development of for specific sites corresponding to selected gcm outputs,using statistical...

Upload: enviropak

Post on 05-Apr-2018

216 views

Category:

Documents


0 download

TRANSCRIPT

  • 8/2/2019 Development of for Specific Sites Corresponding to Selected GCM Outputs,Using Statistical Downscaling Techniques

    1/88

    Research ReportGClSC-RR-09

    Development of Climate Change Scenarios forSpecific Sites Corresponding to Selected GCM Outputs, using Statistical Downscaling Techniques

    Fahad Saeed, Mnhanunad Rehan Anis, Rizwan AslamArshad M. Khan

    June 2009

    Global Change Impact Studies CentreIslamabad, Pakistan

    ,

  • 8/2/2019 Development of for Specific Sites Corresponding to Selected GCM Outputs,Using Statistical Downscaling Techniques

    2/88

    Published by:Global Change Impact Studies Centre (GCISC)

    National Centre for Physics (NCP) ComplexQuaid-i-Azam University CampusP.O. Box 3022, Islamabad-44o00Pakistan

    . @GCISC

    Copyright. This Report, or any part of it, may not be used for resale or any other commercial or gainful purposewithout prior permission of Global Change Impact Studies Centre, Islamabad, Pakistan. For educational or non-profit use, however, any part of the Report may be reproduced with appropriateacknowledgement.

    Published in: June 2009

    This Report may be cited as follows:

    Saeed, F. , M. R. Anis, R Aslam and A.M. Khan, (2009), Development of Climate Change Scenarios for Specific Sites Corresponding to Selected GCM Outputs, using Statistical Downscaling Techniques,GCISC-RR-09, Global Change Impact Studies Centre (GCISC), Islamabad, Pakistan.

  • 8/2/2019 Development of for Specific Sites Corresponding to Selected GCM Outputs,Using Statistical Downscaling Techniques

    3/88

  • 8/2/2019 Development of for Specific Sites Corresponding to Selected GCM Outputs,Using Statistical Downscaling Techniques

    4/88

  • 8/2/2019 Development of for Specific Sites Corresponding to Selected GCM Outputs,Using Statistical Downscaling Techniques

    5/88

    FORWARD

    Global Change Impact Studies Centre (GCISC) Was established in 2002 as a dedicated research centrefor climate change and other global change related studies, at the initiative of Dr. Ishfaq Ahmad, NI, HI,SI .r the then Special Advisor to Chief Executive of Pakistan, The Centre has since been engaged inresearch on past and projected climate change in different sub regions of Pakistan' correspondingimpacts On the country's key sectors; in particular Water and Agriculture; and adaptation measures tocounter the negative impacts.

    The work described in this report was carried out at GCISC and was supported in part by APN(AsiaPacific. Network for Global Change Research), Kobe, Japan, through its CAPaBLE Programme under a 3-year capacity enhancement cum research Project titled "Enhancement of national capabilities inthe application of simulation models for assessment of climate change and its impacts on water resources, and food and agricultural production" awarded to GCISC in 2003 in collaboration withPakistan Meteorological Department (PMD)

    It: is hoped that the report will provide useful information to national planners and policymakers aswell as to academic and research organizations in the country on issues related to impacts of climatechange of Pakistan.

    The keen interest and support by Dr. Ishfaq Ahmad, Advisor (S & T) to the Planning Commission, anduseful technical advice by Dr. Amir Muhammed, Rector, National University for Computer andEmerging Sciences and Member, ScientificPlanning Group throughout the course of his work are gratefully acknowledged.

    Dr. Arshad. KhanExecutive Director, GCISC

    i

  • 8/2/2019 Development of for Specific Sites Corresponding to Selected GCM Outputs,Using Statistical Downscaling Techniques

    6/88

    PREFACE

    Climate Change is one of the most serious problems the world facing today and has become agrowing concern for nationsacross the globe. Human activitieshave now reached a level whereGreen House Gases(GHGs) of anthropogenic origin are significantly affecting the globalenvironment. The research conducted by the Intergovernmental Panel on Climate Change (IPCC)over the last twenty yearshas establishedthat the increasing concentrations of GHGs in theatmosphere caused an increase in average global temperature by about 0.6 C over the Last century.Much larger changes are expec ted in the 21st century, with the average global temperature rising inthe order of 1.1 c - 6.4C.

    The climate change information required for impact studiesshould be of a spatial scalemuch finer than that provided by globalOr regional climate models.Typically, GCMs have : a resolutionof about 300 x 300 km and RCMs of about 30 X 30 km, whereasfor impact assessments studies muchfiner resolution climate observations

  • 8/2/2019 Development of for Specific Sites Corresponding to Selected GCM Outputs,Using Statistical Downscaling Techniques

    7/88

    ii

  • 8/2/2019 Development of for Specific Sites Corresponding to Selected GCM Outputs,Using Statistical Downscaling Techniques

    8/88

    IV

  • 8/2/2019 Development of for Specific Sites Corresponding to Selected GCM Outputs,Using Statistical Downscaling Techniques

    9/88

    1. Introduction:

    Global climate models (GCMs) have resolutions of hundreds of kilometers while Regional climate models(RCMs) may be as fine as tens of kilometers, However, many impact applications require the equivalent o

    point climate observations and are highly sensitive to fine-scale climate variations that are parameterised incoarse-scale models. This is particularly necessary for regions of complex topography} coastal or islandlocations, and. in regions of highlyheterogeneous land-cover.There are several methodologies to bridgethe gap that exists between what the General Circulation Models (GCM.s) are able to simulate with

    enough accuracy and what is needed in climate impact studies. The most straightforward means of obtaining higher spatial resolution scenarios is to apply downscaling technique.

    Downscaling is defined as the creation of a relationship between the large-scale circulation (predictors) andlocal weather variables (predictand), The term downscaling is a bit misleading since the methodology isactually increasing the resolution, therefore scaling up the picture. But downscaling is more referring to the

    proce-ss of moving fromthe large-scale predictor to the local predictand, i.e., moving from. the-large scaleto the small scale. The two frequently used methods are dynamical (physically-based) and statistical(empirical) downscaling.

    In dynamical downscaling a Regional Climate Mode! (RCM) is applied to large-scale circulation usingGCM output (Xu; 1999a). Statistical downscaling is based on the view that the regional climate isconditioned by two factors: the large scale climatic state, and regional/local physiographic features (e.g,topography, land-sea distribution and land use; (Von Storch, 1995, 1999). From this perspective, regional or local climate information is derived by first determining a statistical model which relates large-scale climatevariables (or "predictors") to regional and local variables (or "predictands"). These relationships areempirical (i.e, calibrated from observations). Then the large-scale output of a GCM simulation is fed intothis statistical model to estimate the corresponding local and regional climate characteristics. One of the

    primary advantages of these techniques is that they are computationally inexpensive; and thus Can he easilyapplied to output from different GCM experiments.

