Clim. Past, 7, 527–542, 2011 www.clim-past.net/7/527/2011/ Climate doi:10.5194/cp-7-527-2011 of the Past © Author(s) 2011. CC Attribution 3.0 License.

The construction of a Central Netherlands temperature

G. van der Schrier1, A. van Ulden*, and G. J. van Oldenborgh1 1KNMI, P.O. Box 201, 3730 AE De Bilt, The Netherlands *retired Received: 27 October 2010 – Published in Clim. Past Discuss.: 10 November 2010 Revised: 15 April 2011 – Accepted: 22 April 2011 – Published: 20 May 2011

Abstract. The Central Netherlands Temperature (CNT) is a 1 Introduction monthly daily mean temperature series constructed from ho- mogenized time series from the centre of the Netherlands. In the Netherlands, the earliest temperature observations The purpose of this series is to offer a homogeneous time were made at the end of the 17th century. From 1706 on- series representative of a larger area in order to study large- wards, systematic measurements were made and a contin- scale temperature changes. It will also facilitate a compari- uous record, albeit constructed from several sources, exists son with climate models, which resolve similar scales. (Labrijn, 1945). Because of the lack of standardization in ob- From 1906 onwards, temperature measurements in the servation procedures, instruments and observations screens, Netherlands have been sufficiently standardized to construct the construction of a homogeneous record on the basis of a high-quality series. Long time series have been constructed these early instrumental records is difficult and is not at- by merging nearby stations and using the overlap to calibrate tempted here. the differences. These long time series and a few time series In 1906 a climatological network had become operational of only a few decades in length have been subjected to a ho- in the Netherlands that employed a highly standardized ob- mogeneity analysis in which significant breaks and artificial servation practice and a type of Stevenson screens at all sta- trends have been corrected. Many of the detected breaks cor- tions but one. The exception was the station De Bilt, where respond to changes in the observations that are documented this replacement happened on 17 May 1950 and where the in the station metadata. Stevenson screen replaced a large thermometer screen (called This version of the CNT, to which we attach the version “Pagoda”), which had a thermograph located at 2.20 m on number 1.1, is constructed as the unweighted average of four the peak of the Pagoda’s roof. A set of main stations stations (De Bilt, Winterswijk/Hupsel, Oudenbosch/Gilze- made observations on an hourly basis, while secondary sta- Rijen and Gemert/Volkel) with the stations Eindhoven and tions in the climatological network took measurements thrice Deelen added from 1951 and 1958 onwards, respectively. daily. Around 1950, a new synoptic network was installed. The global gridded datasets used for detecting and attribut- This was operated by the Weather Forecasting department of ing climate change are based on raw observational data. Al- KNMI in parallel to the climatological network, which was though some homogeneity adjustments are made, these are operated by the Climate Division at KNMI. This situation not based on knowledge of local circumstances but only on persisted until around 1990, when the two networks were in- statistical evidence. Despite this handicap, and the fact that tegrated to form a single, fully automated, observation net- these datasets use grid boxes that are far larger then the area work. associated with that of the Central Netherlands Tempera- The locations of the observing stations relevant for this ture, the temperature interpolated to the CNT region shows a study, both the stations that ceased operation and the opera- warming trend that is broadly consistent with the CNT trend tional ones, are shown in Fig. 1. in all of these datasets. The actual trends differ from the CNT trend up to 30 %, which highlights the need to base The aim of this study is twofold. The first is to con- future global gridded temperature datasets on homogenized struct a set of homogeneous monthly averaged records for time series. daily mean temperature at various locations spread over the Netherlands. These records are either based on long contin- uous records from the KNMI network or, when these are not Correspondence to: G. van der Schrier available, on combinations of two records from nearby sta- ([email protected]) tions to obtain time series as long as possible. Next, based on

Published by Copernicus Publications on behalf of the European Geosciences Union. 2 G. van der Schrier et al.: Central Netherlands Temperature 528 G. van der Schrier et al.: Central Netherlands temperature

time) mean temperature. This approach is a refinement of a method used in the 1980s at KNMI and aims to give a better representation of the daily cycle in temperatures. Note that only for the main stations in the climatological network, min- imum and maximum temperatures reached between consec- utive readings rather than those reached in a 24-hour period, are recorded. The approach of Van der Hoeven (1992) was to make five estimates of the daily mean temperature T24: (i) T24 = Ti −Ci(t)(Tx −Tn), i = 1,2,3,x,n (1) where subscripts 1,2,3 refer to temperature readings at 08, 14 and 19 h local time and subscripts x and n to daily maximum and minimum temperatures. The coefficients Ci(t) were sea- sonally dependant and obtained from a comparison with 24- hour temperature observations at De Bilt in the period 1961– 70. The seasonal dependence of the coefficients is introduced to account for the annual variations in the times of sunset and sunrise and are computed for the 36 decades of the year. Fi- (i) nally, the five estimates of the daily mean temperature T24 were averaged to give the best estimate: Fig. 1. Map of the Netherlands with the station locations and station types. Fig. 1. Map of the Netherlands with the station locations and station T24 = (T1 +T2 +T3 +Tx +Tn− a selection oftypes these homogeneous records, a Central Nether- different approaches in arriving at an estimate of temperature lands Temperature (CNT) record is assembled that is, by con- representative for a larger region adds to the confidence in it. (C1 +C2 +C3 +Cx +Cn)(Tx −Tn))/5. (2) struction, representative for a larger area. In a precursor of this study, van Ulden et al. (2009) based Applying this approach to the De Bilt data, gives that the an earlier version of the CNT on the same monthly aver- 2 Construction of long time series aged stationtions records but to used obtain a different time method series to homog- as long as possible. Next, based on value estimated from these five measurements agreed with enize these records and made different choices in the appli- At the secondary stations in the climatological network oper- ◦ cation of theira method selection than what of is these done here. homogeneous Differences ating records, since the early a 20th Central century, temperature Nether- readingsthe have true 24-h mean to within 0.006 C. For the coastal sta- are found inlands the construction Temperature of the reference (CNT) series, aggre- recordbeen is assembledmade at 08:00, 14:00 that and is, 19:00 by LT con- (local time) astion well Den Helder, in the north-west corner of The Nether- gation levels, window size etc. Despite these different ap- as the minimum and maximum temperatures reached in the proaches, westruction, will show that therepresentative locations and sizes of for most a largertime period area. 19:00–19:00 LT. Based on these measurements,lands, the estimate of daily averaged temperatures are ≈ 0.1 of the detected breaks in the station records are similar be- van der Hoeven (1992) made accurate estimates of daily ◦ tween the currentIn study a precursor and that of van Uldenof this et al. study,(2009) van(00:00-00:00 Ulden LT) et mean al. temperature. (2009) basedThis approach isto a re- 0.2 C too low, which is probably related to changes in the and that consequently,an earlier the differences version between of the the CNT CNT as onfinement the same of a method monthly used in the averaged 1980s at KNMI and aimsdaily to cycle of this location compared to that of the De Bilt presented in this study and the one presented by van Ulden give a better representation of the daily cycle in temperatures. et al. (2009)station are very small. records The robustness but of used the CNT a for differentNote that method only for the to main homogenize stations in the climatologicalwhich is more inland. these records and made different choices in the application of For the period up to 1970, daily averaged temperatures for Clim. Past, 7,their 527–542 method, 2011 than what is done here. Differences www.clim-past.net/7/527/2011/ are found the secondary stations are calculated following the method in the construction of the reference series, aggregation lev- described above. els, window size etc. Despite these different approaches, From 1970 until the introduction of the automated weather we will show that the locations and sizes of most of the de- systems, ten measurements a day were made at the sec- tected breaks in the station records are similar between the ondary stations: eight at three-hourly intervals plus the min- current study and that of van Ulden et al. (2009) and that imum and maximum temperatures. Twenty-four hour aver- consequently the differences between the CNT as presented ages were made by an unweighted average of these values. in this study and the one presented by van Ulden et al. (2009) The secondary stations involved in this study are indicated are very small. The robustness of the CNT for different ap- with a ‘G’ in table 1. proaches in arriving at an estimate of temperature represen- Stevenson huts were used in all stations until about 1990, tative for a larger region adds to the confidence in the CNT. with the exception of De Bilt until 1950. From around 1990 a new automated observing system was gradually in- troduced using small multi-plate thermometer screens. This 2 Construction of long records transition has negligible effect on monthly mean tempera- tures (Brandsma and van der Meulen, 2008). At the secondary stations in the climatological network oper- Most records in table 1 were complete. Den Helder and ating since the early 20th century, temperature readings were Sittard had 9 missing months, Winterswijk had 1 missing made at 08, 14 and 19 h local time as well as the minimum month, Hoorn had 6 missing months and Eindhoven missed and maximum temperatures reached in the time period 19– May and June 1952. These missing data were filled with data 19 h local time. Based on these measurements, Van der Ho- from alternative stations with a monthly adjustment to ac- even (1992) made accurate estimates of daily (00-00 h local count for the any climatological difference (see tables 1 and G. van der Schrier et al.: Central Netherlands temperature 529

