Table Of Content

INTERNATIONALJOURNALOFCLIMATOLOGY Int.J.Climatol.(2016) PublishedonlineinWileyOnlineLibrary (wileyonlinelibrary.com)DOI:10.1002/joc.4642 New Azores archipelago daily precipitation dataset and its links with large-scale modes of climate variability Armand Hernández,a* Haim Kutiel,b Ricardo M. Trigo,a Maria A. Valente,a Javier Sigró,c Thomas Croppera and Fátima Espírito Santod aFaculdadedeCiências,InstitutoDomLuiz(IDL),UniversidadedeLisboa,Portugal bLaboratoryofClimatology,DepartmentofGeographyandEnvironmentalStudies,UniversityofHaifa,Israel cCentreforClimateChange(C3),Tarragona,Spain dIPMAIP-ThePortugueseSeaandAtmosphereInstitute,Lisbon,Portugal ABSTRACT:Withinthescopeofthetwomajorinternationalprojectsoflong-termreanalysisforthe20thcenturycoordinated bytheNationalOceanicandAtmosphericAdministrationandEuropeanCentreforMedium-RangeWeatherForecasts,the InstitutoDomLuizfromtheUniversityofLisbonhasdigitizedalargenumberoflong-termdailyandmonthlyclimaterecords fromstationsinPortugalandformerPortuguesecolonies.Wehaverecentlyfinishedthedigitizationofprecipitationvalues from Ponta Delgada (capital of the Azores Archipelago), obtaining an almost complete daily precipitationseries, with the exceptionofsomeyears(1864–1872;1878–1879;1888–1905;1931;1936;and1938)forwhichonlymonthlyvaluesare available.Atdailyresolution,wehaveusedacomprehensiveassessmentondifferentcharacteristicsofrainspells(consecutive days with rainfall accumulation). The distribution of precipitation presents an evident seasonal pattern and reveals large inter-annualandintra-annualvariability,increasingconsiderablyinthelast3decades.Thefrequencyofdryyearsdecreases almostbyhalfbetweenthefirstandthesecondpartoftherecord,whereaswetyearsincreaseuptothreetimes.Thisismainly due to more intense events that are reflected by higher rain-spell yields (amount of precipitation) and rain-spell intensity (amountofprecipitationperday)values.Mostoftheextremeprecipitationeventsoccurredduringthelast2decades,and theygenerallycorrespondtodateswithcyclonicconditionsovertheNorthAtlantic.Wehavealsolookedintotheinfluenceof large-scalemodesofclimatevariabilityontheprecipitationregimeoftheAzoresArchipelago.Asexpected,theNorthAtlantic Oscillation(NAO)hasamajorimpactontheprecipitationregimeofPontaDelgadabothinwinterandsummer.However,our resultsshowanon-stationaryNAOinfluenceandtheimpactofotherlarge-scalemodes(includingtheAtlanticMultidecadal OscillationandElNiño-SouthernOscillation)increaseswhenthisinfluencebecomesweaker. KEYWORDS dailyprecipitation;AzoresArchipelago;teleconnection;seasonality;NorthAtlanticOcean Received27May2015;Revised3December2015;Accepted4December2015 1. Introduction long-termstationsrecordsfromPortugalandformerPor- tuguesecolonies(Stickleretal.,2014). Long-term climate datasets are essential for understand- Island environments are very sensitive to externally ing climate changes (Compo etal., 2011). However, the inducedchanges,andthelimitedareaandresourcesavail- currentlylimitedavailabilityoflong-terminstrumentalcli- able to island states can make it difficult to apply adap- materecordshampersourabilitytoconductmorereliable tation methods (IPCC, 2014, WG3). In particular, small assessments of climate variability and change (Trenberth islands,likethoseintheMacaronesianregionintheeast- etal.,2007).Presentaccesstodailydigitalclimatedatais ern North Atlantic (Figure 1(a)), have long been identi- scarceovermanypartsoftheworld,particularlyinoceanic fied among the most vulnerable to climate change and regions, and this limited coverage of instrumental data climate extremes (Nurse etal., 2014) due to the poten- obstructs the ability to better understand climate changes tialimplicationsintherelocationofresourcesandactivi- and its forcing factors (Brunet and Jones, 2011). In this ties (Santos etal., 2004). Changes in precipitation are of senseandwithinthescopeofthetwomajorinternational particular concern as precipitation, or a significant lack projectsoflong-termreanalysisforthe20thcenturycoor- of it, can be responsible for flash floods, landslides and dinatedbytheNOAA(Compoetal.,2011)andECMWF droughts. The Azores Archipelago (Figure 1(a)) is usu- (Hersbachetal.,2013),theInstitutoDomLuiz(IDL)from ally under the direct influence of the semi-permanent, the University of Lisbon has digitized a large number of high-pressureAzoresanticyclone.Duringmostoftheyear (September–March), if this high-pressure centre is dis- sipated or displaced latitudinally, then the Azores region *Correspondence to: A. Hernández, Faculdade de Ciências, Instituto DomLuiz(IDL),UniversidadedeLisboa,1749-016Lisboa,Portugal. is frequently crossed by the North Atlantic storm track, E-mail:[email protected] the main path of rain-producing weather systems (Trigo ©2016RoyalMeteorologicalSociety A.HERNÁNDEZetal. (a) (b) (c) Figure1.(a)MapoftheMacaronesianregionshowingtheAzoresArchipelagoandPontaDelgadalocation.(b)Annualand(c)monthlymean precipitation(mm),1873–2012.Annualseriesofprecipitationwerecalculatedbysummingdailyvaluesifnomorethantwovaluespermonthwere missing(solidbars).Whenthisoccurs,serieswerecompletedusingmonthlydatasetsfrompreviouslydigitisedmonthlyseriesalreadyusedbyother authors(e.g.CropperandHanna,2014)coveringthistimeperiod(openbars).Thethicklinerepresentsa31-pointmovingmean,whichemphasizes thelow-frequencyvariabilityoftheannualprecipitationseriesandtheshadowindicatesthecorresponding31-pointmovingstandarddeviation.Note thatthevaluesobtainedforbothendswerecalculatedusingasmanyyearsaspossibleinthemovingaveragewhenthenumberofyearswasless than31. etal., 2004). Conversely, during late spring and summer, (Andrade etal., 2008; Marquesetal., 2008; Cropper and theAzoresclimateismainlydominatedbytheAzoresanti- Hanna, 2014). The NAO is the leading mode of climate cyclone (Agostinho, 1948; Santos etal., 2004; Azevedo, variability in the North Atlantic region and is manifested 2006).Thus,therainfallregimefromPontaDelgada(São asameridionaldipoleanomalyinsea-levelpressure,with Miguel Island, Azores Archipelago, Portugal) displays thetwocentresofactionlocatedapproximatelyoverIce- a strong seasonal cycle and large inter-annual variabil- land and the Azores (Hurrell, 1995; Trigo etal., 2004). ity (Marques etal., 2008). The changes in precipitation However, besides the NAO, there are other large-scale areusuallylargerthantemperatureorpressurevariations, modes, such as the East Atlantic pattern (EA), the Scan- and long-term records are essential to accurately present dinavian pattern (SCAND), the Arctic Oscillation (AO), the local rainfall regime (e.g. Esteban-Parra etal., 1998; the Atlantic Multidecadal Oscillation (AMO), the Pacific Kutiel and Trigo, 2014). Variations in total precipitation Decadal Oscillation (PDO) and the El Niño-Southern canbecausedbychangesinthecharacteristicsofthecon- Oscillation(ENSO),thatmayalsocontributesignificantly stituting rain events, including their timing, intensity and to the North Atlantic region’s precipitation regime (e.g. some aspects of the dryness conditions between rainfall ThompsonandWallace,1998;Pozo-Vázquezetal.,2005; episodes(e.g.PazandKutiel,2003).Therefore,inorderto Sutton and Hodson, 2005; Trigo etal., 2008; Comas-Bru improvetheunderstandingofprecipitationbehaviourasan and McDermott, 2014). Furthermore, it is worth noting indicatorofclimatechangesinthelastcentury,dailypre- the existence of inter- and intra-annual variations in the cipitationseriesmustbeanalysed(Brunettietal.,2001). impactoftheselarge-scalemodesontheclimatethrough- Someauthors(e.g.Andradeetal.,2008;Marquesetal., out long periods. In particular, the correlation between 2008;CropperandHanna,2014)havepreviouslyanalysed the NAO and precipitation in the Macaronesian region long-term characteristics of precipitation for Ponta Del- showsanon-stationarybehaviourovertime(Cropperand gadaanditsrelationshipwiththemainlarge-scalemodes Hanna, 2014), and therefore, studies of the long-term in the North Atlantic sector, the North Atlantic Oscil- influence of large-scale modes on rainfall variability are lation (NAO). However, these studies have used shorter required. However, despite the increasing understanding and/or lower-resolution time series, and to the best of of the impact of large-scale modes on the North Atlantic our knowledge, there is no work evaluating the evolu- region climate, to our knowledge, the impact of other tionofdailyprecipitationtime-seriescharacteristicsstart- large-scale modes on the rainfall regime of the Azores ing in the second half of the 19th century. Uninterrupted Archipelagohasnotpreviouslybeenexplored. monthlyprecipitationvaluesfromPontaDelgadaarecur- The influence of large-scale modes on the variability rentlyavailablesince1865withoutmissingdata(Cropper of North Atlantic storms has been the focus of several and Hanna, 2014), and a new, almost complete, corre- previous works (e.g. Kossin etal., 2010). Intense precip- sponding daily precipitation series, with the exception of itation greatly affects human activities as it can cause someyears,hasrecentlybeendigitizedatIDL. catastrophic floods or landslides among other impacts. A large fraction of the inter-annual variability of pre- The analysis of the behaviour of extreme rainfall and its cipitation observed in the Azores Archipelago has previ- links to large-scale modes is essential to better under- ously been ascribed to the NAO influence during winter stand their impact on human endeavours. Days with ©2016RoyalMeteorologicalSociety Int.J.Climatol.(2016) AZORESARCHIPELAGOPRECIPITATIONANDMODESOFCLIMATEVARIABILITY extreme values of precipitation are potential triggers of homogenization method, we have used some long-term catastrophic episodes, andprevious studies ofthe Azores precipitation series available for the Azores Archipelago Archipelago (e.g. Andrade etal., 2008; Marques etal., asreferences[Terceira(AngradoHeroismo,1873–2012) 2008) have focussed their efforts on establishing links andSãoMiguelIsland(Furnas,1935–2012;CasadoJosé, betweencyclonesandtheNAOpattern. 