Probabilistic Streamflow Forecasts Research Articles

Generative deep learning for probabilistic streamflow forecasting: Conditional variational auto-encoder

Probabilistic hydrological forecasting has gained increasing importance in recent years, as it offers essential information for risk-based decision-making and flood management. Traditional hydrological models often produce deterministic forecasts, which do not account for the inherent uncertainties in hydrological systems. Although many previous studies have investigated using deep learning (DL) models for hydrological prediction, the development of probabilistic DL models (especially, generative models) has not yet been thoroughly examined for hydrological forecasting. The present study investigates the efficacy of a generative DL model, namely, conditional variational auto-encoder (CVAE). CVAE is applied for one-seven day(s) ahead probabilistic streamflow forecasting in 75 basins from the Canadian model parameter experiment (CANOPEX) database. The CVAE forecast model, which outputs forecasts in the form of a probability distribution, was benchmarked against two state-of-the-art quantile-based DL models: quantile-based encoder-decoder (ED) and quantile-based CVAE (QCVAE). The latter outputs forecasts for specific quantiles of a probability distribution (here, q = 0.05, 0.5, 0.95). More than 9000 models were developed based on different basins, input variable sets, and model structures. The models were evaluated in terms of point forecast accuracy and forecast reliability. The results indicate that the CVAE model generally outperforms the benchmark models in terms of reliability at a 90 % confidence level (median reliability of 92.49 % compared to 87.35 % and 84.59 % for ED and QCVAE, respectively). However, the quantile-based forecast models produce slightly more accurate point forecasts than the CVAE (median Kling-Gupta efficiency (KGE) of 0.88 compared to 0.90 for both ED and QCVAE). Notably, the CVAE model exhibits superior probabilistic forecasts in basins with poor point forecast accuracy, highlighting its usefulness over benchmark methods in difficult-to-forecast basins. Overall, CVAE is a promising probabilistic DL model for streamflow forecasting, and it can be readily applied for forecasting other hydrological variables (evapotranspiration, water level, etc.). The findings of this study provide a basis for future research on probabilistic forecasting of hydrological variables using generative DL models.

Uncertainty in three dimensions: the challenges of communicating probabilistic flood forecast maps

Abstract. Real-time operational flood forecasting most often concentrates on issuing streamflow predictions at specific points along the rivers of a watershed. However, we are now witnessing an increasing number of studies aimed at also including flood mapping as part of the forecasting system. While this additional new information (flood extent, depth, velocity, etc.) can potentially be useful for decision-makers, it could also be overwhelming. This is especially true for probabilistic and ensemble forecasting systems. While ensemble streamflow forecasts for a given point in space can be visualized relatively easily, the visualization and communication of probabilistic forecasts for water depth and extent pose additional challenges. Confusion typically arises from too much information, counterintuitive interpretation, or simply too much complexity in the representation of the forecast. The communication and visualization of probabilistic streamflow forecasts has been studied in the past, but this is not the case for the probabilistic flood forecast map, which is still an emerging product. In this paper, we synthesize the results of a large-scale survey (28 government representatives, 52 municipalities, 9 organizations, and 38 citizens and farmers, for a total of 140 people) regarding the users' preferences in terms of visualizing probabilistic flood forecasts over an entire river reach. The survey was performed through interviews, during which the interviewees were asked about their needs in terms of hydrological forecasting. We also presented the interviewees with four prototypes representing alternative visualizations of the same probabilistic forecast in order to understand their preferences in terms of colour maps, wording, and the representation of uncertainty. Our results highlight several issues related to the understanding of probabilities in the specific context of visualizing forecasted flood maps. We propose several suggestions for visualizing probabilistic flood maps and also describe potential adaptations for different categories of end users. This study is the first to investigate the visualization of probabilistic flood maps, which are gaining popularity. Given that the interview questions were not tied to a specific geographical location, our findings are applicable outside of the study area and, therefore, to other operational centres interested in providing probabilistic flood forecast maps to decision-making organizations and citizens.

