机器学习方法在预测泉水潜在出露位置中的应用

doi:10.12082/dqxxkx.2021.200522

摘要/Abstract

摘要：

泉水出露受到多种因素影响,在传统地质勘查手段之外,各种模型方法及影响因子预测手段,也被越来越多地应用于泉水的研究中。本文尝试利用机器学习的方法进行泉水出露位置的预测研究。根据北京市野外调查,确定了1378个测试样本点,选取了高程、坡度、坡向、地形湿度指数、径流强度指数、距河流距离、距断裂距离、岩性、归一化植被指数及土地利用类型作为影响因子,对比了2种机器学习方法（随机森林模型、分类回归树模型）和地统计方法（证据权重模型）的预测效果。研究发现：随机森林模型的预测效果最好（Area Under Curve, AUC=0.86）,分类回归树和证据权重模型效果相当（AUC分别为0.81、0.80）;随机森林模型同时揭示,岩性、距断裂距离和距河流距离这3个影响因子对泉潜在出露的影响最大。本研究表明,在强烈人类活动影响下机器学习方法仍然具有较好的泉水出露预测能力,有望为泉水保护、恢复提供新的技术方法。

关键词: 泉水潜在出露, 证据权重, 随机森林模型, 分类回归树模型, 北京市

Abstract:

The exposure of spring is usually difficult to be monitored over mountainous terrain. In this study we investigated the performance of statistical models (Weight of Evidence) and two machine learning models (Random Forest and Classification and Regression Tree) in predicting the potential exposure positions of spring water in Beijing. A total of 1378 springs from field survey were used for model training and validation. The environmental factors included elevation, slope, aspect, topographic wetness index, stream power index, distance to rivers, distance to faults, lithology, normalized difference vegetation index, and land use. The predicted results from the three models are validated using the receiver operating characteristics curve. The area under the curve for the Weight of Evidence model was 0.80, while that for Classification and Regression Tree and Random Forest the AUC was 0.81 and 0.86, respectively. Therefore, the Random Forest model has the best prediction performance. Moreover, the Random Forest model revealed that lithology, distance to faults, and distance to rivers had the greatest impact on the spring exposure. This study shows that the machine learning method has good prediction ability and is expected to be applied in future spring protection and restoration researches.

Key words: Groundwater spring potential map, weight of evidence, random forest, classification regression tree, Beijing

李慧香, 潘云, 宫辉力, 孙颖. 机器学习方法在预测泉水潜在出露位置中的应用[J]. 地球信息科学学报, 2021, 23(6): 1028-1039.DOI:10.12082/dqxxkx.2021.200522

LI Huixiang, PAN Yun, GONG Huili, SUN Ying. Application of Machine Learning Method in Prediction of Potential Exposure Position of Spring Water[J]. Journal of Geo-information Science, 2021, 23(6): 1028-1039.DOI:10.12082/dqxxkx.2021.200522

图/表 13

图1

表1

图2

图3

表2

计算的各影响因子权重"