1935–2012; Santana, 1947–2012)]. We applied annual Our study aims to contribute to the understand- pair-wise detection and joint detection to estimate homo- ing of the rainfall regime in the Azores Archipelago geneity breaks and to determine potential change points. by investigating daily precipitation data recorded at Then,metadatarecordsforeachstationwereusedtocon- Ponta Delgada. Changes in precipitation in the period firmbreakpoints.Twohomogeneitybreaksweredetected 1873–2012 (∼140years) were studied using a method- inthePontaDelgadaprecipitationseriesfortheyears1888 ological approach based on different characteristics of and 1936, the first one coinciding with the beginning of rainspells(e.g.ReiserandKutiel,2008).Thismethodhas a period that was filled with the newly digitized monthly beenshowntoperformwellwhenanalysingthedifferent data(1888–1905)andthesecondwithastationrelocation parameters related to the rainfall regime in other parts of (1936).Theadjustmentofinhomogeneitieswasdonewith the world with strongly seasonal precipitation behaviour, the option log ratio multiplicative correction for cumu- suchastheMediterraneanregion(e.g.ReiserandKutiel, lative parameters in HOMER. This type of adjustment 2007).Additionally,theinfluenceofthelarge-scalemodes by multiplicative ratio is normally used to homogenize on the inter-annual precipitation variability as well as on precipitation series, which is different from the additive the severe rainfall events was investigated. We also anal- factors used in temperature series (Alexandersson, 1986; ysed the trajectory of significant North Atlantic tropical Venema etal., 2012).To adjust daily data, we have cho- storms that are shown to sometimes influence extreme sen the scheme developed by Vincent etal. (2002). This precipitationacrosstheAzoresArchipelago. approach attempts to provide a better time-interpolation procedure that preserves monthly values and does not introduce artificial discontinuities at the beginning and 2. Dataandmethods end of calendar months. The impact of this homogene- ity adjustment is relatively small, and it affects only the 2.1. Datasets period previous to 1936. Between 1889 and 1936, the Daily rainfall data from 1873 until 2012 from the Ponta monthly impact of the adjustment is ∼10mm on average ∘ ∘ Delgadastation(37 44’N,25 40’W,77m)havebeendig- compared with the original data, whereas for the earlier itized from meteorological annals and then quality con- period,1872–1888,theaveragedifferenceis∼5mm. trolled according to the framework presented in Stickler This study has been performed with data from an etal.(2014).Additionally,wetestedthequalitycontrolof Azorean meteorological station, and therefore, an dailydatawithRClimDex_extraQCroutines(Aguilarand extended daily (1850 to present) station-based NAO Prohom, 2011). Data from 1873 to 1946 were obtained index, based on the reconstructed Azores sea-level from the ‘Annaes do Observatório Infante D. Luiz’ (see pressure data presented by Cropper etal. (2015), has Valenteetal.,2008andStickleretal.,2014)andfor1947 been employed. The temporal evolution of the other to 2012 were supplied by the Instituto Português do Mar large-scale modes was obtained through the numeri- e da Atmosfera (IPMA). Thus, we obtained a new daily cal indices stored at the Climate Prediction Center of precipitation series from 1873 to 2012, with the excep- the National Oceanic and Atmospheric Administra- tionofsomeyears(1878–1879;1888–1905;1931;1936; tion (http://www.cpc.ncep.noaa.gov). Monthly resolution and1938).Inaddition,annualandmonthlyseriesofpre- datasetswereemployedwithdifferentdurationsaccording cipitation were calculated by summing daily values if no to the available data (Table S1, Supporting Information). morethantwovaluespermonthweremissing.Whenthis TheAtlantictropicalstormsdatawereretrievedfromthe occurs, monthly and annual series were completed using NationalHurricaneCenterfromtheNationalOceanicand the monthly datasets from previously digitized monthly AtmosphericAdministration(http://www.nhc.noaa.gov/). series(IDL-IPMA)coveringthistimeperiod(Cropperand Hanna,2014) 2.2. Rainfallanalyses Long-term climate time series have been influenced by non-climaticfactorsquiteoften,mainlyrelatedtochanges The daily rainfall regime was explored using a specially in instrumentation, instrumental exposure, station loca- designed statistical model, entitled the Rainfall Uncer- tions and local environments, observing practices and/or tainty Evaluation Model (RUEM; Reiser and Kutiel, data processing (e.g. Aguilar etal., 2003; Trewin, 2010). 2008). Allthesefactorscanintroducegradualorabruptbreaksin To avoid splitting the Azores rainy period into two thehomogeneityofclimaterecords. different years, we use the hydrological year rather than To detect and estimate adjustments for these problems, the traditional meteorological year. Most definitions for thePontaDelgadaannualandmonthlyprecipitationseries a hydrological year start on September 1 or October 1. wereexaminedforhomogeneityusingthesoftwarepack- However, in this study, we will use the hydrological year ageHOMER(Mestreetal.,2013)todetectpossiblebreaks definition starting on 1 July and ending on 30 June of in the homogeneity of series. As HOMER is a relative the next calendar year as proposed by Reiser and Kutiel ©2016RoyalMeteorologicalSociety Int.J.Climatol.(2016) A.HERNÁNDEZetal. (2008)forregionswithdrysummersandlongwetseasons. 1(b)).Thisbreakpoint(1942–1943)waschosenasitcor- Thenumberoftheyearreferstowhenitstarted,e.g.1873 respondstoapotentialprecipitationregimechangeinthe referstotheperiodthatspansbetween1July1873and30 regionalseries(CropperandHanna,2014).Additionally,it June1874. wasalsoasuitablepointastheSubperiod1(SP1)andSub- For our analysis, we focussed on rain spells instead of period 2 (SP2) needed to be of near similar lengths. The rainy days as the former better accommodates the total firsthalf(SP1)spansfrom1873until1942butonlycom- length of any rainy period. A rain spell is defined as a prises42years(outofapossible70)duetothelackofdata; period of consecutive days with rainfall above a given thesecondhalf(SP2)spansfrom1943until2012,includ- predetermined daily rainfall threshold (e.g. Reiser and ing all 70years. To evaluate the representativeness of the Kutiel,2012).Eachrainspellisprecededandfollowedby 42years in the first period, appropriate statistical t- and atleastonedrydayoradaywitharainfallamountlessthan F-tests were applied to the mean and standard deviation thedailyrainfallthreshold(1.0mminthisstudy).Foreach values of SP1 obtained with the different length (70-year rainspell,itsduration(RSD),yield(RSY,thetotalrainfall vs 42-year) datasets. Results do not show significant dif- accumulated during the rain spell) and mean intensity ferencesbetweenthemonthly(70-year)anddaily(42-yr) (RSI, the average daily rainfall in the rain spell) were datasets with a confidence level of 99%, which suggests calculated. Additionally, the total number of rain spells itisfeasibletomakevalidcomparisonsbetweenthedaily (NRS) and the rainy season length (RSL) in each year precipitationregimeofSP1(42-year)andSP2(70-year). was also considered. The RSL is defined as the number The rainfall regime in the Azores Archipelago at of days elapsed between the beginning and the end of daily resolution comprises a time window of 140years the rainy season. The beginning and the end dates are (1873–2012) with five gaps lacking daily data (1878– when10and90%oftheannualrainfallwereaccumulated, 1879;1888–1905;1931;1936;and1938).Thus,117years respectively. These dates vary from year to year, and the wereinitiallyavailable,and,asweusedhydrologicalyears RSL varies accordingly. This was done in order to avoid insteadofmeteorologicalyears,atotalof112hydrologi- somesporadicrainswithminuteamountsatthebeginning calyears(July1–30)couldbeanalysedwiththeRUEM. and/orendoftherainyseasonthatwillcauseaverylong WedecidedtoreproducetheRUEManalysisfortheentire RSL consisting of long dry periods in between (Reiser period,thetwosubperiodsandthecategorisedverydryto and Kutiel, 2009). Following Kutiel and Trigo (2014), verywetyears. annual totals in all years/seasons were standardized, and the years/seasons were categorized accordingly into five 2.3. Atmosphericcirculationpatterns’influence categories:verydry(z<−1),dry(−1≤z<−0.5),normal onrainfallregime (−0.5≤z≤0.5),wet(0.5<z≤1)andverywet(1<z).For Theextendedwinter(DJFM)andsummer(JJAS)influence each category, the above rain-spell parameters were also ofeachconsideredlarge-scalemode(NAO,EA,SCAND, calculatedseparately. AO, AMO, PDO and ENSO) on precipitation has been Intra-annualvariabilitywasanalysedbycalculatingand explored. This was performed according to Spearman’s comparing the dates of accumulated rainfall percentage. rank correlation coefficients (𝜌) and associated p-values. For each day in each hydrological year, the accumulated Unless otherwise stated, significance (p-value) is always percentageofthetotalannualrainfallofthatyearwascal- discussedatp<0.01. culated,andthesevalueswerethensortedinanascending order for each Julian day, thus enabling the values to be presentedinaprobabilisticway. 3. Overallrainfallregime Drynesswasanalysedusingthedrydayssincelastrain (DDSLR)approach(e.g.Kutiel,1985;Aviadetal.,2009; The rainfall regime in Ponta Delgada shows clear Lana etal., 2012; Kutiel and Trigo, 2014). According to inter-annual and intra-annual variability (Figures 1(b) the DDSLR methodology, each rainy day is attributed a and (c)). The mean annual precipitation for 1873–2012 ‘ 0’ value, the first dry day is attributed a ‘1’ value, the is 960.6mm, with a standard deviation of 201mm, next consecutive dry day a ‘2’ value and so on until the and almost 75% of the total precipitation was accu- nextrainyday,whichisagainattributeda‘0;value.Values mulated between October and March. The mean rainy carry over from 1year to the next. Once all days in the seasonlengthis277days(Table1).Although theAzores analysedperiodwereattributedwithavaluethatdescribes Archipelago is geographically distant from the Mediter- the distance in days of that specific day from the last ranean basin, the rainfall regime presents some similar rainyday,allvaluesforeachJuliandayweresortedinan characteristics with the Mediterranean region, with ascending order, thus enabling the values to be presented maximum precipitation during the cold season and min- againinaprobabilisticway. imum precipitation during summers (Reiser and Kutiel, The long-term inter-annual rainfall evolution through- 2009). Precipitation occurred (on average) in ∼122 rainy out the entire period (1873–2012) was studied by apply- days consisting of 60 rain spells with a mean duration ing a 31-point moving mean window to the annual mean of ∼2days. Each rain spell yields 16mm on average precipitationcalculatedfromthedailytimeseries(Figure with a mean rate of 7.9mmday−1. About 42% of the 1(b)).Inaddition,thisperiodwasdividedintotwohalves annual total is due to rain spells longer than 3days thatcorrespondtoaregimechangeinprecipitation(Figure (Table1). ©2016RoyalMeteorologicalSociety Int.J.Climatol.