Long-Lead Forecasting of Runoff Season Flows in the Colorado River Basin Using a Random Forest Approach

David Woodson,Balaji Rajagopalan,Edith Zagona

DOI: https://doi.org/10.1061/jwrmd5.wreng-6167

IF: 3.457

2024-02-02

Journal of Water Resources Planning and Management

Abstract:Journal of Water Resources Planning and Management, Volume 150, Issue 4, April 1, 2024.

water resources,engineering, civil

What problem does this paper attempt to address?

Assessing Different Roles of Baseflow and Surface Runoff for Long-Term Streamflow Forecasting in Southeastern China

Hao Chen,Yue-Ping Xu,Ramesh S. Teegavarapu,Yuxue Guo,Jingkai Xie

DOI: https://doi.org/10.1080/02626667.2021.1988612

2021-01-01

Hydrological Sciences Journal

Abstract:Accurate long-term streamflow forecast is essential to alleviate and solve the water security problems related to flood and drought disaster warnings. In this study, a new strategy for forecasting monthly streamflow is proposed and four scenarios are designed for the evaluation of different roles of baseflow and surface runoff on performances of long-term streamflow forecasting. The developed models are evaluated at multiple streamflow sites located in the Zhejiang Province of China. The results show that artificial intelligence (AI)-based models with two predictor variables (i.e. baseflow and surface runoff) performed better than that with a single predictor (streamflow) for all the months in a year, and the prediction accuracy of annual peak and monthly streamflow values is improved. Based on the comprehensive evaluations of all the models, the baseflow and surface runoff values are recommended as inputs to AI-based models for an improved prediction accuracy of streamflows.
Subseasonal to seasonal streamflow forecasting in a semiarid watershed

Patrick D. Broxton,Willem J. D. Leeuwen,Bohumil M. Svoma,James Walter,Joel A. Biederman

DOI: https://doi.org/10.1111/1752-1688.13147

2023-07-27

Abstract:Operational streamflow forecasting is critically important to managers of river basins that supply water, hydropower, and flood protection. While seasonal water supply forecasts (WSFs) are important for long‐term water resources planning operations, shorter term (e.g., 1–5 weeks) streamflow forecasts are critical for balancing water conservation with flood risk during wet periods. In this study, we designed a streamflow forecasting system with the water resources group at the Salt River Project (SRP), a provider of water and power to millions of customers in central Arizona (AZ), to provide streamflow forecasts for a diverse and operationally important set of watersheds in AZ. The forecast system uses machine learning to make seasonal WSFs, a rainfall–runoff model driven by ensemble meteorological forecasts to make 35‐day streamflow forecasts, and an innovative approach to improve the WSFs based on the 35‐day streamflow forecasts. This model integration allows for an assessment of the impact of different meteorological forecasts on WSFs, helping SRP to balance water conservation goals with shorter term flood risks. In addition, seasonal WSFs are improved in the early winter when they incorporate the 35‐day streamflow predictions. Furthermore, these improvements are larger than when they incorporate 7‐day streamflow predictions, demonstrating the value of using subseasonal to seasonal (S2S, >1–2 weeks) forecasts to improve seasonal WSFs in these watersheds.
A Medium and Long-Term Runoff Forecast Method Based on Massive Meteorological Data and Machine Learning Algorithms

