Modeling of hydrological processes in cryospheric watersheds based on machine learning

doi:10.13745/j.esf.sf.2023.2.52

Abstract

Abstract:

Machine learning models are widely used in hydrological research for their high predictive accuracy, however, their application in high-altitude cryospheric watersheds is seldom mentioned. In this study, machine learning models for two typical cryospheres, Yarkant and Shule river basins, were developed using BP neural network (BP), GRNN neural network (GRNN), RBF neural network (RBF), support vector regression (SVR), genetic optimization BP neural network (GA-BP) and double-layer long-term and short-term memory neural network (LSTM) algorithms, and model performance was evaluated using evaluation indexes NSE, RMSE and R and runoff frequency curves. The double-layer LSTM model performed much better than and similar to other models for the Yarkant and Shule River Basins, respectively; and overall the double-layer LSTM algorithm was more suitable for modeling hydrological processes in cryosphere basins. The loss function was used to evaluate the model parameterization scheme. It was found that the performance of the LSTM models was mainly affected by the optimizer, followed by the learning attenuation rate and initial learning rate for the Yarkant River Basin, and by the initial learning rate for the Shule River Basin. Model testing under abrupt changes in runoff suggested that climatic factors could have different hydrological impacts, meanwhile, precipitation and heavy precipitation R95p had the greatest impacts on the hydrological process in the study area, followed by temperature, during the entire study period.

Key words: climate change, machine learning, cryospheric watersheds, hydrological process simulation, parameterization

CLC Number:

P338

SONG Xuanyu, XU Min, KANG Shichang, SUN Liping. Modeling of hydrological processes in cryospheric watersheds based on machine learning[J]. Earth Science Frontiers, 2023, 30(4): 451-469.

Figures/Tables 17

Fig.1 Location (left) and geological sketch maps of the Yarkant (top right) and Shule (bottom right) river basins

Table 1 Information on hydrometeorological stations

站名	经度	纬度	海拔/m	时间段
莎车	75°23'E	37°77'N	1 231 m	1954-2015年
托勒	98°25'E	38°48'N	3 368 m	1970-2006年
卡群^*	76°90'E	37°98'N	1 370 m	1954-2015年
昌马堡^*	96°51'E	39°49'N	2 080 m	1970-2006年

Fig.2 Flow chart of hydrological modeling based on machine learning

Fig.3 Diagram of a LSTM unit

Fig.4 Annual temperature, precipitation, R95p and runoff variations in the Yarkant (a) and Shule (b) river basins. Meteorological data from [38].

Fig.5 Monthly temperature, precipitation, R95p and runoff in the Yarkant (a) and Shule (b) river basins. Meteorological data from [38].

Fig.6 Correlation relationships between monthly temperature, precipitation, R95p and runoff in the study area

Fig.7 Hydrological modeling results for the study area using 6 machine learning algorithms

Fig.8 Taylor diagrams to compare 6 machine learning algorithms in hydrological modeling of the Yarkant River Basin during training (a) and model testing (b)

Fig.9 Taylor diagrams to compare 6 machine learning algorithms in hydrological modeling of the Shule River Basin during training (a) and model testing (b)

Table 2 Accuracy evaluation of machine learning models

方法	叶尔羌河流域						疏勒河流域
	训练期			验证期			训练期			验证期
	NSE	RMSE/mm	R	NSE	RMSE/mm	R	NSE	RMSE/mm	R	NSE	RMSE/mm	R
BP	0.71	7.19	0.84	0.69	7.92	0.83	0.82	2.85	0.91	0.81	4.28	0.92
GRNN	0.71	7.18	0.84	0.58	9.13	0.76	0.74	3.41	0.89	0.72	5.19	0.90
RBF	0.71	7.15	0.84	0.67	8.18	0.82	0.85	2.60	0.92	0.79	4.43	0.91
SVR	0.70	7.35	0.84	0.65	8.41	0.81	0.85	2.59	0.92	0.78	4.62	0.92
GA-BP	0.71	7.16	0.84	0.69	7.85	0.83	0.84	2.66	0.92	0.82	4.18	0.93
LSTM	0.90	4.24	0.95	0.88	4.80	0.94	0.85	2.57	0.92	0.72	5.17	0.91