(2016) AZORESARCHIPELAGOPRECIPITATIONANDMODESOFCLIMATEVARIABILITY dh D 42)anoreac RatioVW/V 2.02 1.37 1.04 2.12 0.99 1.35 1.94 1.48 1.51 9f 1s (1873–wetyear RatioW/D 1.38 1.13 1.08 1.21 0.97 1.05 1.27 1.21 1.08 eriods1ndvery RatioP2/SP1 1.12 1.03 1.03 1.03 0.97 1.00 1.10 1.11 0.97 pa S subdry –2012),ndvery wet62mm)ears) SD 113.6 13 4 2 34 0.2 2.2 0.7 9.4 1873weta Very>(11(16y X 302.5 141 62 10 276 2.3 21.1 9.3 48.4 (d 1 hn gta entireserieslend2,betweendry Wet(1061–1162mm)(13years) XSD 1114.225.1 13110 647 92 27024 2.10.2 17.72.2 8.50.7 44.713.2 en ha t nthisstudyforeensubperiod1 Normal(860–1061mm)(47years) XSD 963.061.3 12011 604 82 27728 2.00.2 16.21.6 8.10.9 41.911.9 fferentprecipitationindicesusediownistheratioofthemeansbetwprecipitationindex. VerydryDry<(760mm)(760–860mm)(13years)(23years) XSDXSD 643.580.9810.033.0 103711710 606596 5282 2803527835 1.71.32.00.2 10.91.613.91.5 6.30.97.00.6 32.010.741.49.8 dish variationforthewetyears.Also 1943–2012SP2(70years) XSD 1000.9184.4 12214 615 82 27332 2.00.2 16.63.2 8.31.1 41.212.1 fy ndcoefficientoverydrytover 1873–1942SP1(42years) XSD 893.3211.5 11916 595 83 28327 2.00.2 15.13.4 7.51.2 42.511.5 ad Table1.Mean,standarddeviation2(1943–2012),andthecategorize 1873–2012(112years) XSD Annualrainfall960.6201.0(mm)Numberofrain12115days(NRD)Numberofrain605spells(NRS)Numberofrain83spellslongerthe>3days(NRS3)Rainyseason27731length(RSL,days)Rainspell2.00.2duration(RSD,days)Rainspellyield16.03.3(RSY,mm)Rainspell7.91.2intensity(RSI,mm/day)Relative41.911.8contributionofrainspellslongerthan3days(%) ©2016RoyalMeteorologicalSociety Int.J.Climatol.(2016) A.HERNÁNDEZetal. Theprecipitationrecordshowsanincreaseintherainfall 4. Inter-annualvariability conditions between a drier earlier period (1873–1942; Table 1 presents the analysed precipitation parameters SP1) and a wetter recent period (1943–2012; SP2). The studied for the five ‘wetness/dryness’ categories (Section meanannualtotalofSP1is893.3mm,whereasSP2shows 2.2).Itcanbeobservedthatthemeantotalrainfallduring avalueof1001mm,butwiththeRSL,itreducedby9days. very wet years was 1302mm, around twice the mean The precipitation increase is mainly a consequence of an of the very dry years, which had a mean of 644mm. increase of the RSY, RSI and, to a lesser extent, in the All parameters, with the exception of the RSL, show an NRS.Conversely,theRSDisunchangedbetweenthetwo increase from the very dry years towards the very wet periods (Table 1). The mean values of all the analysed years.TheNRShasthelowestverywet/verydryratiowith parameters (with the exception of the RSL and RSD) are valuesof1.04.Thismeanstherewasaslightmeanincrease slightlyhigherduringSP2butpresentalowercoefficient of4%inthenumberofrainspells.Incontrast,whenonly of variation, which highlights a larger heterogeneity and longrainspells(RSD>3)wereconsidered,theirnumber variabilityoftherainfallregimethroughoutSP1. increasedconsiderably,fromameanof4.9intheverydry As mentioned above, the Ponta Delgada precipitation years to 10.4 in the very wet years, corresponding to an regimedisplayssimilarcharacteristicswithMediterranean increase of more than two times. These long rain spells, climate behaviour (e.g. Reiser and Kutiel, 2009) in spite whichconsistonlyof28.6%ofthetotalNRSinthevery of the ca. 3000km between them, although both regions dryyears,nearlydoubledtheirpercentageto42.4%inthe are within similar latitudinal bands. Wigley and Farmer very wet years (Figure 2). The RSD and RSI represent a (1982) defined the Mediterranean climate as having win- moderate increase between those two categories of years ter season rainfall more than three times that of summer. between1.3and1.5.TheRSYrepresentsthehighestratio KutielandTrigo(2014)laterdefinedthisastheMediter- betweenthetwo(besidestheNRS>3)andwasdoubledin ranean Climate Index (MCI). The MCI is an index that theverywetyearscomparedwiththeverydryyears,21.1 describesthelevelofseasonalityoftherainfall.Thisindex and 10.9mm, respectively. Hence, these five categories calculates the ratio between winter and summer rains. According to it, an MCI<1 means summer wetter than have very different RSY values. However, even though winter, 1≤MCI<3 winter slightly wetter than summer, their NRS values slightly increase from very dry to very 3≤MCI<10 dry summer and 10≤MCI extremely dry wet years, they overlap to a large extent (Figure 3). This summer.AlocationwithanMCI≥3isconsideredtohave implies that while the RSY can be used as a good index of the rainfall variability, the NRS cannot serve as an aso-calledMediterraneanclimate.TheMCIinPontaDel- gada is 3.02 and therefore, fits well, albeit by a minimal indicator to differentiate between these years and has an margin,withinthedefinitionoftheMediterraneanclimate insignificantimpactontheannualtotal. typepreviouslydefined.TheMCIofSP1is2.69,whereas The percentage of the total rainfall contributed by rain SP2 shows a MCI of 3.34, indicating a change towards spells of various durations is represented by the rela- more‘Mediterranean’conditions,withalargerconcentra- tive contribution. The relative contribution of long rain tionofprecipitationinwinter. spells(RSD>3days)isusuallyhigherinmorehumidsta- A priori, dryness is not a problem in the Azores tions, whereas, in drier locations, the main contribution Archipelago, but islands are very sensitive to climate to the annual total is made by very short rain spells of changes, especially to changes in the hydrological avail- 1 or 2days (Reiser and Kutiel, 2012; Kutiel and Trigo, ability.Thus,thedrynessconditions(severity,consistency 2014; Kutiel etal., 2014). This is reflected in this study andtemporaluncertainty)wereanalysedusingtheDDSLR by dividing the entire dataset into five rainfall categories approach[seeKutielandTrigo(2014)forfurtherdetails]. as explained earlier. Although Ponta Delgada is consid- In Ponta Delgada, the longest DDSLR (severity) occurs ered to be a humid location, the maximum contribution at the end of August and the beginning of September, varies according to the different categories. In very dry anditspans63days(FigureS1,SupplementaryInforma- or dry years, the maximum contribution is by rain spells tion). We can expect a DDSLR longer than 30days on of 1day as in arid regions, whereas in normal, wet and average once every 25years. The temporal uncertainty, very wet years, the distributions resemble humid regions i.e. the span of time when such a DDSLR has been (Figure 2). observed,isfrom14Aprilto21September,overaperiod If we compare the relative contribution of short of 161days. Similarly, the only observed example of a rain spells (RSD≤3days) and very long rain spells DDSLRlongerthan60daysoccurredduringthe3days30 (RSD>7days),wefindthatthefirstdecreasesfrommore August–1September,andtherefore,themostlikelytime thantwo-thirds(65.2%)inverydryyearstoslightlyover for very high DDSLR values would be late August/early half (51.3%) in very wet years. By contrast, the relative September (Figure S1). The dry period started 60days contributionofverylongrainspellsisdoubledfrom4.8% earlier (around the end of June or beginning of July) inverydryyearsto9.5%inverywetones(Figure 2). and reached its maximum (more than 60days) during These results are in agreement with those obtained these 3days. These DDSLR values are considerably recently by Reiser and Kutiel (2012) with an application lower compared with those obtained for Lisbon (Kutiel fortheentireMediterraneanarea,butconsideringamuch and Trigo, 2014), indicating wetter conditions in Ponta shorterperiod,andbyKutielandTrigo(2014)forasim- Delgada. ilar long-term period in Lisbon. This means that wet or ©2016RoyalMeteorologicalSociety Int.J.Climatol.(2016) AZORESARCHIPELAGOPRECIPITATIONANDMODESOFCLIMATEVARIABILITY Figure2.Meanrelativecontributionofrainspellsofvariousdurationsintheverydry,dry,normal,wetandverywetyears. 5. Intra-annualvariability Previous works (e.g. Paz and Kutiel, 2003; Kutiel and Trigo, 2014) have presented the range of dates for which a certain rainfall percentage was accumulated and/or the rangeofaccumulatedpercentageobtainedforagivendate of the hydrological year. This kind of analysis allows us to extract a considerable amount of information from the daily series (it would be unfeasible with monthly values only) and represents a highly instructive way to analyse theintra-annualrainfallvariability. In Ponta Delgada, the median date when 50% of the Figure3.The distribution of very dry, dry, normal, wet and very wet annual (hydrological year) rainfall was accumulated years according to their NRS and RSY. The curved lines represent (mid-seasondate)is1January.However,therangeofthe isohyetsofannualrainfalldelimitingverydry,dry,normal,wetandvery dates, from the year when the accumulation of 50% was wetconditions. the earliest to the year when it happened latest, ranged from 16 October to 21 February. This means that the very wet years in Ponta Delgada are caused primarily by mid-seasondatevariedoverarangeof128days,ormore aconsiderableincreaseintheRSYorviceversa,whereas than 4months, which is almost half of the entire rainy dryorverydryyearsarecausedmainlybyaconsiderable season in Ponta Delgada. However, when this range is reductionoftheRSYandnotsomuchbyastrongreduc- examined between more conservative percentiles (10th tionoftheNRS. and the 90th), its temporal span is reduced by half and ©2016RoyalMeteorologicalSociety Int.J.Climatol.