Adaptability of machine learning methods and hydrological models to discharge simulations in data-sparse glaciated watersheds

doi:10.1007/s40333-021-0066-5

Journal of Arid Land

2021, Vol. 13

Issue (6): 549-567 DOI: 10.1007/s40333-021-0066-5

Research article

Adaptability of machine learning methods and hydrological models to discharge simulations in data-sparse glaciated watersheds

JI Huiping¹^,², CHEN Yaning¹^,²^,^*(

), FANG Gonghuan¹^,², LI Zhi¹^,², DUAN Weili¹^,², ZHANG Qifei¹^,²

¹State Key Laboratory of Desert and Oasis Ecology, Xinjiang Institute of Ecology and Geography, Chinese Academy of Sciences, Urumqi 830011, China
²University of Chinese Academy of Sciences, Beijing 100049, China

Download:

HTML

PDF(1487KB)
Export: BibTeX | EndNote (RIS)

Abstract

The accurate simulation and prediction of runoff in alpine glaciated watersheds is of increasing importance for the comprehensive management and utilization of water resources. In this study, long short-term memory (LSTM), a state-of-the-art artificial neural network algorithm, is applied to simulate the daily discharge of two data-sparse glaciated watersheds in the Tianshan Mountains in Central Asia. Two other classic machine learning methods, namely extreme gradient boosting (XGBoost) and support vector regression (SVR), along with a distributed hydrological model (Soil and Water Assessment Tool (SWAT) and an extended SWAT model (SWAT_Glacier) are also employed for comparison. This paper aims to provide an efficient and reliable method for simulating discharge in glaciated alpine regions that have insufficient observed meteorological data. The two typical basins in this study are the main tributaries (the Kumaric and Toxkan rivers) of the Aksu River in the south Tianshan Mountains, which are dominated by snow and glacier meltwater and precipitation. Our comparative analysis indicates that simulations from the LSTM shows the best agreement with the observations. The performance metrics Nash-Sutcliffe efficiency coefficient (NS) and correlation coefficient (R²) of LSTM are higher than 0.90 in both the training and testing periods in the Kumaric River Basin, and NS and R ² are also higher than 0.70 in the Toxkan River Basin. Compared to classic machine learning algorithms, LSTM shows significant advantages over most evaluating indices. XGBoost also has high NS value in the training period, but is prone to overfitting the discharge. Compared with the widely used hydrological models, LSTM has advantages in predicting accuracy, despite having fewer data inputs. Moreover, LSTM only requires meteorological data rather than physical characteristics of underlying data. As an extension of SWAT, the SWAT_Glacier model shows good adaptability in discharge simulation, outperforming the original SWAT model, but at the cost of increasing the complexity of the model. Compared with the oftentimes complex semi-distributed physical hydrological models, the LSTM method not only eliminates the tedious calibration process of hydrological parameters, but also significantly reduces the calculation time and costs. Overall, LSTM shows immense promise in dealing with scarce meteorological data in glaciated catchments.

Key words： hydrological simulation long short-term memory extreme gradient boosting support vector regression SWAT_Glacier model Tianshan Mountains

Received: 07 April 2021 Published: 10 June 2021

Corresponding Authors:

About author: CHEN Yaning (E-mail: chenyn@ms.xjb.ac.cn)

	Service
	E-mail this article
	Add to my bookshelf
	Add to citation manager
	E-mail Alert
	RSS
	Articles by authors
	Huiping JI
	Yaning CHEN
	Gonghuan FANG
	Zhi LI
	Weili DUAN
	Qifei ZHANG

Cite this article:

JI Huiping, CHEN Yaning, FANG Gonghuan, LI Zhi, DUAN Weili, ZHANG Qifei. Adaptability of machine learning methods and hydrological models to discharge simulations in data-sparse glaciated watersheds. Journal of Arid Land, 2021, 13(6): 549-567.

