土地利用/土地覆盖变化（LUCC）在全球变化中起着不可忽视的作用，不仅改变着地表的土地覆被方式，同时通过改变土壤性质、土壤碳排放等影响全球或区域的气候和环境。耕地是人为改变LUCC的重要途径之一，是人类影响全球或区域气候和环境不可忽视的环节之一。受限于历史耕地资料等原因，对历史耕地分布的研究一直是过去LUCC研究的重要组成部分之一。本研究将中国已有的历史文献重建耕地结果进行空间和时间订正，得到基于行政单元的中国1850年耕地空间分布。基于2000年数据，对不同格网分辨率下影响耕地分布的因子进行了筛选，构建了分区的耕地格网化分配模型，并将分配后的1850年格网化耕地结果同CHCD数据集和HYDE3.2数据集进行对比，得到以下结果： （1）基于行政单元的1850年耕地分布集成重建。本文收集了已有研究成果，通过对行政边界的调整和1850年时间断面的归一，集成重建了1850年中国东北地区（Ⅰ区）、华北地区（Ⅱ区）、新疆维吾尔族自治区（Ⅴ区）和青藏高原地区（Ⅵ区）的县域垦殖率数据，以及西南地区（Ⅲ区）和中部地区（Ⅳ区）的省域垦殖率数据。 （2）对不同地区不同格网分辨率下影响耕地分布因子的选取，以及耕地格网分配模型建立。本文基于2000年遥感数据，分析中国6个区域3种分辨率（10km, 60km和100km）耕地的空间分布与气候生产潜力、海拔、坡度以及土壤有机质含量的关系，发现在不同区域不同分辨率下影响耕地分布的因素均有所不同。并在此基础上建立分区的耕地格网化分配模型，模型分配的耕地数据与遥感数据相对差异在±10%以内的网格均在75%以上。在不同地区，其最优的格网分辨率有所不同，东北地区、华北地区、西南地区和中部地区在100km分辨率网格数量最多，分别占约93.12%、97.99%、98.03%和98.60%；新疆地区和青藏高原地区在60km分辨率的网格数量最多，分别占97.89%和100%。可见本文建立的不同区域不同分辨率的耕地分配模型对耕地格局的重建具有可靠性，可以为历史时期的耕地格网化处理提供参考。 （3）集成重建的耕地数量与同类研究的比较。在全国尺度上，CHCD数据集在1851年的耕地总量比本文重建的数据集（简称“1850CD数据集”）高出约9.12%，1850CD数据集比HYDE3.2数据集在1850年的耕地总量高出约32.38%。在区域尺度上，CHCD数据集在Ⅰ区、Ⅱ区、Ⅳ区、Ⅴ区和Ⅵ区分别比1850CD数据集高出约79.73%、11.82%、12.31%、24.68%和48.08%，在Ⅲ区，CHCD数据集比1850CD数据集低约4.68%。HYDE3.2数据集在Ⅰ区分配的耕地较1850 CD数据集高出约23.46%，而在Ⅱ区、Ⅲ区、Ⅳ、Ⅴ区和Ⅵ区则是1850CD数据集分配的较多，高出的耕地分别约为41.55%、35.90%、27.48%、558.70%和42.37%。在省域尺度方面，三套数据耕地格网分配在各个省份的分布趋势大致相同，数量稍有差别。产生差异的主要原因为：耕地重建所使用的源数据不同，时间断面归一方法不同等原因。 （4）耕地格网化空间分布与国内外同类研究的比较。将1850CD数据集的结果与HYDE3.2数据集和CHCD数据集进行对比。1850CD数据集的结果与HYDE3.2数据集之间的绝对差异在-100km2—-50km2、-50km2—-10km2、-10km2—10km2、10km2—50km2和50km2以上的格网分别占1.49%、17.10%、73.86%、7.51%和0.03%。1850CD数据集与CHCD数据集之间的绝对差异在-100km2—-50km2、-50km2—-10km2、-10km2—10km2、10km2—50km2和50km2以上的格网分别占0.58%、13.90%、67.04%、16.93%和1.55%。HYDE3.2数据集与本研究结果相比存在差异的主要原因是重建时所用的耕地源数据不同以及HYDE3.2数据集沿河流分配较多造成的；与CHCD数据集相比存在差异的主要原因是重建时所用的耕地源数据的差异、1850年数据订正的差异以及模型因子的差异造成的。1850CD数据集与HYDE3.2耕地分配的相对差异在±20%以内、±20%-±40%、 ±40%-±60%、±60%-±80%、±80%-±100%和100%以上的格网分别占16.03%、16.78%、17.56%、15.47%、21.89%和12.27%；1850CD数据集与CHCD数据集耕地分配的相对差异在±20%以内、±20%-±40%、 ±40%-±60%、±60%-±80%、±80%-±100%和100%以上的格网分别占14.44%、13.33%、12.11%、12.66%、27.64%和19.82%。1850CD数据集与CHCD数据集的相对差异较小，主要在华北平原地区、四川盆地地区、长江中下游平原部分地区以及辽宁南部地区的相对差异较小，在山地广泛分布的西南地区和青藏高原地区以及新疆地区的相对差异较大。而与HYDE3.2数据集的相对差异较大。

Land Use/Land Cover Change (LUCC) plays a significant role in global change. It not only changes the way of the land cover, but also affects the climate and environment of the global or regional by changing soil properties and soil carbon emissions. Cultivated land is one of the important ways to artificially change LUCC, and it is one of the links that humans can influence global or regional climate and environment. Due to the availability of historic cultivated land data and other reasons, the research of the distribution of historical farmland has always been one of the important components of LUCC in the past. This paper screened the factors and the grid resolution that affecting the distribution of cultivated land by the 2000 datas, and established a grid distribution model of cultivated land of partition. And compare the result with CHCD dataset and HYDE3.2 dataset and get the following result: 1. Integrated reconstruction of arable land distribution based on administrative units in 1850. This article has collected the existing research results, through the adjustment of the administrative boundary and the normalization of the 1850 time section, integrated reconstruction the cultivated land data of the counties in Northeast China's (I), North China (II), Xinjiang Uygur Autonomous Region (V) and Tibetan Plateau (VI) in 1850, and the cultivated land data of the provinces in Southwest (III) and Central Region (IV). 