Implications of climate change for river regimes in Wales
– a comparison of scenarios and models
J. A. A. JONES1, N. C. MOUNTAIN2, C. G. PILLING3 & C. P. HOLT4
1 Institute of Geography and Earth Sciences, University of Wales Aberystwyth, UK. 2 Environment Agency, Rivers House, East Quay, Bridgwater, UK. 3 Meteorological Office, Fitzroy Road, Exeter, UK. 4 School of Environmental Science, University College Northampton, UK. BackgroundBackground
•• There is still considerable uncertainty about the likely impact of future climate change on water resources •• Researchers are actively trying to quantify the uncertainties •• And to develop ever more accurate and reproducible results •• Current debate concentrates particularly on the relative merits of different downscaling methods – especially multivariate regression, artificial neural networks and stochastic weather generators AimsAims
•• To compare the results from various approaches we have taken over the last decade to assessing probable impacts on Welsh water resources •• soAs havethe methods the scenarios have become more sophisticated, INITIALINITIAL SENSITIVITYSENSITIVITY STUDIESSTUDIES
At the first Celtic Hydrology Colloquium in Rennes: Holt and Jones (1996) presented results from a sensitivity study based on: i) 3 national scenarios - wettest, best estimate and driest (UK Climate Change Impacts Research Group) ii) The Hadley Centre’s (1992) high resolution model run in transient mode (UKTR) for 2055 ScenariosScenarios usedused inin sensitivitysensitivity studystudy
Winter Spring Summer Autumn
UK Climate Change Impacts Research Group scenarios Temperature +2.3oC +2.2oC* +2.1oC +2.2oC*
Rainfall
- wettest +16% +16% +16% +16% -Best +8% +8% 0% +8% estimate - driest 0% 0% -16% 0%
Transient UKHI GCM scenario for 2055** Temperature +1.0oC +1.8oC +1.0oC +1.5oC Rainfall +45 mm +70.8 mm +22.5 mm +4.8 mm
* interpolated by Holt and Jones ** from UK Hadley Centre and Climatic Research Unit MethodologyMethodology
RiversRivers •• statistical transfer functions derived by correlating: - effectiveeffective precipitationprecipitation from the Met Office’s MORECS cells with - recordedrecorded riverflowsriverflows for 1979 to 1991
ReservoirsReservoirs •• Flows from the regulating reservoir modelled using Welsh Water’s spreadsheet models UnsupportedUnsupported riverriver abstractionabstraction sitessites
SUMMER
20
- wettest scenario n 0 i - Best estimate
nge -20 a f h - driest scenario c -40 unof ge r a
t - 1984 design drought
n -60 e c
r - transient GCM e -80 P
-100 Erch Yscir Taf
Reductions of the order of 25% in summer discharges under the best estimate equilibrium scenario for rivers in North Wales (Erch), mid-Wales (Ysgir) and South Wales (Taf) DirectDirect supplysupply reservoirsreservoirs
SUMMER
20 - wettest scenario
n 0 i - Best estimate
nge -20 a h w - driest scenario o c l -40 f ge
in - 1984 design drought a t
n -60 e
c - transient GCM r
e -80 P
-100 Craig y Pistyll Preseli Crai
Direct supply reservoirs also showed at least 25% reduction in summer inflows, but those in mid-Wales and the Brecon Beacons suffer less than the Preseli reservoir, which receives only half the rainfall of the mid-Wales reservoir (Craig y Pistyll) RiverRiver regulationregulation schemescheme –– damdam inflowinflow andand abstractionabstraction pointpoint
SUMMER
20 - wettest scenario ow
l 0 f
n - Best estimate i -20
nge - driest scenario a
h -40
c - 1984 design drought
ge -60 a t - transient GCM n - transient GCM e
c -80 r e P -100 Llyn Brianne Nantgaredig
Summertime response at the two sites on the River Towy in South Wales, regulated for public water supply by the Llyn Brianne reservoir, was very similar to natural rivers DOWNSCALINGDOWNSCALING CLIMATECLIMATE && LINKINGLINKING HYDROLOGICALHYDROLOGICAL MODELSMODELS
Mathematical downscaling
• Despite the improved spatial resolution in UKHI, the scenarios were far from basin-specific