Fig.10 Comparison of average monthly runoffs between the measured values and machine learning results for the Yarkant (a) and Shule (b) river basins

Fig.11 Runoff frequency curves using different machine learning algorithms for the Yarkant (a, b) and Shule (c,d) river basins

Fig.12 Plots of loss function vs. iterations under different parameterization schemes in training and testing runs. (a-f) Yarkant River Basin; (g-l) Shule River Basin.

Table 3 Parameterization schemes in LSTM models

参数	叶尔羌河流域	疏勒河流域
初始学习率	0.005	0.005
学习递减速率	0.9	0.9
优化器	Adam	Adam
批大小	12	12
迭代次数	25	25
LSTM层数	2	1
第一隐含层单元数目	100	100
第二隐含层单元数目	200

Fig.13 Model testing under abrupt change in mean annual runoff in the study area

Table 4 Percent contributions of temperature, precipitation and heavy precipitation R95p to runoff changes

参数	叶尔羌河流域						疏勒河流域
	1954-2015年		1954-1998年		1999-2015年		1970-2006年		1970-1998年		1999-2006年
	SR	MLR	SR	MLR	SR	MLR	SR	MLR	SR	MLR	SR	MLR
t	1%	2%	2%	2%	80%	80%	17%	11%	6%	6%	11%	11%
p	49%	49%	50%	50%	19%	19%	66%	66%	54%	54%	44%	44%
R95p	50%	49%	48%	48%	1%	1%	17%	23%	40%	40%	45%	45%