    The major theoretical weakness of SD (Statistical Downscaling) methods is that their basic assumption isnot verifiable, i.e, the statistical relationships developed for the present day climate also hold under thedifferent forcing conditions of possible future climates, a limitation that also applies to the physical

    parameterizations of dynamical models.Before going to the main tasks accomplished in the study, a brief preview of climate change is provided inthe next sections.

    1.1 Climate change

    Climate refers to the- average, Or typical, weather conditions observed Over a long period of time for a. givenarea, while weather is the condition of the atmosphere at any given place and time involving factors such astemperature, precipitation, direction and speed of wind, and the amount of water vapor in the air. Theoverall state of the global climate is determined by the balance of solar and terrestrial radiation budgets.How this energy balance is regulated depends upon the fluxes of energy ~ moisture. Mass and momentumwithin the global climate system.

  • 8/2/2019 Development of for Specific Sites Corresponding to Selected GCM Outputs,Using Statistical Downscaling Techniques

    10/88

    In the last 200 years. through increased utilization of the world's resources, humans have begun to influence theglobal climate system, primarily by increasing the Earth's natural greenhouse effect which is widely believed tocause climate change, Climate change) as defined in IPCC, refers to statistically significant variations that

    persist for an extended period, typically decades or longer, It includes shifts in the frequency and magnitude of sporadic weather events as wen as the slow continuous rise in global mean surface temperature (IPCC 3rdAssessment report, 2001), The natural greenhouse effect, in which certain gases in the atmosphere. prevent someof the energy that. otherwise goes. from earth to space, has for millennia warmed the earth's surface (Bruce, et al,2000), However, the earth's climate system has demonstrated unprecedented changes both in global and regionalscales since the pre-industrial era, with Some of these changes attributed to human activities. Emission of greenhouse gases, aerosols and their increasing concentration in the atmosphere continue to alter the atmospherein wa ys that are expected to affect the climate. In terms of radiative forcing by greenhouse gases emitted throughhuman activity, carbon dioxide (C0 2) and methane (CH4) arc the first and second most important,respectively. The increase in atmospheric CO 2 is believed to be caused mainly by fossil-fuel burning and landuse change including deforestation. The increase in CF[~ can be identifies with emissions from energy use,livestock, rice agriculture, and landfill, Increase in concentration of other greenhouse gases such as troposphericozone (0 3) are also attributed to fossil-fuel combustion as well as other industrial and agricultural emissions,

    An increasing number of observations give a collective picture of a warming Earth and other changes in theclimate system. The mean annual global surface temperature has increased by about 0,6 C since fhc late 19thcentury and it is anticipated to further increase by 1.5 - 4.8 C over the next 1 00 years (IPCC, 2007). At thesame time, changes in sea level, snow cover, ice extent, and precipitation are consistent with a warming climatenear the Earth's surface. However, substantial differences are projected in regional changes in climate comparedto the global mean change; All these changes are expected to continue under all IPCC emissions scenariosduring the 21 st century. As the climate system shifts to a new equilibrium, each of its components andsub-components will respond to the complex interaction of feedback loops. Climate impacts can therefore beexpected to occur throughout the system. Some of the major impact areas identified (IPCC 1990b) include.agriculture, forestry, natural ecosystems, hydrology and water resources, human settlements, health, oceans andcoastal zones. Global-mean surface temperature, regional temperature increases, precipitationincreases/decreases. Soil moisture availability, climatic variability and the occurrence of extreme events such ashurricanes are all likely to influence the nature of these impacts.

    2. Downscaling Methods: An Overview

    Downscaling is the means of relating the large scale atmospheric predictor variables to local or station-scalemeteorological records, Such downscaled data could then be used to generate different climate changescenarios. There is a variety of down scaling techniques in the literature, but iu practice two major approachescan be identified at the moment, namely} dynamic downscaling and empirical (statistical) downscaling,

    2

  • 8/2/2019 Development of for Specific Sites Corresponding to Selected GCM Outputs,Using Statistical Downscaling Techniques

    11/88

    1.1 Dynamical Downscaling

    Dynamical downsealinginvolves the nesting of a higher resolutionRCM within acoarser resolution GCM(McGregor~1997; Giorgi and Mearns, 1999). The RCM uses the GCM to define time- varyingatmospheric boundary conditions around a finite domain, within which the physical dynamics of theatmosphere are modeled using horizontal grid spacing of 20~50 km, The main limitation of RCMs is thatthey are as computationally demanding as GCMs (placing constraints on the feasible domain size. number of experiments and duration of simulations). The scenarios produced by RCMs are also sensitive to thechoice of boundary conditions (such as soil moisture) used to initiate experiments, The. main advantage of RCMs is that they can resolve smaller-scale atmospheric features such as orographic precipitation or low-level jets better than the host GCM.

    2.2 Statistical Downscaling

    Statistical downscaling involves developing quantitative relationships between large-scale atmosphericvariables (predictors) and local surface variables (predictands). The most common formhas the predictandas a function of the predictors). Moststatistical downscalingwork has focusedon single-site (i.e .. pointscale) daily precipitation asthe predictand becauseit is themost important inputvariable for many naturalsystems models and cannot be obtained directly fromclimate model output. Predictor sets are typicallyderived fromsea level pressure, geopotentia1 height} wind fields, absolute or relative humidity, andtemperature variables. These variables are archived at the grid resolution of operational and re-analysisclimate models. with the horizontal resolution typically 300--300 km, However, the grid spacing of theobserved climate fields and GCM climate change projection output do not always correspond. Therefore,driving a statistical downscaling model with GCM output often requires interpolation of the GCM fields tothe grid resolution of the atmospheric predictor sets used in fitting.

    2.2.1 Regression models

    Regression-baseddownscaling methods rely on empirical relationships between local-scale predietands andregional-scale predictors). Individual downscaling schemes differ according to the choice of mathematicaltransfer function, predictor variables or statistical fitting procedure. There are different Regressionmethodsfor downscaling such as Linear and non ~ linear regression, artificial neural networks, canonical correlationand principal components analyses (Conway et al., 1996; Schubert and Henderson-Sellers.1997~ Crane andHewitson,1998). The main strength of the regression downscaling is the relative ease of application. 'Themain weakness of regression-based methods is that the models often explainonly a fraction of the observed

    climatevariability (especially in precipitation series). In common with weather typingmethods, regressionmethods also assume validity of the model parameters under future climate conditions, andregression-based downscaling is highly sensitive to the choice of predictor variables and statistical transfer function. Furthermore, downscaling future extreme events using regression methods is problematic sincethese phenomena, by defination, tend to lie at the margins or beyond the range of the calibration data set.

    The steps in regression methods (Heyen et at. 1996) are:

    3

  • 8/2/2019 Development of for Specific Sites Corresponding to Selected GCM Outputs,Using Statistical Downscaling Techniques

    12/88

    1. Identify a large-scale parameter G (predictor) that controls the local parameter L (predictand). If the intent isto calculate L for climate experiments, G should be simulated well by climate models.2. Find a statistical relationship between Land G.3. Validate therelationship with independent data.4. If the relationship is confirmed, G can be derived from GCM experiments to estimate L.