(2016) A.HERNÁNDEZetal. (a) (b) (c) Figure4.Mid-seasondatesofearliest,10%,median,90%andlatestyearsandtherespectivelyaccumulatedpercentages,for(a)theentireperiod (1873–2012),(b)SP1(1873–1942)and(c)SP2(1943–2012). varies between 28 November and 30 January, around 2 thisspanofpercentagesisreducedtolessthanhalf(24%), months(Figure4). varyingfrom39.9to63.9%(Figure4). The intra-annual variability can also be analysed by These results show that when the intra-annual variabil- examining the range of accumulated percentage for any ity is examined – without the influence of the earliest given date, e.g. at the mid-season date. When the entire and latest 10th percentile years – the series variability is rangeofyearsareconsidered,inthefirsthalfoftherainy reduced.Thereisanuncertaintyof2monthsintheoccur- season, between 22.2 and 76.6% of the annual total was rence of the mid-season date and about a quarter of the accumulated, i.e. a range of 54.4% of the annual rainfall annualtotalbythemedianmid-seasondate.Nonetheless, forthatdate.Fortherangeofthe10thand90thpercentiles, when all years are considered, these ranges are doubled, ©2016RoyalMeteorologicalSociety Int.J.Climatol.(2016) AZORESARCHIPELAGOPRECIPITATIONANDMODESOFCLIMATEVARIABILITY Table2.Spearman’s rank correlation coefficients associated with seasonal series of precipitation and large-scale modes (only large-scalemodeswithsignificantvaluesareshown). Precipitation 1873–2012 1873–2012 1900–2012 1873–2012 NAO |NAO|>0.5 AMO |AMO|>0.5 PDO |PDO|>0.5 ENSO |ENSO|>0.5 Winter(DJFM) −0.66 −0.74 0.14 0.17 −0.06 −0.10 0.13 0.09 Summer(JJAS) −0.37 −0.51 0.21 0.23 0.17 0.19 0.18 0.19 Grey,regularandboldnumbersindicatevaluesatp>0.05,p<0.05andp<0.01,respectively. representingslightlymorethan4monthsintheoccurrence winter and −0.37 for summer (Figures 5(a) and (b)). of the mid-season date and more than half of the annual These 𝜌 values rise to −0.74 and −0.51 if only seasons total by the median mid-season date. Therefore, the vari- withnon-neutralNAOivaluesareconsidered.Thesecond abilityinducedby20%oftheextremeyearsisalmostequal large-scalemodewithsignificantcorrelationcoefficientsis tothatinducedbytheremaining80%(Figure4). theAMOindex(AMOi).The𝜌valuebetweentheAMOi Differencesinthemid-seasondatebetweenSP1andSP2 andtheDJFMprecipitationis0.14(p-value<0.05),rising aresignificant(Figure 4).Thesedifferencesareespecially to 0.17 (p-value<0.05) if only seasons with non-neutral accentuated in the range of the dates of the mid-season AMOi values are taken into account. The JJAS 𝜌 val- date and in the range of rainfall percentage accumulated ues obtained with the AMOi are higher, with 𝜌=0.21 bythatdate.Despitethesmallernumberofyearsincluded, and 𝜌=0.23, respectively. The PDO index (PDOi) also SP1showshigherrangesofvariabilityofthedatesofthe displays relatively small albeit significant correlations mid-season date than SP2. This range is 78days when it (p-value<0.05) but only for JJAS with 𝜌 values of 0.17 iscalculatedbetweenthe10and90%percentilesand121 and0.19ifweconsiderallthesummersoronlythesum- days if the extreme years, from the earliest to the latest, merswithnon-neutralPDOivalues,respectively.Finally, are also included. These values decrease considerably to theENSOindex(ENSOi)isthelastlarge-scalemodethat 53and113days,respectively,forSP2.Iftheaccumulated shows significant correlation (p-value<0.05) with sea- rainfall by the median mid-season date is considered, sonal precipitation in Ponta Delgada. The summer pre- SP1 displays a higher range (27.1%), without the 10% cipitationcorrelatespositivelywiththeENSOibothifwe earliest and latest years, than SP2 (22.7%). However, if considereverysummer(𝜌=0.18)andifweconsideronly weconsideralltheyears,theSP2rangeishigher(54.4%) the summers with non-neutral ENSOi values (𝜌=0.19). than the SP1 range (39.6%). Thus, the range of the dates By contrast, the EA, the SCAND and the AO indexes do of the mid-season date for SP1 presents a ratio of 1.55 notshowanysignificantcorrelationbetweenthemandthe betweenalltheyearsanddisregardsthe10%earliestand seasonalprecipitation.ThislackintheinfluenceoftheEA latest years, whereas this ratio is 2.13 for SP2. Similarly, andSCANDisamajordifferencetowhatcanbeseenover therangeofrainfallpercentageaccumulatedbythemedian the Iberian Peninsula where these two large-scale mode mid-season date for SP1 shows a ratio of 1.46 between patternsplayamajorroleinshapingtheclimate,asshown alltheyearsandrejectsthe10thandthe90thpercentiles, with low (Trigo etal., 2008) and high (Jerez and Trigo, whereas this ratio is 2.40 for SP2. This would mean and 2013)resolutionclimatedatasets.Thismaysimplyreflect confirm that the so-called ‘normal’ conditions are more the fact that the SCAND major regions of influence are unstableforSP1,whiletheextremeyearsaremorevariable locatedtoofarfromtheAzoresArchipelagoandalsothat anduncertainforSP2(Figure4). the island falls inside the area of neutral influence of the EApattern,i.e.theareawheretheinfluenceofEAinpre- cipitationswitchesfrompositivetonegative. Nevertheless, the link between the Ponta Delgada 6. Impactoflarge-scalemodesofclimatevariability precipitation and the large-scale modes has not been onrainfallregime constant through time (Figures 6(a) and (b)). The Spearman’s correlation coefficients (𝜌) between the non-stationary relationship between the NAOi and extended (DJFM and JJAS) seasonal large-scale modes environmental variables has already been pointed out and precipitation were computed with the seasonal data in the Azores Archipelago (Cropper and Hanna, 2014), of all the available years (Table S1). However, only the North Atlantic sector (García Molinos and Dono- large-scale modes displaying significant correlations hue, 2014) and continental Europe (e.g. Trigo etal., (p-value<0.05)arepresentedanddiscussed(Table2).In 2004; Vicente-Serrano and Lopez-Moreno, 2008). addition,wedefinedtheseasonalhighandlowlarge-scale Here, we also identify a non-stationary pattern on the mode composites to be constituted by all winters and precipitation–NAOi relationships. The running corre- summers, with the corresponding indices higher than lation (19-year sliding window) between the winter 0.5 and lower than −0.5, respectively, i.e. non-neutral (Figure 6(a)) and summer (Figure 6(b)) Ponta Del- conditions. gada precipitation and the NAOi illustrates a clear The NAO index (NAOi) shows the highest correlations non-stationary behaviour of the relationship. The winter of all the large-scale modes with 𝜌 values of −0.66 for NAOi–precipitation correlation shows a fairly constant ©2016RoyalMeteorologicalSociety Int.J.Climatol.(2016) A.HERNÁNDEZetal. (a) (b) (c) (d) (e) (f) Figure5.Correlationbetweenextendedwinter(DJFM)andsummer(JJAS)precipitation(mm)andNAOifortheentirestudiedperiod(aandb), SP1(candd)andSP2(eandf). behaviour until approximately 1930 before declining for observedduringthefirstdecades(1910–1940)andthelast the following 20years. The lowest values of correlation years(1997–2005).Conversely,thelargeinfluenceofthe areobservedaroundthe1950s,withaminimumof−0.37. NAOonwinterprecipitationseemstomasktheroleplayed After 1950, an increasing trend is recorded, achieving by the other large-scale modes for winter. Only prompt the maximum correlation values (−0.84) in 1975 fol- peaksofAMOinfluenceoverwinterprecipitation(around lowed by a period of weaker correlations with the values 1905,1965,1988and1995)seemtooverlapwiththewin- rising again since 2000. The temporal evolution of the terNAOimpact,whiletheimpactoftheENSOandPDO correlation coefficient between summer precipitation and onthewinterprecipitationismuchlower(Figure6(a)). NAOi is more oscillatory. These correlations are also Aspreviousworks(e.g.Trigoetal.,2008;Bladéetal., negative but with smaller values, with more restricted 2012;Comas-BruandMcDermott,2014)pointedout,the significant periods between 1886–1909 and 1941–1997. NAOisthemainlarge-scalemodecontrollingtheseasonal An increasing trend can be observed from 1886 to 1897 precipitation over the North Atlantic region. Our results beforedeclininguntil1909.Asimilarpatterncanbeseen highlightthisfact,showingthatupto50%ofthevariance from1941to1997withanincreasingtrenduntil1970and ofthewinterprecipitationcanbeexplainedbytheNAOfor a continuous weakening until the end of the record. The winterswithpositiveandnegativevalues.Insummer,the range of significant correlations is from −0.35 to −0.72 NAOroleisalsosignificantalthoughitislessprominent, and −0.35 to −0.84, respectively. Several authors (e.g. andtheroleofotherlarge-scalemodes(AMO,ENSOand Pauling etal., 2006; Comas-Bru and McDermott, 2014) PDO)shouldnotbeneglected.Theseresultsalsohighlight linked this changing nature to variations in the spatial thenon-stationarybehaviouroftheNAOandprecipitation configuration of the NAO and the increment of the role relationshipintheAzoresArchipelago,whichwasalready playedbytheotherclimaticcirculationmodes.Theroleof pointedoutforotherclimatevariables(i.e.temperature)in theothersignificantlarge-scalemodes(AMO,ENSOand previousworks(e.g.CropperandHanna,2014). PDO)forperiodsoflowerNAOimpactseemsclearforthe 6.1. TheseasonalNAOinfluence North Atlantic sector in summer. Figure 6(b) highlights thefactthatwhentheNAOimpactdoesnotachieveasta- As outlined above, the 𝜌 values for extended win- tisticallysignificantrole,otherlarge-scalemodesincrease ter and summer precipitation in Ponta Delgada and the theirinfluenceontheprecipitation.Thisbehaviourcanbe correspondingNAOiare−0.66and−0.37forthe140-year ©2016RoyalMeteorologicalSociety Int.J.Climatol.(2016)

Description:
d IPMA IP - The Portuguese Sea and Atmosphere Institute, Lisbon, Portugal . cyclone (Agostinho, 1948; Santos et al., 2004; Azevedo,. 2006). Thus
ebook img

New Azores archipelago daily precipitation dataset and its links with large-scale modes of climate PDF

16 Pages·2016·1.98 MB·English
by  
Save to my drive
Quick download
Download
Upgrade Premium
Most books are stored in the elastic cloud where traffic is expensive. For this reason, we have a limit on daily download.