Table 1. Climatological records analysed in this study. (H-records Table 2. Synoptic records based on 24 hourly observations. based on 24 hourly observations; G-records based on 5 observa- tions per day until 1970 (at 08:00, 14:00 and 19:00 h plus minimum station record id period missing data filled with and maximum), 10 observations per day from 1971 onwards (at 3- Den Helder 235 1971–Jul 1972 hourly intervals plus minimum and maximum). De Kooy 235 Aug 1972–2008 Schiphol 240 1951–2008 station record id period missing data filled with De Bilt 260 1951–2008 Soesterberg 265 1953–2007 De Bilt D001 H 1901–1970 Leeuwarden 270 1951–2008 Den Helder D002 H 1906–1970 Sep 1944–May 1945 Hoorn Deelen 275 1958–2008 Vlissingen D003 H 1906–1970 1918–1930, Excluded Eelde 280 1951–2008 1944–1945 from Hupsel 283 1990–2008 analysis Twenthe 290 1971–2008 Eelde D004 H 1946–1970 Rotterdam 344 1957–2008 Beek D005 H 1946–1970 Gilze-Rijen 350 1971–2008 Groningen D006 H 1906–1951 Eindhoven 370 1951–2008 May, Gemert Maastricht D007 H 1906–1952 June 1952 and De Kooy D009 H 1961–1970 Oudenbosch Winterswijk D020 G 1906–1990 Nov 1944, Oct 1988 De Bilt Hoorn D029 G 1906–1990 Nov 1947–Apr 1948 Den Helder Volkel 375 1953–2008 Oudenbosch D032 G 1906–1992 Beek 380 1971–2008 Gemert D033 G 1906–1990 Sittard D145 G 1906–1948 Apr–Aug 1940, Maastricht Nov 1944–Feb 1945 Gilze-Rijen D132 G 1953–1970 Twenthe D146 G 1947–1970 For the period up to 1970, daily averaged temperatures for the secondary stations are calculated following the method described above. From 1970 until the introduction of the automated weather network, minimum and maximum temperatures reached be- systems, ten measurements a day were made at the sec- tween consecutive readings, rather than those reached in a ondary stations: eight at three-hourly intervals plus the min- 24-h period, are recorded. imum and maximum temperatures. Twenty-four hour aver- The approach of Van der Hoeven (1992) was to make five ages were made by an unweighted average of these values. estimates of the daily mean temperature T24: The secondary stations involved in this study are indicated with a “G” in Table 1. (i) = − − = T24 Ti Ci(t) (Tx Tn), i 1, 2, 3, x, n (1) Stevenson huts were used in all stations until about 1990, with the exception of De Bilt between 1901–1950. From where subscripts 1,2,3 refer to temperature readings at 08:00, around 1990 onwards, a new automated observing system 14:00 and 19:00 LT and subscripts x and n to daily maximum was gradually introduced using small multi-plate thermome- and minimum temperatures. The coefficients C (t) were sea- i ter screens. This transition has had a negligible effect on sonally dependant and obtained from a comparison with 24-h monthly mean temperatures (Brandsma and van der Meulen, temperature observations at De Bilt in the period 1961–1970. 2008). The seasonal dependence of the coefficients is introduced to account for the annual variations in the times of sunset and Most records in Table 1 were complete. Den Helder and sunrise and are computed for the 36 decades of the year. Fi- Sittard had 9 missing months, Winterswijk had 1 missing month, Hoorn had 6 missing months and Eindhoven missed nally, the five estimates of the daily mean temperature T (i) 24 May and June 1952. These missing data were filled with data were averaged to give the best estimate: from alternative stations with a monthly adjustment to ac-

T24 = (T1 + T2 + T3 + Tx + Tn (2) count for any climatological differences (see Tables 1 and 2). The record from Vlissingen appeared to be too incomplete − (C1 + C2 + C3 + Cx + Cn)(Tx − Tn))/5. to be useful for this study. All records have been tested for outliers, but none were found with the exception of Eind- Applying this approach to the De Bilt data results in the value hoven which is discussed in Sect. A12. Eight long records estimated from these five measurements agreeing with the were constructed covering the period 1906–1908 by merg- ◦ true 24-h mean to within 0.006 C. For the coastal station ing the records with records from nearby stations (see Ta- Den Helder, in the north-west corner of The Netherlands, the ble 3). The older parts of these merged records were ad- ◦ estimate of daily averaged temperatures are ≈0.1 to 0.2 C justed to the recent parts using overlapping observation peri- too low, which is probably related to changes in the daily ods. The monthly adjustment factors were smoothed with a cycle of this location compared to that of the De Bilt which 5-point quasi-gaussian filter. is more inland. For the transition Winterswijk to Hupsel, the overlapping period was only 10 months, which is too short for a reliable www.clim-past.net/7/527/2011/ Clim. Past, 7, 527–542, 2011 530 G. van der Schrier et al.: Central Netherlands temperature G. van der Schrier et al.: Central Netherlands Temperature 3

stationTable 3. Long compositerecord id recordsperiod used formissing break and data trend analysis. filled with De Bilt D001 H 1901–1970 Den Helder D002station H 1906–1970 Sept. record 1944 - ids May 1945 period transitionHoorn overlap Vlissingen D003De H Bilt1906–1970 1918–1930, D001+260 1944–1945 1901–2008 JanExcluded 1971 from analysis Eelde D004Den H Helder/De1946–1970 Kooy D002+235 1906–2008 Aug 1972 1961–1970 Beek D005Groningen/Eelde H 1946–1970 D004/006+280 1906–2008 Jan 1946 1946–1951 Groningen D006 H 1906–1951 Maastricht/Beek D005/007+380 1906–2008 Jan 1946 1946–1952 Maastricht D007 H 1906–1952 Winterswijk/Hupsel D020+283 1906–2008 Jan 1991 Mar–Dec 1990 De Kooy D009 H 1961–1970 Hoorn D029+270 1906–1990 No suitable follow-up Winterswijk D020 G 1906–1990 Nov. 1944, Oct. 1988 De Bilt Oudenbosch/Gilze-Rijen D032+350 1906–2008 Jan 1993 1988–1992 Hoorn D029 G 1906–1990 Nov.1947 - April 1948 Den Helder Gemert/Volkel D033+375 1906–2008 Jan 1991 1990 Oudenbosch D032 G 1906–1992 Sittard/Beek D145+380 1906–2008 Jan 1946 1946–1948 Gemert D033 G 1906–1990 Twenthe D146+290 1946–2008 Jan 1971 Sittard D145 G 1906–1948 Apr-Aug 1940, Nov 1944-Feb 1945 Maastricht Schiphol 240 1951–2008 Gilze-Rijen D132 G 1953–1970 Soesterberg 265 1953–2008 Twenthe D146 G 1947–1970 Deelen 275 1958–2008 Table 1. ClimatologicalRotterdam records analysed in this study. 344 (H-records based 1957–2008 on 24 hourly observations; G-records based on 5 observations per day until 1970 (at 08,Eindhoven 14 and 19h plus minimum 370 and maximum), 10 1951–2008 observations per day from 1971 onwards (at 3-hourly intervals plus minimum and maximum)

