基于BP-SVM模型的植被变化模拟研究

doi:10.13866/j.azr.2021.04.20

摘要/Abstract

摘要：

植被是连接生物圈、大气圈、水圈的重要纽带,对流域生态环境和水热状态变化有重要的影响。当前研究大多聚焦于NDVI与气候因子的相关性分析,少数研究预测NDVI时忽视了考虑滞后性以及增加预报因子对提高模型精确度的影响。基于此,本文比较多元线性回归(MLR)、人工神经网络(ANN)和支持向量机模型(SVM),优选出精度较高的模型。在常规因子(降雨和气温)基础上增加了对植被生长有影响的土壤湿度和日照因子,并考虑了不同因子与NDVI之间的滞时效应。研究结果表明:(1) 支持向量机模型(SVM)的拟合能力最强,NDVI预测精度最高,泾河、北洛河流域均方根误差均减少1.8%以上。(2) 加入土壤湿度、日照等因子,泾河流域模型预测精度提高,泾河流域均方根误差减少8.8%。(3) 考虑滞时后,泾河、北洛河流域均方根误差分别减少15%、11%,NDVI预测精度进一步提高,增加了模型的可靠性。预测结果可对未来制定生态保护策略,指导生态修复具有重要参考意义。

关键词: 支持向量机, 预测精度, 预测模型, 滞时效应, 植被覆盖度

Abstract:

Vegetation is an important link that connects the biosphere, atmosphere and hydrosphere, and has an important impact on the watershed in the ecological environment and on the exchange of water and heat. Previous studies focused on correlation analyses between the normalized difference vegetation index (NDVI) and climate factors, but the lag and the increase of forecast factors were rarely used to improve the precision of the NDVI model. Thus, this article compared the prediction performance of the multiple linear regression, artificial neural network and support vector machine, then selected the model with the highest precision. Using conventional factors (rainfall and temperature), the prediction performance was tested when soil moisture and sunshine duration were increased, which affect vegetation growth, and the time lag effect between the different factors and NDVI were considered. (1) SVM had the strongest fitting ability and the highest NDVI prediction accuracy. The root-mean-square error of the Jing River Basin and the Beiluo River Basin were reduced by more than 1.8%. (2) The root-mean-square error (RMSE) of Jing River Basin decreased by up to 8.8% when soil moisture, sunshine and other factors were added. (3) Considering the lag time, the root-mean-square error of the Jing River Basin and the Beiluo River Basin decreased by 15% and 11%, respectively, and the predicting accuracy of NDVI was improved further, thus increasing the reliability of the model. These prediction results have an important reference significance for the formulation of ecological protection strategies and the guidance of ecological restoration in the future.

Key words: support vector machines, prediction accuracy, prediction model, time-lag effect, vegetation coverage

贾松涛,黄生志,王浩,李紫妍,黄强,梁浩. 基于BP-SVM模型的植被变化模拟研究[J]. 干旱区研究, 2021, 38(4): 1085-1093.

JIA Songtao,HUANG Shengzhi,WANG Hao,LI Ziyan,HUANG Qiang,LIANG Hao. Simulation of vegetation change based on BP-SVM mode[J]. Arid Zone Research, 2021, 38(4): 1085-1093.

