I Sensitivity Analysis of California Water Supply

Sensitivity analysis of California water supply: Assessment of vulnerabilities and adaptations

Max Fefer B.S. (University of California, Berkeley) 2016

THESIS

Submitted in partial satisfaction of the requirements for the degree of

MASTER OF SCIENCE

Civil and Environmental Engineering

in the

OFFICE OF GRADUATE STUDIES

of the

UNIVERSITY OF CALIFORNIA

DAVIS

Approved:

Jonathan Herman, Chair

Jay Lund

Samuel Sandoval-Solis

Committee in Charge

2017

i Abstract

Long-term changes in climate and population will have significant impacts on California’s freshwater management. Hydro-economic models can address climate change concerns by identifying system vulnerabilities and exploring adaptation strategies for statewide water operations. This thesis combines the new Python implementation of the CALVIN model, a hydro-economic model describing California water resources, with an ensemble of climate scenarios to identify adaptation strategies for managing water in a range of possible climates. A sensitivity analysis is performed by altering the magnitude and the timing of statewide inflows, defined as water availability and winter index respectively, to emulate changes in precipitation and temperature predicted by climate models. Model results show quadratic increases in shortage cost and marginal value of environmental flows, conveyance expansion, and reservoir expansion as water availability decreases.

Reservoirs adapt to warmer climates by increasing average storage levels in winter and routing excess runoff to reservoirs downstream with available capacity. Both small and large changes to reservoir operations were observed compared to historical hydrology, showing that no single operating strategy achieves optimality for all reservoirs. Increasing the fraction of winter flow incurs small increases in total shortage cost, showing the state’s ability to manage a changing hydrologic regime with adaptive reservoir operations.

ii Table of Contents Abstract ...... ii Table of Contents ...... iii List of Figures ...... iv List of Tables ...... vi Section 1 – Introduction ...... 1 Section 2 – California Background ...... 5 2.1 California’s water network ...... 5 2.2 CALVIN Model Introduction ...... 6 Section 3 – Methods ...... 9 3.1 Climate Change Modeling in CALVIN ...... 9 3.2 Climate Sensitivity Analysis ...... 11 3.3 Removing Infeasibilities ...... 15 3.4 Computational Capacity of Python CALVIN ...... 16 3.5 Analysis of model outputs ...... 17 Section 4 – CEC GCM Analysis ...... 18 Section 5 – Sensitivity Analysis Results ...... 22 5.1 – Shortage Cost ...... 22 5.2 – Water Supply Portfolio ...... 27 5.3 – Reservoir Operations ...... 29 5.3.1 – Reservoirs with major operations shift at -30% WA ...... 30 5.3.2 Gradual Decrease in Storage with decreasing WA ...... 31 5.3.3 Moderate Decrease in Storage with decreasing WA ...... 33 5.4 – Buffering Capacity of San Luis Reservoir ...... 35 5.5 – Environmental Flows ...... 36 5.6 – Infrastructure Expansion ...... 41 5.6.1 – Reservoir Expansion ...... 41 5.6.2 – Conveyance Expansion ...... 42 5.7 - Mitigation of Delivery Disruption ...... 43 Section 6 – Discussion ...... 45 6.1 Optimal Reservoir Operations ...... 45 6.2 Collaboration for Adaptive Operations ...... 46 6.3 Modeling Limitations ...... 47 Section 7 – Conclusion ...... 49 References ...... 50

iii List of Figures

Figure 1. Orders of climate change effects. Adapted from Chalecki and Gleick (1999) and Gain et al. (2012)...... 2

Figure 2. California regions represented in CALVIN. From Dogan (2015)...... 7

Figure 3. Runoff multipliers for CEC scenarios at Shasta Lake inflow ...... 10

Figure 4. Scenarios varying water availability and winter index developed for sensitivity analysis ...... 12

Figure 5. Modified Inflows for Shasta Lake, Lake Oroville, Millerton Lake, and Pine Flat Lake ...... 14

Figure 6. Modified Inflows for North & South Fork of the American River, Folsom Lake, Trinity Lake, and New Bullards Bar Reservoir ...... 14

Figure 7. Automatic Debug Algorithm for CALVIN ...... 16

Figure 8. Comparing CEC scenarios on WI and WA ...... 20

Figure 9. Total shortage costs across all scenarios ...... 23

Figure 10. Shortage costs by region ...... 24

Figure 11. Agricultural/Urban shortage costs by region at -30% WA ...... 25

Figure 12. Contour plots of Statewide and Upper Sacramento Valley ...... 26

Figure 13. Contour plots of Lower Sacramento Valley and Delta and San Joaquin South Bay ...... 26

Figure 14. Contour plots of Tulare Basin and Southern California ...... 27

Figure 15. Statewide monthly average surface reservoir storage ...... 29

Figure 16. Lake Shasta average monthly storage ...... 30

Figure 17. Lake McClure and New Don Pedro Reservoir average monthly storage ...... 31

Figure 18. Lake Oroville average monthly storage ...... 32

Figure 19. Folsom Lake and New Bullards Bar Reservoir average monthly storage ...... 33

Figure 20. Trinity Lake average monthly storage ...... 34

iv Figure 21. New Melones Reservoir and Millerton Lake average monthly storage ...... 34

Figure 22. San Luis Reservoir average monthly storage ...... 35

Figure 23. Average monthly dual values for Clear Creek minimum environmental flow requirement at Whiskeytown Reservoir ...... 36

Figure 24. Quadratic behavior of marginal opportunity costs of environmental flows with decreasing WA ...... 38

Figure 25. Distributions of surface and groundwater deliveries for each temperature scenario ...... 44

v List of Tables

Table 1. Average water availablity for CEC scenarios ...... 19

Table 2. Water supply portfolio for CEC scenarios ...... 21

Table 3. Water supply portfolio across changes in water availability and winter index ..... 28

Table 4. Environmental flow dual values [$/AF/month] ...... 37

Table 5. Enviromental flows with changing water availability at Warm3 WI [TAF] ...... 40

Table 6. Reservoir expansion marginal values [$/AF/year] ...... 42

Table 7. Conveyance expansion marginal values at Warm3 WI [$/AF/month] ...... 43

vi Section 1 – Introduction

Hydrologic records and climate model projections provide abundant evidence that freshwater resources are vulnerable to climate change (IPCC 2008). While climate change models agree on rising temperatures, projections of precipitation remain uncertain. For example, various downscaled global circulation models (GCMs) for California predict a range of both drier and wetter futures (Vicuna and Dracup 2007). Most recently, Allen and Luptowitz (2017) project California to receive more precipitation on average in the future due to increase in the El Niño phenomenon. Uncertainty is inherent in developing climate change projections, but important decisions regarding water infrastructure and operations are required regardless of agreement in model projections (Dettinger 2005).

Figure 1 outlines the progression of climate change of effects, starting from first order climate parameters such as temperature and precipitation through their effects 0n higher order geophysical and socio-economic processes. Both short term and long term changes in water availability due to changes in precipitation, temperature, snowpack accumulation, humidity, and other important factors have the potential to negatively affect local economies. Higher order effects like prolonged drought can depress agricultural production, while intensified storms may impose costly flood damages to infrastructure. Developing strategies to mitigate effects due to climate change will be critical to water managers navigating uncertain climate conditions.

Figure 1. Orders of climate change effects. Adapted from Chalecki and Gleick (1999) and Gain et al. (2012).

A primary element of studying the effects of climate change is identifying system vulnerabilities and measuring system performance under possible projected climates. The

IPCC defines vulnerability as “a function of the character, magnitude, and rate of climate variation (the climate hazard) to which a system is exposed, and of non-climatic characteristics of the system, including its sensitivity, and its coping and adaptive capacity” (IPCC 2007). More specifically for water resources, vulnerability is one of multiple ways to describe system failure which is defined as the severity of the likely consequences of system failure (Hashimoto et al. 1982). Vulnerabilities limit a water system’s ability to perform in a variety of operating conditions; identifying such vulnerabilities is key to developing successful climate adaptation plans. By subjecting a water network to a range of possible operating parameters, eliminating vulnerabilities that are identified establishes a water system better equipped to adapt to extreme hydrologic events.

