Tree-ring reconstructions of streamflow and climate and their application to water management Jeff Lukas Western Water Assessment, University of Colorado Connie Woodhouse University of Arizona & Climate Assessment for the Southwest (CLIMAS) An annotated slide presentation - updated June 2009

Outline of Presentation 1)What is the value of tree-ring reconstructions? 2)How tree rings record climate information 3)Building the tree-ring chronology 4)Generating the reconstruction of streamflow 5)Uncertainty in the reconstructions 6)What the reconstructions can tell us about past drought 7)How the reconstructions are being applied to water management 8)Why the reconstructions are relevant in a changing climate 9)Summary and the TreeFlow web resource

Part 1: What is the value of tree-ring reconstructions?

Learning from experience in water management Colorado at Lees Ferry Gaged (natural flow) record,

Learning from experience in water management Colorado at Lees Ferry Gaged (natural flow) record,

Learning from experience in water management Colorado at Lees Ferry Gaged (natural flow) record,

Tree-ring reconstructions - a surrogate for experience Colorado at Lees Ferry Gaged (natural flow) record

Tree-ring reconstructions - a surrogate for experience Colorado at Lees Ferry Gaged (natural flow) record Tree-ring reconstruction

Tree-ring reconstructions - a surrogate for experience Benefits: - Improved anticipation (not prediction) of future conditions - Improved assessment of risk Tree-ring reconstruction ?

Dendrochronology: the science that deals with the dating and study of annual growth layers in wood Fritts 1976 Main products: - Reconstructions of past conditions; continuous time- series of environmental variables (e.g., climate, hydrology) - Dates of environmental and human events (e.g., fires, infestations, prehistoric settlement)

Tree-ring science and streamflow reconstructions are not new 1900s - Douglass links tree growth and climate in Southwest 1930s - First studies relating tree growth to runoff 1940s - Schulman investigates history of Colorado River flow using tree rings 1960s - Fritts develops modern statistical methods for climate reconstruction Stockton and Jacoby reconstruction of Lees Ferry streamflow 1980s – Further refinement of analytical techniques 1990s 2000s – Many new flow reconstructions for western US; major increase in applications to water management Douglass Schulman

Part 2: How tree rings record climate information

Across much of the western US, annual tree growth is limited by moisture availability So: – a dry year leads to a narrow growth ring – a wet year leads to a wide growth ring Douglas-fir, south-central CO

The moisture signal recorded by trees in the interior West is particularly strong Here, the annual ring widths from one tree are closely correlated to the annual basin precipitation (r = 0.78) from Our goal is to capture and enhance the moisture signal, and reduce noise, through careful sampling and data processing Western CO Annual Precip vs.Pinyon ring width (WIL731) Precipitation (in.) Ring Width (mm)

Main moisture-sensitive trees in the western US Douglas-fir Pinyon PinePonderosa Pine

Stressful sites produce ring series with a stronger moisture signal from Fritts 1976

Regional scale of moisture variability = regional coherence in the moisture signal Image courtesy of K. Kipfmueller (U. MN) and T. Swetnam (U. AZ)

This moisture signal in tree rings can serve as a proxy for multiple moisture-related variables Annual (water-year) or cool season precipitation Drought indices (e.g., summer PDSI) Snow-water equivalent (SWE) Annual (water-year) streamflow These variables are closely correlated in much of the western US, and trees whose ring widths are a good proxy for one tend to be good proxies for all of them

Ring-width and streamflow - an indirect but robust relationship Like ring width, streamflow integrates the effects of precipitation and evapotranspiration, as mediated by the soil Image courtesy of D. Meko (U. AZ)

Part 3: Building a tree-ring chronology Chronology: time-series of site ring-width variability and building block for the reconstruction

Core trees at a site, same species (pinyon, ponderosa, Doug-fir) Goal: maximize the number of samples throughout the chronology ( years) Can also core or cut cross-sections from dead trees 1) Sampling the trees

2) Crossdating the samples Because of the common climate signal, the pattern of wide and narrow rings is highly replicated between trees at a site, and between nearby sites This allows crossdating: the assignment of absolute dates to annual rings (not just ring-counting) Two Douglas-fir trees south of Boulder, CO When cored, the current year of growth is the first ring next to the bark