Yujie Li,Dong Wang,Jing Wei,Bo Li,Bin Xu,Yueping Xu,Huaping Huang

DOI: https://doi.org/10.3390/w13091308

IF: 3.53

2021-05-07

Water

Abstract:Accurate and reliable predictors selection and model construction are the key to medium and long-term runoff forecast. In this study, 130 climate indexes are utilized as the primary forecast factors. Partial Mutual Information (PMI), Recursive Feature Elimination (RFE) and Classification and Regression Tree (CART) are respectively employed as the typical algorithms of Filter, Wrapper and Embedded based on Feature Selection (FS) to obtain three final forecast schemes. Random Forest (RF) and Extreme Gradient Boosting (XGB) are respectively constructed as the representative models of Bagging and Boosting based on Ensemble Learning (EL) to realize the forecast of the three types of forecast lead time which contains monthly, seasonal and annual runoff sequences of the Three Gorges Reservoir in the Yangtze River Basin. This study aims to summarize and compare the applicability and accuracy of different FS methods and EL models in medium and long-term runoff forecast. The results show the following: (1) RFE method shows the best forecast performance in all different models and different forecast lead time. (2) RF and XGB models are suitable for medium and long-term runoff forecast but XGB presents the better forecast skills both in calibration and validation. (3) With the increase of the runoff magnitudes, the accuracy and reliability of forecast are improved. However, it is still difficult to establish accurate and reliable forecasts only large-scale climate indexes used. We conclude that the theoretical framework based on Machine Learning could be useful to water managers who focus on medium and long-term runoff forecast.

environmental sciences,water resources
Geospatial patterns in runoff projections using random forest based forecasting of time-series data for the mid-Atlantic region of the United States

Brandi Gaertner

DOI: https://doi.org/10.1016/j.scitotenv.2023.169211

2024-02-20

Abstract:This research explores the geospatial patterns of historical runoff for the period 1958-2021 in the Mid-Atlantic region and uses these time-series data plus nine external climatic and hydrologic variables to predict future runoff for the period 2022-2031. Gridded, average monthly climatic water balance data were obtained from the TerraClimate dataset. A cluster analysis of the long term (1958-2021) historical runoff found 13 significant temporal trends, which tend to form large contiguous regions associated with climate gradients and topographic patterns. The runoff time-series clusters, and the associated time-series of 9 TerraClimate variables, were used to generate random forest based forecast models to predict future (2022-2031) runoff. The random forest-based forecast with the greatest accuracy included inputs of actual evapotranspiration, climate water deficit, minimum, average, and maximum temperature, and vapor pressure deficit. The final model predicted significantly increasing runoff in nine of the 13 clusters.
Runoff Forecasting using Convolutional Neural Networks and optimized Bi-directional Long Short-term Memory

Junhao Wu,Zhaocai Wang,Yuan Hu,Sen Tao,Jinghan Dong

DOI: https://doi.org/10.1007/s11269-022-03414-8

IF: 4.426

2023-01-11

Water Resources Management

Abstract:Water resources matters considerably in maintaining the biological survival and sustainable socio-economic development of a region. Affected by a combination of factors such as geographic characteristics of the basin and climate change, runoff variability is non-linear and non-stationary. Runoff forecasting is one of the important engineering measures to prevent flood disasters. The improvement of its accuracy is also a difficult problem in the research of water resources management. To this end, an ensemble deep learning model was hereby developed to forecast daily river runoff. First, variational mode decomposition (VMD) was used to decompose the original daily runoff series data set into discrete internal model function (IMF) and distinguish signals with different frequencies. Then, for each IMF, a convolutional neural network (CNN) was introduced to extract the features of each IMF component. Subsequently, a bi-directional long short-term memory network (BiLSTM) based on an attention mechanism (AM) was used for prediction. A Bayesian optimization algorithm (BOA) was also introduced to optimize the hyperparameters of the BiLSTM, thereby further improving the estimation precision of the VMD-CNN-AM-BOA-BiLSTM model. The model was applied to the daily runoff data from January 1, 2010 to November 30, 2021 at the Wushan and Weijiabao Hydrological Stations in the Wei River Basin, and the RMSEs of 3.54 and 15.23 were obtained for the test set data at the two stations respectively, which were much better than those of EEMD-VMD-SVM and CNN-BiLSTM-AM models. Additionally, the hereby proposed model is proven to have better peak flood prediction capability and adaptability under different hydrological environments. Based on this sound performance, the model becomes an effective data-driven tool in hydrological forecasting practice, and can also provide some reference and practical application guidance for water resources management and flood warning.

water resources,engineering, civil
Influence of Subseasonal-to-Annual Water Supply Forecasts on Many-Objective Water System Robustness under Long-Term Change

Kyla Semmendinger,Scott Steinschneider

DOI: https://doi.org/10.1061/jwrmd5.wreng-6205

IF: 3.457

2024-02-24

Journal of Water Resources Planning and Management

Abstract:Journal of Water Resources Planning and Management, Volume 150, Issue 5, May 1, 2024.