The prospect of climate vulnerability requires infrastructure and operations to be robust to a range of possible futures beyond those for which the system was designed.

Recent efforts in this area are discussed in several review papers (Herman et al. 2015;

2 Maier et al. 2016). One such framework is Robust Decision Making (RDM), a systematic, objective approach for developing management strategies that create systems more resilient to impacts from future uncertainty (Lempert 2002; Groves et al. 2013). This framework is useful in climate modeling because of the variety of GCMs available and uncertainty of what future climate will be realized. Accounting for uncertainty in these projections is no excuse for allowing imprecision, rather planning for uncertainty creates more system flexibility to adapt to a range of possible parameters. U.S. government agencies at various levels, such as Metropolitan Water District of Southern California,

U.S. Bureau of Reclamation, Denver Water, California Department of Water Resources, have utilized RDM in their planning processes (Weaver et al. 2013). The US Bureau of

Reclamation commissioned a study from the RAND corporation to evaluate the uncertain effects of climate change on the Colorado River system by using RDM. RAND researchers specifically sought to identify future vulnerable conditions leading to inability to meet water delivery objectives, develop a computer-based tool to define a portfolio of management options reflecting different strategies management, and evaluate these portfolios across a range of simulated future scenarios to quantify the benefits of each scenario (Groves et al. 2013).

This thesis draws inspiration from the RDM framework by implementing an ensemble of climate scenarios to extract climate adaptation strategies for California water supply management. Here we assess the vulnerability of the statewide system to changes in total annual runoff (a function of precipitation) and the fraction of runoff occurring during the winter months (primarily a function of temperature). An ensemble of

3 scenarios is sampled and compared to the most recent available streamflow projections from the state’s 4th Climate Assessment. This technique is based upon classical sensitivity analysis, as a wide span of input parameters is explored and insight is gained from analyzing the optimization results. These scenarios are evaluated using a new open- source version of the CALVIN model, a network flow optimization model encompassing roughly 90% of the urban and agricultural water demands in California, which is capable of running scenario ensembles on a parallel computing cluster. The economic representation of water demand in the model yields several advantages for this type of analysis: optimized reservoir operating policies to minimize shortage cost, and the marginal value of adaptation opportunities, defined by shadow prices on infrastructure and regulatory constraints. Results indicate a shift in optimal reservoir operations in response primarily to changes in water availability, and a variety of operating strategies among reservoirs through changes in flow volume and timing. The collaborative management of reservoirs in CALVIN yields increased storage in downstream reservoirs with available capacity to store the increased winter runoff. This study contributes an ensemble evaluation of a large-scale network model to investigate uncertain climate projections, and an approach to interpret the results of economic optimization for long- term adaptation strategies.

4 Section 2 – California Background

2.1 California’s water network

California has a rich history of engineering solutions to water supply. Beginning in the 1800s when American settlers first came to the region, they learned to adapt to a new arid climate compared to the East Coast (Hanak and Lund 2012). Both water scarcity and surplus in the forms of drought and floods have beleaguered the state’s history. In response to high potential for agricultural production and relative lack of water supply in the summer months, California has developed a vast water infrastructure network of reservoirs, aqueducts, and pipelines over the past 100 years to mitigate the effects of living in an arid climate, high population, and large agricultural water demand. After nearly a century of development, California has concluded the era of constructing big water projects like the State Water Project and is now faced with new challenges of managing high water demand and weathering extreme hydrologic conditions.

Climate change poses a significant challenge to California’s primarily snowpack- fed water resources. Adaptation strategies will be necessary to tackle these challenges.

Water planning models provide a way to emulate California’s complex water management and explore infrastructure and policy alternatives (Draper et al. 2003). In California, various water resources models have been used to explore the effect climate change drawn from GCM results. CALSIM (California Simulation model), WEAP (Water

Evaluation and Planning model), and CALVIN (CALifornia Value Integrated Network) are some of the more comprehensive water planning models for California, but many local

5 and regional models exist for individual irrigation districts and water utilities (California

Department of Water Resources 2014; Lempert and Groves 2010; Connell-Buck et al. 2011;

Yates et al. 2009). In addition to findings developed from water planning models, potential institutional responses to climate change involve legal changes to water rights, changes to water pricing, implementation or expansion of water banking, water transfers, and changes in operation of water infrastructure including dams, reservoirs, conveyance infrastructure, and levees (Loomis et al. 2003; Olmstead 2014). California’s Department of

Water Resources includes in their Water Plan Update 2013 addresses climate adaptation by predicting 2050 water demands and evaluating adaptation strategies like recycled municipal water, conjunctive management of groundwater, and improved water-use efficiency of urban and agricultural users (CA DWR 2014).

2.2 CALVIN Model Introduction

Network flow programing is an apt tool for studying infrastructure systems like transportation, energy, and water resources. CALVIN is a hydro-economic optimization model of California’s water network with an 82-year hydrologic input dataset. Economics are represented by penalty functions developed for urban and agricultural areas throughout California. Figure 2 shows the regions of California included in the CALVIN model. Areas excluded from CALVIN have low populations and water consumption compared to other regions in the state.

CALVIN’s network flow structure is represented by a set of nodes and links subject to constraints. A node in CALVIN is defined by a location element and temporal element.

6 For example, “SR_SHA.2002-10-31” represents Lake Shasta at October 2002. A link connects two nodes and has the following properties: flow (decision variable), unit cost, amplitude (loss factor), lower bound, and upper bound. This structure allows the model to deliver water both across the network within a specified time step and to represent storage at reservoirs as links between two time steps at a specified location. For example,

“C120.2002-10-31-SR_LA.2002-10-31” is the link representing the Los Angeles Aqueduct in

October 2002, while “SR_FOL.2002-10-31-SR_FOL.2002-11.30” represents storage in

Folsom Lake carried over from October to November 2002.

Figure 2. California regions represented in CALVIN. From Dogan (2015).

7 The Python implementation of the CALVIN model solves the network flow problem using linear programming to optimize flow over all links in the network to minimize the combined cost of water shortage and operations. Hydrology-related inputs include surface and groundwater hydrology, environmental flow constraints, and wildlife water deliveries. Model results include valuable management information about combined operations of surface and groundwater reservoirs, environmental flows, and hydropower. In addition, economic data--including shortage costs in drought years, and marginal values of storage and conveyance capacity--allow agricultural and urban water users to develop adaptive management strategies and identify potential upgrades to water supply infrastructure. The CALVIN model uses projected urban and agricultural water demands for the year 2050 that consider changes in land use, per capita water use, and population growth. Climate change will likely alter agricultural and urban water demands, but the projected 2050 demands are unchanged throughout the scenarios presented in this thesis (Dogan 2015).

The CALVIN model provides a unique opportunity to encapsulate the majority of

California’s water system in a single model. It represents economics in a linear programming framework to operate the system in an economically efficient way. Hydro- economic models like CALVIN provide an opportunity for water resources modelers to analyze potential changes to operations without assuming that current reservoir operating rules and conveyance agreements will remain unchanged, providing suggestions on more optimal operations strategies (Olmstead 2014).

8 Section 3 – Methods

3.1 Climate Change Modeling in CALVIN

The 4th Climate Assessment sponsored by the California Energy Commission

(CEC) has provided an ensemble of downscaled GCM models for California, routed through the Variable Infiltration Capacity (VIC) hydrologic model to provide streamflow projections for various Northern California basins. Combining the information provided by the GCM projections with the CALVIN model allows modelers to understand how

California may adapt to changes in timing and magnitude of reservoir inflows. Instead of using the direct streamflow values outputted by the GCM model, a perturbation method is used to modify CALVIN’s historical hydrology to reflect the GCM’s behavior, following an approach developed by California state agencies (Miller, Bashford, and Strem 2003). The application of perturbation ratios to model climate change in the CALVIN model has been implemented in previous studies (Medellín-Azuara et al. 2008; Tanaka et al. 2006; Harou et al. 2010;

Connell-Buck et al. 2011).