• Pilling and Jones (1999) used climatologies downscaled and interpolated to a 10 x 10 km grid for the whole of Britain produced by the Climatic Research Unit, University of East Anglia These comprised: - present-day climatology - equilibrium scenario (UKHI) for 2050 - transient scenario (UKTR) for 2065 Downscaling: - present-day fitting a thin plate smoothing spline to irregularly spaced stations - future scenarios using a standard space-filtering scheme LinkingLinking climatologyclimatology toto hydrologicalhydrological simulationsimulation modelmodel
•• Climatologies were used as input to 17-parameter hydrological simulation model HYSIM
•• Surface characteristics were derived individually for each 10 x 10 km gridcell TransientTransient modelmodel -- annualannual
Precipitation
RUNOFF 2065
% change PET TransientTransient modelmodel -- winterwinter
Precipitation
RUNOFF 2065
PET % change TransientTransient modelmodel -- summersummer
Precipitation
RUNOFF 2065
PET % change StatisticalStatistical andand stochasticstochastic downscalingdownscaling
•• A major failing in mathematically downscaled climatologies is that they cannot reproduce topographictopographic influencesinfluences on a smaller scale than the spacing of the original data points
•• A statistical method of “downscaling” scenarios to catchment scale was adopted, which uses local climatological data
•• This implicitly incorporates the effects of local topography MethodMethod outline:outline:
● establish seasonal relationships between regional meso-scale airflow indices and basin rainfall and evaporation using current daily observations
● input simulated future daily airflow parameters into these equations (transition probabilities and correlations)
● input the calculated daily weather parameters into the physically-based lumped rainfall-runoff model HYSIM WeatherWeather simulationsimulation
1. Correlate airflow indices: - Total shear vorticity (Z)
- Geostrophic wind direction components (Dx, Dy) - Strength of airflow (F) with Observed daily precipitation: - to get equations for Probability + Amount of rain
2. Use derived transitional probabilities and correlations for each 3-month season as input to a daily stochastic weather generator StochasticStochastic weatherweather generationgeneration i) Input daily Z and F values into appropriate equation Previous day wet Pww = a + bZ + cZ² + dF + eF² Previous day dry Pdw = f + gZ + hZ² + iF + jF² ii) Mean wet-day amount (Pa) for Z and F that day is Pa = k + lZ + mZ² + nF + oF² iii)Actual daily rainfall amount from random sample
Pday = -Paln(r2)2r3 iv) Repeat for next day
(similar procedure for potential evapotranspiration) RelationshipsRelationships ofof dailydaily probabilitiesprobabilities ofof precipitationprecipitation (A)(A) andand precipitationprecipitation amountamount (B)(B) toto vorticityvorticity
A) 1
y 0.8 t i l bi Wet day probability 0.6 oba pr
y 0.4 a d t e 0.2 Pdw Pww W
0 -80 -60 -40 -20 0 Z-index 20 40 60 80 100
B) 14
12
Wet day amount 10 ount
m 8 a y 6 da t e 4 w n a
e 2 M 0 -80 -60 -40 -20 0 20 40 60 80 100 Z-index
Transitional probabilities for dry-wet / wet-wet sequences then used in a stochastic sampling procedure to generate local daily weather sequences ObservedObserved andand simulatedsimulated dailydaily precipitationprecipitation probabilitiesprobabilities
A) Obser ved 1969- 1978 0. 35 0. 3
y 0. 25 t i l 0. 2 0. 15 obabi r 0. 1 P 0. 05 0 0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40 42 44 46 48 50>50 P recip itatio n (m m)
B) S imulated 1969- 1978 0. 35 0. 3 0. 25 y it
il 0. 2 b
a 0. 15 b o
r 0. 1 P 0. 05 0 0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40 42 44 46 48 50 >50 P recipita tio n (m m) SimulatedSimulated andand observedobserved dailydaily WyeWye discharge,discharge, 19961996
Simulated Wye at Ddol Farm 1996 (Validation) Observed 50 45 40
35
-1 30 s 3
m 25 20 15
10 5 0
Results fit many of the flow parameters for the period of daily discharge records reasonably well UpperUpper WyeWye basinbasin