2).estimate The record of the from adjustment Vlissingen factors. appeared Therefore to be we too used incom- a 10 yr 0.2 pleteoverlap to be of useful both time for this series study. with AllDeelen records to determine are tested the for ad- outliers,justments. but none The weresmoothed found, adjustment with the exception factors are of shown Eind- in 0 hovenFig. 2 which. We can is discussed see in this in figure § A12. that the Eight adjustment long records factors areare constructed all negative, covering meaning the that period the recent 1906–08 stations by merging are cooler -0.2 thethan records the older with records stations. from This nearby may stations be related (see to table an urban 3). -0.4 Theheat older island parts effect of for these stations merged like records Maastricht were and adjusted Groningen, to thewhere recent observations parts using were overlapping made in observation the centre of periods. the city Th ine the -0.6 monthlyearlier periodadjustment and at factors the airport were latersmoothed on. In with Maastricht, a 5-point the quasi-gaussianmodern station filter. at the airport is also located at a much higher -0.8 andFor exposed the transition position. Winterswijk to Hupsel the overlapping period was only 10 months, which is too short for a reliable -1 estimate of the adjustment factors. Therefore we used a 10 yr overlap of both time series with Deelen to determine the -1.2 3 Method 0 1 2 3 4 5 6 7 8 9 10 11 12 adjustments. The smoothed adjustment factors are shown in figureThe 2.approach We see taken in this to figure identify that possible the adjustment changepoints factors and are es- Fig.Fig. 2. 2.AdjustmentAdjustment factors factors used used to toblend blend series series from from operational operational alltimate negative, the size meaning of the that break the is recent based stations largely are on the cooler two-phase than stationsstations to to those those from from stations stations which which ceased ceased operation. operation. (red=Den (red = Den theregression older stations. technique This suggested may be related by Vincent to an( urban1998). heat Potential is- Helder/DeHelder/De Kooy, Kooy, green=Gemert/Volkel, green = Gemert/Volkel, blue=Groningen/Eelde, blue = Groningen/Eelde, landdiscontinuities effect for stations are detected like Maastricht on 40-yr and sliding Groningen, windows whe ofre the pink=Maastricht/Beek,pink = Maastricht/Beek, magenta=Oudenbosch/Gilze magenta = Oudenbosch/Gilze Rijen, Rijen, yel- yel- ◦ ◦ observationsdifference time were series made between in the centre the target of the series city in and the a earlie (homo-r low=Winterswijk/Hupsel,low = Winterswijk/Hupsel, black=Sittard/Beek) black = Sittard/Beek) [ C] [ C]. periodgeneous) and atreference the airport series. later The on. construction In Maastricht, of the the reference mod- ernseries station is discussed at the airport in Sect. is also4. locatedEasterling at a and much Peterson higher( and1995) exposednote that position. a windowing technique may obscure discontinuities discontinuitiesseries fails to which pass this are test, close three in time, different buthaving models a are slidin usedg to which are close in time but have a sliding window with incre- windowestimate with the increments location and of size one ofyear a potential will (at least step. partially) Every year ments of one year will (at least partially) eliminate this prob- eliminatein the record this problem. is tested In as order a potential to prevent step this with problem, the exception ho- 3lem. Method In order to prevent this problem, homogenized records mogenizedof the first records three areand put last through three years. the detection The regression algorithm is t per-o are put through the detection algorithm to detect possible in- detectformed possible on the inhomogeneities difference series which resulting were from left the undetecte subtractiond Thehomogeneities approach taken which to identify were left possible undetected changepoints in the first and run. inof the the first target run. series For and the a Eindhoven homogeneous record, reference this procedure series. estimateFor the the Eindhoven size of the record, break this is based procedure largely yielded on the a two- break- yieldedTo a test breakpoint if a series which is inhomogeneous, remained undetected a straight in the linefirst is phasepoint regression which remained technique undetected suggested in the by first Vincent run of (1998). the pro- runfitted of the to program. the data. The goodness of fit is quantified using Potentialgram. discontinuities are detected on 40-yr sliding win- theThe Durbin-Watson two-phase regression statistic, technique which applies is a test four for different the corre- dowsThe of the two-phase difference regression time series technique between applies the target four different series models.lation ofThe regression first model residuals determines (Wilks whether, 1995 the, Sect. time 6.2.6). series It andmodels. a (homogeneous) The first model reference determines series. whether The construction the time series of istests homogeneous the null-hypothesis over the tested that the interval residuals of time. are seriallyIf the time inde- theis reference homogeneous series over is discussed the tested in interval § 4. Easterling of time. If and the Pe- time seriespendent fails against to pass the this alternative test, three that different they aremodels consistent are used with terson (1995) note that a windowing technique may obscure to estimate the location and size of potential step. Every year