影响因子	分类		栅格数目/个	训练泉点数目	正相关权重W⁺	负相关权重W^-	空间相关性C	后验概率P₍_s₎
高程/m	<0		175 681	0	-	-	-	0
	0~400		10 485 480	326	-0.5338	0.4460	-0.9798	3.1091E-05
	400~800		5 103 572	486	0.5857	-0.3714	0.9570	9.5234E-05
	800~1200		2 033 781	133	0.2098	-0.0298	0.2396	6.5399E-05
	1200~1600		345 649	20	0.0874	-0.0018	0.0892	5.7865E-05
坡度/°	0~10		9 015 882	379	-0.2321	0.1851	-0.4172	4.2038E-05
	10~20		3 944 854	335	0.4711	-0.1821	0.6532	8.4926E-05
	20~30		3 054 769	171	0.0543	-0.0113	0.0656	5.5981E-05
	30~40		1 688 169	67	-0.2896	0.0254	-0.3150	3.9689E-05
	40~50		438 911	12	-0.6623	0.0119	-0.6742	2.7341E-05
	50~60		54 453	1	-1.0602	0.0020	-1.0622	1.8365E-05
	>60		3 361	0	-	-	-	0
坡向/°	北		2 077 367	100	-0.0966	0.0118	-0.1084	4.8140E-05
	东北		2 214 218	126	0.0707	-0.0102	0.0810	5.6908E-05
	东		2 332 611	178	0.3642	-0.0668	0.4309	7.6314E-05
	东南		2 417 942	146	0.1301	-0.0215	0.1516	6.0385E-05
	南		2 480 078	133	0.0114	-0.0018	0.0132	5.3630E-05
	西南		2 233 444	89	-0.2856	0.0342	-0.3197	3.9850E-05
	西		2 134 969	94	-0.1858	0.0223	-0.2081	4.4030E-05
	西北		2 104 272	95	-0.1607	0.0192	-0.1800	4.5148E-05
	平面		205 498	4	-1.0021	0.0072	-1.0093	1.9465E-05
影响因子		分类	栅格数目	训练泉点数目	正相关权重W ⁺	负相关权重W ^-	空间相关性C	后验概率P₍_s₎
距河流距离/m		0~200	4 093 718	460	0.7512	-0.3928	1.1440	1.1238E-04
		200~400	3 554 978	153	-0.2086	0.0447	-0.2532	4.3040E-05
		400~600	3 051 250	116	-0.3326	0.0554	-0.3880	3.8018E-05
		600~800	2 537 820	88	-0.4246	0.0545	-0.4792	3.4676E-05
		800~1000	1 999 299	66	-0.4738	0.0455	-0.5193	3.3013E-05
		>1000	2 963 334	82	-0.6503	0.0889	-0.7392	2.7672E-05
距断裂距离/m		0~500	2 608 221	276	0.6911	-0.1822	0.8734	1.0583E-04
		500~1000	1 951 677	148	0.3579	-0.0531	0.4110	7.5837E-05
		1000~1500	1 584 723	112	0.2875	-0.0323	0.3197	7.0679E-05
		1500~2000	1 311 105	90	0.2583	-0.0231	0.2815	6.8649E-05
		2000~2500	1 051 075	64	0.1384	-0.0091	0.1476	6.0894E-05
		2500~3000	841 660	41	-0.0847	0.0039	-0.0886	4.8716E-05
