道南部、新甫街道东部、汶南镇东北部、东都镇以及刘杜镇东部。因此，为了更好地保护耕地资源，应该对耕地实现精细化管理，针对不同类型的耕地流失现象制定差异化的空间管控措施。

The cultivated land is an important resource for the survival and development of human beings, but in recent years, our country urbanization is promoting rapidly, THE cultivated land loss phenomenon is more and more common, and the contradiction between cultivated land resource and various kinds of land uses is increasingly prominent. Loss of cultivated land will not only affect agricultural productivity and local climate, but also destroy biological habitats, threaten biodiversity, and even pose a great threat to national food production security and social stability. In this context, this paper takes Xintai City, an important agricultural production demonstration base in the North China Plain, as the case study area. Then, using the land use survey data in 2009 and 2019, we analyzed the status, internal structure, spatial pattern and multiple causes of arable land loss in the study area. Finally, the random forest model was used to predict the probability of cultivated land loss, non-grain conversion of cultivated land and non-agricultural conversion of cultivated land, and ArcGIS was used to visualize the predicted results, forming the risk zoning map of the overall trend of cultivated land loss, the risk zoning map of non-grain conversion of cultivated land and the risk zoning map of non-agricultural conversion of cultivated land. The main findings are as follows:

(1) The loss of cultivated land is very serious in Xintai City. The area of cultivated land decreased by 385.8km², and the loss rate of cultivated land reached 37.18%. The non-agricultural conversion of cultivated land is the main reason for the loss of cultivated land in Xintai City, accounting for 83.13% of the loss area, while the non-agricultural conversion of cultivated land only accounts for 16.87% of the loss area. The two show great differences not only in scale, but also in spatial distribution. From the perspective of township scale, the loss of cultivated land in the western townships is mainly caused by the non-grain conversion of cultivated land, while the loss of cultivated land in the streets and towns near the urban center in the east is mainly caused by the non-agricultural conversion of cultivated land. In Xintai City, the non-grain conversion rate of cultivated land decreased from the center to the north and south, while the non-agricultural conversion rate of cultivated land decreased from the center to the periphery.From the perspective of the scale of administrative villages, the cold spot area of the non-grain conversion rate is mainly distributed in the administrative villages near the district (county) government, while the hot spot area of the non-agricultural conversion rate is mainly distributed in the administrative villages here.From the perspective of plot scale, the spatial agglomeration characteristics of non-grain conversion are more obvious than non-agricultural conversion.This may be because the operation behavior of non-grain conversion needs to reach a relatively large scale to have relatively considerable economic benefits, while the small scale of non-agricultural conversion behavior can produce relatively considerable economic benefits.

(2) The random forest model had a good prediction effect on the loss of cultivated land, the non-grain conversion of cultivated land and the non-agricultural conversion of cultivated land, and the prediction accuracy (AUC) was 75.29%, 75.90% and 97.01%, respectively. In general, the importance of each influencing factor varies significantly. Among them, the spatial correlation factor is the most important factor affecting the cultivated land loss in Xintai City; Slope grade is the lowest factor affecting the importance of cultivated land loss in Xintai City. In addition, there are significant differences in the relative importance of influencing factors from the perspectives of non-grain conversion of cultivated land and non- agricultural conversion of cultivated land.In terms of non-grain conversion of cultivated land, spatial correlation factors have the highest importance, while the importance of basic farmland protection policies is the lowest; In terms of non- agricultural conversion of cultivated land, the distance to rural roads is of the highest importance, while the slope grade is of the lowest importance. Moreover, there are significant differences in the risk trends of factors affecting the non-grain conversion and non-agricultural conversion of cultivated land.

(3) The high-risk area of farmland loss in Xintai City is the largest, with an area of 397.24km², accounting for 38.44% of the arable land area in the study area. In Xintai City, more than half of the townships and streets have a high proportion of arable land loss in high-risk areas, which exceeds the average of the city. The high-risk areas of cultivated land loss in Xintai City are mainly distributed in the east, middle, northwest and southwest of the study area. Therefore, the monitoring of these key areas should be further strengthened. However, there is a large spatial difference between the risk regionalization of cultivated land non-grain conversion and non-agricultural conversion in Xintai City. From the results of risk zoning, it can be seen that compared to the non- agricultural conversion of cultivated land, the high-risk areas of non-grain conversion of cultivated land in Xintai City have a larger area and more obvious clustering characteristics. The high-risk area of non-grain conversion is much larger than the high-risk area of non-agricultural conversion, while the low-risk area of non-agricultural conversion is much larger than the low-risk area of non-grain conversion. Among them, the high-risk area of non-grain conversion of cultivated land in Xintai City is 316.46km², accounting for 30.50% of the cultivated land area in the study area, mainly distributed in the east, northwest, and southwest of the study area; The high-risk area of non-agricultural conversion of cultivated land in Xintai City is only 147.91 km², accounting for 14.25% of the research area's cultivated land area. It is mainly distributed in the south of Qingyun Street, the east of Xinfu Street, the northeast of Wennan Town, Dongdu Town, and the east of Liudu Town. Therefore, in order to better protect cultivated land resources, it is necessary to achieve refined management of cultivated land and develop differentiated spatial control measures for different types of cultivated land loss.