URL:

http://jal.xjegi.com/10.1007/s40333-021-0066-5 OR http://jal.xjegi.com/Y2021/V13/I6/549

Fig. 1 Location of Kumaric and Toxkan river basins as well as distributions of hydrological stations, meteorological stations, and glacier distribution in the Tianshan Mountains

Fig. 2 Basic structure of long short-term memory (LSTM) and detailed calculation process at time step t(a), structural diagram of extreme gradient boosting (XGBoost; b), and diagram of support vector regression (SVR; c). The yellow highlighted area indicates the interval band of 2ϵ, where ϵ means the absolute deviation value.C_t, value of cell state at time step t; C_t_-1, value of cell state at time step t-1; X_t, current input; H_t, output result of hidden layer at time step t; H_t_-1, hidden layer output result of the previous time t-1; tanh, hyperbolic tangent function; σ, sigmoid activation function; T₁, the 1^st tree; T_n, the n^th tree; f(x), the function of the model.

Table 1 Evaluation of the LSTM, XGBoost, SVR, SWAT_Glacier and SWAT models in daily discharge simulation during the training period (2002-2007) and testing period (2008-2011) in the Kumaric and Toxkan river basins

Fig. 3 Observed and simulated discharges of the five models for the Kumaric River Basin during the training period. (a), daily observations and simulations of the machine learning methods (LSTM, XGBoost, and SVR); (b), simulation performance of the traditional hydrological models (Soil and Water Assessment Tool (SWAT) and SWAT_Glacier (an extended SWAT model with glacier melting mechanism)); (c)-(g), scatterplots of observed and simulated values for each model, withR² representing the correlation coefficient.

Fig. 4 Observed and simulated discharges of the five models in the Toxkan River Basin during the training period. (a), daily observations and simulations of the machine learning methods (LSTM, XGBoost, and SVR); (b), simulation performance of traditional hydrological models (SWAT and SWAT_Glacier); (c)-(g), scatterplots of observed and simulated values for each model, with R² representing the correlation coefficient.

Fig. 5 Predictions of the three machine learning algorithms (LSTM, XGBoost, and SVR) and two hydrological models (SWAT_Glacier and SWAT) for discharge in the Kumaric River Basin (a and b) and Toxkan River Basin (c and d) during the testing period

Fig. 6 Distribution of relative errors generated by machine learning algorithms (LSTM, XGBoost, and SVR) and hydrological models (SWAT_Glacier and SWAT) on a monthly scale in the Kumaric River Basin (a and b) and Toxkan River Basin (c and d). The box and whisker plots show the five-number summary of a set of data: the minimum score, first (lower) quartile, median, third (upper) quartile, and the maximum score. The center represents the middle 50%, or 50^th percentile of the data set, and is derived using the lower and upper quartile values. The median value is displayed inside the "box". The maximum and minimum values are displayed with vertical lines ("whiskers") connecting the points to the center box. The circles represent the outliers. The box and whisker plots have the same meaning as Figure 8.

Fig. 7 Relationship between basin area and altitude in the Toxkan River Basin

Fig. 8 Boxplot of precipitation, maximum temperature, and minimum temperature at meteorological observation stations in the Kumaric River Basin (a, c, and e) and Toxkan River Basin (b, d, and f)


[1]	Aggarwal S K, Goel A, Singh V P. 2012. Stage and discharge forecasting by SVM and ANN techniques. Water Resources Management, 26:3705-3724. doi: 10.1007/s11269-012-0098-x

[2]	Ajami N, Gupta H, Wagener T, et al. 2004. Calibration of a semi-distributed hydrologic model for streamflow estimation along a river system. Journal of Hydrology, 298:112-135. doi: 10.1016/j.jhydrol.2004.03.033

[3]	Arnold J G, Srinivasan R, Muttiah R S, et al. 1998. Large area hydrologic modeling and assessment—Part 1: Model development. JAWRA Journal of the American Water Resources Association, 34:73-89. doi: 10.1111/jawr.1998.34.issue-1

[4]	Bengio Y, Simard P, Frasconi P. 1994. Learning Long-Term dependencies with gradient descent is difficult. IEEE Transactions on Neural Networks, 5:157-166. pmid: 18267787

[5]	Chen T Q, Guestrin C. 2016. XGBoost: A scalable tree boosting system. In: Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. San Francisco: Special Interest Group on Management of Data, 785-794.

