Development of a Water Temperature Modeling Platform to Support Short- and Long-Term Water Temperature Management in Reservoir–River Systems

Abstract

The U.S. Department of the Interior, Bureau of Reclamation (Reclamation) supports water temperature management for fishery species protection in downstream river reaches below Central Valley Project (CVP) reservoirs in the Sacramento, American, and Stanislaus River systems. The Water Temperature Modeling Platform (WTMP) Project was initiated to modernize and enhance modeling capabilities to predict summer–fall water temperature through reservoir cold water pool management using temperature control facilities designed for temperature management. The WTMP supports forecasts, historical analyses, and long-term planning efforts and advances previous modeling approaches by using an integrated modeling platform. This platform includes a data management system that acquires real-time data, provides quality assurance methods, and yields model-ready data for simulations; a modeling framework that manages model input file construction for multiple models, controls selected model simulation for reservoir and/or river reaches, and manages model output; and an automated reporting feature providing efficient and comprehensive reporting of tabular and graphical output for assessment and analysis by technical teams and decision-makers. The WTMP takes advantage of technological advancements in simulation models, available software, and databases to support Reclamation’s short- and long-term water temperature management needs. The platform is also adaptive for future integration with new or improved models and tools.

1. Introduction

Water temperature management is an increasingly critical resource management activity, particularly in the western United States where habitat for cold water fishes in many rivers has been restricted due to construction of dams and associated reservoir operations [1]. Temperature regulation in downstream river reaches often includes careful management of the cold water stored in the reservoir through selective withdrawal to achieve downstream temperature objectives throughout a period (e.g., summer and early fall) [2,3]. Reservoir and river simulations models that represent hydrology, operations, meteorology, and water temperature are invaluable tools to assist resource managers in meeting downstream habitat objectives. Such water temperature management modeling tools have been used in supporting the U.S. Department of the Interior, Bureau of Reclamation (Reclamation) Central Valley Project (CVP) operation and planning for decades. Furthermore, temperature management processes and activities in the CVP have evolved over time, reflecting the changing needs of regulatory and legal requirements and societal preferences. Concurrently, improved understanding of the relationships among physical conditions and biological responses of protected species, project operations, and the influence of other factors, coupled with technological improvements (e.g., computational speed and efficiency, improved software, inexpensive real-time monitoring) have occurred.

Computer models representing flow and temperature in river and reservoir systems [4,5,6,7,8,9,10,11,12] have been used for decades to manage water temperature [13]. The Water Temperature Management Platform (WTMP) is an initiative that Reclamation has pursued to improve and modernize the quantitative, science-based tools to support CVP operations. Such tools support evaluation of how CVP operational decisions, hydrology, meteorology, and other factors can affect water temperature in reservoirs (i.e., cold water storage) and downstream river reaches, as well as potential impacts to downstream fishery species sensitive to water temperature.

In a deviation from past modeling practices in the CVP system, the WTMP aims to advance previous modeling approaches using an integrated modeling framework that includes (1) a data management system (DMS) that acquires real-time data and allows staff to provide timely quality assurance to develop model-ready data; (2) a modeling framework that manages construction of model input files (from model-ready data), controls selected model simulation for identified reservoir and/or river reaches, and manages model output for tabular and graphical output assessment and analysis; and (3) an automated reporting feature to provide efficient and comprehensive reporting to technical teams and decision-makers. The WTMP was developed in a transparent and open process using a modeling technical committee (MTC), with diverse practitioners, stakeholders, and interested parties, extensive documentation covering all phases and aspects of model and platform development, and independent review. The WTMP was developed and implemented to create robust communications methods, including latest developments and knowledge, and establish a common understanding and expectations of the resulting tools using the following principles:

• A focus on technical improvement to advance water temperature modeling tools and analytical methods;
• Use of a collaborative model development approach with stakeholders and interested parties;
• Maintaining an open and transparent environment for information sharing and cooperation.

The WTMP is designed to be a rigorous yet flexible toolbox for housing and implementing different temperature management modeling tools and utilities with confidence to support consistent high-quality, inclusive, and transparent analyses to meet Reclamation’s evolving needs, regulatory changes, and technological advancements.

1.1. Background and Project Area

Reclamation operates the CVP in coordination with the California Department of Water Resources (DWR) State Water Project (SWP) (Figure 1). Reclamation and DWR operate in accordance with water right permits and licenses issued by the California State Water Resources Control Board (SWRCB) to appropriate water by directly diverting to use or by diverting to storage and re-diverting releases from storage later in time. As conditions of their water right permits and licenses, the SWRCB requires the CVP and SWP to meet specific water quality, quantity, and operational criteria, including environmental protection [14,15]. Environmental protection is served by several objectives established to assist in stabilizing and improving species/habitat in the tributaries managed by the CVP, with temperature management a key consideration for the protection of species with specific cold-water needs. For example, winter-run Chinook salmon, a listed species under the Endangered Species Act (ESA), has been one of the species impacted by water temperature management challenges associated with the limited supply of water on the Sacramento River, particularly during recent drought periods [16,17].

Environmental goals are achieved, in part, using physically based computational modeling frameworks/tools specifically designed to incorporate hydrology and operations, meteorology, water temperatures, installed facilities and infrastructure, and temperature management objectives [16,17]. Computational models support the development of strategies for water temperature management that allow reservoir operators with limited resource supply (total storage and cold-water storage), to effectively plan, forecast, and operate storage and conveyance systems to meet a wide range of water supply demands while limiting impacts to aquatic species sensitive to temperature [15]. A conceptual representation and description of the multiple elements, activities, and conditions in managing temperature below reservoirs is shown in Figure 2.

