Knowledge Resource Center for Ecological Environment in Arid Area
| 土地利用/覆被数据同化研究 | |
| 其他题名 | Data assimilation of Land Use/Cover in River Basin Scale |
| 胡晓利 | |
| 出版年 | 2018 |
| 学位类型 | 博士 |
| 导师 | 李新 |
| 学位授予单位 | 中国科学院大学 |
| 中文摘要 | 人类活动对地球表面系统环境产生了深远的影响。土地利用/覆被变化是人类活动对地球表层系统影响最为直接的表现方式，通过改变地球表层系统的下垫面，进而影响地球表层系统各大圈层的循环过程。因此，准确模拟和预测过去、现在、未来地球表层真实的土地利用/覆被变化格局和过程，可为土地利用规划、政策制定提供科学依据，也可服务于区域可持续发展。模型和观测是土地利用/覆被变化研究的两种基本手段。模型有助于深入了解土地利用变化成因，重建过去、预测未来发展趋势、评估环境影响以及支持土地利用规划决策分析。目前，元胞自动机（CA）已成为土地利用/覆被变化模拟最常用的工具之一。但元胞自动机模型结构、静态参数以及边界条件、初始场引起的误差和不确定性会随着模型的运行不断的传递和积累；而观测能够快速准确地获取各种时空尺度上观测时刻的地表真实土地利用/覆被状况。但无法了解无观测时的状况，不能提供时间上连续的土地利用/覆被变化情况。如何将观测信息和模型有机融合，来量化和减少土地利用/覆被模拟和预测中的各种不确定性，获取时空连续的土地利用/覆被信息，是目前土地利用/覆被变化研究中面临的一个重大挑战。本文基于对黑河中游张掖绿洲土地利用/覆被变化的分析和讨论，发展了两个以元胞自动机为核心的多维土地利用/覆被变化动态集成模型。在此基础上，提出了一种基于共轭先验贝叶斯推理方法的，多维、离散类型变量的同化策略，以融合观测数据和模型模拟结果。以张掖绿洲的典型区域—甘州区为实验区，进行了土地利用/覆被的模拟同化实验，并对景观斑块空间尺度和同化周期进行了敏感性分析。主要研究内容如下：首先，制备了2000年、2007年以及2011年三期张掖绿洲土地利用/覆被数据集。基于这些数据集对2000年至2011年期间张掖绿洲的土地利用/覆被动态变化，及其对可持续水资源管理的影响进行了分析讨论。结果表明，张掖绿洲土地利用/覆被发生了较大的变化。受经济利益驱使，耕地持续扩张。从2000年到2011年，耕地增加了12.01%。耕地的扩张，增加了水资源压力，导致局部地区地下水超采和生态系统的退化，严重威胁着整个黑河中游乃至整个流域的可持续发展。因此，应加强水资源管理，尤其是地下水的管理，严格控制耕地面积，从而实现生态和社会经济的水资源可持续利用，确保中、下游之间用水的公平和合理。该研究为本论文后续土地利用/覆被变化模型的构建提供了基础数据，对可持续水资源管理的影响分析也可为其它内陆河流域的水土资源可持续管理提供参考。其次，针对传统元胞自动机（CA）的不足，将马尔科夫链（Markov Chain）、逻辑回归（Logistic Regression）、人工神经网络（Artificial Neural Network，ANN）方法分别耦合到CA中，构建了两个以元胞自动机为核心的多维土地利用/覆被变化动态集成模型：多维逻辑回归马尔科夫元胞自动机模型（MLRMCA）和多维人工神经网络马尔科夫元胞自动机模型（MANNMCA）。利用这两个集成模型模拟了张掖绿洲2000-2011年的土地利用/覆被演变过程。总体来看，MANNMCA模型的模拟精度略高于MLRMCA模型。这主要是因为：人工神经网络和元胞自动机相结合能够有效地捕获土地利用/覆被类型转换概率与一系列空间驱动变量之间的非线性的复杂关系。然而，MLRMCA模型也有一些优点。如在局部地区，MLRMCA模型模拟的一些土地利用/覆被类型的变化趋势要优于MANNMCA模型，并且逻辑回归方法可以甄别确定土地利用/覆被变化的空间驱动因素。总之，MLRMCA模型和MANNMCA模型这两个土地利用/覆被变化动态集成模型均能够有效的模拟与预测区域土地利用/覆被的演变过程，为区域的土地利用规划、水资源管理、生态恢复以及区域的可持续发展提供有用的信息。该研究为本论文土地利用/覆被同化研究奠定了模型基础。第三，针对土地利用/覆被同化这一前沿研究，基于目前研究中（1）没有开展对多维的土地利用/覆被模型的同化、以及（2）对离散的类型分布状态变量如何更新这两个主要问题出发，本文利用贝叶斯推理方法，提出了一种离散的多维土地利用/覆被类型状态变量同化策略。实验结果表明：利用基于共轭先验的贝叶斯推理方法，同化后的总体精度（OA）、Kappa系数以及性能指数（FM）均有所提高。数据同化的性能与景观斑块空间尺度及同化周期密切相关。景观斑块越小，性能指数（FM）越高，同化性能越好，但其对应的同化系统误差（RMSE）却增加。这主要是由于较小的景观斑块的空间异质性较大，无形中放大了模拟和观测之间的差距，致使土地利用/覆被同化系统的误差增加。此外，同化周期越短，性能指数（FM）越高，同化性能也越好，并且对应的同化系统误差（RMSE）也越小。总体而言，基于共轭先验的贝叶斯推理方法能够融合观测数据和多维的土地利用/覆被类型模拟值，以此调整模型的运行轨迹，控制模型的误差积累，提高模型的预测精度。总之，本论文从干旱区内陆河流域土地利用/覆被变化分析、土地利用/覆被动态演化模拟、土地利用/覆被数据同化等方面进一步促进了土地利用/覆被变化研究。土地利用/覆被类型状态变量的同化对相关的离散型变量数据同化方法研究具有一定的指导意义和参考价值。 |
| 英文摘要 | Land use/cover change (LUCC) is a key cause of alternations in the Earth’s land-surface system. They impact regional and global climate, freshwater availability and quality, food security and biodiversity by altering biogeochemical and biophysical processes. Simulating the evolution of complex land use/cover can provide scientific support to land use planning and decision making, but also can serve regional sustainable development. Model simulation is the important method for understanding key processes in land use changes, and for backtrack historical trend and predict future scenario, and support land use policy decisions. Model simulation can produce the required land use change continuously and consistently at a variety of spatial and temporal scales. Over the past decades, a wide range of models of land use/cover change have been developed from different disciplinary backgrounds to meet different reseach needs. Among them, CA are the common model in simulating the pattern and process of