[6]	Chen X, Long D, Hong Y, et al. 2017. Improved modeling of snow and glacier melting by a progressive two-stage calibration strategy with GRACE and multisource data: How snow and glacier meltwater contributes to the runoff of the Upper Brahmaputra River basin? Water Resources Research, 53:2431-2466. doi: 10.1002/2016WR019656

[7]	Chen Y N, Takeuchi K, Xu C C, et al. 2006. Regional climate change and its effects on river runoff in the Tarim Basin, China. Hydrological Processes, 20:2207-2216. doi: 10.1002/(ISSN)1099-1085

[8]	Chen Y N, Li W H, Deng H J, et al. 2016. Changes in Central Asia's water tower: past, present and future. Scientific Reports, 6:35458, doi: 10.1038/srep35458. doi: 10.1038/srep35458

[9]	Chen Y N, Li W H, Fang G H, et al. 2017. Review article: Hydrological modeling in glacierized catchments of Central Asia - status and challenges. Hydrology and Earth System Sciences, 21:669-684. doi: 10.5194/hess-21-669-2017

[10]	Cheng C T, Feng Z K, Niu W J, et al. 2015. Heuristic methods for reservoir monthly inflow forecasting: A case study of Xinfengjiang Reservoir in Pearl River, China. Water, 7:4477-4495. doi: 10.3390/w7084477

[11]	Duethmann D, Bolch T, Farinotti D, et al. 2015. Attribution of streamflow trends in snow and glacier melt-dominated catchments of the Tarim River, Central Asia. Water Resources Research, 51:4727-4750. doi: 10.1002/2014WR016716

[12]	Duethmann D, Menz C, Jiang T, et al. 2016. Projections for headwater catchments of the Tarim River reveal glacier retreat and decreasing surface water availability but uncertainties are large. Environmental Research Letters, 11:054024, doi: 10.1088/1748-9326/11/5/054024. doi: 10.1088/1748-9326/11/5/054024

[13]	Fan Y T, Chen Y N, Li W H. 2014. Increasing precipitation and baseflow in Aksu River since the 1950s. Quaternary International, 336:26-34. doi: 10.1016/j.quaint.2013.09.037

[14]	Fang K, Shen C P, Kifer D, et al. 2017. Prolongation of SMAP to spatiotemporally seamless coverage of continental U.S. using a deep learning neural network. Geophysical Research Letters, 44, doi: 10.1002/2017gl075619. doi: 10.1002/2017gl075619

[15]	Fang K, Shen C P. 2019. Near-real-time forecast of satellite-based soil moisture using long short-term memory with an adaptive data integration kernel. Journal of Hydrometeorology, 21:399-413. doi: 10.1175/JHM-D-19-0169.1

[16]	Feng D P, Fang K, Shen C P. 2020. Enhancing streamflow forecast and extracting insights using Long-Short Term Memory networks with data integration at continental scales. Water Resources Research, 56:e2019WR026793, doi: 10.1029/2019WR026793. doi: 10.1029/2019WR026793

[17]	Finger D, Vis M, Huss M, et al. 2015. The value of multiple data set calibration versus model complexity for improving the performance of hydrological models in mountain catchments. Water Resources Research, 51:1939-1958. doi: 10.1002/wrcr.v51.4

[18]	Fu M L, Fan T C, Ding Z A, et al. 2020. Deep learning data-intelligence model based on adjusted forecasting window scale: Application in daily streamflow simulation. IEEE Access, 8:32632-32651. doi: 10.1109/Access.6287639

[19]	Gao S, Huang Y F, Zhang S, et al. 2020. Short-term runoff prediction with GRU and LSTM networks without requiring time step optimization during sample generation. Journal of Hydrology, 589:125188, doi: 10.1016/j.jhydrol.2020.125188. doi: 10.1016/j.jhydrol.2020.125188

[20]	Greff K, Srivastava R K, Koutník J, et al. 2017. LSTM: A search space odyssey. IEEE Transactions on Neural Networks and Learning Systems, 28:2222-2232. doi: 10.1109/TNNLS.2016.2582924

[21]	Greve P, Burek P, Wada Y. 2020. Using the Budyko framework for calibrating a global hydrological model. Water Resources Research, 56:e2019WR026280, doi: 10.1029/2019WR026280. doi: 10.1029/2019WR026280

[22]	Hochreiter S, Schmidhuber J. 1997. Long short-term memory. Neural Computation, 9:1735-1780. pmid: 9377276

[23]	Hsu K, Gupta H V, Sorooshia S. 1995. Artificial neural network modeling of the rainfall-runoff process. Water Resources Research, 31:2517-2530. doi: 10.1029/95WR01955

[24]	Hu C H, Wu Q, Li H, et al. 2018. Deep learning with a Long Short-Term Memory networks approach for rainfall-runoff simulation. Water, 10:1543, doi: 10.3390/w10111543. doi: 10.3390/w10111543

[25]	Hu Y C, Yan L, Hang T, et al. 2020. Stream-flow forecasting of small rivers based on LSTM. arXiv e-prints. arXiv:2001.05681. https://arxiv.org/abs/2001.05681.

