基于GD__SVM__CA-Markov模型的县域景观格局模拟

Landscape pattern simulation within a county based on GD__SVM__CA-Markov model

利用地理探测器探究地理要素变化与驱动因子关系的优势以及支持向量机分类决策的特点,对CA模型进行改进,并结合Markov模型,形成GD__SVM__CA-Markov模型,以期为县城城镇发展规划以及环境保护提供决策参考。以生态脆弱区典型县域内蒙古磴口县为研究区,基于2011年磴口县土地利用数据,应用GD__SVM__CA-Markov模型,对磴口县2016年的土地景观空间分布格局进行模拟预测,以期发现其变化规律,为了保证模拟精度,将模拟结果与传统CA-Markov模型模拟结果进行对比验证。结果表明,CA-Markov模型模拟结果的总体Kappa系数为0.862 8,GD__SVM__CA-Markov模型模拟结果的总体Kappa系数为0.925 0,2个模型模拟结果的精度均较高,但GD__SVM__CA-Markov模型模拟结果的精度更高,结果更优。因此,将GD__SVM__CA-Markov模型应用于当地土地景观空间分布格局模拟预测,具有一定可行性,可为当地生态治理以及相关政策的实施提供参考。

[Background] Dengkou county is a typical arid and semi-arid area with obviously serious desertification. Ecological environment protection and treatment needs to be solved urgently. In the process of urbanization,balancing the three types of land for construction land,sandy land and ecological land is particularly important. Based on the GD_SVM_CA-Markov model,this paper aims to analyze the change of the dynamic distribution of the landscape in Dengkou county from two dimensions of time and space,to explore its change pattern,and carry on the simulation prediction,so as to provide a certain decisionmaking reference for the local urban development planning,the desertification control and the ecological environment protection. [Methods]Based on 10 driving factors（ DEM,slope,aspect,NDVI,groundwater depth,evapotranspiration,population density,the nearest distance to water area,the nearest distance to settlement,the nearest distance to road）,the land use suitability atlas was created by using Geo Detector to explore the relationship between land use change and 10 driving factors and MCE module provided by IDRISI software; Through SVM to define the transformation rules of the cell,thus the improvement of CA model was achieved; Based on the land use data of the two periods of 2006 and 2011,the Markov model was used to generate the land use transfer matrix. The landscape pattern simulation of study area in 2016 based on the GD_SVM_CA-Markov model was implemented with the above process integrated. In order to test the simulation accuracy,the Kappa coefficient was used for the test [Results]From 2006 to 2016,the landscape area of construction land in Dengkou county increased from 5 785. 55 hm2 to 8 952. 67 hm2,the landscape area of sandy land decreased from 76 616. 15 hm2 to 56 460. 50 hm2,the landscape area of water area increased from 23 859. 88 hm2 to 24 679. 10 hm2,the landscape area of woodland and grassland increased from 117 452. 37 hm2 to 128 120. 87 hm2. For construction land,there was 15. 64% probability of conversion to arable land. In the case of water area,there was 11. 56% probability of turning into arable land. In terms of sandy land,there was 18. 37% probability of turning into woodland and grassland. The influence degree of the 10 driving factors on the landscape type change in Dengkou county was 0. 248 816,0. 048 784,0. 134 342,0. 951 212,0. 975 924,0. 873 667,0. 520 317,0. 256 226,0. 413 550,0. 178 658 respectively according to the above order. The Kappa coefficient of the CA-Markov model simulation results of 2016 was 0. 862 8,the Kappa coefficient of the GD_SVM_CA-Markov model simulation result of 2016 was 0. 925 0. Based on the land use data of 2016 and the land use transfer data of 2011—2016,the GD_SVM_CA-Markov model was used to simulate and predict the spatial distribution pattern of landscape in2021. During 2016—2021,the landscape area of construction land increased from 8 952. 67 hm2 to11 610. 21 hm2,the landscape area of sandy land increased from 56 460. 50 hm2 to 67 235. 11 hm2,and the landscape area of ecological land such as water area and woodland and grassland decreased from 152 799. 97 hm2 to 143 670. 04 hm2. [Conclusions]Hydrological conditions,vegetation cover and population factor are the decisive factors that determine the temporal and spatial changes of local landscape types. Thus,at the same time as urban development, it is necessary to pay attention to ecological and environmental protection. The simulation result of 2016 based on GD_ SVM_ CA-Markov model has higher overall simulation accuracy and is better than the simulation result of 2016 based on CA-Markov model. Therefore,it is feasible to use GD_SVM_CA-Markov model to simulate and predict the spatial distribution pattern of landscape in Dengkou county.