water resources,engineering, civil
A new approach of coupled long-range forecasts for streamflow and groundwater level

David E. Robertson,Guobin Fu,Olga Barron,Geoff Hodgson,Andrew Schepen

DOI: https://doi.org/10.1016/j.jhydrol.2024.130837

IF: 6.4

2024-02-04

Journal of Hydrology

Abstract:Methods to produce seasonal to annual forecasts of surface and groundwater availability have been developed and implemented nationally and globally. Such forecasts allow water managers to create effective irrigation schedules and better plan and manage water supply. However, one limitation of existing approaches that the forecasts of surface streamflow and groundwater are independent and therefore neglect inter-variable correlations that allow for reliable forecasting of total water availability. In this study, we develop an approach to jointly forecast streamflow and groundwater level and demonstrate its performance for the Lockyer Valley, Australia, where both surface water and groundwater are critical for irrigation. Informed by analysis of the processes influencing dynamics of groundwater levels in the alluvial aquifer (as a substitute to available groundwater storage) and streamflow dynamics, an existing hydrological model is adapted to better represent the ephemeral nature of streamflow in the catchment and also to simulate changes in groundwater levels. The modified hydrological model is then integrated into a statistical-dynamical forecasting framework to jointly forecast streamflow and groundwater level. Verification results indicate that forecasts skilful to lead times of up to 3 months for streamflow and 12 months for groundwater levels, even though rainfall forecasts are not skilful beyond the first month. Forecasts for both streamflow and groundwater are also statistically reliable with low bias. The approach developed can be extended to forecast the spatial distribution of groundwater levels across a large region, as well as be used to infill missing groundwater observations.

geosciences, multidisciplinary,water resources,engineering, civil
A Comparative Study on Forecasting of Long-term Daily Streamflow using ANN, ANFIS, BiLSTM and CNN-GRU-LSTM

Sajjad M. Vatanchi,Hossein Etemadfard,Mahmoud F. Maghrebi,Rouzbeh Shad

DOI: https://doi.org/10.1007/s11269-023-03579-w

IF: 4.426

2023-08-24

Water Resources Management

Abstract:Long-term streamflow forecasting is a critical step when planning and managing water resources. Advanced techniques in deep learning have been proposed for forecasting streamflow. Applying these methods in long-term streamflow prediction is an issue that has received less attention. Four models, including Adaptive Neuro-Fuzzy Inference Systems (ANFIS), Artificial Neural Networks (ANN), Bidirectional Long-Short Term Memory (BiLSTM), and hybrid Convolutional Neural Network (CNN)-Gated Recurrent Unit (GRU)-LSTM, are applied to forecast the long-term daily streamflow of the Colorado River in the U.S. The proper time lag for input series creation is determined using partial autocorrelation. 60% of the data (1921–1981) is used for training, whereas 40% (1981–2021) is used to evaluate the model's performance. The results of the studied models are assessed by Using four indices: the Mean Absolute Error ( MAE ), the Normalized Root Mean Square Error ( NRMSE ), the Correlation Coefficient ( r ), and the Nash–Sutcliffe Coefficient ( E NS ). As a result of the testing step, the ANFIS model with NRMSE = 0.118, MAE = 26.16 (m3/s), r = 0.966, and E NS = 0.933 was more accurate than other studied models. Despite their complexity, the BiLSTM and CNN-GRU-LSTM models did not outperform the others. Comparing these models to ANN and ANFIS, it is evident that their performance is not superior.

water resources,engineering, civil
Long-term Runoff Forecasting Models Based on the Teleconnection coupled with Machine Learning

Teng Zhang,Zhongjing Wang,Zixiong Zhang

DOI: https://doi.org/10.5194/egusphere-egu2020-1369

2020-03-23

Abstract:<p>Runoff forecast with high precision is important for the efficient utilization of water resources and regional sustainable development, especially in the arid area. The monthly runoff of Changmabao (CMB) station has an upwards trend and an abrupt point in 1998. The impact factor analysis shows that it is highly correlated with the current precipitation and temperature in the wet season while the previous runoff and previous global land temperature in the dry season. Three models including the time-series decomposition model, the model based on teleconnection coupled with the support vector machine, and the model based on teleconnection coupled with the artificial neural network are used to predict the runoff of CMB station. An indicator &#946; is constructed with the correlation coefficient (R) and mean relative deviation (rBias) to evaluate the model performance more conveniently and intuitively. The results suggest that the model based on teleconnection coupled with the support vector machine preforms best. This forecasting method could be applied to the management and dispatch of water resources in arid areas.</p>
Application of Machine Learning and Process-Based Models for Rainfall-Runoff Simulation in DuPage River Basin, Illinois