Fully integrating probabilistic flood forecasts into the decision-making process across southern Quebec, Canada: some factors to consider

Flood forecasts can enable decision-makers to plan for mitigation measures, ensure the safety of people and reduce damages to private and public property. However, while flood forecasting systems are becoming increasingly sophisticated, those technical improvements do not always translate into a reduction of damages. In particular, ensemble and probabilistic streamflow forecasts provide more information regarding the uncertainty, but such forecasts are also more difficult to interpret. The connection between streamflow, water depth and flood extent is also not straightforward, as this connection requires extensive previous experience with a given river reach. Streamflow forecasts can be transformed in predictive flood maps through a hydraulic model, and those maps might have a more intuitive interpretation. However, if the system is designed to also represent the uncertainty of both water depth and extent, the interpretation becomes considerably more complex. Within this context of continuous improvement of flood forecasting systems, a broad consultation (90 participants) of forecast users was conducted across southern Quebec, Canada, with the aim of understanding their perception of flood forecasts and how they use those forecasts for decision-making. Participants responses emphasize the importance of contextualizing the forecasts according to potential consequences, providing an accurate representation of the time evolution of flood events, and facilitating the communication between forecasters and decision-makers. Following this, our main recommendation is to organize workshops to provide training for decision-makers on how to interpret hydrological forecasts correctly. Another recommendation is to implement additional communication channels between forecasters and end-users.

HUP‐BMA: An Integration of Hydrologic Uncertainty Processor and Bayesian Model Averaging for Streamflow Forecasting

AbstractUncertainty quantification and providing probabilistic streamflow forecasts are of particular interest for water resource management. The hydrologic uncertainty processor (HUP) is a well‐known Bayesian approach used to quantify hydrologic uncertainty based on observations and deterministic forecasts. This uncertainty quantification is model‐specific; however, utilizing information from multiple hydrologic models should be advantageous and should lead to better probabilistic forecasts. Using seven, structurally different, conceptual models, this study first aims at evaluating the effects of implementing different hydrologic models on HUP performance. Second, using the concepts of the Bayesian Model Averaging (BMA) approach, a multimodel HUP‐based Bayesian postprocessor (HUP‐BMA) is proposed where the combination of posterior distributions derived from HUP with different hydrologic models are used to better quantify the hydrologic uncertainty. All postprocessing approaches are applied for medium‐range daily streamflow forecasting (1–14 days ahead) in two watersheds located in Ontario, Canada. The results indicate that the HUP forecasts for short lead‐times are negligibly affected by implementing different hydrologic models, while with increasing lead‐time and flow magnitude, they significantly depend on the quality of the deterministic forecast. Moreover, the superiority of the proposed HUP‐BMA method over HUP is demonstrated based on various verification metrics in both watersheds. Additionally, HUP‐BMA outperformed the original BMA in quantifying hydrologic uncertainty for short lead‐times. However, by increasing lead‐time, considering the effects of initially observed flow on HUP‐BMA formulation may not be beneficial. So, its modified version unconditioned on initial observations is preferred.

Evaluation of a stochastic weather generator for long-term ensemble streamflow forecasts

ABSTRACT Resampling historical time series remains one of the main approaches used to generate long-term probabilistic streamflow forecasts, while there is a need to develop more flexible approaches taking into account non-stationarities. One possible approach is to use a modelling chain consisting of a stochastic weather generator and a hydrological model. However, the ability of this modelling chain to generate adequate probabilistic streamflows must first be evaluated. The aim of this paper is to compare the performance of a stochastic weather generator against resampling historical meteorological time series in order to produce ensemble streamflow forecasts. The comparison framework is based on 30 years of forecasts for a single Canadian watershed. Forecasts resulting from the two methods are evaluated using the continuous ranked probability score (CRPS) and rank histograms. Results indicate that while there are differences between the methods, they nevertheless perform similarly, thus showing that weather generators can be used as substitutes for resampling the historical past.

Evaluation of an Application of Probabilistic Quantitative Precipitation Forecasts for Flood Forecasting

Probabilistic streamflow forecasts using precipitation derived from ensemble-based Probabilistic Quantitative Precipitation Forecasts (PQPFs) are examined. The PQPFs provide rainfall amounts associated with probabilities of exceedance for all grid points, which are averaged to the watershed scale for input to the operational Sacramento Soil Moisture Accounting hydrologic model to generate probabilistic streamflow predictions. The technique was tested using both the High-Resolution Rapid Refresh Ensemble (HRRRE) and the High-Resolution Ensemble Forecast version 2.0 (HREF) for 11 river basins across the upper Midwest for 109 cases. The resulting discharges associated with low probability of exceedance values were too large; no events were observed having discharges above the 10% exceedance value predicted from the technique applied to both ensembles, and no events were observed having discharges above the 25% exceedance value from the HREF-based forecast. The large differences are due to using the same precipitation exceedance value at all points; it is unlikely that all watershed points would experience the heavy rainfall associated with the 5% probability of exceedance. The technique likely can be improved through calibration of the basin-average precipitation forecasts based on typical distributions of precipitation within the convective systems that dominate warm-season precipitation events or calibration of the resulting probabilistic discharge forecasts.