Temperature objectives are explicitly considered in the Sacramento [14], Trinity [18], and American River [16] systems. Furthermore, models are explicitly identified as preferred tools to meet temperature requirements in the National Marine Fisheries Service (NMFS) 2019 Biological Opinion (BO) for forecasting seasonal temperature conditions, cold water pool management, and adaptive management [17]. The BO states that the temperature modeling platform “that Reclamation is proposing to consider as a possible Cold Water Management Tool would advance a tool that could provide a more accurate characterization of reservoir temperature conditions and contribute to more efficient use of [the] available cold-water pool, improved temperature conditions, and likely increased species protections” [17] (p. 267).

Water temperature modeling has a long history in the Central Valley Project region, with initial applications in the 1970s and 1980s [19,20,21]. The increasingly widespread use of personal computers in the late 1980s and early 1990s, and the advent of inexpensive, accurate, and reliable remote temperature monitoring technology provided the opportunity not only to model water temperature in a wide range of stream and lake environments, but also to populate and test models with extensive datasets. By the late 1990s and early 2000s most of the Sacramento River [13,22,23,24,25,26,27] and Trinity River basins [28,29,30,31], including Clear Creek [32,33], had been modeled for flow and temperature. Similar efforts occurred in the American River [21,34] and in the Stanislaus River [35,36,37]. With continued increases in computational power, storage, and software improvements, expanding access to high-resolution spatial and temporal datasets (including the availability of economical, high-resolution bathymetric data), and increased expertise in modeling, the models have continued to be expanded and refined in the Sacramento and Trinity [12,38,39,40,41], American [42,43], and Stanislaus River [44] basins, and modeling frameworks began to appear [12,26]. These frameworks ranged from representing the system as a system model (one model representing all system components, e.g., rivers, reservoirs, tunnels) [26] to linked models that may represent different systems with different models (e.g., one model for a reservoir and a different model for a river reach) [12]. Model development that started off decades ago with minimal computational capabilities and a small number of users has expanded to include powerful tools and supporting software used by a community of modelers. Concurrently with these advances, there has been a remarkable expansion of simulation analyses, with extended calibration periods, historical reanalysis (simulation of a past period), forecasts, and long-term planning; the latter including Monte Carlo, position analysis, ensemble analysis, and others.

The development of the WTMP is a logical extension of this long history of model evolution and development. Furthermore, this effort was able to incorporate contemporary approaches in developing decision support tools including the latest models and software; utilizing a modeling framework; improving data acquisition, management, and tracking through a data management system; automating modeling tasks (e.g., development of model boundary conditions, simulation, automated reporting) to improve overall analysis efficiency and reduce error; and providing a means to explore model uncertainty in forecasts, historical reanalysis, and long-term operations analyses. This unique combination of WTMP model development elements included a communications and outreach plan that explicitly involved stakeholders in all project activities [45,46,47]. Another layer of review was an external peer review [48,49] that included two-phase processes, wherein an interim peer review provided invaluable mid-term feedback and “course correction” prior to a final project peer review before project completion.

1.2. Project Goals and Objectives

Development of the WTMP was based on well-defined goals, objectives, and principles. The goal of developing a WTMP was to modernize the analytical tools that Reclamation uses to support activities and decision-making for water temperature management. To support this goal, WTMP capabilities included improved summer and fall water temperature prediction, the ability to address long-term planning efforts, and included effective performance measure reporting.

The WTMP objective set to achieve these goals was to develop and implement water temperature models and associated tools for the Sacramento-Trinity, American, and Stanislaus River systems with the following requirements:

• That developers conform to professional standards of care in analytical tool development and applications for reservoir–river system water temperature management;
• That models should be used consistently for both CVP real-time operations, seasonal and long-term planning purposes;
• That WTMP use be flexible to accommodate future technologic advancements in analytical modeling for reservoir–river system water temperature management.

Ultimately, the elements of a successful WTMP include an effective and efficient database capable of managing data and data quality, calibrated models that reproduce temperature dynamics (e.g., stratification dynamics, cold water pool evolution, river heating rates), and a framework that automates repetitive tasks to minimize errors (e.g., data transfer, data reporting) and improve analysis proficiency.

2. Methods

Development of the WTMP project occurred in four stages:

• Review and selection of modeling framework and temperature models;
• Development of a data management approach;
• Data development, model calibration, validation, and sensitivity analysis for the Sacramento-Trinity, American, and Stanislaus River systems;
• Model application and exploration of model uncertainty.

All stages included active stakeholder outreach and incorporated Reclamation knowledge transfer. While listed in serial order, the above process necessarily included iteration and revisiting of previous decisions and approaches: a necessary and expected process in the development of data and models for complex reservoir–river systems in a collaborative environment.

2.1. Review and Selection of Modeling Framework and Water Temperature Models

Models of large complex reservoir–river systems have been developed for a wide range of applications [50,51,52,53]. Reservoir and river reaches can be modeled with a modeling system as an interconnected network [4,54,55], or as discrete components with individual models for each reservoir or river reach [56]. A framework is a software application that provides a means to represent reservoir–river systems as a suite of linked but discrete models that can be used to streamline model use and automate repetitive tasks, making the modeling process more efficient and more robust [26,43,44,48,49,50]. For the WTMP, there was a need for both high-resolution discrete reservoir and/or river element models with detailed representations and a modeling system that can accommodate system-wide representations in a computationally efficient manner.