Amrit Bhusal,Utsav Parajuli,Sushmita Regmi,Ajay Kalra

DOI: https://doi.org/10.3390/hydrology9070117

2022-06-27

Hydrology

Abstract:Rainfall-runoff simulation is vital for planning and controlling flood control events. Hydrology modeling using Hydrological Engineering Center—Hydrologic Modeling System (HEC-HMS) is accepted globally for event-based or continuous simulation of the rainfall-runoff operation. Similarly, machine learning is a fast-growing discipline that offers numerous alternatives suitable for hydrology research’s high demands and limitations. Conventional and process-based models such as HEC-HMS are typically created at specific spatiotemporal scales and do not easily fit the diversified and complex input parameters. Therefore, in this research, the effectiveness of Random Forest, a machine learning model, was compared with HEC-HMS for the rainfall-runoff process. Furthermore, we also performed a hydraulic simulation in Hydrological Engineering Center—Geospatial River Analysis System (HEC-RAS) using the input discharge obtained from the Random Forest model. The reliability of the Random Forest model and the HEC-HMS model was evaluated using different statistical indexes. The coefficient of determination (R2), standard deviation ratio (RSR), and normalized root mean square error (NRMSE) were 0.94, 0.23, and 0.17 for the training data and 0.72, 0.56, and 0.26 for the testing data, respectively, for the Random Forest model. Similarly, the R2, RSR, and NRMSE were 0.99, 0.16, and 0.06 for the calibration period and 0.96, 0.35, and 0.10 for the validation period, respectively, for the HEC-HMS model. The Random Forest model slightly underestimated peak discharge values, whereas the HEC-HMS model slightly overestimated the peak discharge value. Statistical index values illustrated the good performance of the Random Forest and HEC-HMS models, which revealed the suitability of both models for hydrology analysis. In addition, the flood depth generated by HEC-RAS using the Random Forest predicted discharge underestimated the flood depth during the peak flooding event. This result proves that HEC-HMS could compensate Random Forest for the peak discharge and flood depth during extreme events. In conclusion, the integrated machine learning and physical-based model can provide more confidence in rainfall-runoff and flood depth prediction.
Predicting flood stages in watersheds with different scales using hourly rainfall dataset: A high-volume rainfall features empowered machine learning approach

Lei Qiao,Daniel Livsey,Jarrett Wise,Kem Kadavy,Sherry Hunt,Kevin Wagner

DOI: https://doi.org/10.1016/j.scitotenv.2024.175231

2024-11-10

Abstract:Accurate prediction of instantaneous high lake water levels and flood flows (flood stages) from micro-catchments to big river basins are critical for flood forecasting. Lake Carl Blackwell, a small-watershed reservoir in the south-central USA, served as a primary case study due to its rich historical dataset. Bearing knowledge that both current and previous rainfall contributes to the reservoirs' water body, a series of hourly rainfall features were created to maximize predicting power, which include total rainfall amounts in the current hour, the past 2 h, 3 h, …, 600 h in addition to previous-day lake levels. Notedly, the rainfall features are the accumulated rainfall amounts from present to previous hours rather than the rainfall amount in any specific hour. Random Forest Regression (RFR) was used to score the features' importance and predict the flood stages along with Neural Network - Multi-layer Perceptron Regression (NN-MLP), Support Vector Regression (SVR), Extreme Gradient Boosting (XGBoost), and the ordinary multi-variant linear regression (MLR) together with dimension reduced linear models of Principal Component Regression (PCR) and Partial Least Square Regression (PLSR). The prediction accuracy for the lake flood stages can be as high as 0.95 in R2, 0.11 ft. in mean absolute error (MAE), and 0.21 ft. in root mean square error (RMSE) for the testing dataset by the RFR (NN-MLP performed equally well), with small accuracy decreases by the other two non-linear algorithms of XGBoost and SVR. The linear regressions with dimension reductions had the lowest accuracy. Furthermore, our approach demonstrated high accuracy and broad applicability for surface runoff and streamflow predictions across three different-sized watersheds from micro-catchment to big river basins in the region, with increases of predicting power from earlier rainfall for larger watersheds and vice versa.
Predict Seasonal Low Flows in the Upper Yangtze River Using Random Forests Model