GRAPS: Generalized Multi-Reservoir Analyses using probabilistic streamflow forecasts

A multi-reservoir simulation-optimization model GRAPS, Generalized Multi-Reservoir Analyses using Probabilistic Streamflow Forecasts, is developed in which reservoirs and users across the basin are represented using a node-link representation. Unlike existing reservoir modeling software, GRAPS can handle probabilistic streamflow forecasts represented as ensembles for performing multi-reservoir prognostic water allocation and evaluate the reliability of forecast-based allocation with observed streamflow. GRAPS is applied to four linked reservoirs in the Jaguaribe Metropolitan Hydro-System (JMH) in Ceará, North East Brazil. Results from the historical simulation and the zero-inflow policy over the JMH system demonstrate the model's capability to support monthly water allocation and reproduce the observed monthly releases and storages. Additional analyses using streamflow forecast ensembles illustrate GRAP's abilities in developing storage-reliability curves under inflow-forecast uncertainty. Our analyses show that GRAPS is versatile and can be applied for 1) short-term operating policy studies, 2) long-term basin-wide planning evaluations, and 3) climate-information based application studies.

Understanding the Impact of Observation Data Uncertainty on Probabilistic Streamflow Forecasts Using a Dynamic Hierarchical Model

AbstractEarlier researches have proposed algorithms to quantify the measurement uncertainty in rating curves and found that the magnitude of the uncertainty can be significant enough to impact hydrologic modeling. Therefore, they suggested frameworks to include measurement uncertainty in the rating curve to make it robust. Despite their efforts, a robust rating curve is often ignored in traditional practices, considering the investment of time and money as well as the resulting benefit from it. In the current research, we are interested in understanding the role of the measurement error variance in real‐time streamflow forecasting. Our objectives are (i) to employ a state‐of‐the‐art statistical forecasting model that can handle measurement uncertainty in daily streamflow and (ii) to understand the trade‐off in forecasting performance when substantial knowledge regarding the measurement uncertainty is provided to the modeler. We apply the Bayesian dynamic hierarchical model (BDHM) on four gauging sites in the United States. Results show that the BDHM performs better than the daily climatology and local linear regression model. Also, the forecast variance changes proportionally with the change in the error variance as an input in the observation equation. Following this, we design a simulation‐based study, which assigns the measurement error in the reported streamflow to obtain multiple realizations of the true streamflow. The inclusion of substantial knowledge about the true error improves the BDHM's performance by lowering the CRPS (continuous rank probability score) values. However, the inclusion increases the forecast variance to bring the true streamflow within the sampling variability of the forecasted streamflow. Overall, an improved trade‐off between the success rate of forecasts and the forecast variance can be achieved by including the measurement error in the BDHM for rivers that witness less dispersed streamflow data.

Probabilistic streamflow forecast based on spatial post-processing of TIGGE precipitation forecasts

Ensemble precipitation forecast is effective in reducing the uncertainty and providing reliable probabilistic streamflow forecast. However, for operational applications, precipitation forecasts must go through bias correction in mean and spread. Although post-processing methods, such as BMA, have demonstrated good performance in ensemble-based calibration, the spatial correlation between stations may be altered after post-processing. In this research, ensemble precipitation forecasts of four NWP models, including ECMWF, UKMO, NCEP, and CMA within the TIGGE database, was bias-corrected and post-processed using quantile mapping and BMA for a case study basin in Iran. The ECC method was then used to recover the spatial correlation of ensemble forecasts. Subsequently, probabilistic streamflow forecast was conducted using post-processed precipitation forecasts. The results showed that the errors in the mean and spread of ensemble precipitation forecasts were corrected for each of the four NWP models while the ECC method was effective in maintaining spatial correlation. Furthermore, the results of probabilistic streamflow forecast showed that the performance of the forecast models improved after post processing, with the ECMWF model providing the best forecasts. More work is recommended to improve the impact of the ECC method on NWP models’ performance.