Clim. Past, 7, 527–542, 2011 www.clim-past.net/7/527/2011/ G. van der Schrier et al.: Central Netherlands temperature 531 a first-order autoregressive process. The threshold for the siginificance level is used as a threshold to determine if a Durbin-Watson statistic relates to the 5 % level where the break is significant or not. null-hypothesis of zero serial correlation can either be re- If a fit of a model fails to meet the significance level using jected or where this statistic is indeterminate. If adjacent the Durbin-Watson statistic, it is not considered further. residuals are of similar magnitude, as would occur in a bad A review of modern methods, including the methods used fit, the Durbin-Watson statistic tends to be small. On the here, is given by Reeves et al. (2007); they concluded that other hand, when residiuals are randomly distributed in time, the common trend two-phase regression model seems opti- this statistic tends to be large. Therefore one does not re- mal for most time series. ject the null hypothesis that the residuals are independent A hierarchy is used in determining which changepoint if the Durbin-Watson statistic is sufficiently large. Upper models Eqs. (3), (5), (7) are used to estimate step sizes. and lower bounds for the significance of the Durbin-Watson If a difference series is inhomogeneous, models Eqs. (3) statistic are calculated using the NAG routine g01epf. and (5) are applied and information from model Eq. (7) is If the difference series is judged inhomogeneous, the loca- only used after a visual confirmation that a discontinuous tion and size of the break are estimated using a simple two- trend is present. No distinction is made for information on phase regression model (Vincent, 1998) the step size from models Eqs. (3), (5), the estimate of the continuous trend from model Eq. (5) is not used to correct  µ + ε , 1 ≤ t ≤ c X = 1 t (3) for this trend. The motivation for not correcting for a con- t µ + ε , c < t ≤ n, 2 t tinuous trend is related to the construction of the reference series. Both the construction of the reference series and this where µ1, µ2 are mean values before and after the break and motivation are discussed in Sect. 4. However, discontinuous εt is the zero-mean independent random error with a constant 2 trends (which are output from model Eq. 7) are corrected for. variance σε . The time c is called a changepoint if µ1 6= µ2. The F statistic for a changepoint at time c is: It is possible to formalize the choice between the various model using a statistical test. In an attempt to do this, we (SSEred − SSEfull)/1 noticed that model Eq. (7) was chosen in more cases than Fc = , (4) SSEfull/(n − 2) what could be confirmed by the available metadata. This ob- servation made us change the procedure and adjusted both where SSE is the sum of squared errors of the “full” model full the step and the trend when the metadata indicated that these Eq. (3), which includes the break and SSE is the sum of red adjustments were required. squared errors of the “reduced” model which assumes a con- stant mean. Slightly more complex is the two-phase regression model 4 Reference time series with a common trend (Wang, 2003):  In the absence of homogeneous time series, constructing a µ1 + βt + εt , 1 ≤ t ≤ c Xt = (5) (near-)homogeneous reference series requires a special ap- µ2 + βt + εt , c < t ≤ n, proach. Instead of using a tailored approach where an av- where β is the value of the trend. The F statistic for a erage of a small number of selected time series is used as changepoint at time c is: reference series from the vicinity of the target record, we use the most dominant mode of variability from a Principal (SSEred − SSEfull)/1 Fc = . (6) Component Analysis (PCA) based on all available long time SSEfull/(n − 3) series. The first mode of variability accounts for the maxi- The fourth model allows for a combination of a discontin- mum amount of joint variability of the variance-covariance uous trend and a step (Lund and Reeves, 2002): matrix (Wilks, 1995), which is based, in its turn, on a se- lection of long station data homogeneously spread over the  µ1 + β1t + εt , 1 ≤ t ≤ c country. The principal mode of variability is a weighted av- Xt = (7) µ2 + β2t + εt , c < t ≤ n, erage of the input series and contains a large fraction of the common variability of the series. This time series will have with β1, β2 values of the trend before and after the break. the warming trend common to all time series and, due to the The F statistic for a changepoint at time c is: averaging of all available long records, inhomogeneities in the individual records are damped. However, a reference se- (SSEred − SSEfull)/2 Fc = . (8) ries constructed from time series scattered over the country SSE /(n − 4) full will not reflect any regional signal. Other considerations are Under the null hypothesis of no changepoints and assum- that the tailored approach is more labour-intensive and can ing Gaussian errors εt in models Eqs. (3), (5), (7), tables with hardly be automated, but the expectation is that a tailored the Fmax percentiles are given by Jaruskova´ (1996), Wang approach will provide a reference series capturing more of (2003) and Lund and Reeves (2002) respectively. The 95 % the month-to-month variability, thus reducing the noise in the www.clim-past.net/7/527/2011/ Clim. Past, 7, 527–542, 2011 532 G. van der Schrier et al.: Central Netherlands temperature difference series. This should make it easier to detect smaller we see that breaks. The PCA-based method is a rather straightforward N ! procedure, easily automated, but potentially less suited to ho- 1 X 1T = Aj (t) − ci Ai(t) . (12) mogenize a series which is not near the centre of the coun- 1 − c j i=1 try. The decorrelation length of the interannual variability of monthly mean temperature throughout the Netherlands One can therefore just use the total reference series and mul- varies from about 1000 km in summer to 2000 km in win- tiply corrections by 1/(1−cj ) afterwards. Alternatively, one ter, which is much larger than the size of the Netherlands, so could recalculate the reference series for each target series regional effects are not expected to be very large. by using this PCA-based technique but excluding the target This approach contrasts with that of van Ulden et al. series from this calculation. An adjustment to the corrections (2009), who use the average of nearby stations as the ref- is then not neccessary. erence series. Peterson et al. (1998) note that the construc- To test the merits of the PCA-based approach, the refer- tion of a reference series by simply averaging series from ence series used by van Ulden et al. (2009) and the PCA- surrounding stations has been done earlier by Potter (1981), based reference series are compared for a selection of sta- although he used an average of 18-stations for this. More tions. The RMS between the target series and either of the specifically, Peterson and Easterling (1994) average the three two reference series is calculated (correcting for any offset). best correlating series from the 5 nearest stations to build the It turns out that the RMSs are very similar (not shown), in- reference series. dicating that for this study involving stations that are much Input to the PCA are the time series of the sta- closer to each other than the decorrelation length, a PCA- tions De Bilt, Groningen/Eelde, Winterswijk/Hupsel, Maas- based reference series does not give higher noise-levels in tricht/Beek, Volkel/Gemert, Oudenbosch/Gilze-Rijen and the difference series, compared to a tailored approach. Den Helder/De Kooy. This excludes the station Vlissingen The different models discussed in Sect. 3 are combined in the southwestern corner of the Netherlands, which is too in the approach of this study which has merit in cases where gappy and has seen too many relocations to be allowed into the “true” regression model is unknown (Reeves et al., 2007). the reference time series. Since the reference series used in this study only holds infor- Over the more recent period, from the 1950s onward, more mation associated with country-wide spatial scales and is not stations have become available to construct a reference se- specifically pin-pointed at a certain region, we expect that ries. In order not to introduce inhomogeneities in the ref- difference series may have a continuous trend throughout the erence series, we did not include these stations in the refer- record. This makes the use of model Eq. (5), which includes ence series. Moreover, the seven stations used for the refer- a step and a continuous trend, particularly suited to the ap- ence series over the first part of the 20th century are scattered proach. around the country and should be sufficiently able to pick up The principal mode explains 96.7 % of the variability and on large-scale variations of temperature. includes the warming trend. The dominant mode of vari- When using a weighted sum of series as a reference series, ability is a weighted average of the input series, with the a correction has to be made when one of the series which is weights shown in Table 4. The weights for the various sta- input to the PCA analysis is adjusted. The reference series is tions are very similar, the relative difference between the ex- written as tremes ((maximum−minimum)/maximum) is only 0.16. The N largest weights are found in Maastricht/Beek and Winter- X ci Ai(t), (9) swijk/Hupsel, the lowest is found in Den Helder/De Kooy, i=1 located at the North Sea coast. where ci is the weight associated with the i-th series and Ai(t) is the i-th series at time t. The sum of the weights ci 5 Quality check equals 1. When the adjustments to the j-th series itself needs to be calculated, a reference series excluding the j-th series In order to assess the quality of the various records used in would be required. The difference from which an adjustment this study, a running standard deviation of the difference of is computed is: the annual average of each series with the reference series N is shown over 41 yr sliding windows (Fig. 3). The stan- 1 X 1T = Aj (t) − ci Ai(t). (10) dard deviations vary considerably with time. Fig. 3a shows 1 − c j i=1, i6=j that Sittard has a maximum in the first few decades of the record, which is in part related to a warm bias (not shown). By writing Eq. (9) as Gemert has a very pronounced peak around 1950, which is N N ! related to a very significant break in that period (discussed X X ci Ai(t) = ci Ai(t) + cj Aj (t), (11) in Sect. A6). The running standard deviation for the series i=1 i=1, i6=j composed of Oudenbosch and Gilze Rijen has a maximum

Clim. Past, 7, 527–542, 2011 www.clim-past.net/7/527/2011/ G. van der Schrier et al.: Central Netherlands temperatureG. van der Schrier et al.: Central Netherlands Temperature 533 7

Table 4. Loading of the seven long records on the first Principal 0.3 The long records from the coastal station Den Helder/De De Bilt Kooy, the series from Groningen/Eelde and Leeuwarden in Component. Den Helder/De Kooy Maastricht/Beek the north and Maastricht/Beek and Sittard/Beek in the south 0.25 Gemert/Volkel of the Netherlands have not been included since they are too station loading Sittard/Beek Winterswijk/Hupsel far at the outer extremes of The Netherlands and are therefore De Bilt 0.143 0.2 Oudenbosch/Gilze Rijen less representative of the central Netherlands. The principal Den Helder/De Kooy 0.128 Groningen/Eelde motivation not to include the records from the airports of Rot- Groningen/Eelde 0.144 terdam and Amsterdam (Schiphol) is that these stations are Maastricht/Beek 0.150 0.15 relatively close to the sea and that they may be influenced by Winterswijk/Hupsel 0.149 Oudenbosch/Gilze-Rijen 0.141 the large cities in their vicinity and, in the case of Schiphol, Gemert/Volkel 0.145 0.1 the rapid development of the airport itself. Other long records in The Netherlands have either been a 0.05 discontinued (Hoorn, Soesterberg) or are too gappy (Vlissin- gen) to be included in the CNT. The running standard devia- 1920 1930 1940 1950 1960 1970 1980 1990 in the late 1960s which is related to a break in that period tions discussed in § 5 and fig. 3 indicate that the records from (discussed in Sect. A4). 0.2 Twenthe and Sittard are too noisy to be included. Deelen Although the CNT is constructed from stations roughly De Bilt, which is situated in the centre of the Netherlands, Eindhoven has low standard deviations for the whole observation period. 0.18 Rotterdam in the area centred in the central-south east of the Nether- Schiphol lands, a correlation analysis between winter (DJF) and sum- This relates to the fact that the reference series reflects cli- Soesterberg Twenthe mer (JJA) averages of the CNT and similar averages of the matic conditions of the central part of the Netherlands best. 0.16 Figure 3b shows that Soesterberg, despite its central loca- E-OBS gridded dataset based on daily averaged temperature tion, and Twenthe both show high noise levels. from surrounding stations from the European Climate As- 0.14 sessment & Dataset (Haylock et al., 2008) (fig. 4a, b), shows Around 1990, the noise levels of all stations (except for that the CNT is representative of a much larger area. Corre- Beek) are significantly reduced (not shown). This is prob- lations remain high for stations in the Netherlands, including ably related to the transition to an automated network and 0.12 those in the north and along the coast and high correlations improved observation practices. are found in Belgium and Germany as well. Monthly mean 0.1 values show smaller correlations, especially in summer (not b shown). 6 The Central Netherlands temperature 1965 1970 1975 1980 1985 1990

