evaluation of downscaling and bias correction techniques ... · ministry of education, university...

Research PapersIssue RP0263November 2015

Regional Models andgeo-HydrogeologicalImpacts Division

This report representsthe Product P93c of WorkPackage 2.17 Action A of

the GEMINA Project,funded by the Italian

Ministry of Education,University and Research

and the Italian Ministry ofEnvironment, Land and

Sea. The authors wouldalso like to thank

NEXTDATA project forsupporting this research.

Evaluation of downscaling and biascorrection techniques to link climateand geo-hydrological impacts models

By Paola MercoglianoMeteo System &

Instrumentation Laboratory,CIRA - Capua Regional Models

and geo-HydrogeologicalImpacts Division, CMCC -

[email protected]

Guido RiannaRegional Models and

geo-Hydrogeological ImpactsDivision, CMCC - Capua

[email protected]

Renata VezzoliRegional Models and


[email protected]

Veronica VillaniRegional Models and

geo-Hydrogeological ImpactsDivision, CMCC - [email protected]

and Valentina CoppolaRegional Models and


[email protected]

SUMMARY This document is the product P93c of GEMINA project. It isaimed to collect the main results achieved from the application of biascorrection and weather generators techniques to climate simulations toestimate weather-related geo-hydrological impacts. This activity is carriedout within WP A.2.17 "Analysis of geo-hydrological risk related to climatechange" of GEMINA project. The main goal of WP A.2.17 is the analysis ofclimate change effects on occurrence and magnitude of landslides, floodsand low flows hazards on some specific contexts of the Mediterranean area.To reach this objective, climate data at the same horizontal resolution (<10km) of impacts model are required (WP A.2.6). Thus, in this document, theeffects of downscaling and bias correction techniques onsimulated/projected geo-hydrological hazards are investigated at differentscales on the test case areas identified in WP A.2.17: Po river basin,Orvieto sites and the additional test case of Calore Irpino basin in southernItaly.

Keywords: Bias correction, statistical downscaling, geo-hydrological impacts

02

Cen

tro

Eur

o-M

edite

rran

eosu

iCam

biam

enti

Clim

atic

i

CMCC Research Papers

INTRODUCTION AND MOTIVATION

In the last years, the evaluation of the potentialeffect of climate changes on geo-hydrologicalhazards became one of the main topic in cli-mate change research [44, 7, 38, 39, 6, 8].To this aim, the different proposed modelingchains are generally formed by three macro-elements: climate models providing weather in-puts, impact models differing for hazard, spatialscale, geomorpholological contexts and "linkingapproaches" mode. Basically, such ones fulfiltwo crucial roles: attempting to bridge the spa-tial gap currently existing between usual hor-izontal resolutions adopted in climate models(especially in the global ones); trying to correct(also partly) biases in weather forcing of inter-est currently affecting climate models in orderto allow quantitative estimations of variationsin hazards. In particular, regarding the sec-ond issue, the presence of biases in RCM out-puts is mostly related to their horizontal reso-lution, i.e. for Europe, the state of art is about25 km in ENSEMBLE project (www.ensembles-eu.org) or about 11 km in CORDEX project(http://wcrp-cordex.ipsl.jussieu.fr) that reflectsin insufficiently resolved surface properties andparametrizations of sub-grid scale processes(i.e. deep convection, soil surface balances)strictly linked to occurrence of extreme weatherevents and geo-hydrological hazards. Asshowed in [29, 30, 48], the validation phaseon control period reveals how RCM resolutionsand resulting necessary physical parametriza-tions usually induce errors in proper assess-ment, for example, of cumulative values of pre-cipitation or wet days preventing the direct useof weather variables provided by RCMs as in-put for impact tools. For overcoming such is-sue, usually RCM outputs are subjected to sta-tistical approaches, known as bias correctionmethods able to correct, at least, the errorsassociated to mean value (i.e. delta change

approach) if not, potentially, those associatedto all main statistical moments (i.e. quantilemapping approaches), while weather genera-tors are used to produce synthetic time seriesstarting from observed data. The adoption, incascade to the RCM of a statistical methodol-ogy like bias correction or Weather Generatorsallows to cope mismatching problems and toprovide, at the same time, a substantial correc-tion of weather forcing distribution making themsuitable as input for hydrological and impactmodels [42, 24, 25, 40]. It is worth to note thatthe adoption of such approaches introduces afurther element of uncertainty to be taken intoaccount in impacts studies [9, 19]. However, adeeper understanding about the performancesand constraints of these techniques (either biascorrection and weather generator) is crucial toassess their effects on impacts simulations andprojections; that is the main aim of this work.Beyond an assessment about relative perfor-mances in reproducing weather variables onthe areas, the goal concerns an increasingawareness about how these approaches couldaffect the derived components of soil surfacebudgets strictly governing the occurrence ofgeo-hydrological hazards. So, the main aimsof this research paper are to display the ca-pability of bias correction and weather genera-tor techniques to reproduce observed weather(mostly precipitation and 2m temperature) vari-ables and their statistics and to assess the im-pact of these techniques in the capability of themodeling chain described in [46] to reproduceobserved geo-hydrological hazards.