图/表 8

图1

图2

表1

图3

表2

表3

表4

图4

参考文献 30

[1]	刘家福, 马帅, 李帅, 等. 1982—2016年东北黑土区植被NDVI动态及其对气候变化的响应[J]. 生态学报, 2018, 38(21):7647-7657.
	[ Liu Jiafu, Ma Shuai, Li Shuai, et al. Changes in vegetation NDVI from 1982 to 2016 and its responses to climate change in the black-soilarea of Northeast China[J]. Acta Ecologica Sinica, 2018, 38(21):7647-7657. ]
[2]	田甜, 李绍才, 陈敏, 等. 雅砻江流域植被指数长时间序列变化分析[J]. 水力发电学报, 2012, 31(2):159-164.
	[ Tian Tian, Li Shaocai, Chen Min, et al. Analysis on vegetation index’s long time series dynamics of Yalong River basin[J]. Journal of Hydroelectric Engineering, 2012, 31(2):159-164. ]
[3]	杨峰, 李建龙, 钱育蓉, 等. 天山北坡典型退化草地植被覆盖度监测模型构建与评价[J]. 自然资源学报, 2012, 27(8):1340-1348.
	[ Yang Feng, Li Jianlong, Qian Yurong, et al. Estimating vegetation coverage of typical degraded grassland in the Northern Tianshan Mountains[J]. Journal of Natural Resources, 2012, 27(8):1340-1348. ]
[4]	何辉, 玉素甫江·如素力. 2001—2015年伊犁地区植被NDVI变化及其影响因子的相对作用分析[J]. 中南林业科技大学学报, 2019, 39(10):76-87.
	[ He Hui, Yusufujiang Rusuli. Analysis of the relative role of vegetation cover changes and its influencing factors in Yili area from 2001 to 2015[J]. Journal of Central South University of Forestry & Technology, 2019, 39(10):76-87. ]
[5]	郭铌, 管晓丹. 植被状况指数的改进及在西北干旱监测中的应用[J]. 地球科学进展, 2007, 22(11):1160-1168.
	[ Guo Ni, Guan Xiaodan. An improvment of the vegetation condition index with applications to the drought monitoring in Northwest China[J]. Advances in Earth Science, 2007, 22(11):1160-1168. ]
[6]	杜加强, 高云, 贾尔恒•阿哈提, 等. 近30年新疆植被生长异常值时空变化及驱动因子[J]. 生态学报, 2016, 36(7):1915-1927.
	[ Du Jiaqiang, Gao Yun, Jiaerheng Ahati, et al. Spatio-temporal patterns and driving factors of vegetation growth anomalies in Xinjiang over the last three decades[J]. Acta Ecologica Sinica, 2016, 36(7):1915-1927. ]
[7]	Wu D H, Zhao X, Huang K C, et al. Time-lag effects of global vegetation responses to climate change[J]. Global Change Biology, 2015, 21(9):3520-3531. doi: 10.1111/gcb.2015.21.issue-9
[8]	陶帅, 邝婷婷, 彭文甫, 等. 2000-2015年长江上游NDVI时空变化及驱动力——以宜宾市为例[J]. 生态学报, 2020, 40(14):5029-5043.
	[ Tao Shuai, Kuang Tingting, Peng Wenfu, et al. Analyzing the spatio-temporal variation and drivers of NDVI in upper reaches ofthe Yangtze River from 2000 to 2015: A case study of Yibin City[J]. Acta Ecologica Sinica, 2020, 40(14):5029-5043. ]
[9]	李登科, 郭铌, 何慧娟. 陕北长城沿线风沙区植被指数变化及其与气候的关系[J]. 生态学报, 2007, 27(11):4620-4629.
	[ Li Dengke, Guo Ni, He Huijuan. Vegetation change and its relationship with climate in the region along the Great Wall in northern Shaanxi[J]. Acta Ecologica Sinica, 2007, 27(11):4620-4629. ]
[10]	孙锐, 陈少辉, 苏红波. 黄土高原不同生态类型NDVI时空变化及其对气候变化响应[J]. 地理研究, 2020, 39(5):1200-1214. doi: 10.11821/dlyj020190399
	[ Sun Rui, Chen Shaohui, Su Hongbo. Spatiotemporal variation of NDVI in different ecotypes on the Loess Plateau and its response to climate change[J]. Geographical Research, 2020, 39(5):1200-1214. ] doi: 10.11821/dlyj020190399
[11]	王伦澈. 区域大气辐射变化及其对地表植被生产力的定量影响研究[D]. 武汉: 武汉大学, 2015.
	[ Wang Lunche. Reginal Variations of Atmosphere Radiation and its Quantitative Effects on the Terrestrial Ecosystem Productivity[D]. Wuhan: Wuhan University, 2015. ]
[12]	张满囤, 黄春萌, 米娜, 等. 基于支持向量机回归的NDVI组合预测模型[J]. 河北工业大学学报, 2017, 46(4):39-45.
	[ Zhang Mantun, Huang Chunmeng, Mi Na, et al. Combination forecast model of NDVI based on support vector machine regression[J]. Journal of Hebei University of Technology, 2017, 46(4):39-45. ]
[13]	Zhou Zhaoqiang, Ding Yibo, Shi Haiyun, et al. Analysis and prediction of vegetation dynamic changes in China: Past, present and future[J]. Ecological Indicators, 2020, 117:106642. doi: 10.1016/j.ecolind.2020.106642
[14]	杨曦, 武建军, 闫峰, 等. 基于地表温度植被指数特征空间的区域土壤干湿状况[J]. 生态学报, 2009, 29(3):1205-1216.
	[ Yang Xi, Wu Jianjun, Yan Feng, et al. Assessment of regional soil moisture status based on characteristics of surface temperature vegetation index space[J]. Acta Ecologica Sinica, 2009, 29(3):1205-1216. ]