References 72

[1]	秦大河. 冰冻圈科学概论[M]. 北京: 科学出版社, 2017.
[2]	丁永建, 张世强, 陈仁升. 冰冻圈水文学:解密地球最大淡水库[J]. 中国科学院院刊, 2020, 35(4): 414-424.
[3]	丁永建, 秦大河. 冰冻圈变化与全球变暖: 我国面临的影响与挑战[J]. 中国基础科学, 2009, 11(3): 4-10.
[4]	何秋乐. 长江源冬克玛底河流域冰川水文过程影响与模拟[D]. 北京: 中国地质大学(北京), 2019.
[5]	丁永建, 张世强, 吴锦奎, 等. 中国冰冻圈水文过程变化研究新进展[J]. 水科学进展, 2020, 31(5): 690-702.
[6]	阳勇, 陈仁升. 冻土水文研究进展[J]. 地球科学进展, 2011, 26(7): 711-723.
[7]	黄克威, 王根绪, 宋春林, 等. 基于LSTM的青藏高原冻土区典型小流域径流模拟及预测[J]. 冰川冻土, 2021, 43(4): 1144-1156. DOI
[8]	刘虎, 王磊. 第三极地区冰川径流研究进展[J]. 冰川冻土, 2022, 44(3): 737-752. DOI
[9]	汤秋鸿, 刘星才, 周园园, 等. “亚洲水塔”变化对下游水资源的连锁效应[J]. 中国科学院院刊, 2019, 34(11): 1306-1312.
[10]	BIEMANS H, SIDERIUS C, LUTZ A F, et al. Importance of snow and glacier meltwater for agriculture on the Indo-Gangetic Plain[J]. Nature Sustainability, 2019, 2(7): 594-601. DOI
[11]	张勇, 刘时银. 度日模型在冰川与积雪研究中的应用进展[J]. 冰川冻土, 2006, 28(1): 101-107.
[12]	高鑫, 叶柏生, 张世强, 等. 1961-2006年塔里木河流域冰川融水变化及其对径流的影响[J]. 中国科学: 地球科学, 2010, 40(5): 654-665.
[13]	高红凯, 何晓波, 叶柏生, 等. 1955-2008年冬克玛底河流域冰川径流模拟研究[J]. 冰川冻土, 2011, 33(1): 171-181.
[14]	WANG X Y, YANG T, XU C Y, et al. Understanding the discharge regime of a glacierized alpine catchment in the Tianshan Mountains using an improved HBV-D hydrological model[J]. Global and Planetary Change, 2019, 172: 211-222. DOI URL
[15]	关志成. 寒区流域水文模拟研究[D]. 南京: 河海大学, 2002.
[16]	王中根, 刘昌明, 黄友波. SWAT模型的原理、结构及应用研究[J]. 地理科学进展, 2003, 22(1): 79-86.
[17]	WANG L, ZHOU J, QI J, et al. Development of a land surface model with coupled snow and frozen soil physics[J]. Water Resources Research, 2017, 53(6): 5085-5103. DOI URL
[18]	GAO B, YANG D W, QIN Y, et al. Change in frozen soils and its effect on regional hydrology, upper Heihe Basin, northeastern Qinghai-Tibetan Plateau[J]. The Cryosphere, 2018, 12(2): 657-673. DOI URL
[19]	陈仁升, 吕世华, 康尔泗, 等. 内陆河高寒山区流域分布式水热耦合模型(Ⅰ): 模型原理[J]. 地球科学进展, 2006, 21(8): 806-818. DOI
[20]	陈仁升, 康尔泗, 吕世华, 等. 内陆河高寒山区流域分布式水热耦合模型(Ⅱ): 地面资料驱动结果[J]. 地球科学进展, 2006, 21(8): 819-829. DOI
[21]	陈仁升, 高艳红, 康尔泗, 等. 内陆河高寒山区流域分布式水热耦合模型(Ⅲ): MM5嵌套结果[J]. 地球科学进展, 2006, 21(8): 830-837. DOI
[22]	CHEN R, WANG G, YANG Y, et al. Effects of cryospheric change on alpine hydrology: combining a model with observations in the upper reaches of the Hei River, China[J]. Journal of Geophysical Research: Atmospheres, 2018, 123(7): 3414-3442. DOI URL
[23]	赵求东. WRF+DHSVM融雪径流预报模式研究[D]. 乌鲁木齐: 新疆大学, 2008.
[24]	KAN B Y, SU F G, XU B Q, et al. Generation of high mountain precipitation and temperature data for a quantitative assessment of flow regime in the upper Yarkant Basin in the Karakoram[J]. Journal of Geophysical Research: Atmospheres, 2018, 123(16): 8462-8486. DOI URL
[25]	陈仁升, 康尔泗, 杨建平, 等. 水文模型研究综述[J]. 中国沙漠, 2003, 23(3): 221-229.
[26]	ZOUNEMAT-KERMANI M, BATELAAN O, FADAEE M, et al. Ensemble machine learning paradigms in hydrology: a review[J]. Journal of Hydrology, 2021, 598: 126266. DOI URL
[27]	CANNON A J, WHITFIELD P H. Downscaling recent stream flow conditions in British Columbia, Canada using ensemble neural network models[J]. Journal of Hydrology, 2002, 259(1/2/3/4): 136-151. DOI URL