From Probabilistic Seasonal Streamflow Forecasts to Optimal Reservoir Operations: A Stochastic Programming Approach

Abstract We investigate the potential use of seasonal streamflow forecasts for the realtime operation of Angat reservoir (Philippines). The system is characterized by a strong intra- and inter-annual variability in the inflow process, which is further amplified by the El Nino Southern Oscillation (ENSO). We bank on the relationship between ENSO indices and local hydro-climatological processes to issue probabilistic streamflow forecasts (with a 3-month forecast horizon) and then integrate them within a Multistage Stochastic Programming (MSP) approach. The rolling-horizon, forecast-informed scheme is adopted for the period 1968–2014 and benchmarked against deterministic optimization solutions with perfect forecasts, climatology, and mean forecasts. We also compare its performance with the current operating rules, and the operating rules obtained by solving a Stochastic Dynamic Programming problem. Results show that the MSP approach can help reduce the severity of failures during prolonged droughts caused by ENSO.

Real-Time Flood Control by Tree-Based Model Predictive Control Including Forecast Uncertainty: A Case Study Reservoir in Turkey

Optimal control of reservoirs is a challenging task due to conflicting objectives, complex system structure, and uncertainties in the system. Real time control decisions suffer from streamflow forecast uncertainty. This study aims to use Probabilistic Streamflow Forecasts (PSFs) having a lead-time up to 48 h as input for the recurrent reservoir operation problem. A related technique for decision making is multi-stage stochastic optimization using scenario trees, referred to as Tree-based Model Predictive Control (TB-MPC). Deterministic Streamflow Forecasts (DSFs) are provided by applying random perturbations on perfect data. PSFs are synthetically generated from DSFs by a new approach which explicitly presents dynamic uncertainty evolution. We assessed different variables in the generation of stochasticity and compared the results using different scenarios. The developed real-time hourly flood control was applied to a test case which had limited reservoir storage and restricted downstream condition. According to hindcasting closed-loop experiment results, TB-MPC outperforms the deterministic counterpart in terms of decreased downstream flood risk according to different independent forecast scenarios. TB-MPC was also tested considering different number of tree branches, forecast horizons, and different inflow conditions. We conclude that using synthetic PSFs in TB-MPC can provide more robust solutions against forecast uncertainty by resolution of uncertainty in trees.

Using Meteorological Analogues for Reordering Postprocessed Precipitation Ensembles in Hydrological Forecasting

AbstractMeteorological ensemble forecasts are nowadays widely used as input of hydrological models for probabilistic streamflow forecasting. These forcings are frequently biased and have to be statistically postprocessed, using most of the time univariate techniques that apply independently to individual locations, lead times and weather variables. Postprocessed ensemble forecasts therefore need to be reordered so as to reconstruct suitable multivariate dependence structures. The Schaake shuffle and ensemble copula coupling are the two most popular methods for this purpose. This paper proposes two adaptations of them that make use of meteorological analogues for reconstructing spatiotemporal dependence structures of precipitation forecasts. Performances of the original and adapted techniques are compared through a multistep verification experiment using real forecasts from the European Centre for Medium‐Range Weather Forecasts. This experiment evaluates not only multivariate precipitation forecasts but also the corresponding streamflow forecasts that derive from hydrological modeling. Results show that the relative performances of the different reordering methods vary depending on the verification step. In particular, the standard Schaake shuffle is found to perform poorly when evaluated on streamflow. This emphasizes the crucial role of the precipitation spatiotemporal dependence structure in hydrological ensemble forecasting.

Utilizing Probabilistic Downscaling Methods to Develop Streamflow Forecasts from Climate Forecasts

Abstract Statistical information from ensembles of climate forecasts can be utilized in improving the streamflow predictions by using different downscaling methods. This study investigates the use of multinomial logistic regression (MLR) in downscaling large-scale ensemble climate forecasts into basin-scale probabilistic streamflow forecasts of categorical events over major river basins across the U.S. Sun Belt. The performance of MLR is then compared with the categorical forecasts estimated from the traditional approach, principal component regression (PCR). Results from both cross validation and split sampling reveal that in general, the probabilistic categorical forecasts from the MLR model have more accuracy and exhibit higher rank probability skill score (RPSS) compared to the PCR probabilistic forecasts. MLR forecasts are also more skillful than PCR forecasts during the winter season as well as for basins that exhibit high interannual variability in streamflows. The role of ensemble size of precipitation forecasts in developing MLR-based streamflow forecasts was also investigated. Because of its simplicity, MLR offers an alternate, reliable approach to developing categorical streamflow forecasts.