		3000~3500	699 895	37	-0.0029	0.0001	-0.0030	5.2868E-05
		>3500	8 152 043	197	-0.7857	0.3657	-1.1514	2.4166E-05
地形湿度指数		2~8	14 966 464	663	-0.1797	0.5660	-0.7457	4.4299E-05
		8~14	2 935 668	248	0.4659	-0.1212	0.5870	8.4484E-05
		14~22	217 898	54	1.5423	-0.0455	1.5878	2.4788E-04
		>22	80 369	0	-	-	-	0
径流强度指数		0~10	16 681 699	723	-0.2016	1.1004	-1.3019	4.3341E-05
		10~20	483 216	29	0.1240	-0.0036	0.1276	6.0018E-05
		20~30	199 579	19	0.5854	-0.0089	0.5943	9.5209E-05
		30~40	117 099	9	0.3714	-0.0029	0.3743	7.6864E-05
		40~50	79 251	14	1.2037	-0.0103	1.2139	1.7668E-04
		50~60	58 232	7	0.8187	-0.0041	0.8227	1.2022E-04
		60~70	45 302	8	1.2033	-0.0058	1.2092	1.7662E-04
		70~80	36 436	7	1.2876	-0.0053	1.2929	1.9215E-04
		80~90	30 290	4	0.9127	-0.0025	0.9152	1.3207E-04
		>90	469 295	145	1.7629	-0.1367	1.8996	3.0907E-04
归一化植被指数		-1~-0.1	288 943	2	-2.0360	0.0139	-2.0499	6.9218E-06
		-0.1~0.06	1 429 937	33	-0.8318	0.0470	-0.8788	2.3078E-05
		0.06~0.08	1 617 730	90	0.0482	-0.0048	0.0530	5.5636E-05
		0.08~0.1	2 542 398	173	0.2496	-0.0471	0.2967	6.8050E-05
		0.1~0.12	3 778 473	257	0.2491	-0.0770	0.3261	6.8021E-05
		0.12~0.14	3 809 386	215	0.0625	-0.0172	0.0798	5.6442E-05
		0.14~0.16	2 284 891	132	0.0858	-0.0130	0.0988	5.7774E-05
		0.16~0.18	1 125 212	40	-0.3997	0.0215	-0.4212	3.5550E-05
		0.18~0.4	1 312 834	24	-1.0648	0.0497	-1.1145	1.8281E-05
		0.4~1	10 595	0	-	-	-	0
土地利用类型		水体	250 428	2	-1.8929	0.0118	-1.9047	7.9864E-06
		草地	2 199 456	34	-1.2325	0.0929	-1.3254	1.5459E-05
		城市	5 059 188	206	-0.2640	0.0856	-0.3495	4.0719E-05
		森林	10 003 205	676	0.2426	-0.4081	0.6507	6.7580E-05
		裸地	688 122	48	0.2744	-0.0125	0.2869	6.9760E-05
岩性		松散岩类	7 370 387	15	-3.2601	0.5035	-3.7636	2.0352E-06
		碳酸岩类	3 768 433	198	-0.0090	0.0023	-0.0114	5.2544E-05
		岩浆岩类	4 688 911	558	0.8086	-0.5654	1.3740	1.1901E-04
		碎屑岩类	1 264 728	98	0.3795	-0.0351	0.4146	7.7493E-05
		变质岩类	1 107 940	96	0.4912	-0.0420	0.5332	8.6654E-05