Preview New Azores archipelago daily precipitation dataset and its links with large-scale modes of climate

INTERNATIONALJOURNALOFCLIMATOLOGY Int.J.Climatol.(2016) PublishedonlineinWileyOnlineLibrary (wileyonlinelibrary.com)DOI:10.1002/joc.4642 New Azores archipelago daily precipitation dataset and its links with large-scale modes of climate variability Armand Hernández,a* Haim Kutiel,b Ricardo M. Trigo,a Maria A. Valente,a Javier Sigró,c Thomas Croppera and Fátima Espírito Santod aFaculdadedeCiências,InstitutoDomLuiz(IDL),UniversidadedeLisboa,Portugal bLaboratoryofClimatology,DepartmentofGeographyandEnvironmentalStudies,UniversityofHaifa,Israel cCentreforClimateChange(C3),Tarragona,Spain dIPMAIP-ThePortugueseSeaandAtmosphereInstitute,Lisbon,Portugal ABSTRACT:Withinthescopeofthetwomajorinternationalprojectsoflong-termreanalysisforthe20thcenturycoordinated bytheNationalOceanicandAtmosphericAdministrationandEuropeanCentreforMedium-RangeWeatherForecasts,the InstitutoDomLuizfromtheUniversityofLisbonhasdigitizedalargenumberoflong-termdailyandmonthlyclimaterecords fromstationsinPortugalandformerPortuguesecolonies.Wehaverecentlyfinishedthedigitizationofprecipitationvalues from Ponta Delgada (capital of the Azores Archipelago), obtaining an almost complete daily precipitationseries, with the exceptionofsomeyears(1864–1872;1878–1879;1888–1905;1931;1936;and1938)forwhichonlymonthlyvaluesare available.Atdailyresolution,wehaveusedacomprehensiveassessmentondifferentcharacteristicsofrainspells(consecutive days with rainfall accumulation). The distribution of precipitation presents an evident seasonal pattern and reveals large inter-annualandintra-annualvariability,increasingconsiderablyinthelast3decades.Thefrequencyofdryyearsdecreases almostbyhalfbetweenthefirstandthesecondpartoftherecord,whereaswetyearsincreaseuptothreetimes.Thisismainly due to more intense events that are reflected by higher rain-spell yields (amount of precipitation) and rain-spell intensity (amountofprecipitationperday)values.Mostoftheextremeprecipitationeventsoccurredduringthelast2decades,and theygenerallycorrespondtodateswithcyclonicconditionsovertheNorthAtlantic.Wehavealsolookedintotheinfluenceof large-scalemodesofclimatevariabilityontheprecipitationregimeoftheAzoresArchipelago.Asexpected,theNorthAtlantic Oscillation(NAO)hasamajorimpactontheprecipitationregimeofPontaDelgadabothinwinterandsummer.However,our resultsshowanon-stationaryNAOinfluenceandtheimpactofotherlarge-scalemodes(includingtheAtlanticMultidecadal OscillationandElNiño-SouthernOscillation)increaseswhenthisinfluencebecomesweaker. KEYWORDS dailyprecipitation;AzoresArchipelago;teleconnection;seasonality;NorthAtlanticOcean Received27May2015;Revised3December2015;Accepted4December2015 1. Introduction long-termstationsrecordsfromPortugalandformerPor- tuguesecolonies(Stickleretal.,2014). Long-term climate datasets are essential for understand- Island environments are very sensitive to externally ing climate changes (Compo etal., 2011). However, the inducedchanges,andthelimitedareaandresourcesavail- currentlylimitedavailabilityoflong-terminstrumentalcli- able to island states can make it difficult to apply adap- materecordshampersourabilitytoconductmorereliable tation methods (IPCC, 2014, WG3). In particular, small assessments of climate variability and change (Trenberth islands,likethoseintheMacaronesianregionintheeast- etal.,2007).Presentaccesstodailydigitalclimatedatais ern North Atlantic (Figure 1(a)), have long been identi- scarceovermanypartsoftheworld,particularlyinoceanic fied among the most vulnerable to climate change and regions, and this limited coverage of instrumental data climate extremes (Nurse etal., 2014) due to the poten- obstructs the ability to better understand climate changes tialimplicationsintherelocationofresourcesandactivi- and its forcing factors (Brunet and Jones, 2011). In this ties (Santos etal., 2004). Changes in precipitation are of senseandwithinthescopeofthetwomajorinternational particular concern as precipitation, or a significant lack projectsoflong-termreanalysisforthe20thcenturycoor- of it, can be responsible for flash floods, landslides and dinatedbytheNOAA(Compoetal.,2011)andECMWF droughts. The Azores Archipelago (Figure 1(a)) is usu- (Hersbachetal.,2013),theInstitutoDomLuiz(IDL)from ally under the direct influence of the semi-permanent, the University of Lisbon has digitized a large number of high-pressureAzoresanticyclone.Duringmostoftheyear (September–March), if this high-pressure centre is dis- sipated or displaced latitudinally, then the Azores region *Correspondence to: A. Hernández, Faculdade de Ciências, Instituto DomLuiz(IDL),UniversidadedeLisboa,1749-016Lisboa,Portugal. is frequently crossed by the North Atlantic storm track, E-mail:[email protected] the main path of rain-producing weather systems (Trigo ©2016RoyalMeteorologicalSociety A.HERNÁNDEZetal. (a) (b) (c) Figure1.(a)MapoftheMacaronesianregionshowingtheAzoresArchipelagoandPontaDelgadalocation.(b)Annualand(c)monthlymean precipitation(mm),1873–2012.Annualseriesofprecipitationwerecalculatedbysummingdailyvaluesifnomorethantwovaluespermonthwere missing(solidbars).Whenthisoccurs,serieswerecompletedusingmonthlydatasetsfrompreviouslydigitisedmonthlyseriesalreadyusedbyother authors(e.g.CropperandHanna,2014)coveringthistimeperiod(openbars).Thethicklinerepresentsa31-pointmovingmean,whichemphasizes thelow-frequencyvariabilityoftheannualprecipitationseriesandtheshadowindicatesthecorresponding31-pointmovingstandarddeviation.Note thatthevaluesobtainedforbothendswerecalculatedusingasmanyyearsaspossibleinthemovingaveragewhenthenumberofyearswasless than31. etal., 2004). Conversely, during late spring and summer, (Andrade etal., 2008; Marquesetal., 2008; Cropper and theAzoresclimateismainlydominatedbytheAzoresanti- Hanna, 2014). The NAO is the leading mode of climate cyclone (Agostinho, 1948; Santos etal., 2004; Azevedo, variability in the North Atlantic region and is manifested 2006).Thus,therainfallregimefromPontaDelgada(São asameridionaldipoleanomalyinsea-levelpressure,with Miguel Island, Azores Archipelago, Portugal) displays thetwocentresofactionlocatedapproximatelyoverIce- a strong seasonal cycle and large inter-annual variabil- land and the Azores (Hurrell, 1995; Trigo etal., 2004). ity (Marques etal., 2008). The changes in precipitation However, besides the NAO, there are other large-scale areusuallylargerthantemperatureorpressurevariations, modes, such as the East Atlantic pattern (EA), the Scan- and long-term records are essential to accurately present dinavian pattern (SCAND), the Arctic Oscillation (AO), the local rainfall regime (e.g. Esteban-Parra etal., 1998; the Atlantic Multidecadal Oscillation (AMO), the Pacific Kutiel and Trigo, 2014). Variations in total precipitation Decadal Oscillation (PDO) and the El Niño-Southern canbecausedbychangesinthecharacteristicsofthecon- Oscillation(ENSO),thatmayalsocontributesignificantly stituting rain events, including their timing, intensity and to the North Atlantic region’s precipitation regime (e.g. some aspects of the dryness conditions between rainfall ThompsonandWallace,1998;Pozo-Vázquezetal.,2005; episodes(e.g.PazandKutiel,2003).Therefore,inorderto Sutton and Hodson, 2005; Trigo etal., 2008; Comas-Bru improvetheunderstandingofprecipitationbehaviourasan and McDermott, 2014). Furthermore, it is worth noting indicatorofclimatechangesinthelastcentury,dailypre- the existence of inter- and intra-annual variations in the cipitationseriesmustbeanalysed(Brunettietal.,2001). impactoftheselarge-scalemodesontheclimatethrough- Someauthors(e.g.Andradeetal.,2008;Marquesetal., out long periods. In particular, the correlation between 2008;CropperandHanna,2014)havepreviouslyanalysed the NAO and precipitation in the Macaronesian region long-term characteristics of precipitation for Ponta Del- showsanon-stationarybehaviourovertime(Cropperand gadaanditsrelationshipwiththemainlarge-scalemodes Hanna, 2014), and therefore, studies of the long-term in the North Atlantic sector, the North Atlantic Oscil- influence of large-scale modes on rainfall variability are lation (NAO). However, these studies have used shorter required. However, despite the increasing understanding and/or lower-resolution time series, and to the best of of the impact of large-scale modes on the North Atlantic our knowledge, there is no work evaluating the evolu- region climate, to our knowledge, the impact of other tionofdailyprecipitationtime-seriescharacteristicsstart- large-scale modes on the rainfall regime of the Azores ing in the second half of the 19th century. Uninterrupted Archipelagohasnotpreviouslybeenexplored. monthlyprecipitationvaluesfromPontaDelgadaarecur- The influence of large-scale modes on the variability rentlyavailablesince1865withoutmissingdata(Cropper of North Atlantic storms has been the focus of several and Hanna, 2014), and a new, almost complete, corre- previous works (e.g. Kossin etal., 2010). Intense precip- sponding daily precipitation series, with the exception of itation greatly affects human activities as it can cause someyears,hasrecentlybeendigitizedatIDL. catastrophic floods or landslides among other impacts. A large fraction of the inter-annual variability of pre- The analysis of the behaviour of extreme rainfall and its cipitation observed in the Azores Archipelago has previ- links to large-scale modes is essential to better under- ously been ascribed to the NAO influence during winter stand their impact on human endeavours. Days with ©2016RoyalMeteorologicalSociety Int.J.Climatol.(2016) AZORESARCHIPELAGOPRECIPITATIONANDMODESOFCLIMATEVARIABILITY extreme values of precipitation are potential triggers of homogenization method, we have used some long-term catastrophic episodes, andprevious studies ofthe Azores precipitation series available for the Azores Archipelago Archipelago (e.g. Andrade etal., 2008; Marques etal., asreferences[Terceira(AngradoHeroismo,1873–2012) 2008) have focussed their efforts on establishing links andSãoMiguelIsland(Furnas,1935–2012;CasadoJosé, betweencyclonesandtheNAOpattern. 