The perturbation method requires the modeler to identify two time periods: a historical period and a study period. For the CEC scenarios, the historical period is defined as 1950-2005 and the study period is defined as 2006-2100. A multiplier is calculated by dividing the average streamflow of the study period by historical period for every month, yielding 12 multipliers for each location provided. The locations provided by the CEC are matched with CALVIN inflows. Figure 3 shows the 12 perturbation ratio multipliers for the inflow to Shasta Lake. Some CALVIN inflow locations are not included in the CEC scenarios, so a correlation analysis was performed to

9 determine the most highly correlated rim inflow from CALVIN’s historical hydrology data and apply the same multiplier to the missing rim inflows. These multipliers are then applied to CALVIN historic hydrology and the model is optimized. Multipliers for the inflow at Shasta Lake are shown in Figure 3 with the baseline value (1.0) representing historical hydrology. Multipliers greater than

1.0 indicate a wetter monthly average compared to the scenario’s historic period; multipliers less than 1.0 represent a smaller monthly average compared to the historic period.

Figure 3. Runoff multipliers for CEC scenarios at Shasta Lake inflow

The model results are then analyzed for storage volumes and scarcity costs of urban, agricultural, and environmental flows. The marginal value of increasing storage, water markets, and increased conveyance capacity point to areas for further investigation to improve performance (i.e., reduce future shortage costs). By changing inflow volume and

10 timing, evaluating the response of state’s main reservoirs to these altered climate conditions suggests optimal operating strategies to minimize shortage cost.

3.2 Climate Sensitivity Analysis

In addition to the climate scenarios developed by the CEC, this thesis explores how

California water operations would be optimally altered with shifts in the magnitude and timing of streamflow. In particular, this study focuses on drier scenarios with an increased fraction of runoff arriving in the winter months due to rising temperatures.

To measure the timing and magnitude of inflows, the winter index and water availability are used. The winter index is defined as the fraction of average inflow volume from November through April for each climate scenario divided by the historical CALVIN average inflow from November through April. Winter index (WI) values greater than 1 indicate a larger proportion of inflows arriving in the winter compared to historical.

�� ℎ�� (��) �� (��) �� = �� ℎ�� (��) �� (��)

This metric represents increasing temperatures statewide, which yield more precipitation falling as rain rather than snow compared to current conditions; the snowpack that does accumulate also melts faster and earlier than in recent history.

11 Water availability (WA) is defined as the sum of all rim inflows, representing all water entering the model each year. A drier scenario will see a decrease in average water availability, and a change of 0% represents the average water availability of CALVIN’s original hydrology (35.3 MAF/year statewide). Figure 4 below shows the 25 scenarios developed for this experiment which are organized into four quadrants: Warm-Dry,

Warm-Wet, Cool-Dry, and Cool-Wet. Each axis represents a range of the timing and magnitude scenarios and each square point in the plot represents a sampled scenario with the indicated average water availability and winter index.

Figure 4. Scenarios varying water availability and winter index developed for sensitivity analysis

12 To alter the statewide winter index, each CALVIN rim inflow must be altered individually. Figure 5 and 6 below show the eight largest CALVIN rim inflows by volume and how their average monthly inflows were altered for the winter index permutations.

The winter index values shown in Figure 4 are relabeled as follows: ‘cold’ scenarios refer to a WI of 0.95 (bottom row of 5 in Figure 4), ‘Warm1’ refers to a WI of 1.05 (2nd from bottom row), ‘Warm2’ refers to a WI of 1.10 (2nd from top row), and ‘Warm3’ refers to a WI of 1.15 (top row). The scenarios plotted in Figures 5 and 6 have historical water availability and varying winter indexes.

To apply the changes in timing and magnitude, multipliers for each month were developed to modify CALVIN’s historical hydrology to reflect the changes in timing and water availability for the given scenario. This method is similar to the perturbation ratio, but the multiplier is calculated based on the degree of change in winter index and water availability.

Figure 5. Modified Inflows for Shasta Lake, Lake Oroville, Millerton Lake, and Pine Flat Lake

Figure 6. Modified Inflows for North & South Fork of the American River, Folsom Lake, Trinity Lake, and New Bullards Bar Reservoir

14 3.3 Removing Infeasibilities

The CALVIN model contains over 5 million decision variables with the full 82-year hydrology and piecewise-linear objective functions. To incorporate climate scenarios into

CALVIN, the matrix of links needs to be edited from the historical hydrology. Altering the bounds of links in a heavily constrained network often leads to over-constrained systems and subsequent model infeasibilities. The CALVIN model includes an option to run in

“debug” mode which allows the user to reconcile model infeasibilities. Debug mode adds two additional nodes, called DBUGSRC and DBUGSNK, which are linked to all nodes in the network. These links have an extreme cost ($2 million/acre foot) and are included to add or remove water when the network is otherwise infeasible. The magnitude of the debug flows is used to adjust the bounds of links within the normal network to allow model feasibility. The algorithm includes rules that prevent changes to bounds on specified links such as reservoir carryover storage. In addition, a log of reduced links is maintained as the algorithm progresses to track which links are reduced for quality control of model results. This process increases the runtime considerably due to solving the model several times before outputting a result, but eliminates the need for the user to change the lower bounds manually. Figure 7 outlines the algorithm for the automatic scripting described above.

Figure 7. Automatic Debug Algorithm for CALVIN

3.4 Computational Capacity of Python CALVIN

The original implementation of CALVIN used the HEC-PRM solver package to run the model and needed roughly 7 days to solve without an initial solution. Advances in computing technology like parallel computing and open-source linear programming solvers now allow the CALVIN model to be solved within 2 hours for the historical hydrology. For this study, 45 CALVIN runs (20 representing CEC scenarios and 25 for the sensitivity analysis) were performed on the UC Davis HPC1 high performance cluster with each scenario needing roughly 9-12 hours to solve on average. Some model runs that required fewer debug iterations ran within 4 hours, while other scenarios took 3 days to reach a feasible solution. Through the use of Python scripting, model results are post- processed into CSV format and used to automatically create timeseries reservoir storage,

16 dual values, and water supply portfolio. Overall, runtimes significantly improved by utilizing Python and the HPC1 cluster.

3.5 Analysis of model outputs

The results of the CALVIN scenarios will be analyzed for potential adaptation strategies by investigating supply type and timing. Reservoirs have key roles in managing the state’s water resources, so operations of the state’s largest reservoirs will be closely examined to understand how the model decided optimal reservoir volumes and how these operating strategies change with water availability and winter index. Changes in operations due to winter index from historical hydrology will indicate general adaptation strategies for accommodating warmer climates. Economic outputs such as shortage cost and marginal value will identify infrastructure vulnerabilities and inform on the economic impact of varying water availability and winter index. Conveyance expansion, reservoir expansion, and value of environmental flows will also be analyzed to understand how their economic value changes with the magnitude and timing parameters.

17 Section 4 – CEC GCM Analysis

The sensitivity analysis described in the methods explores how CALVIN allocates water while the timing and magnitude of water is changed, but does not represent any one scenario developed by climate modelers. The CEC utilized 10 different climate models at 2 emissions levels (RCP 4.5 and 8.5), totaling 20 climate scenarios, to analyze how climate change will impact California water resources. The CEC projections provide insight into how climatologists believe California will receive more precipitation and overall be warmer. Although these predictions are wetter, research has predicted that rainfall will arrive in more extreme events, increasing the potential for flooding and presenting an interesting dilemma to dam operators located on rivers and streams heavily influenced by precipitation. Regardless of increased precipitation predictions, California exists in the arid West and drought will always be a part of California’s story.