In the following we describe (a) the testcases areas for landslides and floods/droughtsphenomena; (b) the bias correction and theweather generator techniques used togetherwith the comparison of their impacts on theassessment of the hydrological cycle. Specif-ically, Orvieto test case will be used to com-

Downscaling and bias correction for geo-hydrological impacts

03

Cen

tro

Eur

o-M

edite

rran

eosu

iCam

biam

enti

Clim

atic

i

pare bias correction and weather generator ap-proaches on precipitation and temperature andthe related effects on hydrologically relevantvariables, Calore Irpino river basin test casewill be used to compare six different bias cor-rection approaches and their effects on the hy-drological cycle at basin scale, Po river basintest case provides a comparison between theperformances of raw and bias corrected climateas input to reproduce observed discharges andthe volume of high (a proxy for floods) and low(a proxy for droughts) flows.

TEST CASES

ORVIETO

The test case Orvieto (Central Italy), Fig.1, isof interest to investigate the impacts of climatechange on different types of landslides phe-nomena. Orvieto is an historical town located100 km North to Rome. It rises on top of a50 m thick tuff slab delimited by subvertical lat-eral cliffs overlying overconsolidated clays. Inthe deeper part, these are stiff and intact, butthe shallowest part of the deposit is jointed andfissured. The clayey slopes are blanketed byan irregular cover of talus and slide debris [27].Since prehistoric times, failures and slow move-ments have been affected the Orvieto slopes:the two historically failures events (Porta Cas-sia, on northern slope, 1900 and Cannicella, onsouthern slope, 1979) were induced by man-made changes to slope geometry or hydraulicconditions; ongoing slow movements (transla-tional) are directly related to soil-atmosphereinteraction. Deep movements occur along pre-existing slip surfaces located within the soft-ened part of clay formation (displacement ratesfrom 2 to 6 mm/years ) while, shallow move-ments, superimposed to the deep ones, in-volve the debris cover and show higher dis-placement rate (displacements between 7 and12 mm/month) [33].

Figure 1:Geographical location and schematic representation of

Orvieto site

CALORE IRPINO BASIN

The Calore Irpino River basin covers an areaof about 3058 km2 in Campania, Fig.2. TheRiver length is about 108 km long and the av-erage discharge is 31.8 m3/s at its outlet inthe Volturno River. The Calore Irpino Riverbasin is characterised by a micro-climate that,together with the water availability guarantee bythe River itself, foster the cultivation of vegeta-bles, vineyard and olive trees. The occurrenceof climate changes may alter the equilibrium ofthis ecosystem with impacts on the local econ-omy. The test case is a limited portion of thebasin, i.e. the basin closed at the Montella sec-tion due to the availability of meteorological andhydrological data.

PO RIVER BASIN

Po river is the longest river in Italy with a lengthof 652 km from its source in Cottian Alps (atPian del Re) to its mouth in the Adriatic Sea,in the north of Ravenna and it is the largestItalian river with an average discharge of 1540m3/s. The area covered by Po river basin isabout 71000 km2 in Italy and about about 3000km2 in Switzerland and France. The orogra-phy of the basin is quite complex since it isbounded by Alps, Apennines with the Po val-

04

Cen

tro

Eur

o-M

edite

rran

eosu

iCam

biam

enti

Clim

atic

i


Figure 2:Calore Irpino basin and sub-basin closed at Montella

Figure 3:Po basin: hydrological network and main closure sections

ley between them, Fig.3. In the context of theItalian Law 183/1989, the Po river basin is clas-sified as being of national relevance. Duringthe last centuries, flooding events, due to ex-treme meteorological conditions, of the Po riveror of its tributaries have caused numerous natu-ral catastrophes, two of them, characterized byextraordinary large scale, occurred in the last10 years, [45] and signal of changes in precip-itation and temperature are present in climateobservations [34, 5, 31].

BIAS CORRECTION AND WEATHERGENERATOR TECHNIQUES

A proper reproduction of observed hydrolog-ical conditions (”minimum requirement” [43])through a correct estimate of the componentsof water/energy budgets and of weather forc-ing is needed to investigate the potential ef-fects of climate changes on the hydrologicalcycle and, especially, on weather-induced geo-hydrological hazards. Within GEMINA projectthe focus has been mostly on precipitationand temperature values [35, 36, 48, 28, 38,39], because of their key role in the hydro-logical cycle [14] that regulates the trigger-ing/reacceleration of landslide movements andfloods/droughts occurrence. Furthermore, ad-equate (for length, resolution and quality) ob-served datasets required for implementation ofstatistical methods are often not available forvariables other than precipitation and temper-ature. In the next a brief description of biascorrection techniques and weather generatortested is given. Note that other bias correctiontechniques like linear scaling and analogs havebeen previously tested [48, 37, 35, 36].