land use changes. Acknowledging the inherent uncertainty and complexity of land use/cover change process implies that simulation results of CA will have model errors. Besides, the parameters of most CA models are static, which are not suit for the changeable simulation environment. These uncertainties induced by CA model structure, static model parameters, boundary conditions and initial field are passed from year to year thus producing a propagation and accumulation of errors over time.Observation is another approach for the study of land use/cover change. Observation is more efficient and effective in obtaining the real Earth’s land-surface condition at the time of observation. However, observation on land use/cover data is not available on the continuity, consistency, and historical coverage.On these conditions, the paper is to merge the simulation results of CA model and observation to provide more accurate evolution process of land use/cover. The Zhangye Oasis in the middle reaches of the Heihe River Basin (HRB) was selected as the study area considering the availability of data. The historical land use datasets for the period of 2000-2011 were analyzed to better understand the driving forces of spatial patterns and land use/cover change in Zhangye Oasis. In addition, two LUCC CA integrated models has been developed to overcome the natural drawback of tranditional CA. Based on the LUCC CA integrated model, a new assimilation strategy of multiple discrete categorical variable using Bayesian inference was presented to improve the performance of LUCC CA integrated model. The main contents and conclusions of this paper are listed as follows:Firstly, to provide the basic datasets for the establishment of LUCC CA integrated models following this paper, we present three datasets of land use/cover in the Zhangye Oasis derived from Landsat TM/ETM+ images in 2000, 2007 and 2011. We used these data to investigate changes in land use/cover between 2000 and 2011 and the implications for sustainable water resource management were investigated. The results show that the most significant land use/cover change in the Zhangye Oasis was the continuous expansion of farmland for economic interests. From 2000 to 2011, the farmland area increased by 12.01%. The farmland expansion increased the water stress; thus, groundwater was over-extracted and the ecosystem was degraded in particular areas. Both consequences are negative and potentially threaten the sustainability of the middle reaches of the HRB and the entire river basin. Local governments should therefore improve the management of water resources, particularly groundwater management, and should strictly control farmland reclamation. Then, water resources could be ecologically and socioeconomically sustained, and the balance between upstream and downstream water demands could be ensured. The results of this study can also serve as a reference for the sustainable management of water resources in other arid inland river basin.Secondly, this paper selected CA model to simulate the land use/cover change in study area to provide the model basis for the research on land use/cover assimilation. But the traditional CA has some deficiencies which as quantity forecasting and making conversion rules, etc. And in land use/cover assimilation research, LUCC CA model are binary-state models. Here, a multiple logistic-regression-based