These models were applied to CALVIN using the perturbation ratio described in the methods. Once the multipliers were applied, the average water availability and winter index was calculated for each scenario and shown below in Table 1. Figure 8 shows a scatter plot of the scenarios.

18 AVERAGE ANNUAL WATER AVAILABILITY SCENARIO (MAF/YR) WINTER INDEX HISTORICAL CALVIN 35.3 1 ACCESS1-0_RCP45 38.0 1.09 ACCESS1-0_RCP85 33.7 1.14 CANESM2_RCP45 40.9 1.14 CANESM2_RCP85 50.5 1.22 CCSM4_RCP45 39.1 1.13 CCSM4_RCP85 40.8 1.18 CESM1-BGC_RCP45 37.2 1.09 CESM1-BGC_RCP85 39.3 1.18 CMCC-CMS_RCP45 35.3 1.02 CMCC-CMS_RCP85 37.0 1.00 CNRM-CM5_RCP45 47.5 1.17 CNRM-CM5_RCP85 51.6 1.26 GFDL-CM3_RCP45 36.7 1.07 GFDL-CM3_RCP85 35.3 1.11 HADGEM2-CC_RCP45 38.3 1.06 HADGEM2-CC_RCP85 38.8 1.06 HADGEM2-ES_RCP45 33.1 1.03 HADGEM2-ES_RCP85 37.6 1.07 MIROC5_RCP45 33.3 1.04 MIROC5_RCP85 32.8 1.03

Table 1. Average water availablity for CEC scenarios

Figure 8. Comparing CEC scenarios on WI and WA

All scenarios excluding MIROC5 show the same or greater average annual inflow compared to the historical CALVIN hydrology. Table 2 shows the water supply portfolio for the average year of each CEC scenario compared against the average historical

CALVIN results. The delivery volumes do not change significantly regardless of scenario water availability, some of which are more than 50% greater than historical CALVIN inputs. This result is due to CALVIN’s optimization structure where there is no economic benefit to delivering additional water. CALVIN’s utility lies in learning how to efficiently allocate shortage in times of water scarcity and does not include the necessary tools like optimizing flood pool rule modification to study how California must adapt to manage a wetter climate.

AVG WATER SUPPLY PORTFOLIO [MAF/YEAR] AVG WA SCENARIO [MAF/YR] GWP SWD NPR PR DESAL HISTORICAL CALVIN 35.3 37.7 45.3 0.6 0.6 0.05 CNRM-CM5_RCP85 51.6 37.7 45.6 0.6 0.6 0.05 CANESM2_RCP85 50.5 37.7 45.6 0.6 0.6 0.05 CNRM-CM5_RCP45 47.5 37.7 45.6 0.6 0.6 0.05 CANESM2_RCP45 40.9 37.7 45.5 0.6 0.6 0.05 CCSM4_RCP85 40.8 38.1 44.9 0.6 0.6 0.05 CESM1-BGC_RCP85 39.3 37.7 45.4 0.6 0.6 0.05 CCSM4_RCP45 39.1 37.9 45.0 0.6 0.6 0.05 HADGEM2-CC_RCP85 38.8 37.9 44.8 0.6 0.6 0.05 HADGEM2-CC_RCP45 38.3 38.0 44.9 0.6 0.6 0.05 ACCESS1-0_RCP45 38 37.9 44.6 0.6 0.6 0.05 HADGEM2-ES_RCP85 37.6 37.9 45.0 0.6 0.6 0.05 CESM1-BGC_RCP45 37.2 37.7 45.5 0.6 0.6 0.05 CMCC-CMS_RCP85 37 37.8 44.7 0.6 0.6 0.05 GFDL-CM3_RCP45 36.7 38.0 45.0 0.6 0.6 0.05 CMCC-CMS_RCP45 35.3 37.9 44.9 0.6 0.6 0.05 GFDL-CM3_RCP85 35.3 37.8 44.8 0.6 0.6 0.05 ACCESS1-0_RCP85 33.7 37.5 43.7 0.7 0.7 0.05 MIROC5_RCP45 33.3 37.7 44.7 0.6 0.6 0.05 HADGEM2-ES_RCP45 33.1 37.8 45.0 0.6 0.6 0.05 MIROC5_RCP85 32.8 37.6 44.5 0.7 0.7 0.05

Table 2. Water supply portfolio for CEC scenarios

21 Section 5 – Sensitivity Analysis Results

5.1 – Shortage Cost

Shortage costs increase across all scenarios with decreasing water availability and increasing winter index (Figure 9). Shortage cost for the set of scenarios at 0% and +10% average water availability (WA) is around $200 million per year. The increase in shortage cost appears quadratic with respect to water availability as scarcity cost increases more severely with incremental decreases in average water availability. A slight and moderate increase in shortage cost occurs from 0% to -10% and -10% to -20% average WA. Shortage cost approximately doubles as average water availability is lowered to -30% of the historical average inflows. This spike in shortage cost is the first indicator of a statewide vulnerability to drought occurring between -20% and -30% water availability.

Within a given water availability, increasing winter index yields small increases in shortage cost. Drier scenarios increase shortage costs between each winter index increment. As water becomes scarcer, the model has less flexibility to modify operations because all water has been allocated to a demand (which would incur additional shortage if deliveries were curtailed) or an instream flow requirement, for which curtailments are not permitted in the model.

Figure 9. Total shortage costs across all scenarios

CALVIN’s 5 geographic regions are the Upper Sacramento Valley (USV), Lower

Sacramento Valley & Delta (LSVD), San Joaquin & South Bay (SJSB), Tulare Basin (TB), and Southern California (SC). These regions are shown in Figure 2 on page 8. The statewide total of shortage costs increases quadratically with decreasing water availability, but each region incurs shortage cost at different rates (shown in Figure 10). Southern

California carries the most shortage cost out of all regions and retains significant shortage throughout all scenarios. Tulare Basin assumes the 2nd largest shortage cost among the regions, but sees dramatic increase in shortage costs from -20% to -30% WA. USV, LSVD, and SJSB incur incrementally greater shortage with water reductions but on a much smaller scale compared to TB and SC.

Missing Value

Figure 10. Shortage costs by region

Both urban and agricultural demands in CALVIN can incur shortage costs. Figure

11 shows the difference in shortage costs between urban and agricultural demands within each region at -30% water availability across all winter indexes. Urban shortage costs are overwhelmingly incurred in Southern California with small urban shortage costs in the other regions. Southern California has small agricultural total shortage cost, but Tulare

Basin agriculture burdens the largest share of agricultural shortage costs at -30% water.

USV, LSVD, and SJSB also incur significant agricultural shortage cost. Total agricultural shortage costs are greater than urban shortage costs due to urban areas having a greater willingness to pay for water than agricultural regions, so agricultural demands are curtailed first.

Figure 11. Agricultural/Urban shortage costs by region at -30% WA

Figures 12, 13, and 14 show contour plots of changing shortage costs with winter index and water availability statewide and for each region. The contour plots show that shortage costs are primarily determined by changes in water availability with slight influence from changing winter index. In all regions excluding Southern California, little to no shortage costs are incurred as water availability decreases to -10%. Even with water availability greater than historical, Southern California still incurs shortage costs. As water availability decreases, the gradient of shortage costs with winter index increases as shown by the changing slope of the contour lines.

Figure 12. Contour plots of Statewide and Upper Sacramento Valley

Figure 13. Contour plots of Lower Sacramento Valley and Delta and San Joaquin South Bay

Figure 14. Contour plots of Tulare Basin and Southern California

5.2 – Water Supply Portfolio

Across the state, five water supply types are represented: Groundwater Pumping

(GWP), Surface Water Delivery (SWD), Non-Potable Reuse (NPR), Potable Reuse (PR), and Desalination (DESAL). Each region in CALVIN has access to surface and groundwater, but only urban areas have access to reuse and desalination. Table 3 on the following page outlines how the portfolio of the 5 water supplies changes with the Winter

Index and Water Availability. Deliveries are maintained with little scarcity through the -

10% scenario, but significant shortage is present in the -20% and -30% scenarios.

