Abstract

Understanding the uncertainty in the projected impacts of climate change on California’s Sierra Nevada hydrology will clarify where hydrologic impacts can be expected with higher confidence, and will help address scientific questions related to possible improvements in climate modeling. In this study, we focus on California, a region that is vulnerable to hydrologic impacts of climate change. We statistically bias correct and downscale the monthly temperature and precipitation projections from 10 global climate models (GCMs) from the Coupled Model Intercomparison Project. These GCM simulations include both a control period (with unchanging CO2 and other atmospheric forcing) and a perturbed period with a 1 percent per year increase in CO2 concentration. We force a distributed hydrologic model with bias-corrected and statistically downscaled GCM data, and generate streamflow at strategic points in the Sacramento-San Joaquin River basin. Among our findings are that inter-model variability does not prevent significant detection of decreases in summer low flows, increases in winter flows or the shifting of flow to earlier in the year. Uncertainty due to sampling of a 20-year period in an extended GCM simulation accounts for the majority of inter-model variability for summer and fall months, while varying GCM responses to future (perturbed) temperature and precipitation forcing add to the variability in the winter. Inter-model variation in projected precipitation accounts for most of the uncertainty in winter and spring flow increases in both the North and South regions, with a greater influence in the North. The influence of inter-model precipitation variability on late summer streamflow decreases in later years, as higher temperatures dominate the hydrologic response, and melting snowpack has less influence.

1

Edwin P. MaurerEdwin P. Maurer(1)(1) and Philip B. Duffy and Philip B. Duffy(2)(2)

(1)(1)Department of Civil Engineering, Santa Clara University, Santa Clara, CA 95053Department of Civil Engineering, Santa Clara University, Santa Clara, CA 95053(2)(2)Atmospheric Science Division, Lawrence Livermore National Laboratory, Livermore CA 94551Atmospheric Science Division, Lawrence Livermore National Laboratory, Livermore CA 94551

Poster: U53A-0705

Selection and Use of GCMs In This Study

Output from all 10 GCMs participating in most recent phase of Coupled Model Intercomparison Project. Model simulations included:•Specified control (constant CO2)•Perturbed (1%/year CO2 increase) simulationsAbbrev. Model, Year Sponsor Abbrev. Model, Year SponsorCCCMA CCCMA, 2001 Canadian Centre for

Climate Modelling and Analysis

MD ECHO-G, 1999 Model & Data Group (Germany)

CSIRO CSIRO_Mk2, 1997

Commonwealth Scientific & Industrial Research Organization

MPI ECHAM4_OPYC3, 1996

Max Planck Institut fur Meteorologie

GFDL GFDL_R30_c, 1996

Geophysical Fluid Dynamics Laboratory

MRI MRI_CGCM2.3, 2002

Meteorological Research Institute (Japan)

HadCM2 HadCM2, 1995 UK Meteorological Office NCAR CCSM2.0, 2002 National Center for Atmospheric Research

HadCM3 HadCM3, 1997 UK Meteorological Office PCM PCM, 1999 Department of Energy (USA)

•GCMs have biases on order of anticipated changes•GCM spatial scale is incompatible with hydrologic processes

Example of Bias in GCMs

40-year control period GCM simulations

One grid cell: Latitude 39N Longitude 123W

Biases in both median and variability

Precipitation Temperature

Precipitation and Temperature Projections – 70 years at 1%/year CO2 increase

Regional P, T for California

P displays no apparent trend

T shows increasing trend in all seasons and for all GCMs

Precipitation Temperature

Control Period Perturbed Years 21-40 Perturbed Years 51-70

•Control period: minor variability due to differences in flow sequencing and spatial correlation in GCMs.•Inter-model variation appears within first few decades, reflecting differences in GCM parameterization, resolution, CO2 sensitivity.•Between 30 and 60 years, uncertainty does not appear to increase, except perhaps in early Spring in South.

To correct for the bias in the GCMs, the technique of Wood et al. (2004; 2002) was applied. This uses a quantile mapping technique that constrains the GCM to reproduce all statistical moments of the observed precipitation and temperature for a climatological (control) period, while allowing both the mean, variance, and other moments to evolve in the future as simulated by each GCM.

Bias-corrected HadCM3 Precipitation, mm/d

Bias-corrected, downscaled HadCM3 Precipitation

125% 118% 116%

120% 116% 112%

117% 109% 107%

108% 105%

102%

2 Implementation of Hydrologic Model

VIC Model Features:•Developed over 10 years•Energy and water budget closure at each time step•Multiple vegetation classes in each cell•Sub-grid elevation band definition (for snow)•Subgrid infiltration/runoff variability

•VIC Model is driven with GCM-simulated (bias-corrected, downscaled) P, T

•Reproduces Q for historic period•Produces runoff, streamflow, snow, soil moisture,…

Bias corrected precipitation and temperature are spatially downscaled to a 1/8° resolution by interpolation of scale and shift factors of each month to the 1961-1990 month’s base period average. Downscaling over the study domain is illustrated below.

Future climate for California – Simulation Set 1

Future climate for California – Simulation Set 2

Second set of simulations used same P, T forcing as Set 1, but with PCM simulated P for all GCMs. This helped isolate the contribution of inter-model P variability, generally considered more variable between models. PCM was selected since its showed the greatest correspondence each season between climatological P and also was least sensitive to CO2 changes.

The fraction of streamflow variability attributed to precipitation is calculated as:

TP

TTPFraction

TP indicates both T and P vary between all GCMs (Set 1); T indicates only T varies between GCMs (Set 2)

• 3 northern gauges lumped together – inflows to major reservoirs in Northern Sierra.• 4 southern gauges lumped – inflows from major reservoirs in higher elevation, southern Sierra

Nevada.• Together they account for most of the Sacramento-San Joaquin streamflow originating from the

Sierra Nevada mountains.