6.1 Definition 6.2 Comparison with an earlier version Fig.Fig. 3. Running 3. Running standard standard deviations deviations over over 41 yr 41 windows yr windows of annual of annual averages of the difference between target series and reference series. The Central Netherlands Temperature (CNT) record is averages of the difference between target series and reference series. Preceding this study, van Ulden et al. (2009) constructed UpperUpper panel panel shows shows the longthe long records, records, lower lower panel panel shows shows the shorter the shorter based on homogenized monthly means of daily averaged an earlier version of the Central Netherlands Temperature records.records. temperatures from a selection of series from the cen- record. This earlier version, to which we attach the version tral part of The Netherlands. These series are from number 1.0, is based on the same selection of series as the De Bilt, Winterswijk/Hupsel, Oudenbosch/Gilze-Rijen and 6relatively The Central close Netherlands to the sea and Temperature that they may be influenced by current version 1.1. However, the homogenisation procedure Gemert/Volkel. The record from Eindhoven is included from the large cities in their vicinity and, in the case of Schiphol, between the van Ulden et al. (2009) study and the current 1951 onwards and Deelen is included from 1958 onwards. 6.1the Definition rapid development of the airport itself. study is different. These differences relate to the construc- The CNT is a simple unweighted average of these records. Other long records in The Netherlands have either been tion of the reference series (see § 4) but also to the break and Monthly adjustments were applied to the CNT prior to the Thediscontinued Central Netherlands (Hoorn, TemperatureSoesterberg) (CNT) or are too record gappy is based (Vlissin- spurious trends detection algorithms. For the construction of inclusion of the Eindhoven record in 1951 and to the CNT on homogenizedgen) to be included monthly in the means CNT. of The daily running averaged standard temper- devia- CNT1.0, homogeneity tests based on Easterling and Peterson record from 1951 to the inclusion of Deelen in 1958 to ac- aturestions from discussed a selection in Sect. of5 seriesand Fig. from3 indicate the central that the part records of (1995) for the detection of breaks were used and a method count for the transition from 4 to 5 to 6 stations. These adjust- Thefrom Netherlands. Twenthe and These Sittard series are aretoo noisyfrom Deto be Bilt, included. Winter- based on that of Alexandersson and Moberg (1997) was used ments are calculated over the 1961–2008 period, smoothed swijk/Hupsel,Although Oudenbosch/Gilze-Rijen the CNT is constructed and from Gemert/Volkel stations roughly. to detect spurious trends. In contrast to Easterling and Peter- by a 5-point Gaussian filter, similar to the adjustments in Thecentred record infrom the Eindhoven central Southeast is included of the from Netherlands, 1951 onwards a cor- son (1995), van Ulden et al. (2009) used moving windows, Sect. 2 and are small at O(0.01 ◦C). andrelation Deelen analysisis included between from 1958 winter onwards. (DJF) and The summer CNT is (JJA) a both for breaks and trends. In both tests, the critical signifi- The long records from the coastal station Den Helder/De simpleaverages unweighted of the average CNT and of similar these records. averages Monthly of the E-OBS ad- cance levels derived from Alexandersson and Moberg (1997) Kooy, the series from Groningen/Eelde and Leeuwarden in justmentsgridded were dataset applied based to theon daily CNT prioraveraged to the temperature inclusion of from were used. the north and Maastricht/Beek and Sittard/Beek in the south thesurrounding Eindhoven record stations in 1951 from and the to European the CNT Climate record from Assess- Because of the differences between the current approach of the Netherlands have not been included since they are too 1951ment to the& Dataset inclusion (Haylock of Deelen et al. in, 2008 1958) to(Fig. account4a,b) showsfor the that and that of van Ulden et al. (2009), differences in the ho- far at the outer extremes of The Netherlands and are therefore transitionthe CNT from is representative 4 to 5 to 6 stations. of a much These larger adjustments area. The are corre- mogenized versions of the underlying station data can be ex- less representative of the central Netherlands. The principal calculatedlations overremain the high 1961-2008 for stations period, in the smoothed Netherlands, by a 5-point including pected. Fig. 5 shows the adjustments made to the records Gaussian filter, similar to the adjustments in section 2 and are in comparison with the adjustments made in version 1.0. motivation not to include the records from the airports of Rot- those in the◦ North and along the coast; high correlations are terdam and Amsterdam (Schiphol) is that these stations are smallfound at O in(0.01 BelgiumC). and Germany as well. The monthly mean Below we will argue that the differences between CNT1.1 www.clim-past.net/7/527/2011/ Clim. Past, 7, 527–542, 2011 8 G. van der Schrier et al.: Central Netherlands Temperature

and CNT1.0 are very modest, despite the differences in ap- proaches, which adds to the robustness of the CNT. Figure 6 shows the RMS between CNT1.1 and CNT1.0 as a function of the month. The RMS has been determined over the timeperiod 1906-2008. This figure shows that the RMS 534is near 0.04G. van◦C, der except Schrier for late et al.:spring Central and for Netherlands July to Septem- temperature ber. In the first period, the RMS rises to nearly 0.04◦C, in ◦ the second(1995) to for approximately the detection 0.07 of breaksC. The wererise in used April-May and a method can bebased attributed on that to of differencesAlexandersson in the homogenisation and Moberg (1997 of th)e was used Eindhoven record (fig. 7). The current study has an addi- to detect spurious trends. In contrast to Easterling and Peter- tional correction for a break in 1958. The rise in the pe- riodson July(1995 to September), van Ulden is mainly et al. attributed(2009) used to the moving Winter- windows, swijk/Hupsel,both for breaks Deelen and and trends. Gemert/Volkel In both series tests, (fig.the critical 7). In signifi- the homogenisationcance levels derived of the from first recordAlexandersson (§ A5) a andsmall Moberg break (1997) in 1960were is used. detected, but we have not corrected for this break due toBecause the absence of of the metadata differences related between to this the break. current How- approach ever,and van that Ulden of etvan al. Ulden (2009) et do al. correct(2009 for), differences this break and in the ho- the largestmogenized amplitude versions of the of correctionthe underlying can be station found data in the can be ex- monthspected. June-September. Figure 5 shows In the the Deelen adjustments record, breaks made are to de-the records tectedin which comparison are corrected with for the in adjustments the current study, made but in not version in 1.0. the seriesBelow used we to argue construct that CNT1.0. the differences Finally, the between adjustmen CNT1.1ts and to theCNT1.0 Gemert/Volkel are very record modest, are despite slightlythe different difference in the in cu approach,r- rentwhich study compared adds to the to that robustness of van Ulden of the et CNT. al. (2009). Regarding the trends, table 5 shows that the differences Figure 6 shows the RMS between CNT1.1 and CNT1.0 as between CNT1.0 and CNT1.1 are very small. a function of the month. The RMS has been determined over the time period 1906–2008. This figure shows that the RMS ◦ 7 Comparisonis near 0.04 ofC, the except CNT for v1.1 late withspring the and HadCRUT3, for July to Septem- ◦ GISTEMPber. In the and first NCDC period, datasets the RMS rises to nearly 0.04 C, in the second to approximately 0.07 ◦C. The rise in April– In Fig.May 8 can the be annual attributed mean to CNT1.1 differences time in series thehomogenisation is com- paredof with the Eindhovenother series recordthat are (Fig. frequently7). The used current to estimate study has an the temperatureadditional correction changes in for the aNetherlands. break in 1958. These The are the rise in the CNT1.0,period the July observed to September De Bilt temperature, is mainly attributed and the inter- to the Win- polatedterswijk/Hupsel, temperature of Deelen datasets and used Gemert/Volkel to construct estimate seriess (Fig. 7). of the global mean temperature (CRUTEM3, NOAA/NCDC In the homogenisation of the first record (Sect. A5), a small and NASA/GISS) at the mean of the co-ordinates of the six break in 1960 is detected, but we have not corrected for this CNT stations (using land temperatures only). Bybreak eye these due six to thetime absence series look of very metadata similar. related Two series to it. How- withever, obviousvan errors Ulden are et given al. ( in2009 Figs) 8d,h: do correct the unadjusted for this De break and Biltthe temperature largest amplitude from GHCN of v2 the and correction the value can correspond- be found in the ing tomonths the CNT June–September. in the GISTEMP 250 In thekm dataset, Deelen which record, is not breaks are useddetected to construct which an estimate are corrected of the global for in mean the current temperature study. but not Fig. 4. Correlation of the interannual fluctuations of the CNT series Fig. 4. Correlation of the interannual fluctuations of the CNT series Thein GISTEMP the series datasets used to use construct the GHCN CNT1.0. De Bilt Finally, time series the adjust- withwith the E-OBS v3 v3 (Haylock (Haylock et et al., al. 2008), 2008 temperature) temperature analysis analysis for for three winter months (December, January and February) and three withments an inhomogeneity to the Gemert/Volkel in 1950 of about record 1.5 areK. This slightly inhomo- different in three winter months (December, January and February) and three summer months (June, July and August) over 1950–09. The trend geneitythe iscurrent caused study by a compared combination to that of the of inhomogeneitiesvan Ulden et al. (2009). summer months (June, July and August) over 1950–2009. The trend was removed by taking year-on-year differences. aroundRegarding 1950 discussed the intrends, §A1 and Table a change5 shows in observational that the differences was removed by taking year-on-year differences. practice.between Prior CNT1.0 to 1950, andthree CNT1.1 daily observations are very small. and the min- imum and maximum temperatures were recorded next to the use of a thermograph at the main stations. From the ther- values show smaller correlations, especially in summer (not mograph7 Comparison and the five daily of the observations, CNT v1.1 hourly with the estimates HadCRUT3, of the temperature at De Bilt (and the other main stations) were shown). GISTEMP and NCDC datasets made. These hourly estimates are used in the current study. However, only the three daily measurement (excluding the 6.2 Comparison with an earlier version In Fig. 8, the annual mean CNT1.1 time series is com- minimum and maximum temperatures) were communicated pared with other series that are frequently used to estimate Preceding this study, van Ulden et al. (2009) constructed the temperature changes in the Netherlands. These are the an earlier version of the Central Netherlands Temperature CNT1.0, the observed De Bilt temperature, and the inter- record. This earlier version, to which we attach the version polated temperature of datasets used to construct estimates number 1.0, is based on the same selection of series as the of the global mean temperature (CRUTEM3, NOAA/NCDC current version 1.1. However, the homogenisation procedure and NASA/GISS) at the mean of the co-ordinates of the six between the van Ulden et al. (2009) study and the current CNT stations (using land temperatures only). study is different. These differences relate to the construction By eye these six time series look very similar. Two series of the reference series (see Sect. 4) and also to the break and with obvious errors are given in Fig. 8d,h: the unadjusted De spurious trends detection algorithms. For the construction of Bilt temperature from GHCN v2 and the value correspond- CNT1.0, homogeneity tests based on Easterling and Peterson ing to the CNT in the GISTEMP 250 km dataset, which is