2.1.1. Modeling Framework Review and Selection

Framework selection consisted of reviewing basic framework modeling processes and system architecture, identifying requirements for a WTMP framework, and developing framework selection criteria. A set of evaluation criteria was developed based on Reclamation’s objectives for development of the WTMP. Implementation requirements listed below consider the required framework elements and capabilities as well as how models are used, what data is required, information flow, input and output reporting, management of model versions, how new models are introduced to the framework, and other factors. To support Reclamation’s WTMP, a framework should be able to:

• Efficiently use several models, individually or in a sequence, or use in concert with a system model;
• Support workflows for several typical modeling activities;
• Utilize common boundary conditions and operational controls across models;
• Manage modeling scenarios for record keeping and reference/reuse;
• Create reports using common report formats across models;
• Provide version control of model executable programs and configuration datasets;
• Allow for incorporation of new modeling tools;
• Focus on the efficiency of production modeling activities.

Framework evaluation criteria were developed based on project implementation goals and objectives; i.e., required or necessary framework capabilities. The criteria generally pertain to model support, data management/communications, user interface, and software installation and configuration. Criteria were organized into eight categories. Reclamation assessed criteria on a relative scale as high, medium, or low importance based on project objectives, and then examined the global applicability of each framework. Although all the evaluation criteria were considered in decision-making, meeting all criteria was not necessary. The eight general types of framework selection criteria are listed below, and the full list of criteria and their assigned priorities are listed in Table S1. Specific attributes of criteria are outlined in the WTMP framework selection technical memorandum [57].

• Models or processes utilized in the WTMP—Reservoir and river models, types of models, sources and format of data;
• Model Coupling—Type of model coupling, or interaction, utilized by a potential model framework;
• Workflow Control for Sequenced Model Simulation—Workflow control used for sequenced model simulation (input/output, iteration, ensemble analysis, etc.);
• Model Configuration and Time Series Data Management—Configuration of data communications, data formats, simulation periods, alternative configurations, results posting;
• User Interface Capabilities—Model parameter editing, simulation control, plotting and reporting, and other activities;
• Location of Model and Framework Configurations and Time Series Data Storage—Where model and framework configuration and time series data are stored (e.g., desktop workstation, local server, cloud server);
• Location of Computations—Where computations are performed (e.g., desktop workstation, local server, cloud server);
• Type of software application used by model operators—Desktop application or web application.

The project team identified a range of potential modeling frameworks for WTMP consideration [58]. The HEC-WAT (version 1.1) framework [59] was selected and developed to automate the many modeling activities to reduce the potential for errors, provide common boundary conditions for the models, and to ease model application and interpretation of output. HEC-WAT is a flexible and extendable modeling framework that has already supported basic simulation with CE-QUAL-W2 and ResSim. HEC-WAT includes a software plug-in application program interface (API) that allows for customization of the user interface and addition of computational features needed for the WTMP. Among those additional computational features is an iterative simulation for sensitivity or ensemble analysis and the scripted pre-processing of boundary conditions needed to drive simulations.

Each numerical model has specific requirements for input and output data files. The HEC-WAT model framework software organizes modeling data in a directory hierarchy on the modeling workstation that accommodates the requirements of each numerical model used in the framework [58,60].

A modeling framework implemented with HEC-WAT has been installed as a desktop application on workstations used by the Reclamation modeling team. This approach is expected to require less direct IT support than might be required by an enterprise modeling architecture. The framework interacts with the DMS to extract time-dependent data, post key modeling results, and access and update model executable programs and configuration files. Finally, production modeling, result processing, and report generation are managed by the modeling framework on team workstations.

2.1.2. Water Temperature Model Review and Selection

Model selection criteria are typically used to assess a range of potential models to meet specific project objectives and needs [58,61]. For the WTMP project, this translated to selecting models that provide realistic predictions of reservoir and downstream river water temperatures with sufficient confidence to support seasonal planning and real-time applications while also assessing uncertainty in the Sacramento-Trinity, American, and Stanislaus River basins. Additionally, selected models needed to be specific to CVP operations and environmental needs associated with temperature management and regulatory considerations. Required model elements and capabilities considered in model selection included (but were not limited to):

• Reservoir or river representation;
• Model capabilities (such as flexibility to represent selective withdrawal, submerged dams, or temperature control curtains for reservoir components) and model performance;
• Model spatial and temporal scales and data needs;
• Ability to interface model with other models in a framework;
• Resources required to develop and maintain a model and cost assessment (initial cost or annual maintenance fee);
• Active model support, access to the principal code author and/or open-source code (allowing review and modification), and comprehensive documentation and training available;
• User interface for input file quality control;
• Post-processors available.

Effective model development encompasses many factors, including identifying questions to be addressed through modeling as well as identifying available information (system configuration, boundary conditions, calibration data), period of analysis, spatial extent, and other data [62]. This information, as well as other literature [58,63,64,65] and professional experience were used to select a series of criteria to assist Reclamation in selecting component models (for discrete reservoirs and selected river reaches) as well as a system model that represents a reservoir–river network.