BIAS CORRECTION TECHNIQUES

Bias correction techniques provide a re-scalingof climate model output in order to reduce theeffects of systematic errors [29]. In the lastyears, several approaches [29, 17] have beendeveloped and tested in different geographicaland geomorphological contexts providing keyinformation about actual performances and ca-pabilities. Numerous researches [48, 29, 30,17, 2] identify quantile mapping (or distribu-tion mapping) approaches as the most efficienttools in removing biases, the approach is basedon the following relationship

FO(x) = FS(x) (1)


05

Cen

tro

Eur

o-M

edite

rran

eosu

iCam

biam

enti

Clim

atic

i

where FO(x) and FS(x) are the cumulativedistribution functions of, respectively, the ob-served and simulated datasets for the variableX. The bias corrected value x∗ obtained usingthe equation:

x∗ = F−1O (FS(x)) = h(x). (2)

The main differences between the distributionmapping methods are due to h(x) transforma-tion and they can be classified as [13]:

distribution derived transformations forwhich h(x) adopts Bernoulli distribu-tion to model occurrence and differ-ent optional approaches for intensi-ties, in this study the Bernoulli-Weibull,Bernoulli-Gamma, Bernoulli-Lognormaland Bernoulli-Exponential mixture havebeen considered;

parametric quantile-quantile approachesaccording which h(x) is an algebraic rela-tionship between simulated and observedquantiles;

non parametric transformations (alsoknown as empirical quantiles) (EQ) whereh(x) is the empirical CDF and valuesfalling between -reference percentiles areobtained by interpolation

DISTRIBUTION DERIVEDTRANSFORMATIONS

This approach is valid for adjusting modelledprecipitation and it assumes that F is a mix-ture of the Bernoulli and the Gamma (or Weibullor Lognormal or Exponential) distribution. TheBernoulli distribution models the probability ofprecipitation occurrence (π) and the other dis-tribution G(x > 0) the intensity. The cumulativedistribution function F·, of both observed andsimulation datasets is defined as

F·(x) =

(1− π) + π ×G·(x) if x > 0

1− π if x = 0(3)

PARAMETRIC QUANTILE-QUANTILEAPPROACHES

In parametric quantile-quantile approaches,thedistribution of the modelled data is adjusted tomatch the distribution of the observations usinga parametric transformations to the quantile-quantile relation of observed and modelled val-ues. Different h(x) functions are possible tolink the observed values xO to the simulatedxS ones

xO =

bxS scale

a+ bxS linear

bxSc power

b(xS − x0)c power.x0(a+ bxS)(1− exp−xS/τ ) expasympt

(a+ bxS)(1− exp−(xS−x0)/τ ) expasympt.x0(4)

where a, b, c, x0 and τ are free parameters tobe calibrated. For precipitation, all paramet-ric transformations are fitted to the continuouspart of the distribution function (x>0) and mod-elled values corresponding to the dry part of theobserved empirical distribution function are setequal to zero.

NONPARAMETRICTRANSFORMATIONS

A common approach is to solve Eq.(2) usingthe empirical cumulative distribution function ofobserved and modelled values instead of as-suming a parametric distributions. Different ap-proaches are possible, among them:

SSPLIN: a smoothing spline is used to fit

06

Cen

tro

Eur

o-M

edite

rran

eosu

iCam

biam

enti

Clim

atic

i


the quantile-quantile plot of observed andmodelled timeseries;

QUANT: regularly spaced quantiles areused to characterise the empirical cu-mulative distribution function of observedand modelled timeseries, linear interpola-tion is used for the remaining quantiles;

RQUANT: as in QUANT but using locallinear least square regression.

All these methods are implemented in Clime [4]and are based on the freely available in qmapR-package [12]. Figure 4 reports the compar-ison, among observed, GCM/RCM simulatedand bias corrected precipitation (1972-2001)and temperature (1972-1994) timeseries at thestation of Montella (Calore Irpino test case).

The quantitative comparison between the av-erage monthly precipitation observed, sim-ulated and bias corrected show that para-metric quantile-quantile and non parametrictechniques outperform the distribution derivedtransformations methods based on Bernoullidistribution, while for temperature non paramet-ric techniques reproduces better the observedaverage values across the year. However,the non parametric two-sample Kolmogorov-Smirnov goodness-of-fit hypothesis test (seepvalues in Fig.4’s legend) indicates that, with aconfidence level of 5%, QUANT and RQUANTbias corrected precipitation and QUANT biascorrected temperature are drawn from the"same" underlying continuous population, i.e.have the same distribution, of observations.

WEATHER GENERATOR

Recently, stochastic weather generators areused as tools for statistical downscaling fromGCMs [16, 10]. Weather generators are clas-sified as “Richardson” type when the occur-rence of wet and dry days is modelled accord-ing Markov chain procedure and “Racsko” type

when wet/dry series are estimated as “randomvariables” on the basis of the proportion of ob-served events [26].

In this study, for precipitation only, we appliedthe freely available LARS-WG (Long Ashton re-search Station- Weather Generator) based on“Rackso” type approach. On monthly scale,daily values are selected as random variableschosen by fixed intervals having as selectionprobability the relative proportion of events; af-ter, in each class, an uniform distribution isadopted. Other climate variables like min-imum and maximum temperatures are esti-mated in a subsequent further stochastic pro-cess conditioned on wet/dry status. Figure 5provides, for the 1981-2010 period, the com-parison among observed, GCM/RCM, bias cor-rected and weather generator downscaled pre-cipitation for Orvieto test case.

The quantitative comparison between the cu-mulative monthly values and wet days returnthat the performances of the weather genera-tor are comparable to those of non paramet-ric methods and parametric quantile-quantilemethods. [41] reports a comparison of theperformances of the different approaches us-ing the no parametric two samples Cramer-vonMises (CvM) test. The test results show that(a) CDFs provided by RCM are significantlydifferent by observed ones for almost the en-tire year except in dry months when probablya large occurrence of zero values could partlycover the differences between CDFs; (b) distri-bution derived approaches provide a moderateimprovement in dry months but it seems to fail toadequately correct rainfall patterns during wetmonths probably due to the limited flexibility ofthe approach; (c) non parametric methods out-perform the other approaches both in terms ofcumulative values that wet days.