    Figure 1 : The transfer function approach to statistical downscaling. Blue arrows indicate steps based onobserved climate data. Red arrows indicate the application of GCM data to determine site values correspondingto a particular future time period

    Figure 1 illustrates the transfer function approach to statistical downscaling; The first step in this process is thedefinition of the statistical relationships based on observed climate data (indicated by blue arrows on left-handside of this figure). This requires the identification of the large-scale climate variables (also known as theindependent Or predictor variables) to be used in the transfer function(s), These predictor variables may belarge-scale variables, such as mean sea level pressure (MSLP), or it may be necessary nocalculate area-averagevalues for a region, roughly corresponding to the size of the relevant GCM grid box, using station data

    2.2.2 Multiple linear Regression (MLR)

    The multiple linear regression model (MLR) downscales the data directly from seasonal measures of thelarge-scale circulation. It establishes a separate model for each grid point. Each index is. expressed as a linear

    function of a set of potential predictors which were selected using correlation analysis between the indices andall the available predictors. Predictors for each index are then selected from the potential predictors using thestepwise selection method. The selected predictors for the indices varyfrom season to season and from index toindex, However) for a given season the tendency is that the leading predictors for most of the indices are similar.

  • 8/2/2019 Development of for Specific Sites Corresponding to Selected GCM Outputs,Using Statistical Downscaling Techniques

    13/88

    2.2.3. principal-component analysi s (PCA)

    The principal-component analysis (PCA) is an often used method in meteorology. Basi cally it identifiescorrelated patterns of variability within the predictor field anomalies (Hurh and Kyseiy, 2000~ von Storch andZwiers, 1999). The first step of PCA is to find the principal correlation between a. number of variables bystandardizing the data set and analysing the eigen values and eigenvectors of the variance-covariance matrix(Johnson and Wichern, 1998). The anomaly of a field is defined as the deviation at each grid point from theaverage pressure for a standardisation period. The standard peri odwas 19611990 if nothing el se is stated,

    A successful PCA identifies a number of patterns which explains a major part of the variability. The explainedvariance of each PC (Principal Component) is. expressed in the eigenvalues, A scree-diagram, where thestandardized eigen-values arc plotted from the largest to the smallest, can guide how many PCs to keep. A rule

    of thumb is to discard the PCs whose eigenvalues are below the "elbow" of the diagram,

    2.2.4 Canonical correlation analysis (CCA)

    The canonical correlation analysis (CCA) [Barnett and Preisendorfer, ]987] models the Seasonal Indices directlyusing seasonal means of circulation variables. For each season and precipitation index. CCA was carried outusing all possible combinations of potential predictors, The best set of predictors was selected using crossvalidation. The skill measure was the average Spcarm ... m correlation (rank correlation) over an grid points.The CCA was performed on the cross-covariance matrix. of the leading PCS of the predictor and predictandHeld, Only statistically significant pes were retained [Barnett and Prcisendorfer , 1987].

    3. Input need for Downscaling

    The basic goal of statistical downscaling is toderive a transfer function (i.e, a statistical relationship) betwlarge-scaleGCM data as predictors and small-scale climate variables as predictands. So weneed two things todo downscaling i.e.dependent variable (Predictaud) andindependent variables(Predictors). Predictand basically the local weather variable. Mostcommonly used Predictand arc Daily temperature and dailyPrecipitation butMean monthly temperature and Monthly Total Precipitation or even seasonaldata can be usedfor downscaling,

    The predictors data set is usedas independent variable in statistical downscaliug. Two types of predictare available, one is the observed predictor data andother is GCM~derived predictor variables, Sincefor statistical downscahngwe have to developa statistical relationship between predietand and predictors, so theserelationships aredeveloped using observed dataand assuming that these relationships remain validill thefuture, they can be used to obtain downsealed local informationfor SOme future time periodby deriving therelationships with GCMdcrivcd predictors.

    Both the observedand GCM-derived predictor variableshave been normalisedwith respectto their 1961-1990means and standarddeviations, i.e., the mean and standarddevia ion for the 1961-]990 period were calculateand the mean subtracted from each value before dividingby the standard

    5

  • 8/2/2019 Development of for Specific Sites Corresponding to Selected GCM Outputs,Using Statistical Downscaling Techniques

    14/88

    deviation, Since. GCMs do not always perform well at simulating the climate. of a particular region this meansthat there may he large differences between observed and GCM-simulated conditions (i.e., GCM bias or error)which could potentially violate the statistical assumptions giving poor results if the predictor data were notnormalized. The normalization process ensures that "the distributions of observed and GCM-derived predictorsare in closer agreementthan those of the raw observed and raw GCM data.

    The predictors may vary slightly from GCM to GCM depending on the data available, but in general thefollowing predictors arc used.

    Table 1. Large-scale predictor variables obtained from HADCM3 outputs

    *** refersto different atmospheric levels: the surface (p_) ,850 hPa height CP8)and 500 hPa height (p5)

    The preparation of the potential predictors from NCEP reanalysis data and from the GCMs' output involves dataextraction re-gridding and standardisation, The. re-gridding is needed because the grid-spacing (i.e. horizontalresolution) and/or coordinate systems of re-analysis data sets used for Statistical Downscaling modelcalibration do not generally correspond to the grid-spacing of the GCM outputs. For example, the NCEP/NCAR reanalysis data (e.g. Kaluay et al., 1996) have a grid-spacing of 2.5 0 latitude by 3.75 longitude whereas theCGCM} has a coarser resolution of around 3.7 latitude by 3.7 0 longitude and. the HadCM3 a resolution of 2.5 0

    latitude by 3.75 longitude. Deriving a statistical downscaling model with GCM outputs requires aninterpolation

    6

  • 8/2/2019 Development of for Specific Sites Corresponding to Selected GCM Outputs,Using Statistical Downscaling Techniques

    15/88

    procedure of the reanalysisfieldsusedfor modelcalibration to the grid resolution of the GCM atmospheric predictors. Moreover, the standardisation of GCM predic tors is widely used prior to statisticaldownscaling toreduce biasesin the meanand variance 91GCM atmospheric fields relative toobservations(or reanalysisdata; e,g. Wilbyetal., 2004). The procedure involves subtraction of the mean and division by the standarddeviation of the predictor for a predefined baseline period (i.e. 1961-1990). The NCEP reanalysis predictowere re-gridded to conform to thegrid-spacing of CGCM 1 and HadCM3, using the weighted average of neighbouring grid-points. Mearis and standard deviations usedfor standardization of HadCM3 were derivedfrom the baseline period1961-1990.

    3.1 Assumptions for downscaling

    The following criteria have to be fulfilled to develop. a successful downscaling method (Hcwitsonand Crane,1996):

    Large-scale predictors are adequately reproduced by GCMs i.e predictors relevant to the local predictand should be adequately reproducedby the host climatemodelat the spatialscalesused tocondition the downscaled responses. Prior knowledge of climate model limitations can be

    advantageous when screening potential predictors. Therefore, predictors haveto be chosen on the balance of their relevance to the target predic tandts] and their accurate representation by climatmodels (Wilby and Wigley) 2(00).