[26]	Immerzeel WW, van Beek L, Bierkens M. 2010. Climate change will affect the Asian water towers. Science, 328:1382-1385. doi: 10.1126/science.1183188 pmid: 20538947

[27]	Immerzeel W W, Lutz A F, Andrade M, et al. 2020. Importance and vulnerability of the world's water towers. Nature, 577:364-369. doi: 10.1038/s41586-019-1822-y

[28]	Ji H P, Fang G H, Yang J, et al. 2019. Multi-objective calibration of a distributed hydrological model in a highly glacierized watershed in Central Asia. Water, 11:554, doi: 10.3390/w11030554. doi: 10.3390/w11030554

[29]	Jodar J, Carpintero E, Martos-Rosillo S, et al. 2018. Combination of lumped hydrological and remote-sensing models to evaluate water resources in a semi-arid high altitude ungauged watershed of Sierra Nevada (Southern Spain). Science of the Total Environment, 625:285-300. doi: 10.1016/j.scitotenv.2017.12.300

[30]	Kingma D, Ba J. 2015. Adam: A Method for Stochastic Optimization. In: Proceedings of the 3rd International Conference on Learning Representations. San Diego: Computational and Biological Learning Society.

[31]	Kratzert F, Klotz D, Brenner C, et al. 2018. Rainfall-runoff modelling using Long Short-Term Memory (LSTM) networks. Hydrology and Earth System Sciences, 22:6005-6022. doi: 10.5194/hess-22-6005-2018

[32]	Kratzert F, Klotz D, Herrnegger M, et al. 2019a. Toward improved predictions in ungauged basins: Exploiting the power of machine learning. Water Resources Research, 55:11344-11354. doi: 10.1029/2019WR026065

[33]	Kratzert F, Klotz D, Shalev G, et al. 2019b. Towards learning universal, regional, and local hydrological behaviors via machine learning applied to large-sample datasets. Hydrology and Earth System Sciences, 23:5089-5110. doi: 10.5194/hess-23-5089-2019

[34]	Kratzert F, Klotz D, Hochreiter S, et al. 2020. A note on leveraging synergy in multiple meteorological datasets with deep learning for rainfall-runoff modeling. Hydrology and Earth System Sciences Discussion. [Preprint]. https://doi.org/10.5194/hess-2020-221, in review.

[35]	Li W, Kiaghadi A, Dawson C. 2021. High temporal resolution rainfall-runoff modeling using long-short-term-memory (LSTM) networks. Neural Computing and Applications, 33:1261-1278. doi: 10.1007/s00521-020-05010-6

[36]	Li Z L, Shao Q X, Xu Z X, et al. 2010. Analysis of parameter uncertainty in semi-distributed hydrological models using bootstrap method: A case study of SWAT model applied to Yingluoxia watershed in northwest China. Journal of Hydrology, 385:76-83. doi: 10.1016/j.jhydrol.2010.01.025

[37]	Luo X G, Yuan X H, Zhu S, et al. 2019. A hybrid support vector regression framework for streamflow forecast. Journal of Hydrology, 568:184-193. doi: 10.1016/j.jhydrol.2018.10.064

[38]	Maier H R, Jain A, Dandy G C, et al. 2010. Methods used for the development of neural networks for the prediction of water resource variables in river systems: Current status and future directions. Environmental Modelling & Software, 25:891-909.

