高精度的全国和区域尺度的土地覆盖数据是全球环境变化、陆表过程建模、生态文明构建和区域可持续开发等科学研究的重要基础依据。尽管目前从全球到区域尺度已有众多土地覆盖数据集，然而由于卫星运行方式、分类方法等方面的差异，直接利用目前已有土地覆盖产品进行科学研究仍具有一定不确定性。利用调查数据加以辅助的多源数据融合方法可以克服现有土地覆盖产品在兼容性、准确性、可比较性等方面的限制，并在一定程度上提高数据的精度。

耕地是一种极为重要的土地利用类型，从农业革命至今不断地对地表自然覆盖及物质和能量平衡乃至全球生态环境及气候变化产生影响。本文以东北地区为研究区，选取8套高分辨率全球土地覆盖数据集（Esri Land Cover, ESA World Cover, GLC_FCS30, FROM_GLC, GlobeLand30, CLCD, AGLC和GFSAD30）的耕地数据，以历次耕地调查数据作为总量控制，以各数据集在像元上的重合度及优先级为指标建立像元耕地分配的优先级，将调查数据的耕地面积分配给更可能是耕地的像元，实现上述多源耕地覆盖信息的融合，生成了一套东北地区1985-2020年30m分辨率的耕地融合数据，并对1985-2020年东北地区的耕地变化情况进行了分析。论文的主要结论如下：

（1）对于2020年，5套原始耕地覆盖产品中的耕地面积均高于耕地调查数据。ESA World Cover和GlobeLand30的耕地面积一致性较高，Esri Land Cover和CLCD的耕地面积一致性最低，5套原始耕地覆盖产品耕地完全一致区域面积占全区域面积比例为20.53%，占5套产品合计有耕地存在区域的比例为49.32%。不同耕地覆盖产品的空间格局在土地覆盖均质性较高区域一致性较好，在土地覆盖异质性较高区域空间格局一致性较差。不同原始产品的耕地空间分布不一致区域与地形因子的回归分析结果表明原始土地覆盖产品对耕地的识别差异或由于海拔高度、地表切割深度和地形起伏度这三种地形因子所导致。

（2）本文所融合的1985-2020年东北地区耕地覆盖产品较原始耕地覆盖产品在耕地面积上更接近于官方公布的土地调查数据，在耕地的空间识别上表现优异。本文融合耕地覆盖产品比原始耕地覆盖产品更能够表现出一些破碎化的耕地以及成片耕地中的小路或田埂。利用混淆矩阵评价耕地融合结果，2020年东北地区融合产品的总体精度达到95%。但按坡度分级的耕地面积仍与土地调查数据存在一定差别，此误差很大程度上与参与融合的原始耕地覆盖产品有关。

（3）东北地区的耕地大部分分布在辽河平原、三江平原和松嫩平原，以及山前台地和山间盆地。1985-2020年东北地区耕地面积及垦殖率总体呈增长趋势，年均增长率为1.66%，2020年耕地面积达到37.90×10⁴km²，全区垦殖率达到26.24%；1990年为研究时段内耕地面积最少时点，为23.99×10⁴km²，全区垦殖率为16.61%。总体来看，1985年以来东北地区农耕区以扩张为主，耕地变化频繁区集中在坡度小于15°的区域，该类型区域内耕地的增加和减少量分别占全部耕地变化量的96.44%和98.04%。

（4）东北地区作为国家重要商品粮和农牧业生产基地的战略定位及相关政策是1985-2020年东北地区耕地变化的主要驱动因素。改革开放以来，国家和地方政府在宏观层面上针对东北地区的粮食生产制定了一系列政策措施，东北地区的耕地资源因此在全面和有效的法律保障和管理手段下得到了有效的保护与利用。同时，一系列生态恢复法规政策也对东北地区耕地造成影响。此外，人口、经济和科技等因素也对东北地区耕地变化产生一定的影响。

Credible global and regional remote sensing based land cover classification data provide essential foundational information for research on global change, land surface process modelling, ecological civilization construction, and regional sustainable development. Despite the availability of numerous land cover datasets from global to regional scales, the use of these land cover products for scientific research still has uncertainty due to differences in satellite operation modes, sensor types, classification methods, validation methods, and other aspects. The data integration combining survey data and the remote sensing based land cover products can improve the accuracy of the data to some extent by overcoming the compatibility, comparability, and accuracy limitations of the products.