    The relationship between the predictors and predictand remains valid for periods outside the fittin period (timeinvariance) , this needs careful assessment for future climate projection. asit is obviouslyimpossible to check with observationaldata, A Way aroundthis problem is to valida te the statisticaldcwnscaling model with observational data stemming from periods well separatedfrom the fitting period,i.e, representing a"different"climate regime(Char les et al., 2004).

    The predictor set sufficiently incorporates the future climate change 'signal'. Some approaches mayexclude predictors based on current climate performancethat could be important in future changedclimates.

    The predictors used for determiningfuture local climate should not lie outside the range of theclimatology used to calibrate the Statistical Downscalingmodel.If this is the case, thea technically theStatistical Downscaling model is invalid.

    3.2 Issues For Statistical Downscaling

    3.2.1 Choice of statistical method

    The choice of statistical method is to some extent determinedby the nature of the local predietand. A localvariable that is reasonably normallydistributed,such as monthly mean temperature, will require nothing morecomplicated than (multiple) regression, since large scaleclimate predictors tend to be normally distributed,

    assuming linearity of the relationship. A localvariablethat is highly heterogeneous and discontinuous in spaceand time will probably require a more complicated nonlinear approach or transformation of the raw datFitting such complex models oftenrequireslarge amounts of observational data.

    7

  • 8/2/2019 Development of for Specific Sites Corresponding to Selected GCM Outputs,Using Statistical Downscaling Techniques

    16/88

    3.2.2 Choice of predictors

    Sometimes the best predictorsidentified in the statistical analysis of observations are not completely adequatefor climate change applications.For example, daily rainfallmay be determinedby geopotential heights in theextratropical areas. But changes in geopotential heights caused by global warming will contain anon-dynamical signal, whichwill spuriously affect the estimationof rainfall changes. Thisuon-dynamicalcomponent should be corrected,either by subtracting the average changes of the geopotential height In asufficiently large mea, or by using geopotential thickness, instead of gecpotential heights, as predictors(Burkhardt, 1999). Conversely, exclusion 0f key predictorsfor future change, perhapsdue to a high degreeof covariance with another variableunder current climates. mayresult in a critical loss of informat ion about futureregional response to changes in large scale forcing.

    4. Downscaling Tools

    Although there is no standard' approach to downscaling, i.e., obtaining finer resolution scenarios of climatechange from the coarser resolutionGCM output, there isa softwareavailable whichcan be used to undertake

    downscaling:

    SDSM (StatisticalDownscaling Model) developed byRob Wilby and Christian Dawson in theUK

    4.1 Statistical Downscahng Model (SDSM)

    SDSM permits the spatialdownscaling of daily predictor-predictand relationships using multiple linear regression techniques. The predictor variables provide daily information concerning thelarge-scale stateof theatmosphere> which the predictand describes conditionsat the site scale. The software reduces the task of statistically downscaling dally weather series into a number of discrete processes:

    1. Preliminary screening of potential downscaling predictor variables -identifies those largescale predictor variab 1 es which are signific-antly correlatedwith observed station(predictand) data. Anumber of variables derived from mean sea level pressure fields are included; e. g., air flow strength,meridional and zonal components of air flow; vorticity etc.

    2. Assembly and calibration of statistical downscaling model(s) ~the large-scale predictor variablesidentified in (1) are used in the de terminationof multiple linear regression relationships between thesevariables and the local station data. Statistical models may be built on a monthly, seasonalOr annual

    basis, Information regarding the amount of varianceex plained by the model (s ) and the standard error is given in order to determine thev aralibilty of spatial downscaling for the variable and site inquestion.

    3. Synthesisof ensemblesof current weather data using observed predictor variables ~ Once statisticaldownscaling models have been determined they can be verified by using an independent data set Of observed predictors. The stochastic component of SD SM allows the generation of up to 100 ensemblesof data which have the same statistical characteristics but which vary On a day-to-day basis.

    8

  • 8/2/2019 Development of for Specific Sites Corresponding to Selected GCM Outputs,Using Statistical Downscaling Techniques

    17/88

    4. Generationof ensembles of future weather datausing GCM . -derived predictor varl.ables provision of the appropriate GCM-derived predictor variables allows the generation of ensembles of future weathedata by using the statistical relationships. Calculated in (2).

    5. Diagnostic testing/analysis of observed data and climate change scenarios - it is possible to calculate thstatistical characteristics of both the observed and synthetic data in order for easy comparison and thudetermination of the effect of spatial downscaling,

    5. Case Study

    The objective of the study was to develop climate change scenarios for Pakistan corresponding[0 selectedGCMoutputs, using statistical downscaling technique.

    5.1 Study Area

    Our study area in this work consists of all geographic region of Pakistan, covering an area of 796,095 squakilometers of territory including a wide variety of landscapes, from arid deserts to lush green valleys to stamountain peaks Figure 2. It consists of 50 stations. Geographically, it can be divided into three regions: thlowlands in the south-east and east, the arid plateau in the southwest) and the mountains of the north. With thcomplex topography, the climate of the region varies according to e levation, April through September is the m pleasant months in the mountains, a lthough they bring oppressive heat to the low-lying plains of the Indus Vallwheremid-day temperatures normally exceed 40C. December through February are the coolest months, aslowland temperatures drop to l0-25

    c and the. air temperature in the mountains falls belowfreezing. Monsoonsreach the eastern areas of the country inthe latesummer and persist duringJuly to September.

    Figure2: The DEM (digital elevation model) and climatic observation stations in Pakistan

    9

  • 8/2/2019 Development of for Specific Sites Corresponding to Selected GCM Outputs,Using Statistical Downscaling Techniques

    18/88

    Table 2: List of Stations during 1931 -60 and 1961-90

    10

  • 8/2/2019 Development of for Specific Sites Corresponding to Selected GCM Outputs,Using Statistical Downscaling Techniques

    19/88

    5.2 Data Sources

    5.2.1 Precidant

    The predictand (i. e. the variable to be down scaled)analyzedin this study are obtain from thePakistan

    Meteorological Department(PMD).Thisdata setcontains monthlymeanvaluesof temperature for the period1967-2000 and total precipitation for the period 1967- 2000 obtained at the 50 stations.

    5.2.2 Predictors

    The predictor data set for NCEP andGCM are obtainedfrom the Canadian Climate Impacts Scenarios"(CCIS)Projectwhich was originally funded by the Climate Change ActionFund, starting in October 1999. The predictorsdab set for each station according to its longitude and latitude were downloaded from the sitehttp://www.cics.uvic.cal.

    5.3 Methodology used

    We use multiple linear regression technique for downscaling of temperature and precipitation. Since for mostof the station daily data are not available, so we use monthlymean data for temperature and monthly total precipitation.Since predictor data set of NCEP andGCM are given on daily basis80 we convert the daily datato monthly means data set. SDSM [statistical downscaling model) uses multiple regression technique butit usesdaily data for downscaling, so we did not use SDSM. Wehave calibrated and validated the regression modelsmanually.