Crossdating allows the extension of tree-ring records back in time using living and dead wood Image courtesy of LTRR (U. AZ)

Computer-assisted measurement system with sliding stage –captures position of core to nearest 0.001mm (1 micron) Output from measurement system are ring-width series 3) Measuring the samples stage

4) Detrending the measured ring-width series Ring-width series typically have a declining trend with time because of tree geometry These are low-frequency noise (i.e. non-climatic) Ring-width series are detrended with straight line, exponential curve, or spline functions These standardized series are compiled into the site chronology Image courtesy of LTRR (U. AZ) Before detrending After detrending

Other data treatment may be used to address persistence in tree growth from year to year The climate in a given year (t) can also influence growth in succeeding years (t+1, t+2, etc.) through storage of sugars and growth of needles This persistence can be greater than the persistence in hydrologic time series Terminology: –Standard chronology: persistence in the ring-width series is retained –Residual chronology: first-order persistence is removed

5) Compiling the ring-width series into the chronology Ring width index Van Bibber, CO (ponderosa) 30 series from 15 trees Robust averaging

Moisture-sensitive tree-ring chronologies across North America (as used in the Cook et al PDSI reconstructions) Includes multiple species Variable start and end dates

Subset of recently collected chronologies, including many of those used in the latest Colorado River reconstructions

Part 4: Generating the streamflow or climate reconstruction Reconstruction: estimate of past hydrology or climate, based on the relationship between tree-ring data and an observed record

Overview of reconstruction methodology Adapted from graphic by David Meko Tree Rings (predictors) Statistical Calibration Reconstruction Model Streamflow/climate reconstruction Observed Flow/Climate (predictand) Model validation

Moisture sensitive species Location – From a region that is climatically linked to the gage of interest –Because weather systems cross watershed divides, chronologies do not have to be in same basin as gage Length –Last year close to present for the longest calibration period possible –First year as early as possible (>300 years) but in common with a number of chronologies Significant correlation with observed record Requirements: Tree-ring chronologies

Length – minimum 40 years in common with tree-ring data for robust calibration Natural/undepleted record – flows must be corrected for depletions, diversions, evaporation, etc. Homogeneous (climate record) – inspected for station moves, changes in instrumentation Fraser River at Winter Park Undepleted Flow (from Denver Water) USGS Gaged Flow The reconstruction quality relies on the quality of the observed record. Requirements: Observed streamflow/climate record

Tree-ring data are calibrated with an observed streamflow record to generate a statistical model –Individual chronologies are used as predictors (dependent variables) in a statistical model, or –A set of chronologies is reduced through averaging or Principal Components Analysis (PCA), and the average or principal components (representing modes of variability) are used as predictors in a statistical model –Most common statistical method: Linear Regression –Other approaches: neural networks Alternative: Non-Parametric method uses the relationships within the tree- ring data set to resample years from the observed record Reconstruction modeling strategies Tree Rings (predictors) Statistical Calibration Observed Flow/Climate (predictand)

Are regression assumptions satisfied? How does the model validate on data not used to calibrate the model? How does the reconstruction compare to the gage record? Model validation and skill assessment

How does the model validate on data not used to calibrate the model? CalibrationValidation Split-sample with independent calibration and validation periods Cross-validation: leave- one-out method, iterative process Calibration/validation

Two statistics for model assessment Gage R2R2 RE Boulder Creek at Orodell Rio Grande at Del Norte Colorado R at Lees Ferry Gila R. near Solomon Sacramento R CalibrationValidation What are desirable values? Of course, higher R 2 s are best, and positive value of RE indicates some skill (the closer to R 2 the better) Calibration: Explained variance: R 2 Validation: Reduction of Error (RE): model skill compared to no knowledge (e.g., the calibration period mean)

How does the reconstruction compare to the gage record? The means are the same, as expected from the the linear regression Also as expected, the standard deviation (variance) in the reconstruction is lower than in the gage record Observed vs. reconstructed flows - Lees Ferry

Subjective assessment of model quality Are severe drought years replicated well, or at least correctly classified as drought years? Wet years?

Subjective assessment of model quality Are the lengths and total deficits of multi-year droughts replicated reasonably well?