[28]	LI P H, KWON H H, SUN L Q, et al. A modified support vector machine based prediction model on streamflow at the Shihmen Reservoir, Taiwan[J]. International Journal of Climatology, 2010, 30(8): 1256-1268. DOI URL
[29]	CHEN N C, XIONG C, DU W Y, et al. An improved genetic algorithm coupling a back-propagation neural network model (IGA-BPNN) for water-level predictions[J]. Water, 2019, 11(9): 1795. DOI URL
[30]	YU X, WANG Y H, WU L F, et al. Comparison of support vector regression and extreme gradient boosting for decomposition-based data-driven 10-day streamflow forecasting[J]. Journal of Hydrology, 2020, 582: 124293. DOI URL
[31]	SIHAG P, AL-JANABI A M S, ALOMARI N K, et al. Evaluation of tree regression analysis for estimation of river basin discharge[J]. Modeling Earth Systems and Environment, 2021, 7(4): 2531-2543. DOI
[32]	NI L L, WANG D, SINGH V P, et al. Streamflow and rainfall forecasting by two long short-term memory-based models[J]. Journal of Hydrology, 2020, 583: 124296. DOI URL
[33]	冯童, 刘时银, 许君利, 等. 1968-2009年叶尔羌河流域冰川变化: 基于第一、二次中国冰川编目数据[J]. 冰川冻土, 2015, 37(1): 1-13.
[34]	王鹏赫, 赵成义, 王丹丹, 等. 气候变化对叶尔羌河流域极端水文事件的影响[J]. 生态科学, 2018, 37(6): 1-8.
[35]	盛煜, 李静, 吴吉春, 等. 基于GIS的疏勒河流域上游多年冻土分布特征[J]. 中国矿业大学学报, 2010, 39(1): 32-39.
[36]	NIU L, YE B S, LI J, et al. Effect of permafrost degradation on hydrological processes in typical basins with various permafrost coverage in Western China[J]. Science China Earth Sciences, 2011, 54(4): 615-624. DOI URL
[37]	周嘉欣, 丁永建, 曾国雄, 等. 疏勒河上游地表水水化学主离子特征及其控制因素[J]. 环境科学, 2014, 35(9): 3315-3324.
[38]	中国气象局. 中国地面气候资料数据集(V3.0)[DS]. [2022-2-05]. http://data.cma.cn/site/index.html
[39]	RUMELHART D E, HINTON G E, WILLIAMS R J. Learning representations by back-propagating errors[J]. Nature, 1986, 323(6088): 533-536. DOI
[40]	LI J, CHENG J H, SHI J Y, et al. Brief introduction of back propagation (BP) neural network algorithm and its improvement[C]// JIN D, LIN S. Advances in computer science and information engineering. Berlin, Heidelberg: Springer, 2012: 553-558.
[41]	SPECHT D F. A general regression neural network[J]. IEEE Transactions on Neural Networks, 1991, 2(6): 568-576. PMID
[42]	LI Q, MENG Q L, CAI J J, et al. Predicting hourly cooling load in the building: a comparison of support vector machine and different artificial neural networks[J]. Energy Conversion and Management, 2009, 50(1): 90-96. DOI URL
[43]	LENG Z W, GAO J W, QIN Y, et al. Short-term forecasting model of traffic flow based on GRNN[C]// 2013 25th Chinese control and decision conference (CCDC). Guiyang, China: IEEE, 2013: 3816-3820.
[44]	HUANG G B, SIEW C K. Extreme learning machine: RBF network case[C]// ICARCV 2004 8th Control, automation, robotics and vision conference. Kunming, China: IEEE, 2005: 1029-1036.
[45]	AZAR A T, EL-SAID S A. Probabilistic neural network for breast cancer classification[J]. Neural Computing and Applications, 2013, 23(6): 1737-1751. DOI URL
[46]	AWAD M, KHANNA R. Support vector regression[M]// Efficient learning machines. Berkeley, CA: Apress, 2015: 67-80.
[47]	DING S F, SU C Y, YU J Z. An optimizing BP neural network algorithm based on genetic algorithm[J]. Artificial Intelligence Review, 2011, 36(2): 153-162. DOI URL