3 Results

Northern Gauges

Southern Gauges

Control 1-40Control 41-60 Perturbed 21-40 Perturbed 51-70

Month Mean SD Mean SD CV 1-tprob% Mean SD CV 1-tprob

%

1 25299 4876 27431 6959 0.25 63.9 30188 10444 0.35 75.1

2 30383 3262 35570 9126 0.26 89.1 39464 10048 0.25 93.7

3 30889 3797 33406 4650 0.14 87.4 37358 6815 0.18 96.5

4 26955 3084 28045 3935 0.14 58.9 29717 4220 0.14 83.6

5 21502 2146 20189 2562 0.13 84.4 19542 3166 0.16 83.8

6 15400 1608 13158 1549 0.12 99.8 12059 1849 0.15 99.8

7 8692 610 7780 620 0.08 99.8 7501 798 0.11 99.5

8 5960 249 5668 269 0.05 99.1 5547 358 0.06 98.3

9 5024 155 4972 267 0.05 44.2 4891 181 0.04 77.5

10 5517 598 5126 594 0.12 93.1 5062 389 0.08 92.7

11 10114 3173 10557 1128 0.11 73.0 9752 2053 0.21 36.3

12 17935 3869 22941 8618 0.38 89.9 27334 8076 0.3 96.7

Simulation Set 1: Streamflow statistics for the composite hydrograph of the northern three gauges. Mean and standard deviation (SD) are in ft3/s, tprob is the probability (according to a 2-tailed t-test for differences in mean from two distributions with unequal variances) of claiming the mean is different from the control period mean when they are actually the same. 1-tprob is the confidence level that the mean of the perturbed is different from the mean of the control. CV is the coefficient of variation. Statistics are calculated across different climate models and thus measure the degree of consistency between results of different models.

Control 1-40Control 41-60 Perturbed 21-40 Perturbed 51-70

Month Mean SD Mean SD CV 1-tprob% Mean SD CV 1-tprob

%

1 6074 1313 6876 1901 0.28 78.4 7981 2995 0.38 87.7

2 7925 1088 9815 3380 0.34 88.8 11314 3309 0.29 95.0

3 8516 1296 9513 1445 0.15 93.9 12724 2789 0.22 99.8

4 10524 1155 12394 1628 0.13 99.4 14761 2726 0.18 99.8

5 17004 2017 17542 3086 0.18 40.1 18567 3772 0.2 66.3

6 13743 2087 12190 2357 0.19 92.8 10595 2998 0.28 97.2

7 5877 861 4797 938 0.2 99.3 4301 1196 0.28 99.0

8 2651 211 2383 225 0.09 99.4 2242 249 0.11 99.6

9 2108 105 1990 167 0.08 94.6 1891 108 0.06 99.3

10 2113 185 1966 153 0.08 98.5 1869 158 0.08 99.3

11 3040 724 3243 354 0.11 88.7 2934 469 0.16 41.8

12 4746 923 6182 2247 0.36 92.4 7295 2098 0.29 97.2

Northern Gauges Streamflow Southern Gauges Streamflow

Control Period Perturbed 21-40 Perturbed 51-70

Statistic Mean Mean SD CV1-tprob

%Mea

n SD CV

1-tprob

%

Day of year to runoff centroid:

North 74 67 7 0.11 98.4 63 7 0.11 99.9

Day of year to runoff centroid:

South 119 110 9 0.08 99.4 101 7 0.07 100

Statistical comparison of the day of year to the centroid of the annual (water year)

runoff hydrograph.

•Inter-model variability due to sampling a 20-year time slice (unsynchronized low frequency variability in GCMs ) accounts for much almost all summer and fall intermodel variability. Differing GCM responses to CO2 future forcing plays larger role in winter/spring

•Greater uncertainty of changes during seasonal transitions (November and May), especially late in perturbed period (shown by lower significance).

•Increase in March-April flows more significant in South than North•Shift in timing of annual hydrograph (occurrence of center of mass of Oct-Sep flow volume) 11 days earlier in North, 18 days earlier in South – very robust across models.

Table shows the percent of inter-model variability in monthly streamflow for the composite North and South hydrographs attributable to inter-model variability in precipitation. The remainder is attributable to inter-model temperature variability.

• Inter-model variation in projected precipitation accounts for 72-90% of total inter-model variation for Oct-Feb flow changes.

• Inter-model precipitation variability more dominant than temperature variability for streamflow uncertainty except during May-July in the North and June-August in the South.

• Precipitation variability in September is less important in later period, showing lessened effect on late-summer low flow.

  North, % South, %

MonthPerturbed

21-40Perturbed

51-70Perturbed

21-40Perturbed

51-70

1 90 91 90 90

2 78 76 77 74

3 81 84 70 69

4 77 67 40 72

5 43 52 69 63

6 24 48 35 60

7 48 54 49 54

8 57 61 52 38

9 74 57 66 39

10 91 84 72 72

11 83 97 75 86

12 90 84 88 82

Simulation Set 2 – PCM Precipitation for all GCMs

Simulation Set 1 – Streamflow Simulations with 10 GCMs

Cannot use GCM output directly:

4 Parting Thoughts•Intermodel variability between GCMs does not prevent significant detection of decreases in summer streamflow, even by years 21-40.•Both increases in winter streamflow and decreases in summer low flows exceed intermodel variability by years 51-70, as is the retreat of the midpoint of the annual hydrograph.

•As temperatures continue to rise, lagging effects of snow and soil moisture are less able to persist through summer (due to more winter precipitation falling as rain and higher evapotransipiration), and winter precipitation variability becomes less important for late summer low flow changes.