From model to full reconstruction When the regression model has been fully evaluated, the model is applied to the full period of tree-ring data to generate the reconstruction Tree Rings (predictors) Statistical Calibration Reconstruction Model Observed Streamflow (predictand) Model validation

Part 5: Uncertainty in the reconstructions

Sources of uncertainty in reconstructions Observed streamflow and climate records contain errors Trees are imperfect recorders of climate and streamflow, and the reconstruction model will never explain all of the variance in the observed record (model error) A number of decisions are made in the modeling process, all of which can have an effect on the final reconstruction (model sensitivity)

Using the model error to generate confidence intervals for the reconstruction Colorado R. at Lees Ferry Gray band = 95% confidence interval around reconstruction (derived from mean squared error, RMSE) Indicates 95% probability that the observed flow falls within the gray band

Lees Ferry Reconstruction, Year Running Mean Assessing the drought in a multi-century context Data analysis: Dave Meko Application of model error: using RMSE-derived confidence interval in drought analysis

An alternative approach to generate confidence intervals on the reconstruction Noise-added reconstruction approach A large number of plausible realizations of true flow from derived from the reconstructed values and their uncertainty allow for probabilistic analysis. Meko et al. (2001) One of 1000 plausible ensemble of true flows, which together, can be analyzed probabilistically for streamflow statistics

Sensitivity of the reconstruction to choices made in the reconstruction modeling process the calibration record used the span of years used for the calibration the selection of tree-ring data the treatment of tree-ring data the statistical modeling approach used There is usually no clear best model

Sensitivity to calibration period Calibration data ––– Standard Model ––– Ensemble Mean ––– Ensemble Members ––– Each of the 60 traces is a model based on a different calibration period All members have similar sets of predictors South Platte at South Platte, CO Annual Flow (MAF)

Sensitivity to available predictors How sensitive is the reconstruction to the specific predictor chronologies in the pool and in the model? Best stepwise model R 2 = 0.82 Alternate stepwise model - predictors from best model excluded from pool R 2 = 0.79 Animas River at Durango, CO – two independent models

Sensitivity to available predictors The two models correlate at r = 0.89 over their overlap period, Completely independent sets of tree-ring data resulted in very similar reconstructions Animas River at Durango, CO - two independent reconstructions 0 200, , , ,000 1,000,000 1,200, Alternate Best-fit

Analysis from David Meko Sensitivity to other choices made in modeling process Lees Ferry reconstructions from 9 different models that vary according to chronology persistence, pool of predictors, modeling strategy Lees Ferry Reconstructions, 20-yr moving averages

Lees Ferry reconstructions, generated between 1976 and 2007 Differences due to combinations of all of the factors mentioned 20-year running means calibration Stockton-Jacoby (1976), Michaelson (1990), Hidalgo (2001), Woodhouse (2006), Meko (2007)

Colorado at Lees Ferry, Reconstructed and Gaged Flows Extremes of reconstructed flow beyond the gaged record often reflect tree-ring data outside the calibration space of the model Uncertainty related to extreme values

Uncertainty summary We can measure the statistical uncertainty due to the errors in the reconstruction model, but this does not fully reflect uncertainty resulting from sensitivity to model choices There are other ways to estimate reconstruction uncertainty or confidence intervals (i.e. Meko et al. noise added approach) For a given gage, there may be no one reconstruction that is the right one or the final answer Some reconstructions may be more reliable than others (model validation assessment, length of longer calibration period, better match of statistical characteristics of the gage record) A reconstruction is a plausible estimate of past streamflow

Part 6: What reconstructions can tell us about droughts of the past

Colorado River: The 20 th century contains only a sample of the interannual variability of the last 500 years

Rio Grande: The extreme low flows of the past 100 years (like 2002) were exceeded prior to 1900 Gage record in blue, reconstruction in green 5 reconstructed annual flows before 1900 were likely to have been lower than 2002 gaged flow (1685, 1729, 1748, 1773, 1861) 2002

Rio Grande: Multi-year droughts were clustered in time, with fewer droughts in the 20 th century Reconstructed Rio Grande Streamflow, Periods of below-average flow, of 2 years or more (length of bar shows acre-feet below average)

Rio Grande: The longest observed droughts are exceeded in length by pre-1900 droughts LONGEST OBSERVED (5) (11) (6) (7) (6) (6) (7 years) Reconstructed Rio Grande Streamflow, Periods of below-average flow, of 2 years or more (length of bar shows acre-feet below average)

