Are there suitable global datasets for monitoring of land use and land cover in the tropics? Evidences from mainland Southeast Asia
Are there suitable global datasets for monitoring of land use and land cover in the tropics? Evidences from mainland Southeast Asia
复制标题
DOI：
10.1016/j.gloplacha.2023.104233
复制
发表时间：
2023-09
影响因子：
3.9
通讯作者：
Jiahao Zhai;Chi-wei Xiao;Zhiming Feng;Ying Liu
中科院分区：
文献类型：
--
作者：
Jiahao Zhai;Chi-wei Xiao;Zhiming Feng;Ying Liu
可自由获取的土地利用和土地覆盖(LULC)数据集是跟踪地表变化、生态系统动态和碳循环的有效工具。然而，LULC数据产品的不确定性、适用性和局限性一直是一个挑战，在世界范围内差异很大，特别是在森林损失迅速的热带地区。对8个应用最广泛的LULC数据集，包括500-m MCD12Q1、300-m ESA CCI-LC、30-m GlobelLand30、30-m GLC_FCS30、30-m From-GLC、10-m World Cover、10-m Esri Land Cover和10-m From-GLC10，进行了统计比较和评估东南亚大陆的一致性和可靠性。结果表明，FROM-GLC10、World Cover和GLC_FCS30的准确率较高，总体准确率在90.8~92.2%之间，而FROM-GLC的准确率最低，为77.7%。GlobeLand30、From-GLC和From-GLC最为相似，相关系数为0.999，而欧空局CCI-LC和World Cover的相关系数仅为0.882。8个LULC森林(52.2%-70.4%)和农田(21.6%-43.9%)是MSEA中的主要LULC，其中ESA CCI-LC的估计农田面积最大(84.48万平方公里，43.9%)，这也导致它的森林面积最小(52.2%)。与其他数据集相比，ESRI土地覆盖的建成地分布最广(4.9%)，而World Cover的草地覆盖面积最大(9.8%)。特别是，农田和森林的空间格局高度一致，但存在显著的地方差异，例如在缅甸西部和越南南部。MCD12Q1、ESA CCI-LC和GLC_FCS30显示了过去20年来MSEA五个国家森林变化的总体趋势，而GlobelLand30产生了不同的结果。我们基于MSEA的8个全球LULC数据集的交叉比较和评估突出了一致性和差异，这将有助于为未来的特定需求(例如，区域生态系统对森林损失的反应)选择合适的数据集。
The freely available Land Use and Land Cover (LULC) datasets are effective tool for tracking land surface changes, ecosystem dynamics, and carbon cycle. However, the issue of the uncertainly, applicability, and limitations of LULC data products is always a challenge and varies distinctively worldwide, especially in the tropics where forest loss is rapid. Eight of the most widely used LULC datasets, here, including 500-m MCD12Q1, 300-m ESA CCI-LC, 30-m GlobelLand30, 30-m GLC_FCS30, 30-m FROM-GLC, 10-m World Cover, 10-m Esri Land Cover, and 10-m FROM-GLC10, were statistically compared and evaluated the consistency and reliability in Mainland Southeast Asia (MSEA). The results revealed that the FROM-GLC10, World Cover, and GLC_FCS30 have higher accuracy than the other five datasets, with overall accuracy ranging from 90.8 to 92.2%, while FROM-GLC had the lowest accuracy of 77.7%. GlobeLand30, FROM-GLC, and FROM-GLC10 were the most similar with a correlation coefficient of 0.999 compared to only 0.882 for ESA CCI-LC and World Cover. Eight LULC-based forests (52.2%–70.4%) and cropland (21.6%–43.9%) are the predominant LULCs in MSEA, with the ESA CCI-LC having the largest estimated area (844,800 km2, 43.9%) of cropland, which also results in it having the smallest area (52.2%) of forest. Compared to other datasets, Esri Land Cover had the most extensive distribution of built-up land (4.9%), while World Cover had larger areas of grassland cover (9.8%). In particular, the spatial patterns of cropland and forest are highly consistent, but there are significant local differences, e.g., in western Myanmar and southern Vietnam. The MCD12Q1, ESA CCI-LC, and GLC_FCS30 showed consistent overall trends in forest change across the five countries of MSEA over the past two decades, while GlobelLand30 produced different results. Our cross-comparison and evaluation based on eight global LULC datasets in MSEA highlighted the consistency and differences, which will help to select the suitable dataset for specific needs (e.g., regional ecosystem response to forest loss) in the future.