Model criteria were grouped into six subcategories. Reclamation assessed criteria on a relative scale as high, medium, or low importance based on project objectives, then examined the global applicability of each model. Although all the evaluation criteria were considered in decision-making, meeting all criteria was not necessary. The six general types of model selection criteria are listed below, and the full list of criteria and their assigned priorities are listed in Table S2. Specific attributes of criteria are outlined in the WTMP model selection technical memorandum [66].

• Numerical Model Criteria—Numerical representation of the physical system in a model;
• Model Linkage Capability—Addresses whether models are WTMP compatible and if models are discrete (reach-specific) or system-wide;
• Model Input and Output Capabilities—Model pre- and post-processing capability and I/O data structures;
• Model Support—Addresses whether the model is supported by the developer or some other entity;
• Representation or Parameterization of Current or Planned CVP Facilities—Ability to represent specific features of the CVP;
• Qualitative Modeling Elements—Represent a range of desired non-quantitative attributes desired CVP modeling needs (e.g., ease of use).

The project team identified a range of potential temperature models for WTMP consideration. Through an extensive review process, the U.S. Army Corp of Engineers one-dimensional HEC ResSim (ResSim) (currently using version 4.0) system model for reservoir (vertical representation) and river (longitudinal representation) reaches [4,67,68,69], and the two-dimensional CE-QUAL-W2 model (currently using version 4.5 and version 5.0 (Shasta Lake only)) [6,7,70,71,72] for higher-resolution representation of reservoirs were selected.

CE-QUAL-W2 is an established, widely used reservoir water quality model. This 2D-branched laterally averaged flow and water quality transport model is well documented [6,7,70,71,72], is familiar to other agencies and stakeholders, and can represent detailed outlet works, temperature control devices, and unique features (e.g., temperature control curtains and submerged dams). HEC-ResSim is an established model for river–reservoir simulation that has been recently enhanced to include water quality simulation (including water temperature) and is a successor to the legacy HEC5Q model that has historically been used by Reclamation for temperature management activities. ResSim uses one-dimensional vertical layered representation for reservoirs and one-dimensional longitudinal representation for rivers. ResSim uses the same detailed heat exchange formulation as CE-QUAL-W2 for temperature simulation and has similar abilities to represent detailed outlet works and temperature control devices, including specific logic parameterizing the Shasta Dam temperature control device (TCD) and Folsom Dam shutter system. Key differences between ResSim and CE-QUAL-W2 are representation of entrainment of inflow into the reservoir from tributaries and longitudinal transport, which are explicitly simulated in CE-QUAL-W2 and empirically represented in ResSim. ResSim and CE-QUAL-W2 water quality models were selected as the primary fast-computational speed and detailed representations, respectively. Additional details on the model selection process are presented in Reclamation’s model selection technical memorandum [66].

2.2. Data Management Approach

Data management serves several important purposes in managing information for analysis and modeling [73,74]. The implementation of a successful data management system (DMS) reduces the time and labor required to assemble the high-quality datasets that are used for model input [75,76]. The DMS also provides an inventory system for adding new data as it becomes available. Data management activities include creating a system for data storage; identifying necessary data and data sources; establishing procedures for data acquisition; instituting measures for data quality analysis and data quality control (consistent metadata definitions and formats); gap filling; and other tasks. A database approach facilitates consistent application of data management rules and establishes hierarchical relationships that make the inventory and reporting process more efficient [77]. As new data sources become available, they can be integrated quickly and efficiently either as new data or as replacements for existing data (e.g., updating provisional with approved data). In addition to these important elements of data management, a database to support the WTMP is used to develop and export model-ready data to support models, reducing the potential for error, tracking data/metadata sources for model inputs, providing common boundary conditions for models in the WTMP, and similar tasks that support the goals, objectives, and principals of the project.

The WTMP DMS utilizes a suite of tools to collect and manage water resources and environmental data from a wide range of sources including United States Geological Survey (USGS), California Data Exchange Center (CDEC), Reclamation’s Hydrogeologic Assessment Report (HAR), and others. The system consists of enterprise-level software components, meaning individual components are deployed within an organization and communicate at high speed with each other over one or more networks. DMS architecture organizes applications into three logical and physical computing tiers (Figure 3):

• The data layer, where the data associated with the application is stored and managed;
• The application layer, where data is processed;
• The presentation layer, or user interface.

The data layer includes two Microsoft (MS) SQL Server databases used for data storage and credentialization. The data storage database includes all project data and associated metadata to support the WTMP. The credentialization database defines and manages role-based access credentials to specific datasets and applications.

The application layer defines the computer code that acts on datasets in the database(s) in a rules-based manner to transfer data from database tables to a specific application. For example, the statements or procedures use information in the database to deliver data to a specific tool the user is credentialed to use (i.e., secure transfer).

The presentation layer is where most users, as well as automated processes, interact with the data. Access to the DMS occurs via a web browser (web interface). Web services run in the background to facilitate exchange between multiple devices, acquiring data from external sources or delivering model ready data to the WTMP. Additional hardware and software specifications are included in Reclamation documentation [78].

2.3. Model Development

Model development occurs in several stages and includes data development, model implementation, calibration, and validation. Required data include geometric information, hydrology and operations, water temperature, and meteorology [79,80]. Model implementation comprises steps to construct the numerical model representations of each system. Subsequently, models are calibrated and undergo sensitivity testing of boundary conditions, individual model parameters, specific operations, or other model assumptions. Model calibration, validation, and sensitivity analysis followed standard modeling practices [62]. Parameters that were identified in the testing and sensitivity analysis as important to model response were modified within typical ranges identified in user guides and literature [26,67,72]. Details on calibration approach and results are included in the WTMP model development technical memorandum [79]. Furthermore, previous modeling efforts throughout the reservoir and river systems provided guidance and insight into model parametrization [26,31,36]. The approach to calibration integrated graphical and statistical assessments, model performance metrics, and temperature signatures as discussed below.