[39]	Michaud J, Sorooshian S. 1994. Comparison of simple versus complex distributed runoff models on a midsized semiarid watershed. Water Resources Research, 30:593-605. doi: 10.1029/93WR03218

[40]	Ni L L, Wang D, Wu J F, et al. 2020. Streamflow forecasting using extreme gradient boosting model coupled with Gaussian mixture model. Journal of Hydrology, 586:124901, doi: 10.1016/j.jhydrol.2020.124901. doi: 10.1016/j.jhydrol.2020.124901

[41]	Qin J, Yang K, Liang S L, et al. 2009. The altitudinal dependence of recent rapid warming over the Tibetan Plateau. Climatic Change, 97:321, doi: 10.1007/s10584-009-9733-9. doi: 10.1007/s10584-009-9733-9

[42]	Razavi T, Coulibaly P. 2013. Streamflow prediction in ungauged basins: Review of regionalization methods. Journal of Hydrologic Engineering, 18:958-975. doi: 10.1061/(ASCE)HE.1943-5584.0000690

[43]	Reichstein M, Camps-Valls G, Stevens B, et al. 2019. Deep learning and process understanding for data-driven earth system science. Nature, 566:195-204. doi: 10.1038/s41586-019-0912-1

[44]	Rudin C. 2019. Stop explaining black box machine learning models for high stakes decisions and use interpretable models instead. Nature Machine Intelligence, 1:206-215. doi: 10.1038/s42256-019-0048-x

[45]	Shen C P. 2018. A transdisciplinary review of deep learning research and its relevance for water resources scientists. Water Resources Research, 54:8558-8593. doi: 10.1029/2018WR022643

[46]	Shen C P, Laloy E, Elshorbagy A, et al. 2018. HESS Opinions: Incubating deep-learning-powered hydrologic science advances as a community. Hydrology and Earth System Sciences, 22:5639-5656. doi: 10.5194/hess-22-5639-2018

[47]	Shen Y J, Shen Y J, Fink M, et al. 2018a. Unraveling the hydrology of the glacierized Kaidu Basin by integrating multisource data in the Tianshan Mountains, Northwestern China. Water Resources Research, 54:557-580. doi: 10.1002/wrcr.v54.1

[48]	Shen Y J, Shen Y J, Fink M, et al. 2018b. Trends and variability in streamflow and snowmelt runoff timing in the southern Tianshan Mountains. Journal of Hydrology, 557:173-181. doi: 10.1016/j.jhydrol.2017.12.035

[49]	Tarasova L, Knoche M, Dietrich J, et al. 2016. Effects of input discretization, model complexity, and calibration strategy on model performance in a data-scarce glacierized catchment in Central Asia. Water Resources Research, 52:4674-4699. doi: 10.1002/2015WR018551

[50]	Viviroli D, Kummu M, Meybeck M, et al. 2020. Increasing dependence of lowland populations on mountain water resources. Nature Sustainability, 3:917-928. doi: 10.1038/s41893-020-0559-9

[51]	Xiang Z R, Yan J, Demir I. 2020. A rainfall-runoff model with LSTM-based sequence-to-sequence learning. Water Resources Research, 56, doi: 10.1029/2019wr025326. doi: 10.1029/2019wr025326

[52]	Yang T, Sun F B, Gentine P, et al. 2019. Evaluation and machine learning improvement of global hydrological model-based flood simulations. Environmental Research Letters, 14:114027, doi: 10.1088/1748-9326/ab4d5e. doi: 10.1088/1748-9326/ab4d5e

[53]	Yang T T, Asanjan A A, Welles E, et al. 2017. Developing reservoir monthly inflow forecasts using artificial intelligence and climate phenomenon information. Water Resources Research, 53:2786-2812. doi: 10.1002/wrcr.v53.4

[54]	Zheng Z S, Ma Q, Jin S C, et al. 2019. Canopy and terrain interactions affecting snowpack spatial patterns in the Sierra Nevada of California. Water Resources Research, 55:8721-8739. doi: 10.1029/2018WR023758

[1]	CHANG Chang, CHANG Yu, GUO Meng, HU Yuanman. Modelling the dead fuel moisture content in a grassland of Ergun City, China[J]. Journal of Arid Land, 2023, 15(6): 710-723.

[2]	ZHANG Wensheng, AN Chengbang, LI Yuecong, ZHANG Yong, LU Chao, LIU Luyu, ZHANG Yanzhen, ZHENG Liyuan, LI Bing, FU Yang, DING Guoqiang. Modern pollen assemblages and their relationships with vegetation and climate on the northern slopes of the Tianshan Mountains, Xinjiang, China[J]. Journal of Arid Land, 2023, 15(3): 327-343.