Groundwater pumping is fairly constant throughout all scenarios due to the simplistic representation of groundwater allocation within CALVIN, where the unit cost of pumping

27 does not account for declining water tables. NPR increases from roughly 0 to 48 TAF/year across the scenarios. Surface water deliveries stay similar as water availability decreases to

-10% in yearly inflows, but deliveries are significantly reduced by the 30% decrease scenarios. Delivery volume is negligibly affected by the change in winter index.

WATER SUPPLY PORTFOLIO [TAF/YR] WA WI GWP SWD NPR PR DESAL +10% COLD 37.6 45.5 0.59 0.59 0.046 +10% HISTORICAL 37.6 45.5 0.59 0.59 0.046 +10% WARM1 37.6 45.5 0.59 0.59 0.046 +10% WARM2 37.7 45.5 0.59 0.59 0.046 +10% WARM3 37.7 45.4 0.59 0.59 0.046 0% COLD 37.7 45.4 0.59 0.59 0.046 0% HISTORICAL 37.7 45.4 0.59 0.59 0.046 0% WARM1 37.7 45.3 0.59 0.59 0.046 0% WARM2 37.7 45.3 0.59 0.59 0.046 0% WARM3 37.7 45.3 0.59 0.59 0.046 -10% COLD 37.5 45.1 0.60 0.60 0.046 -10% HISTORICAL 37.5 45.0 0.61 0.61 0.046 -10% WARM1 37.4 45.0 0.61 0.61 0.046 -10% WARM2 37.5 44.8 0.63 0.63 0.046 -10% WARM3 37.4 44.7 0.67 0.67 0.046 -20% COLD 37.4 43.0 1.20 1.20 0.046 -20% WARM1 37.4 41.9 2.06 2.06 0.046 -20% WARM2 37.4 41.2 2.62 2.62 0.046 -20% WARM3 37.4 41.0 2.67 2.67 0.046 -30% COLD 34.7 40.9 2.90 2.90 0.046 -30% HISTORICAL 34.7 40.8 2.90 2.90 0.046 -30% WARM1 34.6 40.7 2.90 2.90 0.046 -30% WARM2 34.6 40.6 2.90 2.90 0.046 -30% WARM3 34.5 40.5 2.90 2.90 0.046

Table 3. Water supply portfolio across changes in water availability and winter index

28 5.3 – Reservoir Operations

The following section analyzes operations for 10 major reservoirs in California.

Figure 15 shows the total statewide storage across all scenarios. The 0% and -10% scenarios have a similar shape, minimum, and maximum. Within each WA scenario, the variation in WI shows a slightly larger peak in total storage in May and a moderate decrease in minimum value in October. Overall, the effect of the warmer scenarios is more pronounced at the 0 and -10% scenarios, whereas the reduction in water availability plays a larger role in the -20% and -30% scenarios.

Figure 15. Statewide monthly average surface reservoir storage

Breaking down the statewide aggregate reveals four categories of behavior shown among the reservoirs. The four categories are as follows.

29 5.3.1 – Reservoirs with major operations shift at -30% WA

Like the statewide total storage, reservoirs including Shasta Lake, Lake McClure, and New Don Pedro Reservoir see similar average storages in the 0%, -10%, and -20% WA scenarios where the WI plays a more significant role in determining end of water year storage. In the -30% scenario, average storage levels drop considerably and a range up to

500 TAF difference in monthly storage across the winter indexes (Figures 16 and 17).

Figure 16. Lake Shasta average monthly storage

Figure 17. Lake McClure and New Don Pedro Reservoir average monthly storage

5.3.2 Gradual Decrease in Storage with decreasing WA

In contrast to Shasta Lake, reservoirs including Lake Oroville, Folsom Lake, and

New Bullards Bar show no clear definition between scenarios at different water availabilities. Average monthly storage values decrease with decreasing water availability, but only gradually in comparison to the large shift in operations at Shasta. For Lake

Oroville (shown in Figure 18), average storage values are closely banded within 200 thousand acre feet [TAF] from March through June, and afterwards the band expands

31 through November. As above, higher winter index/warmer scenarios have decreased end of water year storage values compared to lower winter index/colder scenarios. Due to

Oroville’s large capacity, the band of storage values in November across scenarios is much wider compared to smaller reservoirs like Folsom Lake and New Bullards Bar Reservoir

(Figure 19). Both Folsom Lake and New Bullards Bar Reservoir have a tight band of scenario operations throughout the year and are negligibly influenced by changes in winter index within a given water availability.

Figure 18. Lake Oroville average monthly storage

Figure 19. Folsom Lake and New Bullards Bar Reservoir average monthly storage

5.3.3 Moderate Decrease in Storage with decreasing WA

A third category of reservoir operations is a moderate decrease in average monthly storage across decreases in water availability. Trinity Lake (shown in Figure 20) shows similar behavior to Shasta Lake but with larger decreases in between the 0% and -20%

WA scenarios. The similar operations between Trinity and Shasta makes sense as reservoirs are within close proximity to each other and both have a large storage volume.

Figure 21 shows that smaller reservoirs such as New Melones Reservoir and Millerton

Lake also show similar behavior where there is a steady decline in storage.

Figure 20. Trinity Lake average monthly storage

Figure 21. New Melones Reservoir and Millerton Lake average monthly storage

34 5.4 – Buffering Capacity of San Luis Reservoir

San Luis Reservoir is critical to managing water statewide across the scenarios.

Among the largest reservoirs in the state, San Luis serves as an off-stream storage facility for operating the State Water Project (SWP) and Central Valley Project. Although the reservoir capacity is 2 million acre feet (MAF), reservoir levels are typically far below capacity. This extra reservoir space is utilized in the climate scenarios to capture excess streamflow unable to be stored at upstream facilities like Shasta and Oroville. Unlike other large reservoirs, San Luis storage values increase considerably between the 0% and -

10% scenarios (Figure 22). By the -30% WA scenarios, the operations of San Luis have converged and do not change regardless of how warm or cold the scenario is.

Figure 22. San Luis Reservoir average monthly storage

35 5.5 – Environmental Flows

In CALVIN, minimum instream flow requirements (MIF) for environmental purposes have a fixed constraint where the lower bound and upper bound are equal for each time step while many others have a set nonzero lower bound and large upper bound.

In dry scenarios, the automatic debug flow algorithm will eliminate the model’s debug flows, possibly reducing MIFs in the process to achieve feasibility. The linear programming format allows a unique opportunity to assign a marginal opportunity cost to environmental flow deliveries, providing an informative look into the economic value of the water being allocated to environmental uses if traded to other demands in the network through water trading. Figure 23 below looks at the monthly average marginal value across scenarios for the Clear Creek MIF. Marginal value dramatically increases as water availability is reduced from -20% to -30%. By comparison, the variation in winter index has negligible impact on marginal value within each water availability scenario.

Figure 23. Average monthly dual values for Clear Creek minimum environmental flow requirement at Whiskeytown Reservoir

36 Table 4 below shows the average yearly maximum marginal values for environmental flows with value greater than $100/AF for the warmest scenario at each

WA level. Average yearly maximum was chosen for analysis because the metric captures the variation due to the water availability when the infrastructure is most valuable within the year. Marginal value increases with decreasing water availability but by varying amounts across the links. Consistent with other CALVIN studies, Mono Lake and the

Trinity River instream flow requirements show high value in the driest scenarios (Tanaka et al. 2006).