1935–2012; Santana, 1947–2012)]. We applied annual Our study aims to contribute to the understand- pair-wise detection and joint detection to estimate homo- ing of the rainfall regime in the Azores Archipelago geneity breaks and to determine potential change points. by investigating daily precipitation data recorded at Then,metadatarecordsforeachstationwereusedtocon- Ponta Delgada. Changes in precipitation in the period firmbreakpoints.Twohomogeneitybreaksweredetected 1873–2012 (∼140years) were studied using a method- inthePontaDelgadaprecipitationseriesfortheyears1888 ological approach based on different characteristics of and 1936, the first one coinciding with the beginning of rainspells(e.g.ReiserandKutiel,2008).Thismethodhas a period that was filled with the newly digitized monthly beenshowntoperformwellwhenanalysingthedifferent data(1888–1905)andthesecondwithastationrelocation parameters related to the rainfall regime in other parts of (1936).Theadjustmentofinhomogeneitieswasdonewith the world with strongly seasonal precipitation behaviour, the option log ratio multiplicative correction for cumu- suchastheMediterraneanregion(e.g.ReiserandKutiel, lative parameters in HOMER. This type of adjustment 2007).Additionally,theinfluenceofthelarge-scalemodes by multiplicative ratio is normally used to homogenize on the inter-annual precipitation variability as well as on precipitation series, which is different from the additive the severe rainfall events was investigated. We also anal- factors used in temperature series (Alexandersson, 1986; ysed the trajectory of significant North Atlantic tropical Venema etal., 2012).To adjust daily data, we have cho- storms that are shown to sometimes influence extreme sen the scheme developed by Vincent etal. (2002). This precipitationacrosstheAzoresArchipelago. approach attempts to provide a better time-interpolation procedure that preserves monthly values and does not introduce artificial discontinuities at the beginning and 2. Dataandmethods end of calendar months. The impact of this homogene- ity adjustment is relatively small, and it affects only the 2.1. Datasets period previous to 1936. Between 1889 and 1936, the Daily rainfall data from 1873 until 2012 from the Ponta monthly impact of the adjustment is ∼10mm on average ∘ ∘ Delgadastation(37 44’N,25 40’W,77m)havebeendig- compared with the original data, whereas for the earlier itized from meteorological annals and then quality con- period,1872–1888,theaveragedifferenceis∼5mm. trolled according to the framework presented in Stickler This study has been performed with data from an etal.(2014).Additionally,wetestedthequalitycontrolof Azorean meteorological station, and therefore, an dailydatawithRClimDex_extraQCroutines(Aguilarand extended daily (1850 to present) station-based NAO Prohom, 2011). Data from 1873 to 1946 were obtained index, based on the reconstructed Azores sea-level from the ‘Annaes do Observatório Infante D. Luiz’ (see pressure data presented by Cropper etal. (2015), has Valenteetal.,2008andStickleretal.,2014)andfor1947 been employed. The temporal evolution of the other to 2012 were supplied by the Instituto Português do Mar large-scale modes was obtained through the numeri- e da Atmosfera (IPMA). Thus, we obtained a new daily cal indices stored at the Climate Prediction Center of precipitation series from 1873 to 2012, with the excep- the National Oceanic and Atmospheric Administra- tionofsomeyears(1878–1879;1888–1905;1931;1936; tion (http://www.cpc.ncep.noaa.gov). Monthly resolution and1938).Inaddition,annualandmonthlyseriesofpre- datasetswereemployedwithdifferentdurationsaccording cipitation were calculated by summing daily values if no to the available data (Table S1, Supporting Information). morethantwovaluespermonthweremissing.Whenthis TheAtlantictropicalstormsdatawereretrievedfromthe occurs, monthly and annual series were completed using NationalHurricaneCenterfromtheNationalOceanicand the monthly datasets from previously digitized monthly AtmosphericAdministration(http://www.nhc.noaa.gov/). series(IDL-IPMA)coveringthistimeperiod(Cropperand Hanna,2014) 2.2. Rainfallanalyses Long-term climate time series have been influenced by non-climaticfactorsquiteoften,mainlyrelatedtochanges The daily rainfall regime was explored using a specially in instrumentation, instrumental exposure, station loca- designed statistical model, entitled the Rainfall Uncer- tions and local environments, observing practices and/or tainty Evaluation Model (RUEM; Reiser and Kutiel, data processing (e.g. Aguilar etal., 2003; Trewin, 2010). 2008). Allthesefactorscanintroducegradualorabruptbreaksin To avoid splitting the Azores rainy period into two thehomogeneityofclimaterecords. different years, we use the hydrological year rather than To detect and estimate adjustments for these problems, the traditional meteorological year. Most definitions for thePontaDelgadaannualandmonthlyprecipitationseries a hydrological year start on September 1 or October 1. wereexaminedforhomogeneityusingthesoftwarepack- However, in this study, we will use the hydrological year ageHOMER(Mestreetal.,2013)todetectpossiblebreaks definition starting on 1 July and ending on 30 June of in the homogeneity of series. As HOMER is a relative the next calendar year as proposed by Reiser and Kutiel ©2016RoyalMeteorologicalSociety Int.J.Climatol.(2016) A.HERNÁNDEZetal. (2008)forregionswithdrysummersandlongwetseasons. 1(b)).Thisbreakpoint(1942–1943)waschosenasitcor- Thenumberoftheyearreferstowhenitstarted,e.g.1873 respondstoapotentialprecipitationregimechangeinthe referstotheperiodthatspansbetween1July1873and30 regionalseries(CropperandHanna,2014).Additionally,it June1874. wasalsoasuitablepointastheSubperiod1(SP1)andSub- For our analysis, we focussed on rain spells instead of period 2 (SP2) needed to be of near similar lengths. The rainy days as the former better accommodates the total firsthalf(SP1)spansfrom1873until1942butonlycom- length of any rainy period. A rain spell is defined as a prises42years(outofapossible70)duetothelackofdata; period of consecutive days with rainfall above a given thesecondhalf(SP2)spansfrom1943until2012,includ- predetermined daily rainfall threshold (e.g. Reiser and ing all 70years. To evaluate the representativeness of the Kutiel,2012).Eachrainspellisprecededandfollowedby 42years in the first period, appropriate statistical t- and atleastonedrydayoradaywitharainfallamountlessthan F-tests were applied to the mean and standard deviation thedailyrainfallthreshold(1.0mminthisstudy).Foreach values of SP1 obtained with the different length (70-year rainspell,itsduration(RSD),yield(RSY,thetotalrainfall vs 42-year) datasets. Results do not show significant dif- accumulated during the rain spell) and mean intensity ferencesbetweenthemonthly(70-year)anddaily(42-yr) (RSI, the average daily rainfall in the rain spell) were datasets with a confidence level of 99%, which suggests calculated. Additionally, the total number of rain spells itisfeasibletomakevalidcomparisonsbetweenthedaily (NRS) and the rainy season length (RSL) in each year precipitationregimeofSP1(42-year)andSP2(70-year). was also considered. The RSL is defined as the number The rainfall regime in the Azores Archipelago at of days elapsed between the beginning and the end of daily resolution comprises a time window of 140years the rainy season. The beginning and the end dates are (1873–2012) with five gaps lacking daily data (1878– when10and90%oftheannualrainfallwereaccumulated, 1879;1888–1905;1931;1936;and1938).Thus,117years respectively. These dates vary from year to year, and the wereinitiallyavailable,and,asweusedhydrologicalyears RSL varies accordingly. This was done in order to avoid insteadofmeteorologicalyears,atotalof112hydrologi- somesporadicrainswithminuteamountsatthebeginning calyears(July1–30)couldbeanalysedwiththeRUEM. and/orendoftherainyseasonthatwillcauseaverylong WedecidedtoreproducetheRUEManalysisfortheentire RSL consisting of long dry periods in between (Reiser period,thetwosubperiodsandthecategorisedverydryto and Kutiel, 2009). Following Kutiel and Trigo (2014), verywetyears. annual totals in all years/seasons were standardized, and the years/seasons were categorized accordingly into five 2.3. Atmosphericcirculationpatterns’influence categories:verydry(z<−1),dry(−1≤z<−0.5),normal onrainfallregime (−0.5≤z≤0.5),wet(0.5<z≤1)andverywet(1<z).For Theextendedwinter(DJFM)andsummer(JJAS)influence each category, the above rain-spell parameters were also ofeachconsideredlarge-scalemode(NAO,EA,SCAND, calculatedseparately. AO, AMO, PDO and ENSO) on precipitation has been Intra-annualvariabilitywasanalysedbycalculatingand explored. This was performed according to Spearman’s comparing the dates of accumulated rainfall percentage. rank correlation coefficients (𝜌) and associated p-values. For each day in each hydrological year, the accumulated Unless otherwise stated, significance (p-value) is always percentageofthetotalannualrainfallofthatyearwascal- discussedatp<0.01. culated,andthesevalueswerethensortedinanascending order for each Julian day, thus enabling the values to be presentedinaprobabilisticway. 3. Overallrainfallregime Drynesswasanalysedusingthedrydayssincelastrain (DDSLR)approach(e.g.Kutiel,1985;Aviadetal.,2009; The rainfall regime in Ponta Delgada shows clear Lana etal., 2012; Kutiel and Trigo, 2014). According to inter-annual and intra-annual variability (Figures 1(b) the DDSLR methodology, each rainy day is attributed a and (c)). The mean annual precipitation for 1873–2012 ‘ 0’ value, the first dry day is attributed a ‘1’ value, the is 960.6mm, with a standard deviation of 201mm, next consecutive dry day a ‘2’ value and so on until the and almost 75% of the total precipitation was accu- nextrainyday,whichisagainattributeda‘0;value.Values mulated between October and March. The mean rainy carry over from 1year to the next. Once all days in the seasonlengthis277days(Table1).Although theAzores analysedperiodwereattributedwithavaluethatdescribes Archipelago is geographically distant from the Mediter- the distance in days of that specific day from the last ranean basin, the rainfall regime presents some similar rainyday,allvaluesforeachJuliandayweresortedinan characteristics with the Mediterranean region, with ascending order, thus enabling the values to be presented maximum precipitation during the cold season and min- againinaprobabilisticway. imum precipitation during summers (Reiser and Kutiel, The long-term inter-annual rainfall evolution through- 2009). Precipitation occurred (on average) in ∼122 rainy out the entire period (1873–2012) was studied by apply- days consisting of 60 rain spells with a mean duration ing a 31-point moving mean window to the annual mean of ∼2days. Each rain spell yields 16mm on average precipitationcalculatedfromthedailytimeseries(Figure with a mean rate of 7.9mmday−1. About 42% of the 1(b)).Inaddition,thisperiodwasdividedintotwohalves annual total is due to rain spells longer than 3days thatcorrespondtoaregimechangeinprecipitation(Figure (Table1). ©2016RoyalMeteorologicalSociety Int.J.Climatol.