2.3.1. Graphical Assessment

Graphical assessment provides an effective qualitative measure of model performance. Both hourly and daily average (based on hourly data) are considered. Comparisons of simulated and observed values provide useful insight to both the analyst and decision-maker, providing a visual assessment of model performance that can be difficult to discern with statistics alone. Basic elements that can be easily conveyed in a graphical comparison include phase, magnitude, response to short (days) or long events (seasonal), and frequency of similarities or differences. Furthermore, short-term (weeks or month) or long-term model (annual or longer) bias, and other statistical measures, are often apparent in comparative graphics. Finally, graphical presentation of results, including multi-axis, multidimensional plots, and animations can provide remarkable insight to assist in conveying information to analysts, decision-makers, and stakeholders.

2.3.2. Statistical Assessment

Statistical assessment provides a quantitative measure of model performance. The selection and use of performance criteria should be sufficiently broad to provide an effective interpretation of results because rarely is one error measure sufficient [81,82,83,84]. Quantitative assessment of model performance for WTMP models included mean bias, mean absolute error (MAE), root mean-squared error (RMSE) and Nash–Sutcliffe (NSE) efficiency coefficient (Equations (1)–(4), respectively).

$$\mathrm{Mean} \mathrm{Bias}, \mathsf{\epsilon }={\displaystyle \frac{1}{n}}{\sum }_{i=1}^n\left(Xsi{m}_i-Xmea{s}_i\right)$$

(1)

$$\mathrm{MAE}={\displaystyle \frac{1}{n}}{\sum }_{i=1}^n\left|Xsi{m}_i-Xmea{s}_i\right|$$

(2)

$$\mathrm{RMSE}=\sqrt{{\displaystyle \frac{{\sum }_{i=1}^n{\left(Xsi{m}_i-Xmea{s}_i\right)}^2}{n}}}$$

(3)

$$\mathrm{NSE}=1-{\displaystyle \frac{{\sum }_{i=1}^n{\left(Xsi{m}_i-Xmea{s}_i\right)}^2}{{\sum }_{i=1}^n{\left(Xmea{s}_i-\overline{{Xmeas}_i}\right)}^2}}$$

(4)

where Xsim is simulated data, Xmeas is measured data, $X\overline{meas}$ is the mean of measured data, n is sample size, and i is summation index.

These selected metrics represent bias (mean bias), absolute error (MAE and RMSE), and goodness-of-fit (NSE) measures, providing a robust means to assess and quantify model performance. Specifically, mean bias provides information relating to systematic model over- or under-prediction. MAE is the average of the absolute value of the bias of paired observations and simulated values and provides an estimate of overall model error. RMSE is a function of the square of the difference between the paired observations and simulated values, and large values indicate that there are periods where model output and field observation differences are appreciable. Finally, the Nash–Sutcliffe efficiency (NSE) is an indication of how well the plot of observed versus simulated data fits the 1:1 line; i.e., a goodness-of-fit parameter [84,85,86,87,88]. Not all statistics yield equally insightful information [89] or are applicable in certain conditions.

2.3.3. Metrics

Model performance metrics associated with each statistic for water temperature, flow, and stage were developed (

Table 1. Model performance metrics for hourly water temperature, flow, and reservoir stage in WTMP models.

Parameter	Mean Bias	MAE	RMSE	NSE
Stage	±0.15 m	≤0.3 m	≤0.45 m	≥0.65
Flow	±4.2 cms	≤8.4 cms	≤14.2 cms	≥0.65
Water Temperature	±0.75 °C	≤1.0 °C	≤1.5 °C	≥0.65

) considering how the WTMP was to be used in decision-making. These metrics considered regulatory temperature targets or objectives in the project domain [14,15,16,17,18], as well as other factors including model structure and process representations, typical data accuracy, and experience. The range of meteorology, hydrology, and operational conditions in the 2000 through 2021 period presents a wide range of conditions for calibration assessment. The objective was to fit all years with a common set of assumptions and calibration parameters (i.e., not changing assumptions and calibration parameters year to year) for each system. Validation years also used the same metrics and included water year types that provided an appropriate range of hydrology, meteorology, and operational conditions.

2.3.4. Specific Temperature Signatures for Reservoirs and Rivers

Based on the peer panel review feedback for the WTMP project [48], specific temperature signatures were considered for reservoirs and rivers to assist in model calibration and improve model performance. Reservoirs included in the WTMP generally fall into three categories:

• Large, long residence time, seasonally stratified impoundments (Shasta Lake, Trinity Lake, Folsom Lake, New Melones Lake)—Calibration objectives for these reservoirs include matching both the simulated thermal profiles and outflow temperatures throughout the year. Capturing both processes simultaneously provides a means to simulate stratification dynamics, cold water pool volumes, temperature management operations, and outflow temperatures.
• Small, short residence time impoundments (Keswick Reservoir, Lewiston Lake, Lake Natoma)—These reservoirs are isothermal or experience weak intermittent stratification. Thus, the calibration objective for small reservoirs is to match outflow temperatures.
• Relatively modest volume and residence time impoundments (Whiskeytown Lake, Tulloch Lake)—These impoundments are directly influenced by seasonal upstream releases from large reservoirs yet are large enough (depth and volume) to stratify seasonably. As with large reservoirs, reproducing vertical temperature profiles and outflow temperatures simultaneously is the important calibration consideration.