DESCRIPTION CALVIN LINK +10% 0% -10% -20% -30% Mono Lake SR_GNT-SR_ML 876 894 975 1267 1431 Trinity River D94-SINK 49 55 96 229 593 Clear Creek SR_WHI-D73 12 15 44 154 478 Sacramento River D5-D73 10 13 41 150 476 @ Keswick Reservoir Calaveras River C41-C42 19 24 62 186 452 San Joaquin River D616-C42 22 28 68 179 405 Mokelumne River SR_CMN-C38 14 20 50 144 372 Sacramento River D507-D509 6 10 40 145 350 @ Rio Vista Feather River C25-C31 4 6 30 109 303 American River D9-D85 6 9 26 85 265 Sacramento River @ Sacramento D61-C301 6 7 16 62 265 Navigation Control Point Yuba River SR_ENG-C28 13 63 39 92 214 Stony Creek SR_BLB-C9 14 17 29 56 211 @ Black Butte Dam Bear River SR_CFW-C33 9 11 24 68 185 Sacramento River D503-D511 7 9 22 71 181 @ Delta Cross Channel Cosumnes River C37-C38 49 51 58 83 136 Stony Creek C9-C12 18 23 33 51 118 @ Tehama-Colusa Canal

Table 4. Environmental flow dual values [$/AF/month]

37 All environmental flows besides Mono Lake show an approximately quadratic increase in marginal opportunity cost with decreasing water availability (shown in Figure

24). Placing economic value on nonrevenue water like environmental flows is difficult.

For example, water delivered to national wildlife refuges provides no direct economic benefit like agriculture but serves an important purpose in maintaining critical ecosystems and fostering biodiversity by providing habitat for endangered and threatened species. However, the model provides an implied opportunity cost for these constraints based on the value of allocating this water to meet demand elsewhere.

Figure 24. Quadratic behavior of marginal opportunity costs of environmental flows with decreasing WA

38 To achieve model feasibility, fixed constraints like some minimum instream flow requirements must be reduced due to the model being over-constrained. MIF deliveries for the Warm3 WI with changing water availability is shown below in Table 5. This table shows increases in most MIF deliveries in the wetter scenario (+10% WA) and gradual reduction in deliveries with decreased water availability. Links with a gradual reduction have a linear relationship with water availability, meaning a 30% reduction in average water availability statewide resulted in a 30% reduction of the MIF requirement. A few

MIF such as Clear Creek, Trinity River, San Joaquin River @ Vernalis, and Bear River have no reduction in the requirement across changes in water availability.

DESCRIPTION CALVIN LINK +10% 0% -10% -20% -30% Clear Creek SR_WHI-D73 135 122 123 122 122 @ Sacramento River Sacramento River D5-D73 7042 6320 5608 4886 4211 @ Keswick Reservoir Sacramento River D77-D75 8612 8132 6968 6210 5561 @ Red Bluff Sacramento River @ D61-C301 7498 6883 5608 4762 4303 Sacramento Navigation Control Point Stony Creek SR_BLB-C9 456 414 373 331 290 @ Black Butte Dam Stony Creek C9-C12 344 303 275 244 227 @ Tehama-Colusa Canal Trinity River D94-SINK 607 607 607 606 606 Bear River Inflow to Lake C35- 385 391 387 388 387 Combie/Rollins Lake SR_RLL_CMB Bear River below Lake N201-N202 294 279 257 238 218 Combie/Rollins Lake Bear River @ Camp Far SR_CFW-C33 380 366 343 324 305 West Reservoir

39 Bear River from DA 68 C33-C308 232 229 252 254 272 American River @ D64-C8 2880 2574 2305 2037 1784 Sacramento River American River D9-D85 2918 2639 2371 2109 1869 @ Nimbus Dam Calaveras River C41-C42 102 103 116 107 101 @ San Joaquin River Feather River @ Kelly Ridge C23-C25 661 554 568 554 547 Feather River C25-C31 4467 4041 3649 3224 2811 @ Thermalito Afterbay Feather River D42-D43 6522 5885 5368 4812 4256 @ Sacramento River Cosumnes River C37-C38 389 361 325 289 253 @ Dry Creek Sacramento River @ Hood D503-D511 17888 16446 14138 12445 11160 Sacramento River D507-D509 6101 12647 5252 5101 4848 @ Rio Vista Yuba River SR_ENG-C28 1827 1660 1494 1327 1160 @ Englebright Dam Yuba River from DA 67 C83-C31 1818 1697 1490 1332 1166 San Joaquin River D616-C42 3073 3072 3073 3077 3098 @ Vernalis San Joaquin River D609-D608 131 126 119 120 119 @Mendota Fresno River D624-C48 44 30 20 14 2 @ Chowchilla Bypass Merced River D645-D646 427 330 242 207 231 @ Lower Merced River Merced River D649-D695 397 288 202 177 199 @ San Joaquin River Stanislaus River @ Ripon D672-D675 703 761 597 551 503 Stanislaus River D675-D676 821 879 714 658 597 @ San Joaquin River Mokelumne River SR_CMN-C38 616 546 492 422 378 @ Camanche Dam Tuolumne River SR_DNP-D662 1575 1438 1277 1102 937 @ New Don Pedro Dam Tuolumne River D662-D663 722 631 520 448 417 @ Lagrange Dam Mono Lake SR_GNT-SR_ML 75 75 74 73 70

Table 5. Enviromental flows with changing water availability at Warm3 WI [TAF]

40 5.6 – Infrastructure Expansion

5.6.1 – Reservoir Expansion

To evaluate the marginal value of reservoir expansion, the maximum dual value for each year was averaged over the 82-year dataset and values for each reservoir are reported in

Table 6 on the following page. Most reservoirs see an increasing marginal value as the scenarios become drier. Reservoirs located in Southern California, such as Lake Success,

Millerton Lake, and Kaweah Lake, decrease in marginal value with decreasing water availability because these reservoirs reach maximum capacity less often than reservoirs north of the Delta. Values reported below are marginal values above $50/AF per year from the warmest of the scenarios (Warm3, WI=1.15). In California, reservoirs typically only fill once a year, thus only one local maximum appears in the yearly time series, so marginal values are reported per year in this section. Variation in winter index is not shown because the marginal values do not increase more than 10% from the coldest to warmest scenario for at a constant water availability. As these values presented are marginal with regard to 1 TAF and assume perfect foresight, true benefit to reservoir expansions requires deeper analysis as agencies would expand reservoirs at much larger volumes, like the

Santa Clara Valley Water District proposing to increase the volume of Pacheco Lake by

124 TAF (Rogers 2017).

41 DESCRIPTION CALVIN LINK +10% 0% -10% -20% -30% Black Butte Lake SR_BLB 19 23 44 134 348 Englebright Lake SR_ENG 2 19 23 90 233 Folsom Lake SR_FOL 2 2 16 69 161 Rollins Reservoir/Combie SR_RLL_CMB 6 8 19 67 161 Lake Camp Far West Reservoir SR_CFW 7 8 18 65 153 New Bullards Bar SR_BUL 3 20 18 58 139 Reservoir Los Vaqueros Reservoir SR_LVQ 7 8 13 40 129 Thermalito SR_TAB 4 4 9 33 80 Fore/Afterbay Clear Lake/Indian Valley SR_CLK_INV 3 3 12 34 70 Reservoir Lake Oroville SR_ORO 1 1 5 25 60 LADWP Reservoir SR_LA 33 33 36 41 54 Lake Crowley/Long SR_CRW 37 35 37 41 52 Valley Reservoir Lake Kaweah SR_TRM 55 57 48 34 23 Lake Success SR_SCC 50 53 47 39 41 Millerton Lake SR_MIL 40 44 39 35 36

Table 6. Reservoir expansion marginal values [$/AF/year]

5.6.2 – Conveyance Expansion

Dual values on upper bounds of conveyance links increased both with warmer and drier scenarios as shown in Table 7. However, the increase in dual value across WI scenarios at the same water availability was small. The upper bounds did not increase more than 10% across the WI scenarios. The SFPUC-Hayward Intertie retains high marginal value throughout all scenarios, suggesting an adaptation to even normal operations that yields economic benefit. Overall, dual values are smaller for conveyance expansion than compared to reservoir expansion because they are monthly rather than yearly dual values.