07

Cen

tro

Eur

o-M

edite

rran

eosu

iCam

biam

enti

Clim

atic

i

(a)

(b)

Figure 4:Comparison among observed, GCM/RCM simulated and

bias corrected (a) precipitation and (b) temperaturetimeseries at Montella station

Figure 5:(a) Monthly cumulative values. (b) Mean monthly numberof wet days. Black line: observed values; magenta line:simulated values through climate models GCM/RCM;

other lines: simulated values through BC or WGapproaches for Orvieto case study. Source [41]

EFFECTS OF BIAS CORRECTION ONHYDROLOGICAL CYCLE AND ONGEO-HYDROLOGICAL HAZARDS

This section is devoted to understand the ef-fects of the above described bias correction ap-proaches in the simulation of the main compo-nents of the soil water balance since soil watercontent is one of the triggering factor for geo-hydrological hazards. The main components ofsoil water balance are precipitation as ingoingflow and evapotranspiration, runoff, infiltrationas main outgoing flows.

ORVIETO TEST CASE

For Orvieto test case, the analysis are per-formed considering raw and Rquant bias cor-rected precipitation and temperature time-series, in particular, Fig.6(a) reports the com-parison with observations for precipitation,maximum and minimum temperature, diurnaltemperature range before and after the biascorrection and Fig.6(b) the effects on runoff inclay and sand soils, potential evaporation andactual evaporation in clay and sand soils es-timated respectively from observed, raw andbias corrected values. The analysis have beencarried out using HELP (Hydrologic Evaluationof Landfill Performance) [1]. For the estimationof the main components of the soil water bal-ance using observed, raw and bias correctedprecipitation and temperature data please referto [41, 40].

Figure 6(a) allows to point out that on controlperiod, climate simulations return cold biasessubstantially different for Tmax and Tmin (forthe first one, 2-4◦C while for the second onenot exceeding 1.5◦C) and therefore probablydepending on different capability to reproducethe atmospheric dynamics during day or night;on the other hand, also for temperature, re-gardless to its value, RQuant bias correction

08

Cen

tro

Eur

o-M

edite

rran

eosu

iCam

biam

enti

Clim

atic

i


reduces the error.

In Figure 6(b) the performances of modelsdriven by raw or bias corrected RCM are fullyconsistent with that displayed for precipitation interms of runoff seasonal cycles (first two rows)being overestimated in the first part of the yearand underestimated in the second one while aperfect overlapping is, generally, achieved us-ing bias corrected climate; a slight worsening ofthe performances is observed in bias correcteddriven model during the second part of dry sea-son probably due to key role of precipitationtemporal distribution that was not considered inthis analysis. For potential evapotranspiration,the underestimation of maximum and minimumtemperature results in an underestimation rang-ing between 60% and 85% with respect to val-ues retrieved starting from observed data. Theestimated actual evaporation (last two rows) isgenerally underestimated and during the dryseason a substantial undervaluation of infiltra-tion induces biases greater than 60%. At thesame time, also models forced by BC valuesdisplay worse performances mainly during thedry season revealing the ”summing” effect ofthe coupled errors (albeit small).

CALORE IRPINO RIVER BASIN TESTCASE

The hydrological cycle of Calore Irpino riverbasin closed at Montella river section is sim-ulated through the physically based hydrolog-ical model TOPKAPI [32] that, for each cou-ple of precipitation and temperature given asinput, returns, among others, the estimated:discharge (m3), net precipitation (mm), snow(mm), potential and actual evapotranspiration(mm), percolation (mm), surface runoff (mm),and soil saturation (%). For this test case, eightcouples, among those in Fig.4, of precipitationand temperature timeseries are considered:observed, raw GCM/RCM, linear, power.x0

and expasympt for parametric quantile-quantilemethods and QUANT, RQUANT and SSPLINfor non parametric methods.

Figure 7 reports the comparison for the pe-riod 1972-1993 of the different simulations forthe following variables: discharge, net precip-itation, actual evapotranspiration, percolation,surface runoff and soil saturation. Note thatonly for discharge, the comparison with the ob-served value is possible, for the other variablesthe reference value will be the one simulated byTOPKAPI driven by observed climate. Resultsshow a good agreement in the average perfor-mances of the different methods with exceptionof QUANT and power.x0.


09

Cen

tro

Eur

o-M

edite

rran

eosu

iCam

biam

enti

Clim

atic

i

(a) (b)

Figure 6:(a) Continuous line ratio (first), difference (the other ones) between bias corrected (blue) or raw RCM (red) and observed value;absolute values are displayed as bars, from top to bottom: precipitation, maximum temperature, minimum temperature, diurnal

temporal range. (b) Continuous line ratio between bias corrected (blue) or raw RCM (red) and observed value; absolute values aredisplayed as bars, from top to bottom: runoff clay soil, runoff sandy soil, potential evaporation, actual evaporation clay soil, actual

evaporation sandy soil. Source [41]

10

Cen

tro

Eur

o-M

edite

rran

eosu

iCam

biam

enti

Clim

atic

i


Figure 7:Comparison over the period 1972-1993 of TOPKAPI outputs in terms of: discharge (observed values are reported too), net

precipitation, actual evapotranspiration, percolation, surface runoff and soil saturation. Climate inputs are: observations (black), rawRCM/GCM (red)), bias corrected with linear, power.x0, expasympt, QUANT, RQUANT and SSPLIN approaches. Parametric

quantile-quantile methods are in green and non parametric methods in magenta.