Stream reaches in the WTMP domain include the Sacramento River, Trinity River, Clear Creek, American River, and Stanislaus River, all of which have a headwater boundary condition for flow and temperature that is modified by one or more upstream reservoirs. Simulating heat gain and sub-daily (e.g., hourly) response of releases from reservoirs is important to Reclamation’s temperature management activities in downstream river reaches. Thus, basic attributes used to assess model performance in rivers are longitudinal heating rates and diurnal variation and range.

2.4. Model Framework Development

The development of a modeling framework required the use of several interrelated components to effectively produce a modeling approach to meet the project objective, goals, and principles of the WTMP. The final WTMP product uses a graphical user interface (GUI), accesses model-ready data through the DMS, can run a single model or several linked models, and produces formatted reports (Figure 4). The WTMP manages simulations of multiple, linked models by passing upstream model output to downstream model input at an hourly time step (regardless of individual model computational time step).

3. Results

The final calibrated WTMP models [79] indicated the models perform well over a range of hydrology, operations, meteorology, and water temperature conditions as assessed using graphical and statistical analysis, performance metrics, and reservoir and river temperature signatures. The WTMP is available to assess a range of conditions consistent with Reclamation activities including forecasting, historical reanalysis, and long-term planning. For the purposes of the WTMP, these activities are defined as:

• Forecasting: Development of annual temperature management plans (TMPs) on the Sacramento and American Rivers. Producing the TMPs involves using forecast hydrology, operations, meteorology, inflow water temperatures, and initial reservoir temperature profiles to simulate seasonal forecasts in early spring through fall, and additional model simulations throughout the temperature management season.
• Historical Reanalysis: Simulating the previous temperature management season (e.g., May through October) with available measured data as model inputs to confirm and assess model performance.
• Long-term Planning: Completing long-term operation (LTO) analyses over many years to assess the implications of different temperature management options for a range of operations, hydrology, and meteorological conditions based on extended historical periods or long-term water resources planning models [90].

In addition, the WTMP can be used to complete iterative simulations, ensemble simulations, position analysis, evaluate risk and uncertainty, and sensitivity analysis—all important in Reclamation’s vision of temperature management activities.

3.1. Seasonal Forecasting

For CVP watersheds included in the WMTP, a seasonal water temperature forecast describes future expected downstream water temperature [2]. This forecast, or simulation of expected water temperature performance, is based on the temperature targets at the reservoir tail bay or at a location downstream as specified by the analyst. Water temperatures in reservoir and river system are forecast using simulation models that utilize estimated future flow, operations, and meteorology (and associated inflow water temperatures). Forecasts may be based on an array of potential conditions (e.g., 50% exceedance, 90% exceedance inflow forecast), desired downstream temperatures, as well as appropriate assumptions regarding the expected range of operations and meteorology.

The WTMP approach to temperature forecasting employs an ensemble simulation. A graphical user interface (GUI) facilitates selection and importing reservoir initial conditions, CVP monthly forecast spreadsheets, local three-month temperature outlook (L3MTO) or position analysis-type meteorologic data, and collections of temperature target time series. Data processing scripts within the WTMP are used to create daily flow and water temperature input time series based on forecast data, whereas in the past Reclamation relied on manual spreadsheet manipulations. Time series representing boundary conditions are automatically saved to DSS collection time series. The final step is for the user to select combinations of boundary conditions, initial conditions (e.g., reservoir temperature profiles), and downstream temperature targets for the forecast simulations. The WTMP then operates over the selected ensemble members, carrying out the simulations and organization results into DSS collections. The automated reporting capability can create summary graphics and tables based on the ensemble results for the simulation.

Simulated Shasta Dam release temperatures for a 1 May and 1 July forecast through temperature management season and simulated TCD gate levels selected by the model to meet these temperatures are shown in Figure 5 and Figure 6, respectively for 2014 and 2018. The year 2014 was a dry year and Shasta Lake did not fill sufficiently to access the upper TCD gate level. The outcome was that temperature management below 12.5 °C was infeasible after approximately 1 September 2018, which was a more typical year, and reservoir storage was sufficient to allow the upper TCD gate level to be employed, thus more efficiently using available cold water to maintain temperatures management below 12.5 °C throughout the season. The differences in temperature management operations of the TCD are apparent in the faster progression to successively lower TCD gate levels in 2014 as compared to 2018 (Figure 6). While these types of simulations have historically been used by Reclamation in the past to develop seasonal temperature management plans and managing temperatures through the management season [91,92], the WTMP improves analysis proficiency of the process, allowing for a more comprehensive assessment of potential outcomes, and eases viewing and interpretation of output.

3.2. Historical Reanalysis

A historical reanalysis is a retrospective simulation completed at the end of the temperature management season that uses the actual or observed input data (e.g., hydrology, operations, meteorology, inflow water temperatures) for the same period as the seasonal forecast. Thus, the simulation uses known boundary conditions versus forecast data (where historical data are not yet available for this simulation, forecast values are assumed). Because forecast conditions are estimates of a future condition and operations deviate from assumed conditions throughout the temperature management season, the historical reanalysis is not an assessment of the quality of a forecast. Rather the analysis provides a means to confirm model performance over the past year and gain insight into the system response and operations to meet temperature management objectives. In this manner the historical reanalysis is a useful exercise to improve all aspects of temperature management including forecasting skill, model performance, temperature management strategies and adjustments, decision-making and communications, and reporting.