表2

图4

表3

图5

表4

表5

图6

图7

表6

参考文献 35

[1]	田秋菊, 牛波, 王现国, 务宗伟. 我国基岩地下水开发利用和研究现状[J]. 地下水, 2004,26(2):88-90.
	[ Tian Q J, Niu B, Wang X G, et al. Present situation of the exploitation, utilization and study of groundwater in bedrock in China[J]. Ground water, 2004,26(2):88-90. ]
[2]	袁文波. 近代地壳运动对基岩裂隙水的控制作用研究[D]. 南京:南京大学, 2018.
	[ Yuan W B. A study of modern movement of the earth crust on groundwater in fractured rocks[D]. Nanjing: Nanjing University, 2018. ]
[3]	潘玉玲, 李振宇, 万乐, 等. 利用核磁共振方法探查基岩裂隙水[J]. CT理论与应用研究, 2000,9(1):22-25.
	[ Pan Y L, Li Z Y, Wan L, et al. Detecting bedrock fissure water with Nuclear Magnetic Resonance (NMR) method[J]. CT Theory and Applications, 2000,9(1):22-25. ]
[4]	潘玉玲, 贺颢, 李振宇, 等. 地面核磁共振找水方法在中国的应用效果[J]. 地质通报, 2003,22(2):135-139.
	[ Pan Y L, He H, Li Z Y, et al. Surface detection of groundwater with the nuclear magnetic resonance method and its application result in China[J]. Geogogical Bullretin of China, 2003,22(2):135-139. ]
[5]	武毅, 封绍武, 王亚清. 应用大地电磁法TE、TM模式勘查构造裂隙水[J]. 物探与化探, 2011,35(3):329-332.
	[ Wu Y, Feng S W, Wang Y Q. The technological application of TE and TM mode to the prospecting for structural fissure water[J]. Geophysical & Geochemical Exploration, 2011,35(3):329-332. ]
[6]	Pourtaghi Z S, Pourghasemi H R. GIS-based groundwater spring potential assessment and mapping in the Birjand Township, southern Khorasan Province, Iran[J]. Hydrogeology Journal, 2014,22(3):643-662. doi: 10.1007/s10040-013-1089-6
[7]	Oh H J, Kim Y S, Choi J K, et al. GIS mapping of regional probabilistic groundwater potential in the area of Pohang City, Korea[J]. Journal of Hydrology, 2011,399(3-4):158-172. doi: 10.1016/j.jhydrol.2010.12.027
[8]	Ghimire M, Chapagain P S, Shrestha S. Mapping of groundwater spring potential zone using geospatial techniques in the Central Nepal Himalayas: A case example of Melamchi-Larke area[J]. Journal of Earth System ence, 2019,128(2).
[9]	Ozdemir A. Using a binary logistic regression method and GIS for evaluating and mapping the groundwater spring potential in the Sultan Mountains (Aksehir, Turkey)[J]. Journal of Hydrology, 2011,405(1-2):123-136. doi: 10.1016/j.jhydrol.2011.05.015
[10]	Naghibi S A, Pourghasemi H R, Dixon B. GIS-based groundwater potential mapping using boosted regression tree, classification and regression tree, and random forest machine learning models in Iran[J]. Environmental Monitoring & Assessment, 2016,188(1):44.
[11]	Chen W, Tsangaratos P, Ilia I, et al. Groundwater spring potential mapping using population-based evolutionary algorithms and data mining methods[J]. Science of The Total Environment, 2019,684(SEP.20):31-49. doi: 10.1016/j.scitotenv.2019.05.312
[12]	Choubin B, Rahmati O, Soleimani F, et al. Regional Groundwater Potential Analysis Using Classification and Regression Trees-ScienceDirect[J]. Spatial Modeling in GIS and R for Earth and Environmental Sciences, 2019: 485-498.
[13]	赵春红, 梁永平, 王维泰, 等. 北京西山泉域岩溶水系统特征探讨[J]. 中国岩溶, 2017,36(5):641-647.
	[ Zhao C H, Liang Y P, Wang X T, et al. Discussion on the characteristics of karst water systems in the Xi Shan area of Beijing[J]. Carsologica Sinica, 2017,36(5):641-647. ]
[14]	秦大军, 孙杰, 郭艺, 等. 永定河对北京西山岩溶水和玉泉山泉的影响[J]. 工程地质学报, 2019,27(1):162-169.
	[ Qin D J, Sun J, Guo Y, et al. Impacts of Yongding River on the Xishan karst aquifer and Yuquan spring in Beijing, China[J]. Journal of Engineering Geology, 2019,27(1):162-169. ]
[15]	唐春雷, 梁永平, 韩凯, 等. 玉泉山泉九龙山—香峪向斜的水文地质意义[J]. 中国岩溶, 2015,34(5):431-437.
	[ Tang C L, Liang Y P, Han K, et al. Hydrogeological significance of the Jiulongshan-Xianyu syncline at Yuquanshan sping[J]. Carsologica Sinica, 2015,34(5):431-437. ]
[16]	刘莉莉, 宋苏林, 崔春梅. 济南泉水的成因及保泉对策研究[J]. 山东水利, 2013(5):17-18.
	[ Liu L L, Song S L, Cui C M. Study on the causes of Jinan spring water and its countermeasures[J]. Shandong Water Resources, 2013(5):17-18. ]
[17]	张彩霞, 杨勤科, 李锐. 基于DEM的地形湿度指数及其应用研究进展[J]. 地理科学进展, 2005,24(6):116-123.
	[ Zhang C X, Yang Q K, Li R. Advancement in Topographic Wetness Index and Its Application[J]. Progress in Geography, 2005,24(6):116-123. ]
[18]	美国航空航天局喷气推进实验室（JPL）和日本经济产业省（METI）数据[DB/OL]. https://lpdaac.usgs.gov/products/aster_products_table/astgtm/.
	[ NASA Jet Propulsion Lab (JPL) and Japan's Ministry of Economy Trade and Industry (METI)[DB/OL]. https://lpdaac.usgs.gov/products/aster_products_table/astgtm/. ]
[19]	中国地质调查局[DB/OL]. http://geocloud.cgs.gov.cn/.