(2016) AZORESARCHIPELAGOPRECIPITATIONANDMODESOFCLIMATEVARIABILITY dh D 42)anoreac RatioVW/V 2.02 1.37 1.04 2.12 0.99 1.35 1.94 1.48 1.51 9f 1s (1873–wetyear RatioW/D 1.38 1.13 1.08 1.21 0.97 1.05 1.27 1.21 1.08 eriods1ndvery RatioP2/SP1 1.12 1.03 1.03 1.03 0.97 1.00 1.10 1.11 0.97 pa S subdry –2012),ndvery wet62mm)ears) SD 113.6 13 4 2 34 0.2 2.2 0.7 9.4 1873weta Very>(11(16y X 302.5 141 62 10 276 2.3 21.1 9.3 48.4 (d 1 hn gta entireserieslend2,betweendry Wet(1061–1162mm)(13years) XSD 1114.225.1 13110 647 92 27024 2.10.2 17.72.2 8.50.7 44.713.2 en ha t nthisstudyforeensubperiod1 Normal(860–1061mm)(47years) XSD 963.061.3 12011 604 82 27728 2.00.2 16.21.6 8.10.9 41.911.9 fferentprecipitationindicesusediownistheratioofthemeansbetwprecipitationindex. VerydryDry<(760mm)(760–860mm)(13years)(23years) XSDXSD 643.580.9810.033.0 103711710 606596 5282 2803527835 1.71.32.00.2 10.91.613.91.5 6.30.97.00.6 32.010.741.49.8 dish variationforthewetyears.Also 1943–2012SP2(70years) XSD 1000.9184.4 12214 615 82 27332 2.00.2 16.63.2 8.31.1 41.212.1 fy ndcoefficientoverydrytover 1873–1942SP1(42years) XSD 893.3211.5 11916 595 83 28327 2.00.2 15.13.4 7.51.2 42.511.5 ad Table1.Mean,standarddeviation2(1943–2012),andthecategorize 1873–2012(112years) XSD Annualrainfall960.6201.0(mm)Numberofrain12115days(NRD)Numberofrain605spells(NRS)Numberofrain83spellslongerthe>3days(NRS3)Rainyseason27731length(RSL,days)Rainspell2.00.2duration(RSD,days)Rainspellyield16.03.3(RSY,mm)Rainspell7.91.2intensity(RSI,mm/day)Relative41.911.8contributionofrainspellslongerthan3days(%) ©2016RoyalMeteorologicalSociety Int.J.Climatol.(2016) A.HERNÁNDEZetal. Theprecipitationrecordshowsanincreaseintherainfall 4. Inter-annualvariability conditions between a drier earlier period (1873–1942; Table 1 presents the analysed precipitation parameters SP1) and a wetter recent period (1943–2012; SP2). The studied for the five ‘wetness/dryness’ categories (Section meanannualtotalofSP1is893.3mm,whereasSP2shows 2.2).Itcanbeobservedthatthemeantotalrainfallduring avalueof1001mm,butwiththeRSL,itreducedby9days. very wet years was 1302mm, around twice the mean The precipitation increase is mainly a consequence of an of the very dry years, which had a mean of 644mm. increase of the RSY, RSI and, to a lesser extent, in the All parameters, with the exception of the RSL, show an NRS.Conversely,theRSDisunchangedbetweenthetwo increase from the very dry years towards the very wet periods (Table 1). The mean values of all the analysed years.TheNRShasthelowestverywet/verydryratiowith parameters (with the exception of the RSL and RSD) are valuesof1.04.Thismeanstherewasaslightmeanincrease slightlyhigherduringSP2butpresentalowercoefficient of4%inthenumberofrainspells.Incontrast,whenonly of variation, which highlights a larger heterogeneity and longrainspells(RSD>3)wereconsidered,theirnumber variabilityoftherainfallregimethroughoutSP1. increasedconsiderably,fromameanof4.9intheverydry As mentioned above, the Ponta Delgada precipitation years to 10.4 in the very wet years, corresponding to an regimedisplayssimilarcharacteristicswithMediterranean increase of more than two times. These long rain spells, climate behaviour (e.g. Reiser and Kutiel, 2009) in spite whichconsistonlyof28.6%ofthetotalNRSinthevery of the ca. 3000km between them, although both regions dryyears,nearlydoubledtheirpercentageto42.4%inthe are within similar latitudinal bands. Wigley and Farmer very wet years (Figure 2). The RSD and RSI represent a (1982) defined the Mediterranean climate as having win- moderate increase between those two categories of years ter season rainfall more than three times that of summer. between1.3and1.5.TheRSYrepresentsthehighestratio KutielandTrigo(2014)laterdefinedthisastheMediter- betweenthetwo(besidestheNRS>3)andwasdoubledin ranean Climate Index (MCI). The MCI is an index that theverywetyearscomparedwiththeverydryyears,21.1 describesthelevelofseasonalityoftherainfall.Thisindex and 10.9mm, respectively. Hence, these five categories calculates the ratio between winter and summer rains. According to it, an MCI<1 means summer wetter than have very different RSY values. However, even though winter, 1≤MCI<3 winter slightly wetter than summer, their NRS values slightly increase from very dry to very 3≤MCI<10 dry summer and 10≤MCI extremely dry wet years, they overlap to a large extent (Figure 3). This summer.AlocationwithanMCI≥3isconsideredtohave implies that while the RSY can be used as a good index of the rainfall variability, the NRS cannot serve as an aso-calledMediterraneanclimate.TheMCIinPontaDel- gada is 3.02 and therefore, fits well, albeit by a minimal indicator to differentiate between these years and has an margin,withinthedefinitionoftheMediterraneanclimate insignificantimpactontheannualtotal. typepreviouslydefined.TheMCIofSP1is2.69,whereas The percentage of the total rainfall contributed by rain SP2 shows a MCI of 3.34, indicating a change towards spells of various durations is represented by the rela- more‘Mediterranean’conditions,withalargerconcentra- tive contribution. The relative contribution of long rain tionofprecipitationinwinter. spells(RSD>3days)isusuallyhigherinmorehumidsta- A priori, dryness is not a problem in the Azores tions, whereas, in drier locations, the main contribution Archipelago, but islands are very sensitive to climate to the annual total is made by very short rain spells of changes, especially to changes in the hydrological avail- 1 or 2days (Reiser and Kutiel, 2012; Kutiel and Trigo, ability.Thus,thedrynessconditions(severity,consistency 2014; Kutiel etal., 2014). This is reflected in this study andtemporaluncertainty)wereanalysedusingtheDDSLR by dividing the entire dataset into five rainfall categories approach[seeKutielandTrigo(2014)forfurtherdetails]. as explained earlier. Although Ponta Delgada is consid- In Ponta Delgada, the longest DDSLR (severity) occurs ered to be a humid location, the maximum contribution at the end of August and the beginning of September, varies according to the different categories. In very dry anditspans63days(FigureS1,SupplementaryInforma- or dry years, the maximum contribution is by rain spells tion). We can expect a DDSLR longer than 30days on of 1day as in arid regions, whereas in normal, wet and average once every 25years. The temporal uncertainty, very wet years, the distributions resemble humid regions i.e. the span of time when such a DDSLR has been (Figure 2). observed,isfrom14Aprilto21September,overaperiod If we compare the relative contribution of short of 161days. Similarly, the only observed example of a rain spells (RSD≤3days) and very long rain spells DDSLRlongerthan60daysoccurredduringthe3days30 (RSD>7days),wefindthatthefirstdecreasesfrommore August–1September,andtherefore,themostlikelytime thantwo-thirds(65.2%)inverydryyearstoslightlyover for very high DDSLR values would be late August/early half (51.3%) in very wet years. By contrast, the relative September (Figure S1). The dry period started 60days contributionofverylongrainspellsisdoubledfrom4.8% earlier (around the end of June or beginning of July) inverydryyearsto9.5%inverywetones(Figure 2). and reached its maximum (more than 60days) during These results are in agreement with those obtained these 3days. These DDSLR values are considerably recently by Reiser and Kutiel (2012) with an application lower compared with those obtained for Lisbon (Kutiel fortheentireMediterraneanarea,butconsideringamuch and Trigo, 2014), indicating wetter conditions in Ponta shorterperiod,andbyKutielandTrigo(2014)forasim- Delgada. ilar long-term period in Lisbon. This means that wet or ©2016RoyalMeteorologicalSociety Int.J.Climatol.(2016) AZORESARCHIPELAGOPRECIPITATIONANDMODESOFCLIMATEVARIABILITY Figure2.Meanrelativecontributionofrainspellsofvariousdurationsintheverydry,dry,normal,wetandverywetyears. 5. Intra-annualvariability Previous works (e.g. Paz and Kutiel, 2003; Kutiel and Trigo, 2014) have presented the range of dates for which a certain rainfall percentage was accumulated and/or the rangeofaccumulatedpercentageobtainedforagivendate of the hydrological year. This kind of analysis allows us to extract a considerable amount of information from the daily series (it would be unfeasible with monthly values only) and represents a highly instructive way to analyse theintra-annualrainfallvariability. In Ponta Delgada, the median date when 50% of the Figure3.The distribution of very dry, dry, normal, wet and very wet annual (hydrological year) rainfall was accumulated years according to their NRS and RSY. The curved lines represent (mid-seasondate)is1January.However,therangeofthe isohyetsofannualrainfalldelimitingverydry,dry,normal,wetandvery dates, from the year when the accumulation of 50% was wetconditions. the earliest to the year when it happened latest, ranged from 16 October to 21 February. This means that the very wet years in Ponta Delgada are caused primarily by mid-seasondatevariedoverarangeof128days,ormore aconsiderableincreaseintheRSYorviceversa,whereas than 4months, which is almost half of the entire rainy dryorverydryyearsarecausedmainlybyaconsiderable season in Ponta Delgada. However, when this range is reductionoftheRSYandnotsomuchbyastrongreduc- examined between more conservative percentiles (10th tionoftheNRS. and the 90th), its temporal span is reduced by half and ©2016RoyalMeteorologicalSociety Int.J.Climatol.