    5.3.1 Calibration of model)

    Firstof all we calibratethe model by usingmean monthly temperatureas dependent variable (Predictand) andall 26 predictors, taken from NCEP reanalysis data, as independent variables, The calibration period for temperature is 1967-2000. We have to select appropriate predictors, this remains one of the most challengingstages in the development of statistical downscaling model since the choice of predictors largely determines thecharacter of the downscaled climate scenario(Winkler. et al., 1997; Charles et al., 1999). We use stepwise procedure as predictor selection method, This procedureselected different numbers of predictors for different stations. We calculated the correlation of each selected predictor with the predict-and to check whichvariable ishighly correlated, and also calculated the correlation among the selected predictors to check the presence of multicollinearity. Soby keeping in mind these two points, we chose only that predictor whicharehighly correlatedwithdependent variable but are not highly correlated with other predictor,The model of mean monthly (Wilby ctal., 1999) temperature is

    11

  • 8/2/2019 Development of for Specific Sites Corresponding to Selected GCM Outputs,Using Statistical Downscaling Techniques

    20/88

    Where OJ is monthly total precipitation, o and j are model parameters and e, is error term, The model

    was calibrated for each station separately-

    5.3.2 Validation .of model

    After the models have been calibrated, we run the regression for the validation period using the same predictors selected in the calibration period. FOT validation period we check whether the model produces theR 2 as in calibration period. If the validated model produces high R 2 then it means that model is good for generating scenarios.

    5.3.3 Generate downscaled scenarios

    Having calibrated and verified the Statistical Downscaling model performance, it is then necessary togenerate ensembles of monthly weather series given standardised atmospheric predictor variables suppliedby the GCM (representing either the present or future climate). The GCM used in this study i.'SHadCM3(resolution of 2.5 0 latitude by 3.75 longitude). We have used the A2 Scenarios of HadCM3_ For generationof scenarios, we use the calibrated. model for downscaling by putting the GCM predictors to tile calibratedmodel and obtaining the predicted values which are the downscaled values of Temperature and Precipitation,We divide the whole period (1961- 2099) into four parts, namely the current (1961-2000), the 2020s(2010-2039), the 2050s (2040- 2069\ the 20805 (2070-2099).

    5.4 Downscaling Results

    The downscaling simulation experiment was conductedwith multiple linear regression method. Selectingthe most relevant predictor variables is the first and important task in the downscaling process. In this case, thescreening is achieved with stepwise method. The results show that) predictor variables such as p500 (500hpageo-potential height), mslp (Mean Sea level pressure) and nceptempas (Mean temperature at 2m) areidentified as the most relevant predictors for the temperature while 500hpa meridional velocity (ncepp5 _vas),500hpa vorticity (nceppf _zas)} Relative humidity at 500 hpa (ncepr 500as) are most relevant predictors for

    precipitation, The calibration results of all stations for temperature and precipitation are given in table 3 andtable 4 respectively.

    5.4.1 Downscaling models calibration results

    The calibration performance of multiple lineal' regression model in downscaling the precipitation andtemperature data at the Gupis station is presented in Figure 2. The bar charts in this figure clearly show the

    performance of the model in terms of average monthly statistics of the observed and simulated precipitation and temperature data. In general, the graph shows that the downscaling method simulated themean monthly temperature reasonably well, while precipitation model is not calibrate-dwell. This is

    because of the fact. that precipitation data is heterogeneous, so it is difficult to calibrate the model. Also wedo not have daily data for model calibration and validation and monthly data is not enough for this purpose,

    12

  • 8/2/2019 Development of for Specific Sites Corresponding to Selected GCM Outputs,Using Statistical Downscaling Techniques

    21/88

    Figure 3: Calibration performance in downscaling a) mean monthly temperature data as well as b) monthly total precipitation at Gupis,

    Next we present the calibration result for another station i.e. Astor. Astor is located in orthcrn area oPakistan. The graph shows that calibrated model. simulated the values well for mean monthly temperatureand also for precipitation, The R

    2 of calibration model. for temperature is 0.98 while for precipitation it is

    0.44. Figure 3, shows the bar chartfor temperature and precipitation. The calibration results for remainingstations are given at the end.

    Figure 4; Calibration performance in downscaling a) mean monthly temperature data as well as b) Monthly total precipitation at Astor.

    5.4.2. Downscaling models validation resultsOne of the main assumptions of downscaling is that the relationship between the predictors and predictandremains valid for periods outside the fitting period (time invariance), For this assumption we have tovalidate tile model. We have taken validation period 1987-2000 for temperature and 1986-2000 for

    precipitation. We develop the model for validation period using the predictors selected in calibration. Weuse R 2 and bar chart as validation criterion. The validation performance of multiple linear regression modelin downscaling the precipitation and temperature data at the Gupis station is presented in Figure 4. The bar charts in this figure

    13

  • 8/2/2019 Development of for Specific Sites Corresponding to Selected GCM Outputs,Using Statistical Downscaling Techniques

    22/88

    show that the calibrated model performs well for the period outside the fitting period. The R 2 of calibratedmodel for temperature is 0.96.

    Figure 5: Validation performance in downscaling) mean monthlytemperature data as well as b) monthly total precipitation at Gupis. The bar charts for Astor are given in Figure 5. The R 2 of validatedmodel for temperature is 0.97 while it was 0.98 is the calibration model and R 2 for precipitation is 0.49while it was 0.44 for the calibration model- The validation results for remaining stations are given at theend.

    Figure 6: Validation performance in downscaling a) mean monthly temperature data as wenas b)monthly total precipitation at Astor.

    5.4.3 Climate change scenario

    The next step is to use the calibrated models to downscale precipitation and temperature data correspondingto the future climate change scenario simulated by the GCM. The large-scale predictor variables taken

    from the Canadian GCM (HADCM3) simulation output are used as input to each of the downscalingmodels.

    The monthly means of the downscaling results for Gnpis are summarized and plotted in Figures 6. Thefigure shows an. increasing trend in the Mean monthly temperature values for almost all months of the year while for precipitation; the figure shows no significant trend ill the future.

    14

  • 8/2/2019 Development of for Specific Sites Corresponding to Selected GCM Outputs,Using Statistical Downscaling Techniques

    23/88

    Climate. change scenario graph for Astor are shown in.Figure7. Herefigurealso shows anincreasing trend for temperature butfor precipitationthe trend is somewhat decreasing.Particularly in winter months there issignificantdecrease in precipitation.

    Figure 8: General trend in a) temperature and b) precipitation(It Astor correspondingto a climatechange scenario downscaled with multiple linear regressions.

    This report presentsthe results of a study ondowns-eatingof large scaleatmosphericvariablessimulated withGlobal Climate Models (GCM) to meteorological variables at regionaland local scale in order to investigatethe future climate change scenario.

    Figure 7: General trend in a ) temperatureand b) precipitation a tGupis corresponding to a climatechange scenariodownscaled with multiple linear regressions.