Clim. Past, 7, 527–542, 2011 www.clim-past.net/7/527/2011/ G. van der Schrier et al.: Central Netherlands Temperature 9 G. van der Schrier et al.: Central Netherlands temperature 535

0.4 De Bilt 1950 0.4 Winterswijk/Hupsel 1940 De Bilt 1968 Winterswijk/Hupsel 1950 De Bilt 1950 (WR) Winterswijk/Hupsel 1940 (WR) 0.3 De Bilt 1976 (WR) 0.3 Winterswijk/Hupsel 1950 (WR)

0.2 0.2

0.1 0.1

0 0

-0.1 -0.1

-0.2 -0.2

-0.3 -0.3 a b -0.4 -0.4 Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec

0.5 0.8 Gemert/Volkel 1950 Oudenbosch/Gilze Rijen 1966 Gemert/Volkel 1950 trend 0.4 Oudenbosch/Gilze Rijen 1984 Gemert/Volkel 1949 (WR) Oudenbosch/Gilze Rijen 1966 (WR) 0.6 Gemert/Volkel trend (WR) Oudenbosch/Gilze Rijen 1984 (WR) 0.3 0.4 0.2 0.2 0.1 0 0

-0.2 -0.1

-0.4 -0.2

-0.6 -0.3 c d -0.8 -0.4 Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec

0.35 0.5 Maastricht/Beek 1931 Eindhoven 1958 Maastricht/Beek 1920 (WR) 0.4 Eindhoven 1969 0.3 Maastricht/Beek 1993 (WR) Eindhoven 1985 Eindhoven 1969 (WR) 0.3 Eindhoven 1988 (WR) 0.25 0.2

0.2 0.1

0.15 0

-0.1 0.1 -0.2 0.05 e -0.3 f 0 -0.4 Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec

Fig. 5. Break and and trend trend corrections corrections for for various various stations stations for for this this study study and and those thoseof vanof van Ulden Ulden et al. et (2009), al. (2009 the),latter the latter marked marked with with (WR). (WR). Shown are corrections for for Winterswijk/Hupsel Winterswijk/Hupsel (a),(a) Oudenbosch/Gilze, Oudenbosch/Gilze Rijen Rijen (b),(b) Gemert/Volkel, Gemert/Volkel (c) and(c) and Eindhoven Eindhoven (d),(d) Maastricht/Beek, Maastricht/Beek (e) (e) andand Schiphol (f).(f).

1906–08 1950–08 1975–08 not used toCNT1.0 constructCNT1.1 an estimateCNT1.0 of the globalCNT1.1 meanCNT1.0 temper- CNT1.1main stations) were made. These hourly estimates are used ature.DJF The0.111 GISTEMP0.114 datasets0.334 use the GHCN0.332 De0.544 Bilt time 0.540in the current study. However, only the three daily measure- seriesMAM with0.126 an inhomogeneity0.130 0.308 in 1950 of0.314 about 1.50.655 K. This 0.662ment (excluding the minimum and maximum temperatures) inhomogeneityJJA 0.154 is caused0.148 by a0.241 combination0.240 of the0.381 inhomo- 0.391were communicated to international databases. After 1950, geneitiesSON around0.127 19500.123 discussed0.149 in Sect.0.149 A1 and0.284 a change 0.27724 hourly observations were communicated. Daily tempera- in observational practice. Prior to 1950, three daily obser- tures in the GHCN De Bilt record prior to 1950 simply aver- Table 5. Trends (◦C/10 yr) in CNT1.0 and CNT1.1 for winter (DJF), spring (MAM), summer (JJA) and autumn (SON), calculated over the vations and the minimum and maximum temperatures were ages the three measurement without application of the esti- periods 1906–08, 1950-08 and 1975-08. recorded next to the use of a thermograph at the main sta- mate of the daily mean temperature of Eq. (2). This inhomo- tions. From the thermograph and the five daily observations, geneity is reflected in the GISTEMP 250 km dataset, which hourly estimates of the temperature at De Bilt (and the other uses only stations within a 250 km radius of each grid box. www.clim-past.net/7/527/2011/ Clim. Past, 7, 527–542, 2011 10 G. van der Schrier et al.: Central Netherlands Temperature