Multi-parametric variational data assimilation for hydrological forecasting

Abstract Ensemble forecasting is increasingly applied in flow forecasting systems to provide users with a better understanding of forecast uncertainty and consequently to take better-informed decisions. A common practice in probabilistic streamflow forecasting is to force deterministic hydrological model with an ensemble of numerical weather predictions. This approach aims at the representation of meteorological uncertainty but neglects uncertainty of the hydrological model as well as its initial conditions. Complementary approaches use probabilistic data assimilation techniques to receive a variety of initial states or represent model uncertainty by model pools instead of single deterministic models. This paper introduces a novel approach that extends a variational data assimilation based on Moving Horizon Estimation to enable the assimilation of observations into multi-parametric model pools. It results in a probabilistic estimate of initial model states that takes into account the parametric model uncertainty in the data assimilation. The assimilation technique is applied to the uppermost area of River Main in Germany. We use different parametric pools, each of them with five parameter sets, to assimilate streamflow data, as well as remotely sensed data from the H-SAF project. We assess the impact of the assimilation in the lead time performance of perfect forecasts (i.e. observed data as forcing variables) as well as deterministic and probabilistic forecasts from ECMWF. The multi-parametric assimilation shows an improvement of up to 23% for CRPS performance and approximately 20% in Brier Skill Scores with respect to the deterministic approach. It also improves the skill of the forecast in terms of rank histogram and produces a narrower ensemble spread.

Design and implementation of an operational multimodel multiproduct real-time probabilistic streamflow forecasting platform

Abstract The task of real-time streamflow monitoring and forecasting is particularly challenging for ungauged or sparsely gauged river basins, and largely relies upon satellite-based estimates of precipitation. We present the design and implementation of a state-of-the-art real-time streamflow monitoring and forecasting platform that integrates information provided by cutting-edge satellite precipitation products (SPPs), numerical precipitation forecasts, and multiple hydrologic models, to generate probabilistic streamflow forecasts that have an effective lead time of 9 days. The modular design of the platform enables adding/removing any model/product as may be appropriate. The SPPs are bias-corrected in real-time, and the model-generated streamflow forecasts are further bias-corrected and merged, to produce probabilistic forecasts that are computed via several model averaging techniques. The platform is currently operational in multiple river basins in Africa, and can also be adapted to any new basin by incorporating some basin-specific changes and recalibration of the hydrologic models.

Evaluation of Probabilistic Streamflow Forecasts Based on EPS for a Mountainous Basin in Turkey

When designing water structures or managing a watershed it is a challenging task to determine the response of a basin to storm and/or snowmelt. In this study, the Upper Euphrates Basin (10,275 km2 area and elevation range of 1125-3500 m) located at the headwater of Euphrates River, one of Turkey's most important rivers, is selected as the application area. In this region, snowmelt runoff constitutes approximately 2/3 in volume of the total yearly runoff, therefore, runoff modeling and forecasting during spring and early summer is important in terms of energy and water resources management. The aim of the study is to make a forward-oriented, medium-range flow forecasting using Ensemble Prediction System (EPS) which is a pioneer study for Turkey. Conceptual hydrological model HBV, which has a common usage in the literature, is chosen to predict streamflows. According to the results, Nash-Sutcliffe model efficiencies are 0.85 for calibration (2001-2008) and 0.71 for validation (2009-2014) respectively. After calibrating/validating the hydrologic model, EPS data including 51 different combinations produced by ECMWF is used as probability based weather forecasts. Melting period during March-June of 2011 is chosen as the forecast period. The probabilistic skill of EPS based hydrological model results are analyzed to verify the ensemble forecasts.