2. This paper selected factors affecting the distribution of cultivated land in different different regions and different grid resolutions, and established a grid allocation model of arable land. Based on remote sensing data from 2000, this study analyzed the relationship between the spatial distribution of three resolution (10km, 60km, and 100km) arable lands in 6 regions of China and climate potential, elevation, slope, and organic matter content, and found that that the factors affecting the distribution of cultivated land in different areas are different. Based on this, we established a partitioned arable grid distribution model. The relative differences in the distribution of arable land data and remote sensing data within ±10% were all above 75%. In different regions, the optimal grid resolution is different: Northeast China, North China, Southwest China and Central China have the largest number of resolution grids at 100km, about 93.12%, 97.99%, 98.03%, and 98.60%; In Northwest and Qinghai Tibet Plateau, the number of grids with 60km resolution is the largest, accounting for 97.89% and 100% respectively. Therefore, the cultivated land allocation model with different resolutions and different regions established in this paper is reliable for the reconstruction of cultivated land pattern. 3. Comparison of the amount of cultivated land integrated with reconstruction and similar research. On the national scale, the total amount of CHCD data in 1851 is about 9.12% higher than that of 1850CD data set. 1850CD data set is about 32.38% higher than that of the HYDE3.2 data set in 1850. On the regional scale, the CHCD data set is about79.73%、11.82%、12.31%、24.68% and 48.08% higher than 1850CD data set in Ⅰ、Ⅱ、Ⅳ、Ⅴ and Ⅵ, and in Ⅲthe 1850CD data set is about 4.68% higher than the CHCD data set. The HYDE3.2 dataset allocated more land in the I than in 1850CD data set, about 23.46%, while in the II , III, IV, V and VI is the 1850CD dataset more allocated higher cultivated land, about 41.55%, 35.90%, 27.48%, 558.70%, and 42.37%. In terms of provincial scale, the distribution trend of the three sets of data in the provinces is basically the same, and the number is slightly different. The main reasons for the differences are: different sources of cultivated land data and different methods for normalizing time sections. 4. Comparison of spatial distribution of cultivated land gridization and similar research at home and abroad. Comparing the results of 1850CD data set with the HYDE3.2 data set and the CHCD dataset, the absolute error between the results of 1850CD data set and the HYDE3.2 data set accounts for 1.49%, 17.10%, 73.86%, 7.51% and 0.03% of the grid of -100km2 - -50km2, -50km2 - -10km2, -10km2 - 10km2, 10km2 - 50km2 and 50km2-100km2. The errors between 1850CD data set and the CHCD data set are 0.58%%, 13.90%, 67.04%, 16.93% and 1.55% in -100km2 - -50km2, -50km2 - -10km2, -10km2 - 10km2, 10km2 - 50km2 and 50km2-100km2. The main reason for the error of HYDE3.2 data set and this study is the difference of cultivated land source data and the distribution of HYDE3.2 data set more along the river. The main reasons for the error of this research and CHCD data set are the difference of the source data of cultivated land, the difference of data correction in 1850 and the difference of the model factors. The relative difference between 1850CD dataset and HYDE3.2 farmland allocation is within ±20%, ±20%-±40%, ±40%-±60%, ±60%-±80%, ±80%-±100% and more than 100% accounted for 16.03%, 16.78%, 17.56%, 15.47%, 21.89% and 12.27% respectively. The relative difference between 1850CD dataset and CHCD farmland allocation is within ±20%, ±20%-±40%, ±40%-±60%, ±60%-±80%, ±80%-±100% and more than 100% accounted for 14.44%、13.33%、12.11%、12.66%、27.64% and 19.82% respectively. The relative difference between the 1850CD dataset and the CHCD dataset is small, and the relative differences in the North China Plain, the Sichuan Basin, parts of the middle and lower reaches of the Yangtze River Plain, and southern Liaoning are relatively small, the relative differences in the southwestern region and the Qinghai-Tibet Plateau region and the Xinjiang region, which are widely distributed in the mountains are relatively large. The relative difference between the 1850CD dataset and the HYDE3.2 dataset is relatively large.