0.08 1906–2010 1975–2010 CNT v1.1 0.013±0.002 0.039±0.011 0.07 CNT v1.0 0.013±0.002 0.038±0.011 536 G. van der Schrier et al.: Central Netherlands temperature De Bilt observed 0.013±0.002 0.049±0.011 0.06 De Bilt GHCN v2 -0.001±0.003 0.049±0.011 CRUTEM3 0.009±0.002 0.036±0.010 ◦ ◦ −1 Table 5. Trends ( C/10 yr) in CNT1.0 and CNT1.1 for winter Table 0.05 6. Trends in C yr of the time series shown in Fig. 8 over NOAA/NCDC 0.010±0.002 0.052±0.011 (DJF), spring (MAM), summer (JJA) and autumn (SON), calculated the periods 1906–2010 and 1975–2010. NASA/GISS 0.011±0.002 0.038±0.010 over the periods 1906–1908, 1950–1908 and 1975–1908. 0.04 NASA/GISS 250 km 0.013±0.002 0.042±0.011 1906–2010 1975–2010 1906–1908 1950–1908 1975–1908 0.03 Table 6. Trends in ◦C/yr of the time series shown in Fig. 8 over the CNT1.0 CNT1.1 CNT1.0 CNT1.1 CNT1.0 CNT1.1 CNT v1.1 0.013 ± 0.002 0.039 ± 0.011 0.02CNT v1.0 0.013 ± 0.002 0.038 ± 0.011 periods 1906–2010 and 1975–2010. DJF 0.111 0.114 0.334 0.332 0.544 0.540 MAM 0.126 0.130 0.308 0.314 0.655 0.662 De Bilt observed 0.013 ± 0.002 0.049 ± 0.011 0.01 JJA 0.154 0.148 0.241 0.240 0.381 0.391 De Bilt GHCN v2 −0.001 ± 0.003 0.049 ± 0.011 SON 0.127 0.123 0.149 0.149 0.284 0.277 CRUTEM3 0.009 ± 0.002 0.036 ± 0.010 0 1946 and overlapped in the periods 1946-1951 and 1946- 10 G. van derNOAA/NCDCJan SchrierFeb etMar al.:Apr CentralMay 0.010 NetherlandsJun ±Jul0.002Aug TemperatureSep 0.052Oct± 0.011Nov Dec 1952 respectively. Furthermore, the relocations and change NASA/GISS 0.011 ± 0.002 0.038 ± 0.010 ± ± of thermometer screen at the De Bilt site around this time 0.08 Fig. 6. RootNASA/GISS Mean Square 2501906–2010 km error of 0.013 the difference 1975–20100.002 between 0.042 CNT1.10.011 ± ± (discussed in §A1) may have added to this break. andCNT CNT1.0. v1.1 0.013 0.002 0.039 0.011 0.07 CNT v1.0 0.013±0.002 0.038±0.011 In Table 6 linear trends over 1906–2010 and 1975–2010 De Bilt observed 0.013±0.002 0.049±0.011 are shown for all the series of Fig. 8. The De Bilt observed 0.06 ± ± De Bilt 0.3 GHCN v2 -0.001 0.003 0.049 0.011 temperature is seen to have a larger trend over the last 36 CRUTEM3 0.009±0.002 0.036±0.010 De Bilt 0.05 Deelen years than the CNT, the difference is due to the inhomogene- NOAA/NCDC 0.010±0.002 0.052±0.011Eindhoven NASA/GISS 0.25 0.011±0.002 0.038±0.010Gemert/Volkel ity in 1968/69 (discussed in §A1). The unadjusted GHCN 0.04 ± Oudenbosch/Gilze± Rijen NASA/GISS 250 km 0.013 0.002 0.042Winterswijk/Hupsel0.011 v2 time series shows no trend over 1906–2010 due to the 0.03 0.2 ◦ 1.5 K inhomogeneity in 1950. Although the inhomogeneity Table 6. Trends in C/yr of the time series shown in Fig. 8 over the is retained in the NASA/GISS 250 km dataset, the trend is 0.02 periods 1906–2010 and 1975–2010. 0.15 not affected due to the trend adjustment used (Hansen et al., 0.01 2010). Finally, it is unknown why the NCDC dataset shows a higher trend over 1975–2010. This may be due to the west- 0 1946 and 0.1 overlapped in the periods 1946-1951 and 1946- Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec 1952 respectively. Furthermore, the relocations and change ern grid box (50–55◦N, 0–5◦E), which includes stations in of thermometer screen at the De Bilt site around this time Belgium and East Anglia (UK). Fig.Fig. 6. 6.RootRoot Mean Mean Square Square error error of the of differencethe difference between between CNT1.1 CNT1.1(discussed 0.05 in §A1) may have added to this break. and CNT1.0. and CNT1.0. In Table 6 linear trends over 1906–2010 and 1975–2010 are shown 0 for all the series of Fig. 8. The De Bilt observed 0.3 Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec 8 Conclusions De Bilt temperature is seen to have a larger trend over the last 36 However, in the 1200 km dataset used for globalDeelen temperatureyears than the CNT, the difference is due to the inhomogene- estimates, the inhomogeneity is not visibleEindhoven due to the large Climate models compute meteorological variables at a typ- 0.25 Gemert/Volkel Fig.ity in 7.Fig. 1968/69Root 7. Root Mean (discussed Mean Square Square in error §A1). error of the ofThe thedifference unadjusted difference between betweenGHCN the the ho- ho- number of stations used. Oudenbosch/Gilze Rijen mogenizedmogenized stations stations records records which which construct construct CNT1.1 CNT1.1 and and CNT1.0. CNT1.0. ical scale of 100 km. Local effects caused by vegetation, Winterswijk/Hupsel v2 time series shows no trend over 1906–2010 due to the The 0.2 CRUTEM3 data shows a discontinuity when com-1.5 K inhomogeneity in 1950. Although the inhomogeneity small lakes or small variations in altitude, are not resolved by pared with the CNT1.1 in 1951 (van Ulden, 2008). Thisis retained in the NASA/GISS 250 km dataset, the trend is the models. In order to make a sensible comparison between the western grid box (50–55◦ N, 0–5◦ E), which includes sta- break 0.15 is probably partially related to the use of the Gronin-tonot international affected due to databases. the trend adjustment Post 1950, used 24 hourly (Hansen observati et al., ons model output and observations, the latter need to be defined gen and Eelde records as one continuous record in the CRUwere2010).tions communicated. Finally, in Belgium it is unknown and Daily East why temperatures Anglia the NCDC (UK). dataset in the shows GHCN De at a spatial scale similar to the model results. The Central data without corrections made for the relocation as done ina higher trend over 1975–2010. This may be due to the west- 0.1 Bilt record prior to 1950 simply averages the three measure- Netherlands Temperature record (CNT) has been designed to ◦ ◦ Sect. 2 and Fig. 2. Similarly, the Maastricht-Beek transi-mentern grid without box (50–55 applicationN, 0–5 ofE), the which estimate includes of thestations daily in mean meet this demand. Additionally, the CNT is expected to be tion is uncorrected for. Both these transitions occurred inBelgium8 Conclusionsand East Anglia (UK). 0.05 temperature of Eq. 2. This inhomogeneity is reflected in the of interest for climate research, being based on high-quality January 1946 and overlapped in the periods 1946–1951 and GISTEMP 250 km dataset, which uses only stations within homogenized records and representative for a larger area. 1946–1952 respectively. Furthermore, the relocations and Climate models compute meteorological variables at a typ- 0 a 250 km radius of each grid box. However, in the 1200 km change of thermometer screen at the De Bilt site around this8 Conclusionsical scale of 100 km. Local effects caused by vegetation, The CNT is based on a selection of homogeneous monthly Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec dataset that is used for global temperature estimates, the in- time (discussed in Sect. A1) may have added to this break. small lakes or small variations in altitude are not resolved by averaged records for daily mean temperature from the KNMI Fig. 7.InRoot Table Mean6 linear Square trends error of over the difference1906–2010 between and 1975–2010 the ho- homogeneityClimatethe models models. is compute not In order visible meteorological to due make to a the sensible large variables number comparison at a of typ- station betweens network. Long records have been constructed by blending mogenizedare shown stations for records all the which series construct in Fig. CNT1.18. The and CNT1.0. De Bilt ob-used.ical scalemodel of output 100 km. and Local observations, effects caused the latter byneed vegetation, to be defined data from nearby station, using the overlap period to calibrate served temperature is seen to have a larger trend over thesmallTheat lakes aCRUTEM3 spatial or small scale variations data similar shows in to altitude, the a discontinuity model are not results. resolve when Thed by com-Central the differences. This resulted in nine records which start in last 36 years than the CNT, the difference is due to the inho-paredthe models.Netherlands with In the order CNT1.1 Temperature to make in a 1951sensible record (van (CNT) comparison Ulden, has been between2008). designed This to the early 1900s. Using seven of these records in a Princi- tomogeneity international in databases. 1968/1969 Post (discussed 1950, 24 in hourly Sect. observati A1). Theons unad-breakmodelmeet isoutput probably this and demand. observations, partially Additionally, related the latter to the theneed CNT use to be ofis definedexpected the Gronin- to be pal Component Analysis, a weighted average of these seven were communicated. Daily temperatures in the GHCN De justed GHCN v2 time series shows no trend over 1906–2010genat a andspatialof interest Eelde scale recordsfor similar climate as to onethe research, model continuous being results. based record The on Central in high-quality the CRU series is obtained which contains a large fraction of the com- Bilt record prior to 1950 simply averages the three measure- Netherlands Temperature record (CNT) has been designed to due to the 1.5 K inhomogeneity in 1950. Although the inho-datahomogenized without corrections records made and representative for the relocation for a larger as done area. in mon variability of the series. This time series contains the ment without application of the estimate of the daily mean meet this demand. Additionally, the CNT is expected to be mogeneity is retained in the NASA/GISS 250 km dataset, the The CNT is based on a selection of homogeneous monthly temperature of Eq. 2. This inhomogeneity is reflected in the §of 2 interest and Fig. for 2.climate Similarly, research, the being Maastricht-Beek based on high-quali transitionty is warming trend common to all time series and, due to the av- GISTEMPtrend is not 250 affected km dataset, due to which the trend uses adjustment only stations used within (Hansenuncorrectedhomogenizedaveraged for. records records Both and for these representative daily transitions mean temperature for aoccurred larger from area. in theJanuar KNMIy eraging of all available long records, inhomogeneities in the et al., 2010). Finally, it is unknown why the NCDC dataset network. Long records have been constructed by blend- a 250 km radius of each grid box. However, in the 1200 km The CNT is based on a selection of homogeneous monthly datasetshows that a higher is used trend for global over 1975–2010.temperature estimates, This may the be in- due toaverageding recordsdata from for daily nearby mean stations, temperature using from the overlapthe KNMI period to homogeneity is not visible due to the large number of stations network. Long records have been constructed by blending used.Clim. Past, 7, 527–542, 2011data from nearby station, using the overlap www.clim-past.net/7/527/2011/ period to calibrate The CRUTEM3 data shows a discontinuity when com- the differences. This resulted in nine records which start in pared with the CNT1.1 in 1951 (van Ulden, 2008). This the early 1900s. Using seven of these records in a Princi- break is probably partially related to the use of the Gronin- pal Component Analysis, a weighted average of these seven gen and Eelde records as one continuous record in the CRU series is obtained which contains a large fraction of the com- data without corrections made for the relocation as done in mon variability of the series. This time series contains the § 2 and Fig. 2. Similarly, the Maastricht-Beek transition is warming trend common to all time series and, due to the av- uncorrected for. Both these transitions occurred in January eraging of all available long records, inhomogeneities in the G. van der Schrier et al.: Central Netherlands Temperature 11