    In this study, multiple regression statistical downscaling methodology was presented using a stepwiseregression. ot unlike other regression approaches the results indicate the strength of statistical downscaling fomodeling temperature and less success [or precipitation. Predictors p500 (500hpa geopotential height). mslp(Mean sealevel pressure) and nceptempas (Mean temperature at 2m) seem to be the most dominant for modeling temperature over Pakistan.

    6. Conclusion

  • 8/2/2019 Development of for Specific Sites Corresponding to Selected GCM Outputs,Using Statistical Downscaling Techniques

    24/88

    For precipitation, results showthat the most important predictor variablesare 500hpa meridional velocity(ncepp 5_ vas), 50 Ohpa vorticity (nccpp5_ zas), Relative humidity at 500hpa (nccprSOOas)_ In general, thedowuscaling methods approximate the observed climatedata corresponding to the current state reasonablywell. However even though the downscaling models indicate an increasing future trend in mean monthlytemperature, while it shows slightly decreasing trend in future precipitation.

    The downscaling calibration results for each station are given below. The calibration period for temperature is1967 -1986 and for precipitation 1961 - 1985.

    I ,

  • 8/2/2019 Development of for Specific Sites Corresponding to Selected GCM Outputs,Using Statistical Downscaling Techniques

    25/88

    17

  • 8/2/2019 Development of for Specific Sites Corresponding to Selected GCM Outputs,Using Statistical Downscaling Techniques

    26/88

    18

  • 8/2/2019 Development of for Specific Sites Corresponding to Selected GCM Outputs,Using Statistical Downscaling Techniques

    27/88

    19

  • 8/2/2019 Development of for Specific Sites Corresponding to Selected GCM Outputs,Using Statistical Downscaling Techniques

    28/88

  • 8/2/2019 Development of for Specific Sites Corresponding to Selected GCM Outputs,Using Statistical Downscaling Techniques

    29/88

  • 8/2/2019 Development of for Specific Sites Corresponding to Selected GCM Outputs,Using Statistical Downscaling Techniques

    30/88

  • 8/2/2019 Development of for Specific Sites Corresponding to Selected GCM Outputs,Using Statistical Downscaling Techniques

    31/88

  • 8/2/2019 Development of for Specific Sites Corresponding to Selected GCM Outputs,Using Statistical Downscaling Techniques

    32/88

    24

  • 8/2/2019 Development of for Specific Sites Corresponding to Selected GCM Outputs,Using Statistical Downscaling Techniques

    33/88

    25

  • 8/2/2019 Development of for Specific Sites Corresponding to Selected GCM Outputs,Using Statistical Downscaling Techniques

    34/88

    26

  • 8/2/2019 Development of for Specific Sites Corresponding to Selected GCM Outputs,Using Statistical Downscaling Techniques

    35/88

    27

  • 8/2/2019 Development of for Specific Sites Corresponding to Selected GCM Outputs,Using Statistical Downscaling Techniques

    36/88

    28

  • 8/2/2019 Development of for Specific Sites Corresponding to Selected GCM Outputs,Using Statistical Downscaling Techniques

    37/88

    29

  • 8/2/2019 Development of for Specific Sites Corresponding to Selected GCM Outputs,Using Statistical Downscaling Techniques

    38/88

    "" ",

    30

  • 8/2/2019 Development of for Specific Sites Corresponding to Selected GCM Outputs,Using Statistical Downscaling Techniques

    39/88

  • 8/2/2019 Development of for Specific Sites Corresponding to Selected GCM Outputs,Using Statistical Downscaling Techniques

    40/88

    The downscaling validation results for each station are given below. The validation period for temperature is 1987-2008 and for precipitation 1986-2000.

  • 8/2/2019 Development of for Specific Sites Corresponding to Selected GCM Outputs,Using Statistical Downscaling Techniques

    41/88

    33

  • 8/2/2019 Development of for Specific Sites Corresponding to Selected GCM Outputs,Using Statistical Downscaling Techniques

    42/88

    34

  • 8/2/2019 Development of for Specific Sites Corresponding to Selected GCM Outputs,Using Statistical Downscaling Techniques

    43/88

    35

  • 8/2/2019 Development of for Specific Sites Corresponding to Selected GCM Outputs,Using Statistical Downscaling Techniques

    44/88

    36

  • 8/2/2019 Development of for Specific Sites Corresponding to Selected GCM Outputs,Using Statistical Downscaling Techniques

    45/88

    37

    ~

  • 8/2/2019 Development of for Specific Sites Corresponding to Selected GCM Outputs,Using Statistical Downscaling Techniques

    46/88

  • 8/2/2019 Development of for Specific Sites Corresponding to Selected GCM Outputs,Using Statistical Downscaling Techniques

    47/88

  • 8/2/2019 Development of for Specific Sites Corresponding to Selected GCM Outputs,Using Statistical Downscaling Techniques

    48/88

  • 8/2/2019 Development of for Specific Sites Corresponding to Selected GCM Outputs,Using Statistical Downscaling Techniques

    49/88

  • 8/2/2019 Development of for Specific Sites Corresponding to Selected GCM Outputs,Using Statistical Downscaling Techniques

    50/88

  • 8/2/2019 Development of for Specific Sites Corresponding to Selected GCM Outputs,Using Statistical Downscaling Techniques

    51/88

  • 8/2/2019 Development of for Specific Sites Corresponding to Selected GCM Outputs,Using Statistical Downscaling Techniques

    52/88

  • 8/2/2019 Development of for Specific Sites Corresponding to Selected GCM Outputs,Using Statistical Downscaling Techniques

    53/88

  • 8/2/2019 Development of for Specific Sites Corresponding to Selected GCM Outputs,Using Statistical Downscaling Techniques

    54/88

  • 8/2/2019 Development of for Specific Sites Corresponding to Selected GCM Outputs,Using Statistical Downscaling Techniques

    55/88

  • 8/2/2019 Development of for Specific Sites Corresponding to Selected GCM Outputs,Using Statistical Downscaling Techniques

    56/88

    47

    I

  • 8/2/2019 Development of for Specific Sites Corresponding to Selected GCM Outputs,Using Statistical Downscaling Techniques

    57/88

    Th e climate change scenarios results for each station are given below

  • 8/2/2019 Development of for Specific Sites Corresponding to Selected GCM Outputs,Using Statistical Downscaling Techniques

    58/88

  • 8/2/2019 Development of for Specific Sites Corresponding to Selected GCM Outputs,Using Statistical Downscaling Techniques

    59/88

  • 8/2/2019 Development of for Specific Sites Corresponding to Selected GCM Outputs,Using Statistical Downscaling Techniques

    60/88

  • 8/2/2019 Development of for Specific Sites Corresponding to Selected GCM Outputs,Using Statistical Downscaling Techniques

    61/88

  • 8/2/2019 Development of for Specific Sites Corresponding to Selected GCM Outputs,Using Statistical Downscaling Techniques

    62/88

  • 8/2/2019 Development of for Specific Sites Corresponding to Selected GCM Outputs,Using Statistical Downscaling Techniques