42 DESCRIPTION CALVIN LINK +10% 0% -10% -20% -30% Los Vaqueros to Contra C310-C70 33 34 51 123 281 Costa Canal SFPUC-Hayward Intertie U209-C78 263 265 264 261 267 Folsom South Canal D9-C173 6 8 24 82 254 Madera Canal C72- 75 83 98 109 145 HSU306C72 Fresno Slough C54-D608 8 13 29 59 74 Chowchilla Bypass D609-C48 7 12 28 42 57 Contra Costa Canal PMP_CC1-C70 52 53 50 47 45 Corning Canal D77- 0 0 2 10 37 HSU102D77 DMC-CAA Intertie D701-D800 12 12 11 12 13 Lower Cherry Creek SR_LL_ENR- 5 5 5 6 8 Aqueduct C44 Delta Mendota Canal D722-D723 0.7 0.1 0.1 1.5 2.7 LA Aqueduct C120-SR_LA 0.1 0.1 0.2 0.3 0.5

Table 7. Conveyance expansion marginal values at Warm3 WI [$/AF/month]

5.7 - Mitigation of Delivery Disruption

Groundwater and surface water make up the majority of water deliveries in

California. Identifying how these deliveries change with water availability and winter index is important. Figure 25 aggregates the results into the 5 winter index categories: cold, historical, warm1, warm2, and warm3. Within each category are the 5 water availability scenarios for the specified winter index scenario. In Figure 25, the 5 distributions on the violin plot each represent all monthly deliveries for the specified supply type and winter index scenario. This plot shows how the volume of each delivery changes with regard to the cooling/warming of the climate.

Figure 25. Distributions of surface and groundwater deliveries for each temperature

scenario

This plot indicates that the distribution of deliveries across scenarios does not change as the winter index changes. Regardless of how the timing of the inflows is varied, the model can dampen the effect of the timing shift through optimal operation of upstream reservoirs to eliminate any significant shift in delivery volumes each month.

This can likely be attributed to perfect foresight and large amounts of groundwater and surface water storage in the model (Draper 2001).

44 Section 6 – Discussion

6.1 Optimal Reservoir Operations

With optimal statewide management, results show that a shift in the timing inflows due to a warmer climate would have a small impact compared to shift in water availability, the effect of which is much more pronounced. In the warmer scenarios,

CALVIN allocates more water to downstream reservoirs in the winter months to account for the additional runoff beyond the capacity of the upstream reservoirs. Large reservoirs like Shasta Lake and Lake Oroville historically maintain high storage levels in winter

(exclusive of flood pool storage) and the additional runoff appearing early in the year cannot be held at these reservoirs. Due to its role as an offstream storage reservoir, San

Luis has capacity to capture this additional water and store until it can be delivered to meet a demand. San Luis Reservoir also shows increased storage levels as the average water availability decreases from historical to -10%.

Both the statewide surface water storage and total shortage cost point towards a sharp increase in shortage and decrease in reservoir volumes occurring between -20% and

-30% water availability. Also, significant reductions in reservoir levels exist compared to the difference between the other scenarios. Results show reservoir storage decreasing with water availability, supporting the fact that reservoir expansion will likely not be the answer to the state’s climate problems. Managing the state’s surface storage capacity is more promising and less costly than constructing new storage facilities or expanding existing ones (Hanak and Lund 2012).

45 6.2 Collaboration for Adaptive Operations

The flexible operations suggested in the CALVIN model requires collaboration among stakeholders in the statewide water network. The results show no “one size fits all” reservoir operating policy for all reservoirs across the state. A combination of adaptation strategies working across the state yielded optimal results; some reservoirs, like Oroville, show minimal changes to operations, while reservoirs such as Trinity Lake and New

Melones Reservoir steadily decreased average monthly storage levels with decreasing water availability.

Resource management and collaborative governance is a popular topic among public policy researchers. Efforts to combine engineering and policy research may benefit water policy in California, where over 400 water utilities control water infrastructure, making cooperation difficult. The overlapping authority of federal, state, and local authorities has led to lack of consensus and inaction on key issues like managing the

Sacramento Delta (Madani and Lund 2012). Moreover, the passage of the Sustainable

Groundwater Management Act (SGMA) in 2014 empowers new entities and expands responsibilities of many districts to establish groundwater regulations, further complicating the chain of command over water control. When multiple governmental units are making policy decisions, this division of power helps control any one body concentrating too much power and can promote competition and innovation. However, it also imposes inefficiencies, as decisions by one governmental unit impose positive and negative externalities on others (Feiock 2009). In conjunction with modeling, effective strategies for climate adaptation require collaboration. CALVIN’s lack of institutional

46 boundaries allows exploration of new policies establishing more optimal operations, but more work is required of policy makers to establish a collaborative governance structure to California water resources (Pahl-Wostl 2006).

6.3 Modeling Limitations

Climate adaptation studies have typically focused on drought and minimizing shortage cost, but recent climate studies suggest increased precipitation and intensity of storms in California’s future (Dettinger et al. 2011). To address this, existing climate adaptation models can include can include ways to measure flood potential and provide flood management strategies. In particular, the CALVIN model’s main utility lies in long- term drought and scarcity cost management based on monthly hydrologic input and does not include extensive hydrologic modeling of peak flood volume and timing on rim inflows. A model with hourly, daily, or weekly hydrologic input data would be more apt than CALVIN to explore optimal flood management.

Climate adaptation modeling is also constrained by the uncertainty of the atmospheric modeling performed. Historically, climatologists have not agreed whether

California will receive more or less precipitation from predictions ranging from -11% to

+28% from the historical average (Connell-Buck et al. 2011). Most recently, University of

California Riverside researchers have proposed California will receive more precipitation as a result of warming in the Pacific Ocean similar to the El Niño phenomena (Allen and

Luptowitz 2017). Moreover, the climate models provided by the California Energy

Commission all point to a wetter climate in California as represented in Figure 8.

47 A significant challenge in hydro-economic modeling is perfect foresight, which allows the model to base its decisions at each timestep on the information of the whole record rather than a narrow forecast window (Draper 2001). This leads to the model preparing for drought and wet periods well before our current meteorological prediction ability would allow. In the real world, the timing and length of droughts is unknown, which motivates the need for limited foresight models. Although CALVIN’s perfect foresight across the 82 year time period hinders the direct application of its reservoir operating rules in the real world, the results still yield useful trends and point to the potential for more optimal system operations that can be studied in detail with a more representative model.

Another improvement to the CALVIN model would emulate other California simulation models like CALSIM to provide “a business as usual” baseline that would emulate current institutional operating rules to compare the optimized management strategies to. Harou et. al. (2010) implements this idea, but did not become included in the primary CALVIN model. The value of this format in CALVIN to emulate current operating procedures would show the economic benefit of cooperative management of reservoirs by identifying the reductions in shortage, shortage cost, and marginal values across the network.

48 Section 7 – Conclusion

Climate change poses a significant threat to California, but proper planning and adaptive water resources management strategies will provide significant relief from future stressors on the water network. Modeling plays a critical role in better understanding climate effects, including a combination of global circulation models water resources planning models to simulate or optimize operations. Furthermore, hydro-economic modeling can enlighten optimal strategies for adaptation and suggest potential operational improvements to the water network.

While reservoir and conveyance expansion remain optional and costly, and reduction to environmental flows remains riddled with legal concerns, adaptive operation of reservoirs statewide yields positive results to manage a variety of hydrologic scenarios.

Flexible management of the statewide network of reservoirs controlled by various institutions requires collaborative governance of water resources within California, which history has shown is a difficult task. Through continued efforts to identify vulnerabilities and develop adaptation strategies, California can adjust to drier and warmer climates by working towards collaboratively managing water at a statewide scale.

