地球信息科学学报 ›› 2020, Vol. 22 ›› Issue (3): 531-542.doi: 10.12082/dqxxkx.2020.190359

• “土地利用模拟”专栏 • 上一篇    下一篇

纳入空间自相关的FLUS模型在土地利用变化多情景模拟中的应用

张经度1, 梅志雄1,2,*(), 吕佳慧1, 陈进钊1   

  1. 1. 华南师范大学地理科学学院, 广州 510631
    2. 华南师范大学粤港澳村镇可持续发展研究中心, 广州 510631
  • 收稿日期:2019-07-04 修回日期:2019-12-19 出版日期:2020-03-25 发布日期:2020-05-18
  • 通讯作者: 梅志雄 E-mail:zhixiongmei76@126.com
  • 作者简介:张经度(1995— ),男,江西赣州人,硕士生,主要从事空间分析与建模、土地利用模拟等研究。E-mail:cuitzjd@163.com
  • 基金资助:
    国家自然科学基金(41001078)

Simulating Multiple Land Use Scenarios based on the FLUS Model Considering Spatial Autocorrelation

ZHANG Jingdu1, MEI Zhixiong1,2,*(), LV Jiahui1, CHEN Jinzhao1   

  1. 1. School of Geography, South China Normal University, Guangzhou 510631, China
    2. Center for Sustainable Development of Rural and Town in Guangdong-Hong Kong-Marco Greater Bay Area, South China Normal University, Guangzhou 510631, China
  • Received:2019-07-04 Revised:2019-12-19 Online:2020-03-25 Published:2020-05-18
  • Contact: MEI Zhixiong E-mail:zhixiongmei76@126.com
  • Supported by:
    National Natural Science Foundation of China(41001078)

摘要:

FLUS模型是一种新型的土地利用变化模拟模型,应用前景广阔。本文通过在FLUS模型的人工神经网络(Artificial Neural Network,ANN)训练模块中引入空间自相关因子来改进模型,以珠江三角洲地区为例,基于2009年、2015年土地利用数据和一系列驱动因子对改进的模型进行了验证,并利用该改进的FLUS模型模拟了2035年研究区在3种情景下土地利用变化格局。结果表明:① 引入空间自相关因子后各地类发生概率分布的预测精度更高,耕地、林地、建设用地、水体和未利用土地的拟合优度ROC值分别从0.819、0.928、0.885、0.855和0.861提高到0.857、0.934、0.890、0.863和0.978;② 改进的FLUS模型的模拟精度有一定的提高,Kappa系数从0.732提高到0.744,FOM系数从0.077升到0.106;③ 情景模拟表明,3种情景下珠江三角洲建设用地和林地均将增加、而耕地均呈减少趋势。但不同情景下模拟的土地利用格局也存在显著差异: 基准情景下,建设用地明显扩张且大幅侵占耕地。耕地保护情景下,耕地面积保持在合理水平,建设用地蔓延扩张趋势得到遏制,土地利用布局总体趋向合理。生态保护情景下,耕地、林地和水体得到较好保护,建设用地布局更为合理,土地利用可持续性明显提高。

关键词: FLUS模型, 人工神经网络, 系统动力学, 空间自相关, 多尺度, 土地利用变化, 多情景模拟, 珠江三角洲

Abstract:

The Future Land Use Simulation (FLUS) model is a new model for simulating multiple land-use changes, and has a broad application prospect. This paper improved the FLUS model by incorporating a spatial autocorrelation factor into the Artificial Neural Network (ANN) module of FLUS, selected thePearl River Delta region as the case study area, and validated the improved FLUS model based on the land use data of 2009 and 2015, as well as a series of driving factors. Three future land-use scenarios in 2035: the baseline scenario, cultivated protection scenario, and ecological protection scenario, were simulated using the improved model. The results showed that: (1) After incorporating the spatial autocorrelation factor, the model had better predictive powerfor the occurrence probability distribution ofeach land use. The ROC values of cultivated land, forestland, water area,construction land, and unused land increased from 0.819, 0.928, 0.885, 0.855, and 0.861 to 0.857, 0.934, 0.890, 0.863 and 0.978, respectively. (2) The simulation accuracy of the improved FLUS model was improved. The Kappa value increased from 0.732 to 0.744, and the FOM value increased from 0.077 to 0.106. (3) The scenario simulation results indicated that under all three scenarios, forestland and construction land would increase, whereas cultivated land would decrease. Apparent differences also existed in the simulated change sizes and locations of each land use type under different scenarios. Under the baseline scenario, construction land would expand rapidly at the expense of a large amount of cultivated land. Under the cultivated land protection scenario, cultivated land area would remain at a reasonable level, the expansion of construction land would alleviate, and the land use layout would tend to be reasonable. Under the ecological protection scenario, cultivated land, forestland, and water area would be well protected, the layout of construction land would be more rational, and the land use sustainability in the study area would be improved significantly.

Key words: Key Word: FLUS model, artificial neural networks, system dynamics, spatial autocorrelation, multiscale, land use change, multi-scenarios simulation, Pearl River Delta