杨大文,蔡喜明,赵铜铁钢,曹勇

2012-01-01

Abstract:Predictor selection and model construction are two key issues in long-term streamflow forecasting.This study introduces a random forests model for selecting predictor set from measured streamflows and 74 hydro-climatic indices of the period from January to October provided by China National Climate Center,and predicts seasonal low flow in the upper Yangtze for the period from November to next May.The results show that：1） as predicting lead-time increases,streamflow auto-correlation becomes weaker and hydro-climatic teleconnection gets stronger than the auto-correlation;2） the predicted streamflows at the Pingshan and Cuntan gauges show a strong linear relationship with the observations and their relative errors are less than 20%;3） analysis of the prediction uncertainty indicates that the random forests model can also be applied to probabilistic prediction of streamflow.
A hydrological simulation dataset of the Upper Colorado River Basin from 1983 to 2019

Hoang Tran,Jun Zhang,Mary Michael O’Neill,Anna Ryken,Laura E. Condon,Reed M. Maxwell

DOI: https://doi.org/10.1038/s41597-022-01123-w

2022-01-20

Scientific Data

Abstract:Abstract This article presents a hydrological reconstruction of the Upper Colorado River Basin with an hourly temporal resolution, and 1-km spatial resolution from October 1982 to September 2019. The validated dataset includes a suite of hydrologic variables including streamflow, water table depth, snow water equivalent (SWE) and evapotranspiration (ET) simulated by an integrated hydrological model, ParFlow-CLM. The dataset was validated over the period with a combination of point observations and remotely sensed products. These datasets provide a long-term, natural-flow, simulation for one of the most over-allocated basins in the world.

multidisciplinary sciences
Integrating geographic data and the SCS-CN method with LSTM networks for enhanced runoff forecasting in a complex mountain basin

María José Merizalde,Paul Muñoz,Gerald Corzo,David F. Muñoz,Esteban Samaniego,Rolando Célleri

DOI: https://doi.org/10.3389/frwa.2023.1233899

2023-09-25

Frontiers in Water

Abstract:Introduction In complex mountain basins, hydrological forecasting poses a formidable challenge due to the intricacies of runoff generation processes and the limitations of available data. This study explores the enhancement of short-term runoff forecasting models through the utilization of long short-term memory (LSTM) networks. Methods To achieve this, we employed feature engineering (FE) strategies, focusing on geographic data and the Soil Conservation Service Curve Number (SCS-CN) method. Our investigation was conducted in a 3,390 km 2 basin, employing the GSMaP-NRT satellite precipitation product (SPP) to develop forecasting models with lead times of 1, 6, and 11 h. These lead times were selected to address the needs of near-real-time forecasting, flash flood prediction, and basin concentration time assessment, respectively. Results and discussion Our findings demonstrate an improvement in the efficiency of LSTM forecasting models across all lead times, as indicated by Nash-Sutcliffe efficiency values of 0.93 (1 h), 0.77 (6 h), and 0.67 (11 h). Notably, these results are on par with studies relying on ground-based precipitation data. This methodology not only showcases the potential for advanced data-driven runoff models but also underscores the importance of incorporating available geographic information into precipitation-ungauged hydrological systems. The insights derived from this study offer valuable tools for hydrologists and researchers seeking to enhance the accuracy of hydrological forecasting in complex mountain basins.
Long-range streamflow prediction using a distributed hydrological model in a snowfed watershed