Pilling and Jones (2002) used this methodology with the runoff simulation program HYSIM, in order to simulate future runoff in the 10.55 km2 Upper Wye catchment UpperUpper WyeWye basinbasin simulations:simulations:
•• This provided a daily sequence of riverflows, which allowed study of variability and runs and, for the first time, changes in the frequency of extreme events based of physical simulations
•• Pilling (1999) used data from the Hadley Centre’s HadCM2: - HadCM2GHG, a “greenhouse gases only” scenario dubbed the “medium-high” estimate adopted as the industry standard for UK water company estimations - HadCM2SUL, which incorporates the cooling effects of sulphate aerosols HadCM2SULHadCM2SUL resultsresults forfor UpperUpper WyeWye basinbasin • Indicate a marked reduction in amount and frequency of precipitation in summer and autumn by 2080-2099
• Shortening of the mean length of wet spells, most pronounced in autumn
• This is accompanied by an increase in evapotranspiration
• These result in: - mean daily discharges -17% summer, -12% autumn (α = 0.01) - marked increase in number of low flow days in summer and autumn
• Also increases in both high and low flows in the year as a whole: - +5% in mean annual flood & -17% in Q95 discharge level (α = 0.01) HadCM2,HadCM2, HadCM3HadCM3 && SDSMSDSM DOWNSCALINGDOWNSCALING
At the second Celtic Hydrology Colloquium in Aberystwyth, Jones and Mountain (2000) applied the same methodology to the larger, 184 km2 basin of the River Elan above the town of Rhayader in mid-Wales RiverRiver ElanElan andand RiverRiver WyeWye aboveabove RhayaderRhayader NewNew scenariosscenarios :: NewNew model:model: NewNew methodmethod • The latest work by Mountain (2004) uses scenarios recently released from the HadCM3 GCM and the improved downscaling methodology, SDSM (Statistical DownScaling Model)
• HadCM3 has 19 atmospheric levels and improved models for convection, minor gases, aerosols and land surface properties
• Using the IPCC scenarios A2 and B2 - not the most extreme scenarios, reaching 1310 and 915 ppmv CO2, respectively, by 2100
• A2 assumes a world population of 15 billion by 2100 and relatively slow economic growth (higher emissions)
• B2 assumes a population of 10.4 billion and more rapid economic growth (lower emissions)
• A2 & B2 were singled out in IPCC (2001) and are now widely used StatisticalStatistical DownScalingDownScaling ModelModel parametersparameters
Parameter Geopotential height
•• SDSM follows the same Sea level 500 hPa 850 hPa general methodology as the original approach, but uses up to 18 indices Temperature ii i
Pressure ii i
•• Each predictor variable is Relative ii i also normalised and humidity standardised before Velocity ii i
regression Vorticity ii i
Divergence ii i COMPARISONCOMPARISON OFOF
HadCM2HadCM2 && HadCM3HadCM3
forfor RiverRiver WyeWye
There are also some important consistencies, despite some notable differences between scenarios, models and methodology CHANGESCHANGES ININ ANNUALANNUAL FLOWSFLOWS
Parameter 2040-2059 2080-2099 HadCM HadCM HadCM HadCM HadCM HadCM HadCM HadCM 2 2 3 3 2 2 3 3 GHG SUL A2 B2 GHG SUL A2 B2
Total -11 -4 +14 -7 -10 +12 discharge
BUT there are differences of sign between scenarios for total discharge:
• HadCM2SUL and HadCM3A2 show a reduction
• HadCM3B2 predicts an increase
• In both cases, the changes remain similar from mid-century onwards CHANGESCHANGES ININ LOWLOW FLOWSFLOWS
Parameter 2040-2059 2080-2099 HadCM HadCM HadCM HadCM HadCM HadCM HadCM HadCM 2 2 3 3 2 2 3 3 GHG SUL A2 B2 GHG SUL A2 B2 Q95 -20 -18 -33 -14 discharge Total no. +12 +25 +30 +145 +36 low flow days Low flow +27 +89 +38 spell length No. of low +27 +94 flow spells
• Changes in Q95 discharge all negative, number of low flow days & length of spells increase late century (Some support for increase in number of spells)