[20]	地理空间数据云[DB/OL]. http://www.gscloud.cn/.
	[ Geospatial Data Cloud[DB/OL]. http://www.gscloud.cn/. ]
[21]	Cheng A Q. Conditional independence test for weights-of-evidence modeling[J]. Natural Resources Research, 2002.
[22]	Corsini A, Cervi F, Ronchetti F. Weight of evidence and artificial neural networks for potential groundwater spring mapping: An application to the Mt. Modino area (Northern Apennines, Italy)[J]. Geomorphology, 2009,111(1-2):79-87. doi: 10.1016/j.geomorph.2008.03.015
[23]	尚敏, 马锐, 张英莹, 等. 基于GIS的证据权重法的崩塌敏感性分析研究[J]. 工程地质学报, 2018,26(5):1211-1218.
	[ Shang M, Ma R, Zhang Y Y, et al. GIS based weights of evidence method for rock fall susceptibility[J]. Journal of Engineering Geology, 2018,26(5):1211-1218. ]
[24]	张松林. CART-分类与回归树方法介绍[J]. 火山地质与矿产, 1997,18(1):67-75.
	[ Zhang S L. An introduction to the methodology of CART-Classifiction and Regression Trees[J]. Volcanology & Mineral Resources, 1997,18(1):67-75. ]
[25]	方匡南, 吴见彬, 朱建平, 等. 随机森林方法研究综述[J]. 统计与信息论坛, 2011,26(3):32-38.
	[ Fang K N, Wu J B, Zhu J P, et al. A Review of Technologies on Random Forests[J]. Statistics & Information Forum, 2011,26(3):32-37. ]
[26]	周超, 方秀琴, 吴小君, 等. 基于三种机器学习算法的山洪灾害风险评价[J]. 地球信息科学学报, 2019,21(11):1679-1688. doi: 10.12082/dqxxkx.2019.190185
	[ Zhou C, Fang X Q, Wu X J, et al. Risk assessment of mountain torrents based on three machine learning algorithms[J]. Journal of Geo-information Science, 2019,21(11):1679-1688. ]
[27]	赖成光, 陈晓宏, 赵仕威, 等. 基于随机森林的洪灾风险评价模型及其应用[J]. 水利学报, 2015,46(1):58-66.
	[ Lai C G, Chen X H, Zhao S W, et al. A flood risk assessment model based on Random Forest and its application[J]. ournal of Hydraulic Engineering, 2015,46(1):58-66. ]
[28]	Pourghasemi, Reza H, Rahmati, et al. Application of GIS-based data driven random forest and maximum entropy models for groundwater potential mapping: A case study at Mehran Region, Iran[J]. Catena: An Interdisciplinary Journal of Soil Science Hydrology-Geomorphology Focusing on Geoecology and Landscape Evolution, 2016.
[29]	Jenks G F, Coulson M R. Class intervals for itatistical maps[J]. International Yearbook of Cartography, 1963: 119-134.
[30]	李乃强, 徐贵阳. 基于自然间断点分级法的土地利用数据网格化分析[J]. 测绘通报, 2020(4):106-110.
	[ Li N Q, Xu G Y. Grid analysis of land use based on natural breaks (jenks) classification[J]. Bulletin of Surveying and Mapping, 2020(4):106-110. ]
[31]	王运生, 谢丙炎, 万方浩, 肖启明, 戴良英. ROC曲线分析在评价入侵物种分布模型中的应用[J]. 生物多样性, 2007,15(4):365-372. doi: 10.1360/biodiv.060280
	[ Wang Y S, Xie B Y, Wan F H, et al. Application of ROC curve analysis in evaluating the performance of alien species' potential distribution models[J]. Biodiversity Science, 2007,15(4):365-372. ]
[32]	Leshowitz, Barry. Comparison of ROC curves from one- and two-interval rating-scale procedures[J]. Journal of the Acoustical Society of America, 1969,46(2B):399. doi: 10.1121/1.1911703
[33]	章旭, 郝红兵, 刘康林, 等. 西藏加查象牙泉水文地球化学特征及成因[J]. 水文地质工程地质, 2019,46(4):1-9.
	[ Zhang X, Hao H B, Liu K L, et al. Hydrogeochemical characteristics and formation of the Ivory Spring in Jiacha County of Tibet[J]. Hydrogeology & Engineering Geology, 2019,46(4):1-9. ]
[34]	张婷婷, 侯利朋, 王万平. 青海玛沁野马滩构造岩溶泉成因及开发利用分析[J]. 青海环境, 2016,26(3):119-123.
	[ Zhang T T, Hou L P, Wang W P. Analysis on genesis and development of karst spring in yematan structure of Maqin, Qinghai Province[J]. Qinghai Huanjing, 2016,26(3):119-123. ]
[35]	韩东银, 常元存. 山西孟县张城堡泉成因研究[J]. 大地构造与成矿学, 1992(1):85-86.
	[ Han D Y, Chang Y C. Study on origin of Zhang Baocheng spring in Meng County, Shan Xi Province[J]. Geotectonica et Metallogenia, 1992,16(1):85-86. ]

影响因子	VIF	TOL
高程	1.305	0.766
坡度	1.410	0.709
坡向	1.021	0.979
距断裂距离	1.349	0.741
距河流距离	1.078	0.927
地形湿度指数	1.461	0.684
径流强度指数	1.375	0.727
土地利用	1.349	0.741
归一化植被指数	1.023	0.977
岩性	1.403	0.713

等级	证据权重模型
等级	范围	面积/%	训练泉点	验证泉点
低	0.000 197 8~0.000 500 9	42.01	40	16
中等	0.000 500 9~ 0.000 731 4	53.46	671	301
高	0.000 731 4 ~0.001 286 3	4.53	254	93

等级	分类回归树				随机森林
等级	范围	面积/%	训练泉点	验证泉点	范围	面积/%	训练泉点	验证泉点
低	0.00~0.23	39.76	10	5	0.00~0.23	41.95	2	10
中等	0.23~0.55	46.16	416	200	0.23~0.58	38.97	218	169
高	0.55~1.00	14.08	539	205	0.58~1.00	19.08	745	231