Abdul Moiz,Akiyuki Kawasaki

DOI: https://doi.org/10.5194/piahs-386-217-2024

2024-04-19

Abstract:Abstract. Inadequate planning of spring snowmelt discharge can lead to wastage of water resources for various purposes such as hydropower and lead to reduced capacity of the dams to control floods during rainy season. In this research we analyze how much can the predictive skill of long-range forecasts be improved by using a distributed hydrological model. We used the Water and Energy Budget-based Distributed Hydrological Model with improved snow physics (WEB-DHM-S) for generating long-range forecasts with a lead time of up to 3 months for the case of Kurobe River Basin in Japan. The predictive skills of two sets of simulations were compared (i) climatology and (ii) ensemble stream flow prediction (ESP). In the case of ESP, the initial conditions of WEB-DHM-S are updated using real-time datasets from Radar-AMeDAS, AMeDAS and JRA55. We found that the model initial conditions are particularly important during the spring snowmelt season and can improve the forecast predictive skills quite significantly compared with the climatology.
Seasonal prediction of end-of-dry-season watershed behavior in a highly interconnected alluvial watershed in northern California

Kouba, Claire

DOI: https://doi.org/10.5194/hess-28-691-2024

IF: 6.3

2024-02-17

Hydrology and Earth System Sciences

Abstract:In undammed watersheds in Mediterranean climates, the timing and abruptness of the transition from the dry season to the wet season have major implications for aquatic ecosystems. Of particular concern in many coastal areas is whether this transition can provide sufficient flows at the right time to allow passage for spawning anadromous fish, which is determined by dry season baseflow rates and the timing of the onset of the rainy season. In (semi-) ephemeral watershed systems, these functional flows also dictate the timing of full reconnection of the stream system. In this study, we propose methods to predict, approximately 5 months in advance, two key hydrologic metrics in the undammed rural Scott River watershed in northern California. The two metrics are intended to characterize (1) the severity of a dry year and (2) the relative timing of the transition from the dry to the wet season. The ability to predict these metrics in advance could support seasonal adaptive management. The first metric is the minimum 30 d dry season baseflow volume, V min , which occurs at the end of the dry season (September–October) in this Mediterranean climate. The second metric is the cumulative precipitation, starting 1 September, necessary to bring the watershed to a "full" or "spilling" condition (i.e., initiate the onset of wet season storm- or baseflows) after the end of the dry season, referred to here as P spill . As potential predictors of these two metrics, we assess maximum snowpack, cumulative precipitation, the timing of the snowpack and precipitation, spring groundwater levels, spring river flows, reference evapotranspiration, and a subset of these metrics from the previous water year. Though many of these predictors are correlated with the two metrics of interest, we find that the best prediction for both metrics is a linear combination of the maximum snowpack water content and total October–April precipitation. These two linear models could reproduce historical values of V min and P spill with an average model error (RMSE) of 1.4 Mm 3 per 30 d (19.4 cfs) and 25.4 mm (1 in.), corresponding to 49 % and 37 % of mean observed values, respectively. Although these predictive indices could be used by governance entities to support local water management, careful consideration of baseline conditions used as a basis for prediction is necessary.

geosciences, multidisciplinary,water resources
Leveraging Next‐Generation Satellite Remote Sensing‐Based Snow Data to Improve Seasonal Water Supply Predictions in a Practical Machine Learning‐Driven River Forecast System