[15]	Tian Siyuan, Van Dijk A I J M, Paul Tregoning, et al. Forecasting dryland vegetation condition months in advance through satellite data assimilation[J]. Nature Communications, 2019, 10:469. doi: 10.1038/s41467-019-08403-x pmid: 30692539
[16]	裴志林, 杨勤科, 王春梅, 等. 黄河上游植被覆盖度空间分布特征及其影响因素[J]. 干旱区研究, 2019, 36(3):546-555.
	[ Pei Zhilin, Yang Qinke, Wang Chunmei, et al. Spatial distribution of vegetation coverage and its affecting factors in the upper reaches of the Yellow River[J]. Arid Zone Research, 2019, 36(3):546-555. ]
[17]	刘静. 退耕还林后黄土高原植被覆被变化过程及未来分布预测[D]. 北京: 中国科学院大学, 2019.
	[ Liu Jing. Vegetation Cover Change Process and Future Distribution Prediction on the Loess Plateau after Grain to Green[D]. Beijing: University of Chinese Academy of Science, 2019. ]
[18]	陈末, 卢文喜, 侯泽宇, 等. 基于支持向量机的吉林西部地下水水质评价[J]. 节水灌溉, 2013, 38(5):29-33.
	[ Chen Mo, Lu Wenxi, Hou Zeyu, et al. The assesement of groundewater quality based on support vector machine in Western Jilin[J]. Water Saving Irrigation, 2013, 38(5):29-33. ]
[19]	毛慧慧, 延耀兴, 张杰. 水文预报方法研究现状与展望[J]. 科技情报开发与经济, 2005, 15(19):172-173.
	[ Mao Huihui, Yan Yaoxing, Zhang Jie. The present situation and prospect of the hydrographic forecasting methods[J]. Journal of Library and Information Science, 2005, 15(19):172-173. ]
[20]	梁浩, 黄生志, 孟二浩, 等. 基于多种混合模型的径流预测研究[J]. 水利学报, 2020, 51(1):112-125.
	[ Liang Hao, Huang Shengzhi, Meng Erhao, et al. Runoff prediction based on multiple hybrid models[J]. Journal of Hydraulic Engineering, 2020, 51(1):112-125. ]
[21]	代萌, 黄生志, 黄强, 等. 干旱多属性风险动态评估与驱动力分析[J]. 水力发电学报, 2019, 38(8):15-26.
	[ Dai Meng, Huang Shengzhi, Huang Qiang, et al. Dynamic assessments of drought multi-attribute risks and analysis of its driving force[J]. Journal of Hydroelectric Engineering, 2019, 38(8):15-26. ]
[22]	王秀英, 廖留峰, 王俊杰. 基于多元线性回归的滇西南短时强降水预报模型研究[J]. 气象与环境学报, 2019, 35(2):15-22.
	[ Wang Xiuying, Liao Liufeng, Wang Junjie. A forecast model for flash heavy rainfall in southwestern Yunnan province based on a multiple linear regression method[J]. Journal of Meteorology and Environment, 2019, 35(2):15-22. ]
[23]	孟二浩, 黄生志, 黄强, 等. 融合大气环流异常因子的径流预报研究[J]. 水力发电学报, 2017, 36(8):34-42.
	[ Meng Erhao, Huang Shengzhi, Huang Qiang, et al. Runoff prediction incorporating anomalous atmospheric circulation factors[J]. Journal of Hydroelectric Engineering, 2017, 36(8):34-42. ]
[24]	张俊. 中长期水文预报及调度技术研究与应用[D]. 大连: 大连理工大学, 2009.
	[ Zhang Jun. Mid-and-Long Term Hydrological Forecasting and Operation Techniques Research and Application[D]. Dalian: Dalian University of Technology, 2009. ]
[25]	张霞, 李占斌, 张振文, 等. 两种预测模型在地下水动态中的比较与应用[J]. 生态学报, 2012, 32(21):6788-6794.
	[ Zhang Xia, Li Zhanbin, Zhang Zhenwen. Application and comparison of two prediction models for groundwater dynamics[J]. Acta Ecologica Sinica, 2012, 32(21):6788-6794. ]
[26]	徐冬梅, 赵晓慎. 中长期水文预报方法研究综述[J]. 水利科技与经济, 2010, 16(1):1-7.
	[ Xu Dongmei, Zhao Xiaoshen. Review on study of mid and long-term hydrological forecasting technique[J]. Water Conservancy Science and Technology and Economy, 2010, 16(1):1-7. ]
[27]	钱剑平. 基于人工神经网络的策勒河流域径流预测研究[D]. 乌鲁木齐: 新疆大学, 2018.
	[ Qian Jianping. A Study on Runoff Prediction in Qira River Basin Based on Artificial Neural Network[D]. Urumqi: Xinjiang University, 2018. ]
[28]	杨汉波, 吕华芳, 胡庆芳, 等. 华北平原的大气逆辐射参数化方法比较[J]. 清华大学学报(自然科学版), 2014, 54(5):590-595.
	[ Yang Hanbo, Lyu Huafang, Hu Qingfang, et al. Comparison of parametrization methods for calculating the downward long-wave radiation over the North China Plain[J]. Journal of Tsinghua University (Science and Technology), 2014, 54(5):590-595. ]
[29]	丛晓红, 拾兵, 于西达, 等. 基于PSO-BP神经网络的黄河利津站输沙量预测[J]. 人民黄河, 2020, 42(1):1-8.
	[ Cong Xiaohong, Shi Bing, Yu Xida, et al. Prediction of sediment discharge at Lijin station of the Yellow River based on PSO-BP neural network[J]. Yellow River, 2020, 42(1):1-8. ]
[30]	Anderson L O, Malhi Y, Aragão Luiz E O C, et al. Remote sensing detection of droughts in Amazonian forest canopies[J]. New Phytologist, 2010, 187(3):733-750. doi: 10.1111/j.1469-8137.2010.03355.x