11

Cen

tro

Eur

o-M

edite

rran

eosu

iCam

biam

enti

Clim

atic

i

PO RIVER BASIN TEST CASE

Po river basin is more complex to model thanCalore Irpino due to its dimension and the highanthropogenic pressure on the water balance[21]: the industrial/civil water demand is esti-mated in 5 km3/year and the volume regulatedfrom Alpine lakes is about 1.3 km3. Thesefluxes are modelled through the basin waterbalance model RIBASIM [15]. In this case, thebias in simulated precipitation and temperatureis relevant for both the “natural” and “humandriven” component of the basin water balanceand their cumulated effects are synthesised bythe error in simulated discharges at the clo-sure section of Pontelagoscuro. Consideringthe results presented in [47, 50], for the Poriver basin test case the most adequate biascorrection techniques is the distribution derivedquantile mapping: in particular, precipitation isassumed to follow a Gamma distribution andtemperature a Gaussian one [39]. Figure 8 re-ports the comparison among seasonal and an-nual cycle for observed, raw and bias correctedtemperature and precipitation in the control pe-riod 1982-2011.

The CMCC-CM driven simulation is affected bya general cold bias in all seasons more pro-nounced in spring, where peaks of -5◦C arereached; it is partially due to the general ten-dency of the Atmosphere-Ocean General Cir-culation Models to underestimate the temper-ature [18]; in particular, [11] have verified thatCMCC-CM is generally affected by a cold biasup to -2◦C over the Mediterranean area. Theprecipitation in CMCC-CM driven simulation isoverestimated over Alps, instead, an under-estimation occurs over the plain area. Theperformances of the GCM/RCM couple overthe Italian territory have been investigated in[3, 22, 23, 47, 49]. With the application of biascorrection, the average seasonal temperaturefield is correctly reproduced, with a mean bias

close to 0◦C in all seasons. Concerning pre-cipitation, the most evident result is the reduc-tion of the spring overestimation over Alpine arcand of autumn underestimation over Po plain(about 0.4 mm/day). A comparison betweencorrected, not-corrected values and observa-tions in terms of annual cycles highlights thestrong improvement of results due to the appli-cation of the quantile mapping technique: theCOSMO-CLM/CMCC-CM cold bias is almosttotally removed in all the months, as well asthe winter and summer precipitation bias, witha strong reduction of the error characterizingspring months. It is worth noting that the appli-cation of quantile mapping allows an improve-ment in the representation of the seasonal spa-tial pattern, especially for precipitation.

The bias removal has impact on the simu-lated Po river discharges as shown in Fig.9where the discharges obtained from TOP-KAPI/RIBASIM simulations driven by CMCC-CM/COSMO-CLM and by bias corrected cli-mate are compared with observations in termsof annual cycle and average high and low flowsvalume that are of interest for river water man-agement.

The discharge (QCMCC−CM ) simulated usingraw precipitation and temperature timeseriesfollow the precipitation temporal distributionpattern with a delay in the spring peak while theautumnal peak is correctly located but underes-timated; discharges results to be overestimatedwhen precipitation exceeds the observed ones,e.g. in June and July, and underestimate if pre-cipitation is less than the observed one, e.g.in November and December. The comparisonbetween QCMCC−CM and observed extremeflows shows an overestimation of the highestflow in summer and of lowest in winter, mostlyrelated to the temporal distribution of precip-itation. The discharges simulated using thebias corrected climate QCMCC−CM/QM show

12

Cen

tro

Eur

o-M

edite

rran

eosu

iCam

biam

enti

Clim

atic

i


Figure 8:First row. Climatology of Po river basin: seasonal (a) 2 meter mean temperature in ◦C and (b) precipitation in mm/day; for

observations, COSMO-CLM driven by CMCC-CM, and bias corrected simulations. Second Row. Monthly areal averaged (a) 2meter mean temperature in ◦C and (b) precipitation in mm/day for observations (black), COSMO-CLM driven by CMCC-CM (red),

and bias corrected CMCC-CM/COSMO-CLM/QM (orange) simulated time series


13

Cen

tro

Eur

o-M

edite

rran

eosu

iCam

biam

enti

Clim

atic

iFigure 9:

First row. Climatology of Po river basin: seasonal (a) 2meter mean temperature in ◦C and (b) precipitation in

mm/day; for observations, COSMO-CLM driven byCMCC-CM, and bias corrected simulations. Second Row.

Monthly areal averaged (a) 2 meter mean temperaturein ◦C and (b) precipitation in mm/day for observations(black), COSMO-CLM driven by CMCC-CM (red), andbias corrected CMCC-CM/COSMO-CLM/QM (orange)

simulated time series

a good agreement with observed ones eitherin terms of average values and high/low flowsvolume and temporal distribution.