Sean W. Fleming,Karl Rittger,Catalina M. Oaida Taglialatela,Indrani Graczyk

DOI: https://doi.org/10.1029/2023wr035785

IF: 5.4

2024-04-02

Water Resources Research

Abstract:Seasonal predictions of spring‐summer river flow volume (water supply forecasts, WSFs) are foundational to western US water management. We test a new space‐based remote sensing product, spatially and temporally complete (STC) MODSCAG fractional snow‐covered area (fSCA), as input for the Natural Resources Conservation Service (NRCS) operational US West‐wide WSF system. fSCA data were considered alongside traditional SNOTEL predictors, in both statistical and AI‐based NRCS operational hydrologic models, throughout the forecast season, in four test watersheds (Walker, Wind, Piedra, and Gila Rivers in California, Wyoming, Colorado, and New Mexico). Outcomes from over 200 WSF models suggest fSCA‐enabled accuracy gains are most consistent and explainable for short‐lead, late‐season forecasts (roughly 10%–25% improvements, typically), which in operational practice can be challenging as snowlines rise above in situ measurement sites. Gains are roughly proportional to how thoroughly spring‐summer runoff is dominated by snowmelt, and how poorly in situ networks monitor late‐season snowpack. fSCA also improved accuracy for long‐lead, early‐season forecasts, which are similarly problematic in WSF practice, but not for WSFs issued around the time of peak snow accumulation, when in situ measurements reasonably characterize mountain snowpack available for upcoming spring‐summer snowmelt. The AI‐based hydrologic model generally outperformed the statistical model and, in some cases, better‐capitalized on satellite remote sensing. Additionally, preliminary analyses suggest reasonable WSF skill in many cases using fSCA as the sole predictor, potentially useful in sparsely monitored regions; and that combining satellite and in situ products in data‐driven hydrologic models using genetic algorithm‐based predictor selection could help guide new SNOTEL site selection.

environmental sciences,water resources,limnology
Predicting Streamflow Duration From Crowd‐Sourced Flow Observations

David A. Peterson,Stephanie K. Kampf,Kira C. Puntenney‐Desmond,Matthew P. Fairchild,Sam Zipper,John C. Hammond,Matthew R. V. Ross,Megan G. Sears

DOI: https://doi.org/10.1029/2023wr035093

IF: 5.4

2024-01-04

Water Resources Research

Abstract:Streamflow duration is important for aquatic ecosystems and assigning stream protection status. This study predicts streamflow duration, represented as the fraction of time with flow each year, using a combination of sensor data and crowd‐sourced visual observations for a study area in northern Colorado, USA. We used 11 stream stage sensors and 177 visual monitoring points to examine how frequently streams should be sampled to compute flow fractions accurately. This showed that the number of visual observations needed to compute accurate flow fractions increases with decreasing flow duration. We then developed random forest models to predict mean annual flow fractions using climate, topographic, and land cover predictors and found that snow persistence, summer precipitation, and drainage area were important predictors. Model performance was best when using sites with ≥10 visual observations. Our model predicts that almost all (98%) of streams in the study region are non‐perennial, about 10% more than the amount of non‐perennial streams in the National Hydrography Dataset. Stream type maps are sensitive to the time period of data collection and to thresholds used to represent perennial versus non‐perennial flow. To improve maps of non‐perennial streams, we recommend moving beyond categorical classification of streams to a continuous variable like flow fraction. These efforts can be best supported with frequent observations in time that span streams with a wide range of flow fractions and drainage area attributes.

environmental sciences,water resources,limnology
Long-run forecasting surface and groundwater dynamics from intermittent observation data: An evaluation for 50 years

M T Vu,A Jardani,N Massei,J Deloffre,M Fournier,B Laignel

DOI: https://doi.org/10.1016/j.scitotenv.2023.163338

2023-07-01

Abstract:The accurate prediction of water dynamics is critical for operational water resource management. In this study, we propose a novel approach to perform long-term forecasts of daily water dynamics, including river levels, river discharges, and groundwater levels, with a lead time of 7-30 days. The approach is based on the state-of-the-art neural network, bidirectional long short-term memory (BiLSTM), to enhance the accuracy and consistency of dynamic predictions. The operation of this forecasting system relies on an in-situ database observed for over 50 years with records gauging in 19 rivers, the karst aquifer, the English Channel, and the meteorological network in Normandy, France. To address the problem of missing measurements and gauge installations over time, we developed an adaptive scheme in which the neural network is regularly adjusted and re-trained in response to changing inputs during a long operation. Advances in BiLSTM with extensive learning past-to-future and future-to-past further help to avoid time-lag calibration that simplifies data processing. The proposed approach provides high accuracy and consistent prediction for the three water dynamics within a similar accuracy range as an on-site observation, with approximately 3 % error in the measurement range for the 7 day-ahead predictions and 6 % error for the 30 d-ahead predictions. The system also effectively fills the gap in actual measurements and detects anomalies at gauges that can last for years. Working with multiple dynamics not only proves that the data-driven model is a unified approach but also reveals the impact of the physical background of the dynamics on the performance of their predictions. Groundwater undergoes a slow filtration process following a low-frequency fluctuation, favoring long-term prediction, which differs from other higher-frequency river dynamics. The physical nature drives the predictive performance even when using a data-driven model.
Forecasting excessive rainfall with random forests and a deterministic convection-allowing model