[3]	ZHANG Shubao, LEI Jun, TONG Yanjun, ZHANG Xiaolei, LU Danni, FAN Liqin, DUAN Zuliang. Temporal and spatial responses of ecological resilience to climate change and human activities in the economic belt on the northern slope of the Tianshan Mountains, China[J]. Journal of Arid Land, 2023, 15(10): 1245-1268.

[4]	WEI Tianfeng, SHANGGUAN Donghui, TANG Xianglong, QIN Yu. The role of glacial gravel in community development of vascular plants on the glacier forelands of the Third Pole[J]. Journal of Arid Land, 2022, 14(9): 1022-1037.

[5]	ZHANG Yin, GULIMIRE Hanati, SULITAN Danierhan, HU Keke. Monitoring and analysis of snow cover change in an alpine mountainous area in the Tianshan Mountains, China[J]. Journal of Arid Land, 2022, 14(9): 962-977.

[6]	LI Hongliang, WANG Puyu, LI Zhongqin, JIN Shuang, XU Chunhai, MU Jianxin, HE Jie, YU Fengchen. Effect of topography on the changes of Urumqi Glacier No. 1 in the Chinese Tianshan Mountains[J]. Journal of Arid Land, 2022, 14(7): 719-738.

[7]	ABAY Peryzat, GONG Lu, CHEN Xin, LUO Yan, WU Xue. Spatiotemporal variation and correlation of soil enzyme activities and soil physicochemical properties in canopy gaps of the Tianshan Mountains, Northwest China[J]. Journal of Arid Land, 2022, 14(7): 824-836.

[8]	SU Yuan, GONG Yanming, HAN Wenxuan, LI Kaihui, LIU Xuejun. Dependency of litter decomposition on litter quality, climate change, and grassland type in the alpine grassland of Tianshan Mountains, Northwest China[J]. Journal of Arid Land, 2022, 14(6): 691-703.

[9]	HUANG Xiaoran, BAO Anming, GUO Hao, MENG Fanhao, ZHANG Pengfei, ZHENG Guoxiong, YU Tao, QI Peng, Vincent NZABARINDA, DU Weibing. Spatiotemporal changes of typical glaciers and their responses to climate change in Xinjiang, Northwest China[J]. Journal of Arid Land, 2022, 14(5): 502-520.

[10]	PENG Jiajia, LI Zhongqin, XU Liping, MA Yuqing, LI Hongliang, ZHAO Weibo, FAN Shuang. Glacier mass balance and its impacts on streamflow in a typical inland river basin in the Tianshan Mountains, northwestern China[J]. Journal of Arid Land, 2022, 14(4): 455-472.

[11]	CHEN Haiyan, CHEN Yaning, LI Dalong, LI Weihong, YANG Yuhui. Identifying water vapor sources of precipitation in forest and grassland in the north slope of the Tianshan Mountains, Central Asia[J]. Journal of Arid Land, 2022, 14(3): 297-309.

[12]	ZHANG Xueting, CHEN Rensheng, LIU Guohua. Economic losses from reduced freshwater under future climate scenarios: An example from the Urumqi River, Tianshan Mountains[J]. Journal of Arid Land, 2022, 14(2): 139-153.

[13]	QIU Dong, TAO Ye, ZHOU Xiaobing, Bagila MAISUPOVA, YAN Jingming, LIU Huiliang, LI Wenjun, ZHUANG Weiwei, ZHANG Yuanming. Spatiotemporal variations in the growth status of declining wild apple trees in a narrow valley in the western Tianshan Mountains, China[J]. Journal of Arid Land, 2022, 14(12): 1413-1439.

[14]	SUN Chen, MA Yonggang, GONG Lu. Response of ecosystem service value to land use/cover change in the northern slope economic belt of the Tianshan Mountains, Xinjiang, China[J]. Journal of Arid Land, 2021, 13(10): 1026-1040.

[15]	Xueting ZHANG, Xuemei LI, Lanhai LI, Shan ZHANG, Qirui QIN. Environmental factors influencing snowfall and snowfall prediction in the Tianshan Mountains, Northwest China[J]. Journal of Arid Land, 2019, 11(1): 15-28.

Viewed

Full text

Abstract

Cited

Shared

Discussed