Monthly streamflow forecasting using Gaussian Process Regression

Streamflow forecasting plays a critical role in nearly all aspects of water resources planning and management. In this work, Gaussian Process Regression (GPR), an effective kernel-based machine learning algorithm, is applied to probabilistic streamflow forecasting. GPR is built on Gaussian process, which is a stochastic process that generalizes multivariate Gaussian distribution to infinite-dimensional space such that distributions over function values can be defined. The GPR algorithm provides a tractable and flexible hierarchical Bayesian framework for inferring the posterior distribution of streamflows. The prediction skill of the algorithm is tested for one-month-ahead prediction using the MOPEX database, which includes long-term hydrometeorological time series collected from 438 basins across the U.S. from 1948 to 2003. Comparisons with linear regression and artificial neural network models indicate that GPR outperforms both regression methods in most cases. The GPR prediction of MOPEX basins is further examined using the Budyko framework, which helps to reveal the close relationships among water-energy partitions, hydrologic similarity, and predictability. Flow regime modification and the resulting loss of predictability have been a major concern in recent years because of climate change and anthropogenic activities. The persistence of streamflow predictability is thus examined by extending the original MOPEX data records to 2012. Results indicate relatively strong persistence of streamflow predictability in the extended period, although the low-predictability basins tend to show more variations. Because many low-predictability basins are located in regions experiencing fast growth of human activities, the significance of sustainable development and water resources management can be even greater for those regions.

CFSv2-Based Seasonal Hydroclimatic Forecasts over the Conterminous United States

AbstractThere is a long history of debate on the usefulness of climate model–based seasonal hydroclimatic forecasts as compared to ensemble streamflow prediction (ESP). In this study, the authors use NCEP's operational forecast system, the Climate Forecast System version 2 (CFSv2), and its previous version, CFSv1, to investigate the value of climate models by conducting a set of 27-yr seasonal hydroclimatic hindcasts over the conterminous United States (CONUS). Through Bayesian downscaling, climate models have higher squared correlation R2 and smaller error than ESP for monthly precipitation, and the forecasts conditional on ENSO have further improvements over southern basins out to 4 months. Verification of streamflow forecasts over 1734 U.S. Geological Survey (USGS) gauges shows that CFSv2 has moderately smaller error than ESP, but all three approaches have limited added skill against climatology beyond 1 month because of overforecasting or underdispersion errors. Using a postprocessor, 60%–70% of probabilistic streamflow forecasts are more skillful than climatology. All three approaches have plausible predictions of soil moisture drought frequency over the central United States out to 6 months, and climate models provide better results over the central and eastern United States. The R2 of drought extent is higher for arid basins and for the forecasts initiated during dry seasons, but significant improvements from CFSv2 occur in different seasons for different basins. The R2 of drought severity accumulated over CONUS is higher during winter, and climate models present added value, especially at long leads. This study indicates that climate models can provide better seasonal hydroclimatic forecasts than ESP through appropriate downscaling procedures, but significant improvements are dependent on the variables, seasons, and regions.

A Bayesian approach to probabilistic streamflow forecasts

One-month-ahead streamflow forecasting is important for water utilities to manage water resources such as irrigation water usage and hydropower generation. While deterministic streamflow forecasts have been utilized extensively in research and practice, ensemble streamflow forecasts and probabilistic information are gaining more attention. This study aims to examine a multivariate linear Bayesian regression approach to provide probabilistic streamflow forecasts by incorporating gridded precipitation forecasts from climate models and lagged monthly streamflow data. Principal component analysis is applied to reduce the size of the regression model. A Markov Chain Monte Carlo (MCMC) algorithm is used to sample from the posterior distribution of model parameters. The proposed approach is tested on gauge data acquired during 1961–2000 in North Carolina. Results reveal that the proposed method is a promising alternative forecasting technique and that it performs well for probabilistic streamflow forecasts.

Data mining methods for hydroclimatic forecasting

Skillful streamflow forecasts at seasonal lead times may be useful to water managers seeking to provide reliable water supplies and maximize hydrosystem benefits. In this study, a class of data mining techniques, known as tree-structured models, is investigated to address the nonlinear dynamics of climate teleconnections and screen promising probabilistic streamflow forecast models for river–reservoir systems. In a case study of the Lower Colorado River system in central Texas, a number of potential predictors are evaluated for forecasting seasonal streamflow, including large-scale climate indices related to the El Niño–Southern Oscillation (ENSO), Pacific Decadal Oscillation (PDO), North Atlantic Oscillation (NAO), and others. Results show that the tree-structured models can effectively capture the nonlinear features hidden in the data. Skill scores of probabilistic forecasts generated by both classification trees and logistic regression trees indicate that seasonal inflows throughout the system can be predicted with sufficient accuracy to improve water management, especially in the winter and spring seasons in central Texas.