[48]	HOCHREITER S, SCHMIDHUBER J. Long short-term memory[J]. Neural Computation, 1997, 9(8): 1735-1780. DOI PMID
[49]	GERS F A, SCHMIDHUBER J, CUMMINS F. Learning to forget: continual prediction with LSTM[J]. Neural Computation, 2000, 12(10): 2451-2471. PMID
[50]	山世光, 阚美娜, 刘昕, 等. 深度学习: 多层神经网络的复兴与变革[J]. 科技导报, 2016, 34(14): 60-70.
[51]	YU Y, SI X S, HU C H, et al. A review of recurrent neural networks: LSTM cells and network architectures[J]. Neural Computation, 2019, 31(7): 1235-1270. DOI PMID
[52]	LIU Y Q, HOU G B, HUANG F, et al. Directed graph deep neural network for multi-step daily streamflow forecasting[J]. Journal of Hydrology, 2022, 607: 127515. DOI URL
[53]	BARRON J T. A general and adaptive robust loss function[C]// 2019 IEEE/CVF conference on computer vision and pattern recognition (CVPR). Long Beach, CA, USA: IEEE, 2020: 4326-4334.
[54]	RUDER S. An overview of gradient descent optimization algorithms[EB/OL]. 2016: arXiv: 1609.04747. https://arxiv.org/abs/1609.04747.
[55]	何晓群. 应用回归分析(R语言版)[M]. 北京: 电子工业出版社, 2017.
[56]	GILES D E A. The interpretation of dummy variables in semilogarithmic equations[J]. Economics Letters, 1982, 10(1/2): 77-79. DOI URL
[57]	XU J H, LI W H, JI M H, et al. A comprehensive approach to characterization of the nonlinearity of runoff in the headwaters of the Tarim River, Western China[J]. Hydrological Processes, 2009, 13: 136-146.
[58]	杨春利, 蓝永超, 王宁练, 等. 1958-2015年疏勒河上游出山径流变化及其气候因素分析[J]. 地理科学, 2017, 37(12): 1894-1899. DOI
[59]	蓝永超, 胡兴林, 肖生春, 等. 近50年疏勒河流域山区的气候变化及其对出山径流的影响[J]. 高原气象, 2012, 31(6): 1636-1644.
[60]	程玉菲, 程文举, 胡想全, 等. 疏勒河流域极端水文事件对极端气候的响应[J]. 高原气象, 2019, 38(3): 583-592. DOI
[61]	CHEN Y N, LI W H, FANG G H, et al. Review article: hydrological modeling in glacierized catchments of central Asia: status and challenges[J]. Hydrology and Earth System Sciences, 2017, 21(2): 669-684. DOI URL
[62]	KUMAR R, SINGH S, KUMAR R, et al. Development of a glacio-hydrological model for discharge and mass balance reconstruction[J]. Water Resources Management, 2016, 30(10): 3475-3492. DOI URL
[63]	孙美平, 张磊, 姚晓军, 等. 1954-2016年疏勒河上游径流变化特征及影响因素[J]. 冰川冻土, 2022, 44(2): 657-666. DOI
[64]	陈亚宁, 杨青, 罗毅, 等. 西北干旱区水资源问题研究思考[J]. 干旱区地理, 2012, 35(1): 1-9.
[65]	高鑫, 张世强, 叶柏生, 等. 1961-2006年叶尔羌河上游流域冰川融水变化及其对径流的影响[J]. 冰川冻土, 2010, 32(3): 445-453.
[66]	沈永平, 丁永建, 刘时银, 等. 近期气温变暖叶尔羌河冰湖溃决洪水增加[J]. 冰川冻土, 2004, 26(2): 234.
[67]	HAEMMIG C, HUSS M, KEUSEN H, et al. Hazard assessment of glacial lake outburst floods from Kyagar glacier, Karakoram Mountains, China[J]. Annals of Glaciology, 2014, 55(66): 34-44.
[68]	纪守领, 李进锋, 杜天宇, 等. 机器学习模型可解释性方法、应用与安全研究综述[J]. 计算机研究与发展, 2019, 56(10): 2071-2096.
[69]	FAN H X, JIANG M L, XU L G, et al. Comparison of long short term memory networks and the hydrological model in runoff simulation[J]. Water, 2020, 12(1): 175. DOI URL
[70]	YIN Z L, FENG Q, LIU S Y, et al. The spatial and temporal contribution of glacier runoff to watershed discharge in the Yarkant River Basin, northwest China[J]. Water, 2017, 9(3): 159. DOI URL
[71]	赵军, 黄永生, 宋阁庆, 等. SRM融雪径流模型在疏勒河流域上游的应用[J]. 水资源与水工程学报, 2015, 26(1): 72-76, 80.
[72]	邹悦. 基于SWAT模型的疏勒河中游径流模拟研究[D]. 兰州: 西北师范大学, 2012.