(2016) A.HERNÁNDEZetal. (a) (b) (c) Figure4.Mid-seasondatesofearliest,10%,median,90%andlatestyearsandtherespectivelyaccumulatedpercentages,for(a)theentireperiod (1873–2012),(b)SP1(1873–1942)and(c)SP2(1943–2012). varies between 28 November and 30 January, around 2 thisspanofpercentagesisreducedtolessthanhalf(24%), months(Figure4). varyingfrom39.9to63.9%(Figure4). The intra-annual variability can also be analysed by These results show that when the intra-annual variabil- examining the range of accumulated percentage for any ity is examined – without the influence of the earliest given date, e.g. at the mid-season date. When the entire and latest 10th percentile years – the series variability is rangeofyearsareconsidered,inthefirsthalfoftherainy reduced.Thereisanuncertaintyof2monthsintheoccur- season, between 22.2 and 76.6% of the annual total was rence of the mid-season date and about a quarter of the accumulated, i.e. a range of 54.4% of the annual rainfall annualtotalbythemedianmid-seasondate.Nonetheless, forthatdate.Fortherangeofthe10thand90thpercentiles, when all years are considered, these ranges are doubled, ©2016RoyalMeteorologicalSociety Int.J.Climatol.(2016) AZORESARCHIPELAGOPRECIPITATIONANDMODESOFCLIMATEVARIABILITY Table2.Spearman’s rank correlation coefficients associated with seasonal series of precipitation and large-scale modes (only large-scalemodeswithsignificantvaluesareshown). Precipitation 1873–2012 1873–2012 1900–2012 1873–2012 NAO |NAO|>0.5 AMO |AMO|>0.5 PDO |PDO|>0.5 ENSO |ENSO|>0.5 Winter(DJFM) −0.66 −0.74 0.14 0.17 −0.06 −0.10 0.13 0.09 Summer(JJAS) −0.37 −0.51 0.21 0.23 0.17 0.19 0.18 0.19 Grey,regularandboldnumbersindicatevaluesatp>0.05,p<0.05andp<0.01,respectively. representingslightlymorethan4monthsintheoccurrence winter and −0.37 for summer (Figures 5(a) and (b)). of the mid-season date and more than half of the annual These 𝜌 values rise to −0.74 and −0.51 if only seasons total by the median mid-season date. Therefore, the vari- withnon-neutralNAOivaluesareconsidered.Thesecond abilityinducedby20%oftheextremeyearsisalmostequal large-scalemodewithsignificantcorrelationcoefficientsis tothatinducedbytheremaining80%(Figure4). theAMOindex(AMOi).The𝜌valuebetweentheAMOi Differencesinthemid-seasondatebetweenSP1andSP2 andtheDJFMprecipitationis0.14(p-value<0.05),rising aresignificant(Figure 4).Thesedifferencesareespecially to 0.17 (p-value<0.05) if only seasons with non-neutral accentuated in the range of the dates of the mid-season AMOi values are taken into account. The JJAS 𝜌 val- date and in the range of rainfall percentage accumulated ues obtained with the AMOi are higher, with 𝜌=0.21 bythatdate.Despitethesmallernumberofyearsincluded, and 𝜌=0.23, respectively. The PDO index (PDOi) also SP1showshigherrangesofvariabilityofthedatesofthe displays relatively small albeit significant correlations mid-season date than SP2. This range is 78days when it (p-value<0.05) but only for JJAS with 𝜌 values of 0.17 iscalculatedbetweenthe10and90%percentilesand121 and0.19ifweconsiderallthesummersoronlythesum- days if the extreme years, from the earliest to the latest, merswithnon-neutralPDOivalues,respectively.Finally, are also included. These values decrease considerably to theENSOindex(ENSOi)isthelastlarge-scalemodethat 53and113days,respectively,forSP2.Iftheaccumulated shows significant correlation (p-value<0.05) with sea- rainfall by the median mid-season date is considered, sonal precipitation in Ponta Delgada. The summer pre- SP1 displays a higher range (27.1%), without the 10% cipitationcorrelatespositivelywiththeENSOibothifwe earliest and latest years, than SP2 (22.7%). However, if considereverysummer(𝜌=0.18)andifweconsideronly weconsideralltheyears,theSP2rangeishigher(54.4%) the summers with non-neutral ENSOi values (𝜌=0.19). than the SP1 range (39.6%). Thus, the range of the dates By contrast, the EA, the SCAND and the AO indexes do of the mid-season date for SP1 presents a ratio of 1.55 notshowanysignificantcorrelationbetweenthemandthe betweenalltheyearsanddisregardsthe10%earliestand seasonalprecipitation.ThislackintheinfluenceoftheEA latest years, whereas this ratio is 2.13 for SP2. Similarly, andSCANDisamajordifferencetowhatcanbeseenover therangeofrainfallpercentageaccumulatedbythemedian the Iberian Peninsula where these two large-scale mode mid-season date for SP1 shows a ratio of 1.46 between patternsplayamajorroleinshapingtheclimate,asshown alltheyearsandrejectsthe10thandthe90thpercentiles, with low (Trigo etal., 2008) and high (Jerez and Trigo, whereas this ratio is 2.40 for SP2. This would mean and 2013)resolutionclimatedatasets.Thismaysimplyreflect confirm that the so-called ‘normal’ conditions are more the fact that the SCAND major regions of influence are unstableforSP1,whiletheextremeyearsaremorevariable locatedtoofarfromtheAzoresArchipelagoandalsothat anduncertainforSP2(Figure4). the island falls inside the area of neutral influence of the EApattern,i.e.theareawheretheinfluenceofEAinpre- cipitationswitchesfrompositivetonegative. Nevertheless, the link between the Ponta Delgada 6. Impactoflarge-scalemodesofclimatevariability precipitation and the large-scale modes has not been onrainfallregime constant through time (Figures 6(a) and (b)). The Spearman’s correlation coefficients (𝜌) between the non-stationary relationship between the NAOi and extended (DJFM and JJAS) seasonal large-scale modes environmental variables has already been pointed out and precipitation were computed with the seasonal data in the Azores Archipelago (Cropper and Hanna, 2014), of all the available years (Table S1). However, only the North Atlantic sector (García Molinos and Dono- large-scale modes displaying significant correlations hue, 2014) and continental Europe (e.g. Trigo etal., (p-value<0.05)arepresentedanddiscussed(Table2).In 2004; Vicente-Serrano and Lopez-Moreno, 2008). addition,wedefinedtheseasonalhighandlowlarge-scale Here, we also identify a non-stationary pattern on the mode composites to be constituted by all winters and precipitation–NAOi relationships. The running corre- summers, with the corresponding indices higher than lation (19-year sliding window) between the winter 0.5 and lower than −0.5, respectively, i.e. non-neutral (Figure 6(a)) and summer (Figure 6(b)) Ponta Del- conditions. gada precipitation and the NAOi illustrates a clear The NAO index (NAOi) shows the highest correlations non-stationary behaviour of the relationship. The winter of all the large-scale modes with 𝜌 values of −0.66 for NAOi–precipitation correlation shows a fairly constant ©2016RoyalMeteorologicalSociety Int.J.Climatol.(2016) A.HERNÁNDEZetal. (a) (b) (c) (d) (e) (f) Figure5.Correlationbetweenextendedwinter(DJFM)andsummer(JJAS)precipitation(mm)andNAOifortheentirestudiedperiod(aandb), SP1(candd)andSP2(eandf). behaviour until approximately 1930 before declining for observedduringthefirstdecades(1910–1940)andthelast the following 20years. The lowest values of correlation years(1997–2005).Conversely,thelargeinfluenceofthe areobservedaroundthe1950s,withaminimumof−0.37. NAOonwinterprecipitationseemstomasktheroleplayed After 1950, an increasing trend is recorded, achieving by the other large-scale modes for winter. Only prompt the maximum correlation values (−0.84) in 1975 fol- peaksofAMOinfluenceoverwinterprecipitation(around lowed by a period of weaker correlations with the values 1905,1965,1988and1995)seemtooverlapwiththewin- rising again since 2000. The temporal evolution of the terNAOimpact,whiletheimpactoftheENSOandPDO correlation coefficient between summer precipitation and onthewinterprecipitationismuchlower(Figure6(a)). NAOi is more oscillatory. These correlations are also Aspreviousworks(e.g.Trigoetal.,2008;Bladéetal., negative but with smaller values, with more restricted 2012;Comas-BruandMcDermott,2014)pointedout,the significant periods between 1886–1909 and 1941–1997. NAOisthemainlarge-scalemodecontrollingtheseasonal An increasing trend can be observed from 1886 to 1897 precipitation over the North Atlantic region. Our results beforedeclininguntil1909.Asimilarpatterncanbeseen highlightthisfact,showingthatupto50%ofthevariance from1941to1997withanincreasingtrenduntil1970and ofthewinterprecipitationcanbeexplainedbytheNAOfor a continuous weakening until the end of the record. The winterswithpositiveandnegativevalues.Insummer,the range of significant correlations is from −0.35 to −0.72 NAOroleisalsosignificantalthoughitislessprominent, and −0.35 to −0.84, respectively. Several authors (e.g. andtheroleofotherlarge-scalemodes(AMO,ENSOand Pauling etal., 2006; Comas-Bru and McDermott, 2014) PDO)shouldnotbeneglected.Theseresultsalsohighlight linked this changing nature to variations in the spatial thenon-stationarybehaviouroftheNAOandprecipitation configuration of the NAO and the increment of the role relationshipintheAzoresArchipelago,whichwasalready playedbytheotherclimaticcirculationmodes.Theroleof pointedoutforotherclimatevariables(i.e.temperature)in theothersignificantlarge-scalemodes(AMO,ENSOand previousworks(e.g.CropperandHanna,2014). PDO)forperiodsoflowerNAOimpactseemsclearforthe 6.1. TheseasonalNAOinfluence North Atlantic sector in summer. Figure 6(b) highlights thefactthatwhentheNAOimpactdoesnotachieveasta- As outlined above, the 𝜌 values for extended win- tisticallysignificantrole,otherlarge-scalemodesincrease ter and summer precipitation in Ponta Delgada and the theirinfluenceontheprecipitation.Thisbehaviourcanbe correspondingNAOiare−0.66and−0.37forthe140-year ©2016RoyalMeteorologicalSociety Int.J.Climatol.(2016)

Description:
d IPMA IP - The Portuguese Sea and Atmosphere Institute, Lisbon, Portugal . cyclone (Agostinho, 1948; Santos et al., 2004; Azevedo,. 2006). Thus
See more

The list of books you might like

Upgrade Premium
Most books are stored in the elastic cloud where traffic is expensive. For this reason, we have a limit on daily download.
DMCA & Copyright: Dear all, most of the website is community built, users are uploading hundred of books everyday, which makes really hard for us to identify copyrighted material, please contact us if you want any material removed.
Copyright @ PDFdrive.to, 2022 | We love our users