• These all add up to reasonably consistent predictions of worsening water supply and environmental problems due to low flows as the century progresses CHANGESCHANGES ININ HIGHHIGH FLOWSFLOWS
Parameter 2040-2059 2080-2099 HadCM HadCM HadCM HadCM HadCM HadCM HadCM HadCM 2 2 3 3 2 2 3 3 GHG SUL A2 B2 GHG SUL A2 B2 Q5 -8 -8 -7 -7 discharge MAF -7 +8 +14
• Changes in high flows are of a lower order with some inconsistencies
• Half of scenarios show a 7-8% reduction in Q5 discharge, remaining consistent through latter half century
• Mean annual flood is less consistent, but both the newest scenarios show increases by 2080
• This suggests possible increases in more extreme floodflows but not in “run-of-the-mill” high flows. SeasonalSeasonal changeschanges inin riverflowriverflow parametersparameters (%)(%) forfor thethe RiverRiver WyeWye
• Even greater consistency than annual results
• Trend towards increased seasonality found in the previous modelling for Welsh rivers confirmed
• The most constant trend is for lower discharges in summer & autumn PercentPercent changechange inin seasonalseasonal dischargedischarge totalstotals forfor thethe RiverRiver WyeWye
2040 to 2059 2080 to 2099 Parameter & season CM2 CM2 CM3 CM3 CM2 CM2 CM3 CM3 GHG SUL A2 B2 GHG SUL A2 B2
Total Win -12 +7 -10 +9 +5 discharge
Spr
Sum -11 -18 -9 ⇐ -32 -20
Aut -14 -5 -14⇐ -30 -13
• Total yields in summer display a significant negative trend in all four model scenarios for the 2080s, and HadCM2SUL suggests this might begin mid-century • Winter less consistent SeasonalSeasonal lowlow flowsflows -- 11
2040 to 2059 2080 to 2099 Parameter & season CM2 CM2 CM3 CM3 CM2 CM2 CM3 CM3 GHG SUL A2 B2 GHG SUL A2 B2 Q95 Win discharge Spr -13 -5 +14 -12 -6 ⇐ Sum -14 -26 -27 -16 ⇐ Aut -10 -27 -19 -9 -45 -25 ⇐ No. of low Win flow days Spr -80 Sum +25 +76 +8 +102 +40 ⇐ Aut +12 +44 +63 +52 +309 +64 ⇐
• Reduction in total discharge matched by reduction in Q95 (many models & scenarios show this beginning in spring and carries on to autumn)
• Marked increases in the number of days with low flows in summer & autumn, rising to over 100% in some scenarios by the 2080s SeasonalSeasonal lowlow flowsflows -- 22
2040 to 2059 2080 to 2099 Parameter & season CM2 CM2 CM3 CM3 CM2 CM2 CM3 CM3 GHG SUL A2 B2 GHG SUL A2 B2 Low flow Win spell Spr length Sum +63 +98 ⇐ Aut +150 +50 ⇐ No. of low Win flow Spr spells Sum +33 +49 +36 ⇐ Aut +38 +191 +55 ⇐
• In HadCM3 the summer & autumn low spells are both longer and more frequent 2040 to 2059 2080 to 2099 Parameter & season CM2 CM2 CM3 CM3 CM2 CM2 CM3 CM3 GHG SUL A2 B2 GHG SUL A2 B2 Q5 Win -7 +4 -8 +7 +6 SeasonalSeasonal highhigh flowsflows discharge
Spr
Sum -35 -21
Aut -11 -12 -15 High flows reduced Max daily Win -8 +7 +10 -9 +7 +11 flow in summer, Spr increased in winter
Sum -9 -14 -28 -18
Aut -9 +10 +20 -9 -9
POT Win +23 +30 +17 ⇐ Marked increases in peaks- over-threshold in winter Spr
Sum -18 -21 -12 -56 -34 ⇐ Consistent reduction in summer peaks-over-threshold Aut -25 -28 CONCLUSIONSCONCLUSIONS
•• In spite of the wide range in models and scenarios, the results are remarkably similar:
•• IncreaseIncrease inin seasonalityseasonality - markedly lower flows in summer/autumn and lesser increases in winter
•• IncreasesIncreases inin extremeextreme flowsflows at both ends - more low flows in summer, higher flows in winter
•• ChangesChanges tendtend toto balancebalance outout overover thethe yearyear - little change in overall water resources •• The main problems for water management identified in all models and scenarios are twofold: - drierdrier summerssummers and - increasedincreased riskrisk ofof floodsfloods && lowlow flowsflows
•• This message appears remarkably robustrobust and independentindependent ofof modellingmodelling procedureprocedure
•• Although the more sophisticated modelling has been confined to rivers in mid-Wales, they are in keeping with earlier results from sensitivity studies on rivers and dams in north and south Wales TheThe EndEnd