In the WTMP, historical reanalysis simulations follow the same workflow as a historical period simulation wherein all boundary conditions, initial conditions, and temperature targets are known (e.g., Shasta Dam on the Sacramento River, Folsom Dam on the American River). Key elements of this assessment for Lake Shasta include comparing the initial forecast and historical reanalysis for Shasta Dam TCD operations (duration and timing of TCD gate levels) (Figure 7) and Shasta Dam temperature releases (Figure 8). Other parameters that are useful to assess include changes in cold water pool volumes throughout the season and downstream temperature conditions at important locations. Similar metrics are used in simulations to support temperature management on the American River.

3.3. Long-Term Operations

The WTMP long-term operations are intended to support planning simulations based on the CalSim 3 operations model. The CalSim 3 operations model represents California’s water system with a detailed computational network with the primary purpose “to evaluate CVP and SWP operations at current or future levels of development, with and without various assumed future facilities, various regulatory requirements, and with different facility management options.” [90] (p. 1-1). CalSim 3 reproduces monthly operations based on historic hydrology for the 100-year period from October 1921 through September 2021 [93]. Beyond monthly hydrology and operations, the WTMP requires inflow temperatures and meteorology. Inflow temperatures can be estimated using the current method based on flow and water temperatures [79,80]. However, the detailed long-term meteorological data required to simulate water temperature at a sub-daily time step have not been developed at this time. Herein an initial proof-of-concept simulation demonstrating the ability of the WTMP modeling framework to configure and carry out simulation utilizing CalSim 3 model outputs is presented for the period 2000–2019—the period of available meteorology in the current DMS. This example uses the Upper Sacramento System model, which includes Shasta, Keswick, Trinity, Lewiston, and Whiskeytown reservoirs, and the results presentation focuses on operation of the Shasta Lake temperature control device to meet a downstream temperature target, which is an intended use of the WTMP in long-term planning to support Reclamation’s various activities [94]. The WTMP utilized ResSim for long-term planning (vs. the computationally intensive CE-QUAL-W2 model), and the workflow for the process is outlined in Figure 9.

Assumptions and considerations for this analysis include:

• Flows for boundary conditions and operations at system reservoirs are derived from CalSim 3;
• Monthly average input hydrologic time series for operations are used;
• Historical meteorology data for the Trinity River and Sacramento River basins (hourly);
• Historical tributary inflows (daily) and inflow temperatures (daily) to the Trinity River downstream of Lewiston Dam and Sacramento River downstream of Keswick Dam.

The temperature target is specified as a constant 12.2 °C at the Sacramento River above Clear Creek temperature compliance location (Figure 1) from May 1 through October 31. In this example simulation, the Shasta Dam tail bay temperature requirement is based on an equilibrium temperature model using the downstream temperature target, flow, and meteorologic conditions. The Shasta TCD operation is determined by the ResSim scripted rule logic developed for seasonal temperature forecast [60]. The WTMP used the data identified above, the TCD ResSim scripted rule logic, and model linkage to simulate the entire Shasta–Trinity system of reservoirs and river reaches in a single simulation. Simulated long-term water temperature below Shasta Dam, Keswick Dam, and at the compliance locations in the Sacramento River above Clear Creek are shown in Figure 10, and reservoir isotherms through depth and time are presented in Figure 11. The model effectively represents both downstream and in-reservoir temperatures from 2000–2019. TCD operations are effectively represented in the WTMP, and downstream temperatures were met in all years except the drier years of 2014 and 2015, when carryover storage limited access to the upper TCD gate levels, hampering the ability to maintain the downstream temperature throughout the temperature management season.

Reclamation is currently planning the development of long-term meteorology and water temperature boundary conditions for the entire 100-year CalSim 3 period. Subsequently the simulation below will be updated to include the entire period. Furthermore, the WTMP is capable of long-term operations ensemble analysis based on different/multiple CalSim 3 simulations.

4. Discussion

The WTMP is a unique and powerful framework to significantly improve and modernize tools Reclamation uses to support CVP water temperature management activities. The platform represents a rigorous yet flexible toolbox for housing and implementing multiple temperature modeling tools and utilities for producing consistent, useful, and reliable results in an efficient manner. This project offers insight and direction on the review and selection of modeling frameworks and models for river–reservoir systems through objective-based selection criteria to ensure appropriate tools meet the temperature management challenges (e.g., unique infrastructure) and needs (e.g., spatial and temporal resolution, flow and temperature representations). Models are housed in a framework that facilitates all elements of analysis, from model selection to development of boundary conditions to simulation and reporting. All historical model input data is managed through the DMS, providing a platform for managing existing data, automatically adding new data, providing data quality analysis and quality control (QA/QC), and tools for user review and development of model-ready data. Model-ready data is exported to the WTMP on a common hourly time step, reducing the potential for error and providing metadata for tracking dataset QAQC and sources. Additionally, comprehensive model calibration and validation approaches include objective-based calibration targets, and multiple assessment techniques (graphical assessment, statistical metrics, and system “signatures”) that are applied to reservoir–river systems for large systems simulating multiple decades.