Colorado River: At decadal time scales, the 20 th century is notable for wet periods, but not dry periods

626, , , , , , , , , , , s1600s1700s1800s1900s Annual flow, acre-feet Rio Grande: On century time scales, the 20 th century was overall wetter than the previous four centuries Mean annual flow, by century Reconstructed Rio Grande Streamflow,

25-yr running means of reconstructed and observed annual flow of the Colorado River at Lees Ferry, expressed as percentage of the observed mean (Meko et al. 2007). Reconstructed flow of Colorado River at Lees Ferry, Medieval period Colorado River: The Medieval Period (~ ) had multi-decade dry periods with no analog since

Part 7: How the reconstructions are being used in water management Reconstruction data Policy analysis

Applications of the reconstructions – three main types 1)As qualitative guidance for water managers, stakeholders and decisionmakers 2)For quantitative assessments of long-term hydrologic variability For example, assessing the frequency of a recent drought event in the gage record in the context of the longer reconstruction 3)As direct inputs into hydrologic models of a water system This allows water managers to model system performance under the tree-ring reconstructed hydrology, as they would do with the gaged hydrology Use of the tree-ring data in a water model usually requires further processing of the data (e.g., time-disaggregation)

1) As qualitative guidance for water managers, stakeholders and decisionmakers Example: Tri-fold brochure developed for Rio Grande Water Conservation District to educate water users about long-term variability in water supply

Example: Analysis of lowest mean reconstructed flows for n-length droughts, Boulder Creek, Graphic by Lee Rozaklis, AMEC Earth and Environnmental 2) For quantitative assessments of long-term hydrologic variability

Example: Salt River Project (SRP), AZ. Test of an allotment/pumping strategy SRP recognized that the 1950s design drought (6 years) was shorter than the worst expected future droughts An 11-year reconstructed drought in the Salt-Verde-Tonto basin ( ) was used to test SRPs current allotment and pumping strategy A simple model, using annual inflows, was used (3) Input into a system model, to assess management scenarios

Salt River Project: test of allotment/pumping strategy The 11-year drought reduced reservoir storage to zero in year 11 (blue) A slight change in the allotment/pumping scenario increased it above zero (green)

Example: Denver Water: supplemental approach to water supply yield analyses Standard approach uses only a 45-year period ( ), and the design drought ( ) probably doesnt represent a true worst- case scenario So this supplemental approach uses two tree-ring flow reconstructions for the main supply basins (common period: ) However, Denver Waters system model (PACSM) requires daily model input from 450 locations So an analogue year approach was used which matches each year in the reconstructed flows with one of the 45 model years ( ) with known hydrology and use that years daily hydrology

Denver Water: water supply yield analyses Two paleo-droughts (1680s, 1840s) deplete contents lower than 1950s design drought Reservoir contents with 345 KAF demand and progressive drought restrictions

Example: Bureau of Reclamation: analyses for Colorado River Shortage EIS Appendix N Analyses of Hydrologic Variability Sensitivity …to evaluate the potential effects to the hydrologic resources of alternative hydrologic inflow sequences. Alternative hydrologies: - Two hydrologies based on tree-ring reconstructions of Lees Ferry flow - Block resampling of observed flow - Stochastic manipulation of observed flow

Generate paleo-flow conditionally (K-NN resampling of observed flow) Nonhomogeneous Markov model with kernel smoothing to generate system state Flowchart of paleohydrologic analyses Tree-ring reconstruction of annual streamflow at Lees Ferry Block resample paleo record, retaining paleo flow magnitudes Convert to binary (wet-dry), calculate transition probabilities Non-parametric spatial and temporal disaggregation into monthly flows at 29 model nodes Input into CRSS for policy analyses Direct PaleoPaleo-Conditioned Adapted from Jim Prairie, Reclamation

Direct Paleo sequence based on Meko et al. Lees Ferry reconstruction Modeled Lake Powell (orange) and Lake Mead (green) year-end elevations under No Action (dashed) and Preferred Alternative (solid) Model output from Reclamation Shortage EIS, 2007 No power from Powell