CONCLUSIONS

This study illustrates how the application ofthe appropriate statistical downscaling tech-nique may represent a valuable tool to correctlyreproduce quantitatively and qualitatively theoccurrence and evolution of weather-relatedgeo-hydrological impacts in control period and,mainly, how these techniques could improvethe evaluation of the soil water balance (char-acterized by highly non linear processes) that,in turns, modifies the geo-hydrological hazardsoccurrence and severity. The results indicatethat the hybrid (dynamical and statistical) down-scaling approach, generally, improves the ca-pability of the climate-impact modelling chainin reproducing both qualitatively and quantita-tively the observed variables. In particular, wetest different bias correction and weather gen-erator based techniques on different, in termsof climate and geo-hydrological hazard to beinvestigated, test cases to identify the pros andcons of each technique finding that non para-metric approaches seems to outperform theothers methods but attention should be paidwhen they are used out of their calibrationrange. Between parametric and distribution de-rived methods the latter has the advantage ofcorrecting all the moments of the distributionfunction making this approach more suitable todeal with extremely high (low) precipitation andtemperature values.

As last we would like to remark that, these tech-niques, even if they introduce an additional el-ement of uncertainty to the overall modellingchain [9, 19], do not alter the climate signalthus are a useful tool to provide qualitative andquantitative estimates of the climate change im-pacts on geo-hydrological hazards as shownin [39, 20, 48, 50, 49].

14

Cen

tro

Eur

o-M

edite

rran

eosu

iCam

biam

enti

Clim

atic

i


Bibliography

[1] The Hydrologic Evaluation of Landfill Perfor-mance (HELP) model: Engineering documen-tation for version 3. U.S. Environmental Pro-tection Agency Office of Research and De-velopment, 1994.

[2] J. Boe, L. Terray, F. Habets, and E. Mar-tin. Statistical and dynamical down-scaling of the Seine basin climate forhydro-meteorological studies. InternationalJournal of Climatology, 27(12):1643–1655,2007.

[3] E. Bucchignani, P. Mercogliano, M. Mon-tesarchio, M.P. Manzi, and A.L. Zollo. Per-formance evaluation of COSMO-CLM overItaly and climate projections for the XXIcentury. In Climate change and its implica-tions on ecosystem and society: Proceedings ofI SISC (Societa Italiana di Scienze del Clima)Conference, pages 78–89, 2013.

[4] L. Cattaneo, V. Rillo, M.P. Manzi, V. Villani,and P. Mercogliano. Clime: climate dataprocessing in gis environment. CMCC -Research Paper, RP0257, 2015.

[5] N. Ciccarelli, J. von Hardenberg,A. Provenzale, C. Ronchi, A. Vargiu,and R. Pelosini. Climate variability innorth-western Italy during the second halfof the 20th century. Global and PlanetaryChange, 63:185–195, 2008.

[6] J.A. Coe and J.W. Godt. Review of ap-proaches for assessing the impact of cli-mate change on landslide hazards. InLandslides and Engineered Slopes, Protect-ing Society Through Improved Understanding,volume 1, pages 311–377, 2012.

[7] E. Coppola, M. Verdecchia, F. Giorgi,V. Colaiuda, B. Tomassetti, and A. Lom-bardi. Changing hydrological conditions inthe Po basin under global warming. Sci-ence of Total Environment, 493:1183–1196,2014.

[8] M.J. Crozier. Deciphering the effect of cli-mate change on landslide activity: A re-view. Geomorphology, 124:260–267, 2010.

[9] U. Ehret, E. Zehe, V. Wulfmeyer,K. Warrach-Sagi, and J. Liebert. Shouldwe apply bias correction to global and re-gional climate model data? Hydrol. EarthSyst. Sci., 16:3391–3404, 2012.

[10] S. Fatichi, V.Y. Ivanov, and E. Caporali.Simulation of future climate scenarios witha weather generator. Advances in Water Re-sources, 34:448–467, 2011.

[11] S. Gualdi, S. Somot, L. Li, V. Artale,M. Adani, A. Bellucci, A. Braun, S. Cal-manti, A. Carillo, A. DellAquila, M. Deque,C. Dubois, A. Elizalde, A. Harzallah, D. Ja-cob, B. L’Heveder, W. May, P. Oddo, P. Ruti,A. Sanna, G. Sannino, E. Scoccimarro,F. Sevault, and A. Navarra. The CIRCEsimulations: Regional climate change pro-jections with realistic representation of theMediterranean Sea. Bulletin of the AmericanMeteorological Society, 94:65–81, 2013.

[12] L. Gudmundsson. qmap: Statistical trans-formations for post-processing climate modeloutput, 2014. R package version 1.0-3.

[13] L. Gudmundsson, J.B. Bremnes, J.E. Hau-gen, and T. Engen-Skaugen. Technicalnote: Downscaling RCM precipitation tothe station scale using statistical transfor-mations - a comparison of methods. Hy-drology and Earth System Sciences, 16:3383–3390, 2012.

[14] S. Hagemann, C. Chen, O.J. Haerter,J. Heinke, D. Gerten, and C. Piani. Im-pact of a statistical bias correction on theprojected hydrological changes obtainedfrom three GCMs and two hydrology mod-els. Journal of Hydrometeorology, 12:556–578, 2014.