The WMTP provides a framework to explore and quantify uncertainty associated with model construct, data, and hydrology and meteorology forecasts through use of ensemble analysis of simulations with varied boundary conditions and/or forecast parameters. Automated reporting provides consistent and appropriate formats for both data and model results that support quantitative analyses, associated project management activities, and communications with stakeholders. The platform provides model access to a wide array of users, from operators to advanced modelers, to meet Reclamation’s evolving needs, regulatory changes, and technological advancements.

Development of integrated software, data acquisition, and automation are ubiquitous in many industrial processes but less common at this scale in water temperature management modeling. Reclamation recognizes the ongoing maintenance of WTMP activities and has committed resources to support the framework. Staff training, providing a model user guide and similar products (quick start guide, video instruction), ongoing documentation updates, and supporting model user groups are among the activities and strategies to maintain a diverse, active, and effective population of WTMP modelers.

Reclamation is currently utilizing the WTMP for the 2025 annual temperature management planning activities in the Sacramento River and American River basins through a facilitated adoption process. Reclamation’s Science and Technology Facilitated Adoption Program invests internally in the adoption of technology and scientific approaches that demonstrate, or have the potential for, successful results. Facilitated adoption is a process designed to facilitate technology migration, integrate staff into new processes, and incorporate certain targeted technology or scientific practices, aiming to empower users with success from the outset. In other words, facilitated adoption is an onboarding process. This adoption process builds off the multi-year, open and collaborative development process with a modeling technical committee made up of stakeholders, academics, and other interested parties, as well as two rounds of peer review led by the Delta Stewardship Council (https://deltacouncil.ca.gov/). The facilitated adoption provides federal and state agencies and stakeholder groups with an opportunity to concur with the determination and support the anticipated model transition, as well as provide comments and suggestions to improve the deployment and future applications. Activities include stakeholder outreach through the multidisciplinary Sacramento River Group (SRG) and American River Group (ARG)—groups consisting of multi-agency and stakeholder technical teams that coordinate fishery and operational requirements for the Sacramento River below Keswick Dam and lower American River below Nimbus Dam. Specifically, the WTMP framework is used to develop forecasts on these river systems in parallel with existing water temperature tools (e.g., HEC-5Q [26] and the Iterative Coldwater Pool Management Model (ICPMM) [95]). Although the results from the WTMP are not used for temperature management decision-making during the facilitated adoption period, WTMP model outputs are:

• Assessed and compared to outputs from legacy tools to demonstrate the utility and performance of the platform;
• Presented to familiarize SRG and ARG participants with the features of the platform and model;
• Evaluated, with input from SRG and ARG participants, to prepare the WTMP for rollout to a broader set of users (external to Reclamation).

Through the facilitated adoption, Reclamation confirms the value of the WTMP and plans to deploy the platform to support CVP temperature management in the 2026 season.

Reclamation is also exploring several actions aimed at expanding WTMP capacities in future years, including refinements to interface function to improve the user experience, exploring additional data management system features (e.g., automated QA), completing LTO simulations to be compatible with the latest CalSim 3 construct, and initiation of a model user group to allow interested parties to participate in model application. Recalibration of the WTMP models is expected to occur periodically (every 3–5 years). This adaptive approach is important because system operators will experience novel conditions, modify project operations, accommodate new/modified infrastructure, respond to regulatory changes and other actions, and respond to climate change. The WTMP was designed to be adaptable to accommodate changing conditions, such as climate change and related changes to operations.

5. Conclusions

The WTMP was created as a common tool to assist water temperature management activities for three distinct Central Valley Project systems, each with its own distinctive characteristics and challenges. Some of the many unique attributes of the WTMP are:

• A common platform for Reclamation to manage water temperature throughout the CVP.
• A data management system (DMS) that automates data collection, provides a means for data QA/QC, tracks record metadata, and develops model-ready datasets (for 20+ years of data).
• Reduces probability of user error by automating input file construction (based on model ready data) to avoid typical “fragile” model input file development and data management carried out by hand (e.g., spreadsheets, flat files, etc.).
• Provides Reclamation (and other agency) users with a common platform to carry out model simulations for planning, analysis, collaborative decision-making, and regulatory reporting.
• Allows users to select models for application in the various system elements (e.g., system reservoirs can be modeled using the lower spatial resolution ResSim for fast simulation or the higher resolution CE-QUAL-W2 for detailed simulation).
• Provides a means for short-term (e.g., seasonal) forecast, long-term planning (e.g., multi-decadal), and historical simulations at an hourly time step.
• Uses ensemble analysis as a means of efficiently assessing a range of analyses (e.g., different hydrology, meteorology, operations).
• Employs the latest versions of temperature control logic specific to the Folsom Dam temperature control shutter system, the Shasta Dam temperature control device, and temperature control curtains in Lewiston and Whiskeytown reservoirs.
• Automates reporting that assists modelers in completing analyses and reporting to internal technical teams, management, and external agencies and decision-makers.
• Comprehensive model development and application documentation and external modeling technical committee (MTC) collaboration to provide transparency.
• Includes a two-stage, independent, comprehensive peer review.

In sum, while many of these elements have been applied discretely for past projects, or at smaller scales, the incorporation of all these elements in the WTMP illustrates that the development of a comprehensive temperature modeling toolset can be applied to large complex systems over extended time periods and a range of climatic conditions to support collaborative flow and temperature management is clearly feasible. The WTMP is a unique and adaptive tool capable of responding to the challenges of managing flow and water temperature in a complex, multipurpose reservoir and riverine system, within a continually evolving regulatory and natural environment.