Aaron J. Hill,Russ S. Schumacher

DOI: https://doi.org/10.1175/waf-d-21-0026.1

2021-07-21

Weather and Forecasting

Abstract:Abstract Approximately seven years of daily initializations from the convection-allowing National Severe Storms Laboratory Weather Research and Forecasting model are used as inputs to train random forest (RF) machine learning models to probabilistically predict instances of excessive rainfall. Unlike other hazards, excessive rainfall does not have an accepted definition, so multiple definitions of excessive rainfall and flash flooding – including flash flood reports and 24-hr average recurrence intervals (ARIs) – are used to explore RF configuration forecast sensitivities. RF forecasts are analogous to operational Weather Prediction Center (WPC) day-1 Excessive Rainfall Outlooks (EROs) and their resolution, reliability, and skill are strongly influenced by rainfall definitions and how inputs are assembled for training. Models trained with 1-y ARI exceedances defined by the Stage-IV (ST4) precipitation analysis perform poorly in the northern Great Plains and southwest U.S., in part due to a high bias in the number of training events in these regions. Increasing the ARI threshold to 2 years or removing ST4 data from training, optimizing forecast skill geographically, and spatially averaging meteorological inputs for training generally results in improved CONUS-wide RF forecast skill. Both EROs and RF forecasts have seasonal skill – poor forecasts in the late fall and winter and skillful forecasts in the summer and early fall. However, the EROs are consistently and significantly better than their RF counterparts, regardless of RF configuration, particularly in the summer months. The results suggest careful consideration should be made when developing ML-based probabilistic precipitation forecasts with convection-allowing model inputs, and further development is necessary to consider these forecast products for operational implementation.

meteorology & atmospheric sciences

Long-Lead Forecasting of Runoff Season Flows in the Colorado River Basin Using a Random Forest Approach

Assessing Different Roles of Baseflow and Surface Runoff for Long-Term Streamflow Forecasting in Southeastern China

Subseasonal to seasonal streamflow forecasting in a semiarid watershed

A Medium and Long-Term Runoff Forecast Method Based on Massive Meteorological Data and Machine Learning Algorithms

Geospatial patterns in runoff projections using random forest based forecasting of time-series data for the mid-Atlantic region of the United States

Runoff Forecasting using Convolutional Neural Networks and optimized Bi-directional Long Short-term Memory

Influence of Subseasonal-to-Annual Water Supply Forecasts on Many-Objective Water System Robustness under Long-Term Change

A new approach of coupled long-range forecasts for streamflow and groundwater level

A Comparative Study on Forecasting of Long-term Daily Streamflow using ANN, ANFIS, BiLSTM and CNN-GRU-LSTM

Long-term Runoff Forecasting Models Based on the Teleconnection coupled with Machine Learning

Application of Machine Learning and Process-Based Models for Rainfall-Runoff Simulation in DuPage River Basin, Illinois

Predicting flood stages in watersheds with different scales using hourly rainfall dataset: A high-volume rainfall features empowered machine learning approach

Predict Seasonal Low Flows in the Upper Yangtze River Using Random Forests Model

A hydrological simulation dataset of the Upper Colorado River Basin from 1983 to 2019

Integrating geographic data and the SCS-CN method with LSTM networks for enhanced runoff forecasting in a complex mountain basin

Long-range streamflow prediction using a distributed hydrological model in a snowfed watershed

Seasonal prediction of end-of-dry-season watershed behavior in a highly interconnected alluvial watershed in northern California

Leveraging Next‐Generation Satellite Remote Sensing‐Based Snow Data to Improve Seasonal Water Supply Predictions in a Practical Machine Learning‐Driven River Forecast System

Predicting Streamflow Duration From Crowd‐Sourced Flow Observations

Long-run forecasting surface and groundwater dynamics from intermittent observation data: An evaluation for 50 years

Forecasting excessive rainfall with random forests and a deterministic convection-allowing model