15

Cen

tro

Eur

o-M

edite

rran

eosu

iCam

biam

enti

Clim

atic

i

[15] Delft Hydraulics. Ribasim - river basinplanning and management simulation pro-gram. Technical report.

[16] C.G. Kilsby, P.D. Jones, A.A. Burton,C. Ford, H.J. Fowler, C. Harpham,P. James, A. Smith, and R. Wilby. Adaily weather generator for use in climatechange studies. Environmental Modellingand Software, 22:1705–1719, 2007.

[17] T. Lafon, S. Dadson, G. Buys, and C. Prud-homme. Bias correction of daily pre-cipitation simulated by a regional climatemodel: a comparison of methods. Inter-national Journal of Climatology, 33:1367–1381, 2013.

[18] I. Macadam, A.J. Pitman, P.H. Whetton,and G. Abramowitz. Ranking climate mod-els by performance using actual valuesand anomalies: Implications for climatechange impact assessments. GeophysicalResearch Letters, 37(16), 2010.

[19] D. Maraun. Bias correction, quantile map-ping and downscaling. revisiting the infla-tion issue. Journal of Climate, 26:2137–2143, 2013.

[20] P. Mercogliano, G. Rianna, and R. Vezzoli.Frequency variation of geo-hydrologicalhazards under climate change conditionsin italy. CMCC - Research Paper, RP0264,2015.

[21] A. Montanari. Hydrology of the po river:looking for changing patterns in river dis-charge. Hydrology and Earth System Sci-ences, 16(10):3739–3747, 2012.

[22] M. Montesarchio, M.P. Manzi, L. Cattaneo,E. Bucchignani, and P. Mercogliano. Per-formance evaluation of a regional climatesimulation with COSMO-CLM in the Alpinespace. CMCC - Research Paper, RP0173,2012.

[23] M. Montesarchio, A.L. Zollo, E. Bucchig-nani, P. Mercogliano, and S. Castellari.

Performance evaluation of high-resolutionregional climate simulations in the Alpinespace and analysis of extreme events.Journal of Geophysical Research: Atmo-spheres, 119, 2014.

[24] M.J. Muerth, B.G. St-Denis, S. Ricard,J.A. Velazquez, J. Schmid, M. Minville,D. Caya, D. Chaumont, R. Ludwing, andR. Turcotte. On the need for bias correctionin regional climate scenarios to assess cli-mate change impacts on river runoff. Hy-drology and Earth System Sciences, 17:1189–1204, 2013.

[25] I. Portoghese, E. Bruno, N. Guyen-non, and V. Iacobellis. Stochastic bias-correction of daily rainfall scenarios forhydrological applications. Natural Haz-ards Earth System Sciences, 11:2497–2509,2011.

[26] P. Racsko, L. Szeidl, and M.A. Semenov. Aserial approach to local stochastic weathermodels. Ecological Modelling, 57:27–41,1991.

[27] G. Rianna, P. Tommasi, L. Comegna, andP. Mercogliano. Preliminary assessment ofthe effects of climate change on landslideactivity of Orvieto clayey slope. In Climatechange and its implications on ecosystem andsociety: Proceedings of I SISC (Societa Ital-iana di Scienze del Clima) Conference, pages507–522, 2013.

[28] G. Rianna, A.L. Zollo, P. Tommasi,M. Paciucci, L. Comegna, and P. Mer-cogliano. Evaluation of the effects of cli-mate changes on landslide activity of Orvi-eto clayey slope. Procedia Earth and Plane-tary Science, 9:54–63, 2014.

[29] C. Teutschbein and J. Seibert. Regionalclimate models for hydrological impactstudies at the catchment scale: A reviewof recent modeling strategies. GeographyCompass, 4/7:834–860, 2010.

16

Cen

tro

Eur

o-M

edite

rran

eosu

iCam

biam

enti

Clim

atic

i


[30] C. Teutschbein and J. Seibert. Bias cor-rection of regional climate model simula-tions for hydrological climate change im-pact studies: Review and evaluation of dif-ferent methods. Journal of Hydrology, 456-457, 2012.

[31] S. Tibaldi, C. Cacciamani, and S. Pecora.Il Po nel clima che cambia. In P. Viaroli,F. Puma, and I. Ferrari, editors, Atti XVIIIcongresso S. It E., volume 24 of Biologiaambientale, pages 21–28, 2010.

[32] E. Todini and C. Mazzetti. TOPographicKinematic APproximation and IntegrationUser Manual and references, 2007.

[33] P. Tommasi, P. Pellegrini, D. Boldini, andR. Ribacchi. Influence of rainfall regime onhydraulic conditions and movement ratesin the overconsolidated clayey slope ofthe Orvieto hill (central Italy). CanadianGeotechnical Journal, 43:70–86, 2006.

[34] R. Tomozieu, V. Pavan, C. Cacciamani,and M. Amici. Observed temperaturechanges in Emilia-Romagna: mean val-ues and extremes. Climate Research,31:217–225, 2006.

[35] M. Turco, A.L. Zollo, G. Rianna, L. Cat-taneo, R. Vezzoli, and P. Mercogliano.Post-processing methods for COSMO-CLM over Italy. CMCC - Research Paper,RP0171, 2013.