Markov cellular automata (MLRMCA) model and a multiple artificial-neural-network-based Markov cellular automata (MANNMCA) model were developed. Compared with the traditional CA model, the two LUCC CA integrated models not only took advantage of the Markov Chain for quantitative forecasting and the CA model for simulating the spatial distribution of a complex system, but also employed a full Logistic Regression and ANN model for determining the parameter values. On the other hand, the two CA models were capable of integrating natural and socioeconomic factors, which were not considered in some CA models. The two CA model were applied simulate complex land use/cover evolutionary processes in Zhangye Oasis during the period 2000-2011. Results indicated that the MANNMCA model was superior to the MLRMCA model in simulated accuracy. Furthermore, combining the artificial neural network with CA more effectively captured the complex relationships between LUCC and a set of drive variables. But the MLRMCA model also showed some advantages. For example, the MLRMCA model was better than the MANNMCA model in terms of the temporal trends in some land use/cover types in particular areas. In short, the two LUCC CA integrated models were reliable, and could reflect the spatial evolution of regional LUCC. Finally, the data assimilation on multiple LUCC model and the update on discrete categorical variable in land use/cover data assimilation study remain unresolved. Here, a new assimilation strategy of multiple discrete categorical variable based on Bayesian inference was presented to control the model errors which will accumulated continuously in the simulation. And we also analyzed the influence of the landscape patch size and the assimilation cycle on assimlation system. Results indicated that the data assimilation method based on the Bayesian inference can effectively reduce the model error in the simulation process and improve the accuracy of simulation results by combing observations and simulation values on multiple discrete land use/cover types. In addition, the performance of data assimilation was closely related to the landscape patch size and assimilation cycle. The error in assimilation system can be reduced by increasing the size of landscape patches. But the assimilation performace using a coarser landscape patch (e.g. 900 m) was worse than that using a finer landscape patch (e.g. 150 m) because the spatial heterogeneity was weakened using a coarser landscape patch. Moreover, the assimilation performance and the error in assimilation system were significantly improved using a short assimilation cycle (e.g. 2 years). In general, the proposed method can regulate the simulation trajectories and reduce accumulated errors when observation data were integrated. |
| 中文关键词 | 土地利用/覆被变化 ; 土地利用模型 ; 数据同化 ; 元胞自动机 ; 贝叶斯推理 |
| 英文关键词 | Land Use/Cover Change Land Use Model Data Assimilation CA Bayesian inference |
| 语种 | 中文 |
| 国家 | 中国 |
| 来源学科分类 | 地图学与地理信息系统 |
| 来源机构 | 中国科学院西北生态环境资源研究院 |
| 资源类型 | 学位论文 |
| 条目标识符 | http://119.78.100.177/qdio/handle/2XILL650/288136 |
| 推荐引用方式 GB/T 7714 | 胡晓利. 土地利用/覆被数据同化研究[D]. 中国科学院大学,2018. |
| 条目包含的文件 | 条目无相关文件。 | |||||
| 个性服务 |
| 推荐该条目 |
| 保存到收藏夹 |
| 导出为Endnote文件 |
| 谷歌学术 |
| 谷歌学术中相似的文章 |
| [胡晓利]的文章 |
| 百度学术 |
| 百度学术中相似的文章 |
| [胡晓利]的文章 |
| 必应学术 |
| 必应学术中相似的文章 |
| [胡晓利]的文章 |
| 相关权益政策 |
| 暂无数据 |
| 收藏/分享 |
除非特别说明，本系统中所有内容都受版权保护，并保留所有权利。