49 References

Allen, Robert J., and Rainer Luptowitz. 2017. “El Niño-like Teleconnection Increases California Precipitation in Response to Warming.” Nature Communications 8 (July): 16055. doi:10.1038/ncomms16055. California Department of Water Resources. 2014. “Chapter 5 - Managing an Uncertain Future.” California Water Plan Update 2013 Volume 1. Chalecki, Elizabeth L., and Peter H. Gleick. 1999. “A Framework of Ordered Climate Effects on Water Resources: A Comprehensive Bibliography.” JAWRA Journal of the American Water Resources Association 35 (6): 1657–1665. Connell-Buck, Christina R., Josué Medellín-Azuara, Jay R. Lund, and Kaveh Madani. 2011. “Adapting California’s Water System to Warm vs. Dry Climates.” Climatic Change 109 (S1): 133–49. doi:10.1007/s10584-011-0302-7. Dettinger, Michael D. 2005. “From Climate-Change Spaghetti to Climate-Change Distributions for 21st-Century California.” San Francisco Estuary and Watershed Science 3 (1). http://escholarship.org/uc/item/2pg6c039.pdf. Dettinger, Michael D., Fred Martin Ralph, Tapash Das, Paul J. Neiman, and Daniel R. Cayan. 2011. “Atmospheric Rivers, Floods and the Water Resources of California.” Water 3 (4): 445–78. doi:10.3390/w3020445. Dogan, Mustafa. 2015. “Integrated Water Operations in California: Hydropower, Overdraft, and Climate Change.” Master’s Thesis, University of California Davis. Draper, Andrew J. 2001. “Implicit Stochastic Optimization with Limited Foresight for Reservoir Systems.” Dissertation, University of California Davis. Draper, Andrew J., Marion W. Jenkins, Kenneth W. Kirby, Jay R. Lund, and Richard E. Howitt. 2003. “Economic-Engineering Optimization for California Water Management.” Journal of Water Resources Planning and Management 129 (3): 155– 164. Feiock, Richard C. 2009. “Metropolitan Governance and Institutional Collective Action.” Urban Affairs Review 44 (3): 356–77. doi:10.1177/1078087408324000. Groves, David G., Jordan Fischback, and RAND Co. 2013. “Adapting to a Changing Colorado River: Making Future Water Deliveries More Reliable Through Robust Management Strategies.” US Bureau of Reclamation. Hanak, Ellen, and Jay R. Lund. 2012. “Adapting California’s Water Management to Climate Change.” Climatic Change 111 (1): 17–44. doi:10.1007/s10584-011-0241-3. Harou, Julien J., Josué Medellín-Azuara, Tingju Zhu, Stacy K. Tanaka, Jay R. Lund, Scott Stine, Marcelo A. Olivares, and Marion W. Jenkins. 2010. “Economic Consequences of Optimized Water Management for a Prolonged, Severe Drought in California:

50 ECONOMIC CONSEQUENCES OF PROLONGED SEVERE DROUGHT.” Water Resources Research 46 (5): n/a-n/a. doi:10.1029/2008WR007681. Hashimoto, T., R. Stedinger, and D.P. Loucks. 1982. “Reliability, Resiliency, and Vulnerability Criteria for Water Resource System Performance Evaluation.” Water Resources Research, no. 18(1): 14–20. doi:10.1029/WR018i001p00014. Herman, Jonathan D., Patrick M. Reed, Harrison B. Zeff, and Gregory W. Characklis. 2015. “How Should Robustness Be Defined for Water Systems Planning under Change?” Journal of Water Resources Planning and Management 141 (10): 4015012. doi:10.1061/(ASCE)WR.1943-5452.0000509. IPCC. 2008. Climate Change and Water. Edited by Bryson Bates and Zbigniew W. Kundzewicz. IPCC Technical Paper; 6. Lempert, Robert J. 2002. “A New Decision Sciences for Complex Systems.” Proceedings of the National Academy of Sciences 99 (suppl 3): 7309–7313. Lempert, Robert J., and David G. Groves. 2010. “Identifying and Evaluating Robust Adaptive Policy Responses to Climate Change for Water Management Agencies in the American West.” Technological Forecasting and Social Change 77 (6): 960–74. doi:10.1016/j.techfore.2010.04.007. Loomis, J., J. Koteen, and B. Hurd. 2003. Economic and Institutional Strategies for Adapting to Water Resource Effects of Climate Change. Water and Climate in the Western United States. Boulder, Colorado: University Press of Colorado. Madani, Kaveh, and Jay R. Lund. 2012. “California’s Sacramento–San Joaquin Delta Conflict: From Cooperation to Chicken.” Journal of Water Resources Planning and Management 138 (2): 90–99. doi:10.1061/(ASCE)WR.1943-5452.0000164. Maier, H.R., J.H.A. Guillaume, H. van Delden, G.A. Riddell, M. Haasnoot, and J.H. Kwakkel. 2016. “An Uncertain Future, Deep Uncertainty, Scenarios, Robustness and Adaptation: How Do They Fit Together?” Environmental Modelling & Software 81 (July): 154–64. doi:10.1016/j.envsoft.2016.03.014. Medellín-Azuara, Josué, Julien J. Harou, Marcelo A. Olivares, Kaveh Madani, Jay R. Lund, Richard E. Howitt, Stacy K. Tanaka, Marion W. Jenkins, and Tingju Zhu. 2008. “Adaptability and Adaptations of California’s Water Supply System to Dry Climate Warming.” Climatic Change 87 (S1): 75–90. doi:10.1007/s10584-007-9355-z. Miller, Norman L., Kathy E. Bashford, and Eric Strem. 2003. Climate Change Sensitivity Study of California Hydrology. Global Climate Change and California: Potential Implications for Ecosystems, Health, and the Economy. Draft Final Report. February. http://www.energy.ca.gov/reports/500-03-058/2003-10-31_500-03- 058CF_A08.PDF. Olmstead, Sheila M. 2014. “Climate Change Adaptation and Water Resource Management: A Review of the Literature.” Energy Economics 46 (November): 500– 509. doi:10.1016/j.eneco.2013.09.005.

51 Pahl-Wostl, Claudia. 2006. “Transitions towards Adaptive Management of Water Facing Climate and Global Change.” Water Resources Management 21 (1): 49–62. doi:10.1007/s11269-006-9040-4. Rogers, Paul. 2017. “Plans for Major New Reservoir in Santa Clara County Moving Forward.” The Mercury News. April 21. http://www.mercurynews.com/2017/04/21/plans-for-major-new-reservoir-in-santa- clara-county-moving-forward/. Solomon, Susan, Intergovernmental Panel on Climate Change, and Intergovernmental Panel on Climate Change, eds. 2007. Climate Change 2007: The Physical Science Basis: Contribution of Working Group I to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change. Cambridge ; New York: Cambridge University Press. Tanaka, Stacy K., Tingju Zhu, Jay R. Lund, Richard E. Howitt, Marion W. Jenkins, Manuel A. Pulido, Mélanie Tauber, Randall S. Ritzema, and Inês C. Ferreira. 2006. “Climate Warming and Water Management Adaptation for California.” Climatic Change 76 (3–4): 361–87. doi:10.1007/s10584-006-9079-5. Vicuna, S., and J. A. Dracup. 2007. “The Evolution of Climate Change Impact Studies on Hydrology and Water Resources in California.” Climatic Change 82 (3–4): 327–50. doi:10.1007/s10584-006-9207-2. Weaver, Christopher P., Robert J. Lempert, Casey Brown, John A. Hall, David Revell, and Daniel Sarewitz. 2013. “Improving the Contribution of Climate Model Information to Decision Making: The Value and Demands of Robust Decision Frameworks: The Value and Demands of Robust Decision Frameworks.” Wiley Interdisciplinary Reviews: Climate Change 4 (1): 39–60. doi:10.1002/wcc.202. Yates, David, David Purkey, Jack Sieber, Annette Huber-Lee, Hector Galbraith, Jordan West, Susan Herrod-Julius, Chuck Young, Brian Joyce, and Mohammad Rayej. 2009. “Climate Driven Water Resources Model of the Sacramento Basin, California.” Journal of Water Resources Planning and Management 135 (5): 303–313.