    63/88

  • 8/2/2019 Development of for Specific Sites Corresponding to Selected GCM Outputs,Using Statistical Downscaling Techniques

    64/88

  • 8/2/2019 Development of for Specific Sites Corresponding to Selected GCM Outputs,Using Statistical Downscaling Techniques

    65/88

  • 8/2/2019 Development of for Specific Sites Corresponding to Selected GCM Outputs,Using Statistical Downscaling Techniques

    66/88

  • 8/2/2019 Development of for Specific Sites Corresponding to Selected GCM Outputs,Using Statistical Downscaling Techniques

    67/88

  • 8/2/2019 Development of for Specific Sites Corresponding to Selected GCM Outputs,Using Statistical Downscaling Techniques

    68/88

  • 8/2/2019 Development of for Specific Sites Corresponding to Selected GCM Outputs,Using Statistical Downscaling Techniques

    69/88

  • 8/2/2019 Development of for Specific Sites Corresponding to Selected GCM Outputs,Using Statistical Downscaling Techniques

    70/88

  • 8/2/2019 Development of for Specific Sites Corresponding to Selected GCM Outputs,Using Statistical Downscaling Techniques

    71/88

  • 8/2/2019 Development of for Specific Sites Corresponding to Selected GCM Outputs,Using Statistical Downscaling Techniques

    72/88

    64

  • 8/2/2019 Development of for Specific Sites Corresponding to Selected GCM Outputs,Using Statistical Downscaling Techniques

    73/88

  • 8/2/2019 Development of for Specific Sites Corresponding to Selected GCM Outputs,Using Statistical Downscaling Techniques

    74/88

    Analyzing the change in Average MonthlyTemperature and Monthly Total Precipitation bydividing the whole region of Pakistan into three

    parts (Northern, Southern and Central) and findingthe Future Trend ill Temperature and

    Precipitation up to 2099 after StatisticalDownscaling of GCM Data using' Regression

    Model

    66

  • 8/2/2019 Development of for Specific Sites Corresponding to Selected GCM Outputs,Using Statistical Downscaling Techniques

    75/88

    Change of Temperature (

    C) and Precipitation (mm) in Northern, C

    and Southern Area of Pakistan.

    67

  • 8/2/2019 Development of for Specific Sites Corresponding to Selected GCM Outputs,Using Statistical Downscaling Techniques

    76/88

    Change of Temperature (

    C) in Northern Area of Pakistan up to 2099

    Table 6: Change in Temperature (

    C) from Base Period (1967 -2000) in Northern Area of Pakistan.

    Change in 2010-2039 Change in 2040-20'69 Change in 20702099Jan 2.55 4.39 I 6.15Feb 2.70 4.88 5.92Mar 3.15 4.33 6.10

    Apr 1.74 2.76 3.68May 1.64 4.23 I 6.66Jun 1.39 2.93 I 6.61Jul 0.33 2.16 3.84

    Aug -0.42 1.51 3,62Sep -0.04 I 1.67 4.05Oct 1.60 3.24 4.58Nov 2.17 3.62 4.99Dec 1,80 3.15 4.51

    Negative sign = decrease

    68

  • 8/2/2019 Development of for Specific Sites Corresponding to Selected GCM Outputs,Using Statistical Downscaling Techniques

    77/88

    Difference in Ternperature from Base Period {1967-2000} for Northern Area

    Figure 9:Graph for the differenceof Averaged MonthlyTemperature ( C) from base period(I967-2000).

    Change of Precipitation (mm) in Northern Area of Pakistan up to 2099

    69

  • 8/2/2019 Development of for Specific Sites Corresponding to Selected GCM Outputs,Using Statistical Downscaling Techniques

    78/88

    Table 7: Change in Precipitation (nun) from base period (1961-2000) in Northern Area of Pakistan,

    North Change in 20102039 Change in 2040-2069 Change in 20702099Jan -2.80 -3.14 -1.03Feb -17.54 -9.12

    -21.47Mar -33.77 -31.0.5 -37.72I Apr -6.26 -5.12 -10.40

    May -12.28 -13.35 -20.05Jun 18.85 16.04 18.19Jul -27.08 -18.02 -3.10

    Aug -5.33 10.22 37.32Sep 16.07 18.04 35.78Oct 12,54 9.24 12.94

    Nov 24.21 24.00 23.38Dec 5.87 , 4.75 7.15

    Negative sign= decrease

    Difference in Precipitation from Base Period (1961 -2000) InNorthern Area

    Figure 10( a): Graph for the difference of Month 1 y Total. Precipitation (nun) from base period (1961 2000),

    70

  • 8/2/2019 Development of for Specific Sites Corresponding to Selected GCM Outputs,Using Statistical Downscaling Techniques

    79/88

    Figure 10 (b): Graph for the percentage change from the base period (1961-2000)

    71

  • 8/2/2019 Development of for Specific Sites Corresponding to Selected GCM Outputs,Using Statistical Downscaling Techniques

    80/88

    Table 8: Change in Temperature ( C) from B ase Period (1967-2000) in Central Area of Pakistan.

    Central Change in2010-2039 Change in 20402069 Change in 2070-2099Jan 3.20 4.30 I 4.99Feb 3.13 4.27 5.33 I Mar 2.89 3.56 5.16Apr 0.68 1.27 2.84May -0.78 0.30 1.61Jun -1' .51 -0.46 0.44Jul -1.07 -0.22 0.46

    Aug -2.34 -1.47 -0.74SeD -2.09 -1,18 0,02Oct 0,16 0.96 1,28

    Nov 1,16 1.97 2.72D'ec 2.33 3.39 4,21

    Negative sign = decrease

    Figure 11: Graph for the difference of Averaged Monthly Temperature ( C) from base period (1967-2000)

    72

  • 8/2/2019 Development of for Specific Sites Corresponding to Selected GCM Outputs,Using Statistical Downscaling Techniques

    81/88

    Change of Precipitation (mm ) in Central Area of Pakistan up to 2099

    Table 9: Change in Precipitation (mm) from base period (1961-2000) in Central Area of Pakistan.