研究区域	类别	多元线性回归				人工神经网络				支持向量机
研究区域	类别	纳什系数	均方根误差	平均相对误差		纳什系数	均方根误差	平均相对误差		纳什系数	均方根误差	平均相对误差
泾河	训练期	0.820	0.066	0.203		0.936	0.038	0.118		0.928	0.040	0.115
泾河	验证期	0.731	0.092	0.224		0.750	0.088	0.183		0.761	0.086	0.177
北洛河	训练期	0.832	0.077	0.187		0.953	0.040	0.092		0.953	0.040	0.083
北洛河	验证期	0.544	0.128	0.205	0.611		0.120	0.191	0.663		0.103	0.184

研究区域	模型类别	降雨气温两因子				降雨、气温、土壤湿度、日照多因子
研究区域	模型类别	纳什系数	均方根误差	平均相对误差		纳什系数	均方根误差	平均相对误差
泾河	训练期	0.928	0.040	0.115		0.942	0.036	0.099
泾河	验证期	0.761	0.086	0.177		0.729	0.079	0.167
北洛河	训练期	0.953	0.040	0.083		0.959	0.041	0.084
北洛河	验证期	0.663	0.103	0.184	0.670		0.118	0.188

研究区域	气候因子	滞后时间/月
研究区域	气候因子	0	1	2	3	4	5	6	7	8	9	10	11	12
泾河	NDVI	-	0.80	0.42	0.01	-0.42	-0.78	-0.90	-0.78	-0.41	0.01	0.42	0.78	0.97
	降雨	0.90	0.88	0.62	0.19	-0.30	-0.70	-0.91	-0.88	-0.62	-0.18	0.31	0.69	0.90
	气温	0.68	0.60	0.32	-0.04	-0.35	-0.59	-0.69	-0.61	-0.37	0.01	0.37	0.59	0.68
	土壤湿度	-0.15	-0.32	-0.41	-0.38	-0.23	-0.05	0.15	0.33	0.43	0.41	0.26	0.03	-0.20
	日照	0.20	0.33	0.35	0.31	0.17	0.01	-0.17	-0.30	-0.38	-0.36	-0.24	-0.02	0.18
北洛河	NDVI	-	0.42	0.00	-0.43	-0.79	-0.92	-0.78	-0.42	0.01	0.43	0.79	0.98	0.79
	降雨	0.90	0.87	0.61	0.17	-0.31	-0.71	-0.92	-0.88	-0.61	-0.16	0.32	0.71	0.90
	气温	0.75	0.64	0.33	-0.08	-0.41	-0.64	-0.73	-0.63	-0.38	0.03	0.42	0.66	0.75
	土壤湿度	-0.20	-0.37	-0.45	-0.40	-0.22	-0.01	0.17	0.33	0.42	0.40	0.26	0.01	-0.22
	日照	0.16	0.30	0.32	0.29	0.15	-0.01	-0.15	-0.25	-0.32	-0.31	-0.21	-0.02	0.14

研究区域	类别	考虑滞时的预测模型				不考虑滞时的预测模型
研究区域	类别	纳什系数	均方根误差	平均相对误差		纳什系数	均方根误差	平均相对误差
泾河	训练期	0.982	0.020	0.048		0.940	0.037	0.103
泾河	验证期	0.880	0.062	0.140		0.787	0.082	0.183
北洛河	训练期	0.989	0.020	0.038		0.947	0.042	0.085
北洛河	验证期	0.927	0.056	0.112	0.687		0.115	0.185