[36] M. Turco, A.L. Zollo, V. Rillo, and P. Mer-cogliano. GCM driven COSMO-CLM post-processed precipitation over Italy: controland future scenarios. CMCC - Research Pa-per, RP0179, 2013.

[37] M. Turco, A.L. Zollo, R. Vezzoli, C. Ronchi,and P. Mercogliano. Daily precipitationstatistics over the Po basin: observationand post-processed RCM results. In Cli-mate change and its implications on ecosys-tem and society: Proceedings of I SISC (So-cieta Italiana di Scienze del Clima) Confer-ence, pages 222–232, 2013.

[38] R. Vezzoli, M. Del Longo, P. Mercogliano,M. Montesarchio, S. Pecora, F. Tonelli,and A.L. Zollo. Hydrological simulationsdriven by RCM climate scenarios at basinscale in the Po river, Italy. In A. Castel-larin, S. Creola, E. Toth, and A. Monta-nari, editors, Evolving Water Resources Sys-tems: Understanding, Predicting and Manag-ing Water-Society Interactions Proceedings ofICWRS2014, volume 364 of IAHS Red Book,pages 128–133, 2014.

[39] R. Vezzoli, P. Mercogliano, S. Pecora, A.L.Zollo, and C. Cacciamani. Hydrologicalsimulation of po river (north italy) dis-charge under climate change scenariosusing the {RCM} cosmo-clm. Science ofThe Total Environment, 521-522:346–358,2015.

[40] V. Villani, G. Rianna, P. Mercogliano, andA.L. Zollo. Generator to downscale rcmoutputs to slope scale for stability assess-ment: A comparison of performances.EJGE, 20:1495–1515, 2015.

[41] V. Villani, G. Rianna, P. Mercogliano,A.L. Zollo, and P. Schiano. Statisticalapproaches versus weather generator todownscale RCM outputs to point scale:A comparison of performances. Jour-nal of Urban and Environmental Engineering,8(2):142–154, 2014.

[42] M. Vrac, P. Drobinski, A. Merlo, M. Her-rmann, C. Lavaysse, L. Li, and S. Somot.Dynamical and statistical downscaling ofthe french mediterranean climate: uncer-tainty assessment. Natural Hazards EarthSystem Sciences, 12:2769–2784, 2012.

[43] A.W. Wood, L.R. Leung, V. Sridhar, andD.P. Lettenmaier. Hydrologic implicationsof dynamical and statistical approachesto downscaling climate outputs. ClimateChange, 62(1-3):189–216, 2004.


17

Cen

tro

Eur

o-M

edite

rran

eosu

iCam

biam

enti

Clim

atic

i

[44] C.Y. Xu. Climate change and hydrologicmodels: A review of existing gaps and re-cent research developments. Water Resour.Management, 13:369–382, 1999.

[45] D. Zanchettin, P. Traverso, andM. Tomasino. Po river discharge: apreliminary analysis of a 200-year timeseries. Climatic Change, 89:411–433,2008.

[46] A.L. Zollo, P. Mercogliano, M. Turco,R. Vezzoli, G. Rianna, E. Bucchignani,M.P. Manzi, and M. Montesarchio. Archi-tectures and tools to analyse the impactsof climate change on hydrogeological riskon Mediterranean area. CMCC - ResearchPaper, RP0129, 2012.

[47] A.L. Zollo, M. Montesarchio, M.P. Manzi,L. Cattaneo E. Bucchignani, and P. Mer-cogliano. Assessment of COSMO-CLMperformances in simulating the past cli-mate of Italy. CMCC - Research Paper,RP0145, 2012.

[48] A.L. Zollo, G. Rianna, P. Mercogliano,P. Tommasi, and L. Comegna. Valida-

tion of a simulation chain to assess cli-mate change impact on precipitation in-duced landslides. In K. Sassa, P. Canuti,and Y. Yin, editors, Landslide Science fora Safer Geoenvironment, volume 1 of Pro-ceedings of World Landslide Forum 3, pages287–292, Beijing, 2014.

[49] A.L. Zollo, V. Rillo, E. Bucchignani,M. Montesarchio, and P. Mercogliano.Temperature and precipitation extremeevents over italy: assessment of high res-olution simulations with COSMO-CLM andfuture scenarios. International Journal ofClimatology, 2015.

[50] A.L. Zollo, M. Turco, and P. Mercogliano.Assessment of hybrid downscaling tech-niques for precipitation over the Po Riverbasin. In G. Lollino, A. Manconi, J. Clague,W. Shan, and M. Chiarle, editors, Engineer-ing Geology for Society and Territory, vol-ume 1, pages 193–197. Springer Interna-tional Publishing, 2015. doi:10.1007/978-3-319-09300-0 37.

18

Cen

tro

Eur

o-M

edite

rran

eosu

iCam

biam

enti

Clim

atic

i


c© Centro Euro-Mediterraneo sui Cambiamenti Climatici 2015

Visit www.cmcc.it for information on our activities and publications.

The Euro-Mediteranean Centre on Climate Change is a Ltd Company with its registered office andadministration in Lecce and local units in Bologna, Venice, Capua, Sassari, Viterbo, Benevento and Milan.The society doesn’t pursue profitable ends and aims to realize and manage the Centre, its promotion, andresearch coordination and different scientific and applied activities in the field of climate change study.

evaluation of downscaling and bias correction techniques ... · ministry of education, university...

Documents