    Central Change in 20 1 O~2039 Chan (]e in 2040-2069 ChanC!e in20702099Jan 0.37 -0.86 -0.02

    Feb -7.55 -2.28 -14.28Mar -20.44 -14.13 23.31Apr -2.26 -0.63 -5.76May 11.10 10.00 5.81Jun 16.27 14.58 16.31Jul -32.93 -23.60 -8.05

    Aug 4.61 9.35 35.39Sept 14.76 16.60 32.27Oct 11.53 9.88 12.34

    Nov 21.37 20.71 21.26Dec 7.75 6.57 8.15

    Negative sign = decrease

    73

  • 8/2/2019 Development of for Specific Sites Corresponding to Selected GCM Outputs,Using Statistical Downscaling Techniques

    82/88

    Figure 12(a): Graph for the difference of Monthly Total Precipitation (mm) from base period (1961- 2000)

    Figure l2(b}: Graph for the percentage change from the base period (l961-2000)

    74

  • 8/2/2019 Development of for Specific Sites Corresponding to Selected GCM Outputs,Using Statistical Downscaling Techniques

    83/88

    Change of Temperature ( C) in Southern Area of Pakistan up to 2099

    Table 10:

    Pakistan

    Change in Temperature ( C) from Base Period (1967-2000) in Southern Area of

    South Change in 20102039 Change in 2040-2069 Change in 20702099Jan 2.47 3.48 4.18Feb 2.89 3.93 4.96Mar 2,64 3.31 4.82Apr 1,02 1.57 3,24May -0.42 0.42 1.75Jun -0.97 -0.24 0.22Jul -0.43 0.44 1,11

    Aug -0044 0.48 1,38Sep -0.67 0,27 1.52Oct

    0.17 0,92 1.80Nov 0.47 0.98 2.66Dee 1.17 2.15 3.43

    egat ve s gn = ecrease

    .~

    75

  • 8/2/2019 Development of for Specific Sites Corresponding to Selected GCM Outputs,Using Statistical Downscaling Techniques

    84/88

    Figure 13: Graph for the difference of Averaged Monthly temperature ( C) from base period (1974- 2000)

    Change of Precipitation (mm) in Southern Area of Pakistan up to 2099

    76

  • 8/2/2019 Development of for Specific Sites Corresponding to Selected GCM Outputs,Using Statistical Downscaling Techniques

    85/88

    Table 11: Change in Precipitation(mm) from base period. (1961-2000) inSouthernArea of Pakistan

    South Change in 2010 -2039 Change in 2040-2069 Change i n 2010~2099Jan 4.88 2,31 1.32Feb -3.57 -2.93 -6,52Mar -7.05 -7.21 ~8,62Apr 1.00 2.74 -1.39May 9.93 8.93 7.36Jun 11.36 11.36 12.43Jul -17,08 -12.78 -5.49Aug 1.71 5.71 17.06Sep 6,59 7.47 12.73Oct 5.12 5.85 6.45Nov 13.32 11,96 14,31

    Dec 4.96 3,71 4,04egat ve s gn = ecrease

    77

  • 8/2/2019 Development of for Specific Sites Corresponding to Selected GCM Outputs,Using Statistical Downscaling Techniques

    86/88

    Figure 14(b): Graph for the percentage change from the base period (1961-2000)

    Results and Discussion

    In Northern Area of Pakistan the Mean Averaged Monthly Temperature shows increasing trend except inthe month of August and September (2010-2039) as shown in Table 6. The Monthly Total Precipitation inthe first five months and July shows the decreasing trend up to 2099. March shows the highest decrease and

    November shows the highest increase in monthly total precipitation in Northern Area of Pakistan (Figure l O b].

    In Central Area of Pakistan the Mean Monthly Temperature shows increasing trend except for the period of monsoon i.e. July - September (Table 8). The Monthly Total Precipitation in the first four months and Julyshows the decreasing trend up to 2099 (Table 9). July shows the highest decrease and November shows thehighest increase in monthly total precipitation in Central Area of Pakistan (Figure 12 b).

    In Southern Area of Pakistan the Mean Monthly Temperature shows increasing trend except for the Periodof Monsoon (201.0-2039) as shown ill Table 10. The Monthly Total Precipitation in the months of February,March and. July show the decreasing trend up to 2099 (Table 11). March shows the highest decrease and.ovember shows the highest increase in monthly total precipitation in Southern Mea of Paki stan (figure 14

    b).

    The results show that maximum rise in air temperature will Occur in Northern Areas of Pakistan ascompared to the Centre and Southern part

    78

  • 8/2/2019 Development of for Specific Sites Corresponding to Selected GCM Outputs,Using Statistical Downscaling Techniques

    87/88

    79

    REFERENCES

    Bardossy, A.. Plate, E.J.~ 1992: Space-time model for dailyrainfall using atmospheric circulation patterns. Water Resources Research 28, 1247-l259.

    Barnett, T., and R. Preisendorfer, 1987: Origins and levels of monthly and seasonal forecase skill for United-States surfaceair temperatures dtermined by canonical correlation analysis, Mon. Wca, Rev. 115.1825-1850.

    Burkhardt. 0., 1999: Alpine precipitation in a tripled C02~d.imate. Tellus, A 51~ 289 w303.

    Charles, S.P., Bates, B.C,) Whetton, P.H.j Hughes, J.P.) 1999: Validation of downscaling models for changed climate condidons: case studyof sou thwestern Australia.Climate Research 12, 1-14.

    Charles, S.P.~ Bates B.C. and Hughes J.P., 2004: Statistical downscaling of daily precipitation from o bserved and modelled atmospheric fields. Hydro logical Processes 18, 13 73 -13 94.

    Conway> D.. Wilby,R.L., Jones P .D., 1996:Precipitation and air flow indices over the British Isles.Climate Research 7, 169-183"

    Crane, R.G., Hewitson,B.c., 1998: Doubled CO2 precipitation changes for the SusqueaennaBasin;downscaling from tho GENESIS general circulation model. International Journal of Climatology ]8,65-76.

    Giorgi, F., Mearns, L.O., 1999: Introduction to special section: regional climate modeling revisited.Journal of GeophysicalResearch 104, 6335-6352.

    Hay, LB., McCabe~ G.J., Wolock, D.M., Ayers, M.A., 1991: Simulation of precipitation by weather type analysis, Water Resources Research 27, 493-501.

    Hewi tson, B.C. and Crane R.G, 1996:Climate downscaling: techniq ues and app 11 cation. ClimateResearch, 7 j 85-95.

    Heyen. R., Zorita, E.. Cubasch, U., 1996: Statistical downscaling of monthly mean North Atlantic air- pressure tosea level anomaliesin the Baltic Sea}Tellus, 48A: 312-323.

    Ruth, R.. Kyscly, J., 2000: Constructing Site-Specific Climate Change Scenarios on a Monthly Scale

    Using Statistical Downscaling, Theoretical Applied Climatology, 66: 13-27.Johnson, RA., Wichern].D.W.; 1998: Applied Multivariate Statistical Analysis, New Jersey,Prentice-Hall, QSA, 799 pp,

  • 8/2/2019 Development of for Specific Sites Corresponding to Selected GCM Outputs,Using Statistical Downscaling Techniques

    88/88

    Global Change Impact Studies Centre (GCISC)

    Global change science is being aggressively pursued around the world. The Global Change ImpactStudies Centre was created in May 2002 to initiate this multidisciplinary effort in Pakistan. The mainobjective of the Centre is to comprehend the phenomenon of global change, scientifically determine itslikely impacts on various socio-economic sectors in Pakistan and develop strategies to counter theadverse effects,if any. Another function of the Centre is to establish itself as a national focal point for

    providing cohesion to global change related activities at the national level and for linking it withinternational global research. An important function of the Centre is to help develop manpower that iscapable of studying and participating in the international effort to study the global change phenomenon.The Centre also works to increase the awareness of the public, the scientific community and the policy

    planners in the country to globalchange. .