Whos using tree-ring reconstructions? (a partial list) Denver Water Colorado Water Conservation Board Northern Colorado Water Conservancy District Colorado River Water Conservation District Rio Grande Water Conservation District U.S. Bureau of Reclamation – Aspinall Unit City of Boulder City of Westminster New Mexico Interstate Stream Commission Salt River Project (Phoenix) City of Chandler California Department of Water Resources Southern Nevada Water Authority U.S. Bureau of Reclamation - Lower Colorado Colorado New Mexico Arizona California Nevada Multi-state

Responses about use of tree-ring data, from Web survey of tree-ring workshop participants in 2008 (n = 30) from Rice et al, in review

Part 8: How the reconstructions are relevant in a changing climate

There is no reason to think that the range of natural hydrologic variability, particularly periods of drought, documented in the past will not be repeated in the future The difference will be that events such as drought will likely occur under warmer conditions. Although projected changes in precipitation are still uncertain in many area, especially in mountain watershed, projections for temperature are robust. Using the reconstructed flows, rather than just the observed record, as the frame of reference for planning can lead to fewer surprises as we head into a climatically uncertain future Even as humans exert a stronger influence on climate, this influence will still be superimposed on natural variability

One can assume a future warming (e.g, 2 degrees C) and run a hydrologic model to estimate the reduction in flow from that warming, then adjust the reconstructed flows accordingly Or, output of projected Temperature and Precipitation from Global Climate Models (GCMs) can be combined with information from flow reconstructions using several different techniques, to generate future-climate-perturbed hydrology Either way, the resulting hydrology reflects the joint risk of natural climate variability (as seen in the tree-ring data) and future climate change Adding warming to the natural flow variability seen in the reconstructions can provide useful scenarios for the future

Example: Integration of tree-ring reconstructed flows with GCM projections - City of Boulder, CO (with Stratus Consulting, University of Colorado, AMEC Earth & Environmental, and NOAA) Disaggregated the annual reconstructed streamflows into monthly precipitation and temperature so that those variables could be manipulated independently Changes in temperature and precipitation projected from climate models are combined with the tree-ring-derived data to produce simulations of past hydrology under plausible future climate conditions Then the simulated monthly temperature and precipitation were input into a snowmelt-runoff (SRM) and water-balance (WATBAL) model to produce modeled Boulder Creek flows see Southwest Hydrology, Jan/Feb 2007 for an overview Final project report released February 2009

Lee Rozaklis, AMEC Earth and Environmental Worst case scenario: A dry GCM projection imposed on the reconstruction (blue bar = modeled reduction in water delivery) Example: Integration of tree-ring reconstructed flows with GCM projections – City of Boulder, CO

Combining climate projections with tree-ring reconstructions - other applications and references US Bureau of Reclamation: Long-Term Planning Hydrology based on Various Blends of Instrumental Records, Paleoclimate, and Projected Climate Information –Study which evaluated combinations of reconstructions and climate model output for the Gunnison (CO) and Upper Missouri (WY, MT) –review report released April 2009 State of Colorado Water Conservation Board – Colorado River Water Availability Study –Tasks 6 and 7 will combine paleohydrology with climate projections Climate Change and Water Resources Management: A Federal Perspective – report by USGS and other agencies (Brekke et al. 2009) –p. 19: Paleoclimate information…can be useful for developing climate scenarios that include a wide range of potential hydroclimatic conditions.

Part 9: Summary and the TreeFlow web resource

Summary, Part I 1) Tree-ring reconstructions are valuable in that they provide much more hydrologic experience than the observed hydrology 2) Tree growth can be particularly sensitive to variations in moisture availability and thus streamflow 3) The methods to develop tree-ring chronologies and streamflow reconstructions are designed to robustly capture and enhance this moisture signal 4) A reconstruction is a best-estimate based on the relationship between tree-growth and gaged flows; there is always uncertainty in the reconstructed flows

Summary, Part II 5) The reconstructions show greater variability than the observed record, including drought events more severe and sustained 6) Reconstructions can be used in a number of different ways to provide guidance to water managers and decision-makers 7) Reconstructions can be combined with projections from climate models to provide plausible scenarios for future hydrology 8) For more information, and easy access to reconstruction data, please visit the TreeFlow web resource:

The TreeFlow web resource - User-friendly, direct, and basin-organized data access for the western US Data Access PLUS Instructional materials Applications information Workshop archives Analysis tools