Cropland, as an important land use type, has been continuously altering the natural land cover, and changing the material and energy balance since the Agricultural Revolution, which have exerted profound impacts on regional and even global environment and climate changes. In this study, focusing on Northeast China, we selected eight sets of high-resolution global land cover products （Esri Land Cover, ESA World Cover, GLC_FCS30, FROM_GLC, GlobeLand30, CLCD, AGLC and GFSAD30）for cropland data to generate a set of 30m cropland integration data in the Northeast China from 1985 to 2020. We established the priority of pixel cropland allocation based on the overlap and weight of each dataset on pixels, using the previous national cropland survey data as the total control, allocated the cropland area of survey data to pixels that are more likely to be cropland to achieve the integration of multi-source cropland cover. Then we analyzed the changes in cropland in the Northeast China from 1985 to 2020. The results are the following.

(1) For the year 2020, the cropland areas in all five sets of original cropland cover products were higher than the cropland survey data. The consistency of cropland areas between ESA World Cover and GlobeLand30 was relatively high, while the consistency between Esri Land Cover and CLCD was the lowest. The proportion of completely consistent cropland areas among different original products was 20.53% of the total area, and it accounted for 49.32% of the total area where cropland existed in all five products combined. The spatial patterns of different cropland cover products showed better consistency in regions with high land cover homogeneity, while the consistency was lower in regions with high land cover heterogeneity. Regression analysis results indicated that the differences in cropland identification among original land cover products were caused by three terrain factors of elevation, surface cut depth, and terrain undulation.

(2) The cropland cover product integrated in this study in the Northeast China from 1985 to 2020 is closer to the high-reliability national land survey data compared to the original cropland cover products in cropland area. The integrated cropland product in this study performed excellently in spatial identification of cropland, showing better capability than the original cropland cover products in identifying fragmented cropland areas and small paths or ridges within contiguous cropland areas. Using a confusion matrix to evaluate the cropland integrated results, the overall accuracy of the integrated product in the Northeast China reached 95% in 2020. However, there is still a certain gap in slope grading compared to the national land survey data, which is largely related to the five sets of original cropland cover products involved in the integration.

(3) In the Northeast China, most croplands are distributed across the Liaohe Plain, Sanjiang Plain, and Songnen Plain, as well as the piedmont terraces and intermountain basins. From 1985 to 2020, the cropland area and cultivation rate in the Northeast China showed an overall increasing trend, with an average annual growth rate of 1.66%. In 2020, the cropland area reached its peak at 37.90×10⁴ km², with a cultivation rate of 26.24% for the entire region. The year 1990 marked the period with the smallest cropland area during the study period, at 23.99×10⁴ km², with a cultivation rate of only 16.61% for the entire region. Overall, since 1985, expansion has been the predominant trend in the agricultural areas of the Northeast China. Changes in cropland have been most frequent in areas with slopes less than 15°. Within these areas, the increase and decrease in cropland area respectively accounted for 96.44% and 98.04% of the total cropland change.

(4) The driving factors of cropland changes in the Northeast China from 1985 to 2020 were analyzed. Since the reform and opening-up, the central and local governments have formulated a series of reform measures at the macro level. The cropland resources in Northeast China have been effectively protected and utilized under the comprehensive and effective legal guarantees and management measures. At the same time, a series of ecological restoration policies and regulations have also had an impact on the cropland in Northeast China. In addition, factors such as population, economy, and technology have also had a certain influence on the changes in cropland in Northeast China.