G. van der Schrier et al.: Central Netherlands temperature 537

Jan-Dec temperature CNT v1.1 Jan-Dec Index CRUTEM3 T2m anom 5.6E 51.8N 11.5 2 11 1.5 10.5 1 10 0.5 9.5 0 9 CNT [Celsius] 8.5 temp [Celsius] -0.5 8 -1 7.5 -1.5 1900 1920 1940 1960 1980 2000 2020 a 1900 1920 1940 1960 1980 2000 2020 e Jan-Dec temperature CNT Jan-Dec Index NCDC T2m anom 5.6E 51.8N 11.5 2 11 1.5 10.5 1 10 0.5 9.5 0 9 CNT [Celsius] 8.5 temp [Celsius] -0.5 8 -1 7.5 -1.5 1900 1920 1940 1960 1980 2000 2020 b 1900 1920 1940 1960 1980 2000 2020 f Jan-Dec Tdebilt Jan-Dec Index GISS 1200 T2m anom 5.6E 51.8N 11.5 2 11 1.5 10.5 1 10 0.5 9.5 0 9 tair [Celsius] 8.5 -0.5

8 tempanomaly [Celsius] -1 7.5 -1.5 1900 1920 1940 1960 1980 2000 2020 c 1900 1920 1940 1960 1980 2000 2020 g

Jan-Dec GHCN mean temperature Jan-Dec Index GISS 250 T2m anom 5.6E 51.8N 11.5 2.5 11 2 10.5 1.5 10 1 9.5 0.5 9 0

temp [Celsius] 8.5 -0.5

8 tempanomaly [Celsius] -1 7.5 -1.5 1900 1920 1940 1960 1980 2000 2020 d 1900 1920 1940 1960 1980 2000 2020 h

Fig. 8. Various measures of the temperature in the Central Netherlands. (a) CNT1.1, (b) CNT1.0, (c) observed De Bilt temperature. (e-g) The Fig. 8. Various◦ measures◦ of the temperature in the Central Netherlands. (a) CNT1.1, (b) CNT1.0, (c) observed De Bilt temperature. (e–g)pointThe 51.8 pointN, 5.6 51.8E◦ bilinearlyN, 5.6◦ E interpolated bilinearly ininterpolated land temperature in land datasets temperature used to datasets construct used globa tol mean construct temperature global estimates. mean temperature (e) CRUTEM3, estimates. (f) NOAA/NCDC, (g) NASA/GISS. For reference the GHCN v2 unadjusted series for De Bilt is shown in (d) and the 250 km GISTEMP (e) CRUTEM3, (f) NOAA/NCDC, (g) NASA/GISS. For reference, the GHCN v2 unadjusted series for De Bilt is shown in (d) and the dataset in (h). The green lines denote 10-yr running means. 250 km GISTEMP dataset in (h). The green lines denote 10-yr running means.

calibrate the differences. This resulted in nine records which the individual records are damped. The weighted average is start in the early 1900s. Using seven of these records in a used as reference series to homogenize the available records. Principal Component Analysis, a weighted average of these Based on an assessment of the noise levels of each differ- seven series is obtained which contains a large fraction of the ence record, the location of the record and whether or not common variability of the series. This time series contains the station is still operational, a selection of four records is the warming trend common to all time series and, due to the made which span the period from 1906 onwards. Two ad- averaging of all available long records, inhomogeneities in ditional records are included from 1951 and 1958 onwards. www.clim-past.net/7/527/2011/ Clim. Past, 7, 527–542, 2011 538 G. van der Schrier et al.: Central Netherlands temperature

The CNT is constructed as an unweighted average of these There is no information from the metadata which could ex- records. plain the breaks near 1918–1920 and 1976. The break at The CNT series presented in this study,to which we at- 1950 is clearly related with a string of modifications which tach the version number 1.1, shows only small differences happened in the early 1950s. On 17 May 1950, the large to an earlier version of the CNT put forward by van Ulden thermometer screen (the so-called “Pagoda”) was replaced et al. (2009) (to which version number 1.0 is attached). The by a Stevenson screen. On 16 September 1950, a relocation differences between CNT1.1 and CNT1.0 are modest despite westwards had taken place, subsequently followed by a the differences in approaches to reach homogeneous input 300 m relocation southwards on 27 August 1951. series. This observation adds to the robustness of the CNT. The break at 1968–1969 is most probably related to In a comparison between the CNT1.1 and various other the uprooting of the nearest part of the neighbouring or- measures of the daily averaged temperature for central chard (which was turned into a parking lot) at the end of Netherlands, we conclude that the trends in most datasets March 1967 and/or the construction of a new 25 m high that are used to construct global mean temperatures are com- KNMI office which started in November 1967 and had patible with the trend of the Central Netherlands Tempera- reached its highest point on 2 January 1969. ture within about 30 %. This is despite the fact that these The only metadata which could possible relate to the de- datasets use grid boxes that are far larger then the area asso- tected break at 1919–1920 are routine replacements of ther- ciated with that of the Central Netherlands Temperature and mometers. The detected break near 1918–1920 is left un- that these datasets are based on unhomogenized time series. corrected due to the absence of metadata indicating possible The global datasets have their various methods to homoge- causes for this jump. nize input data, but they lack knowledge on detailed metadata After correcting for the 1950 and 1968 jumps, a new run records. This shows that the warming trend in these datasets through the program did detect the possible inhomogeneity is robust, but also points to the need to base these datasets on at 1976 but it failed to exceed the 95 % significance level. homogenized time series of temperature. Brandsma (KNMI, personal communication, 2010) ho- The CNT series and the ten homogenized station series mogenized the De Bilt record using more physical methods are updated monthly and can be downloaded from the KNMI rather than the statistical approach used here. He corrected web site, http://www.knmi.nl. for the replacement of the screen to the Stevenson screen and its transition (September 1950), a relocation of the Steven- son screen (August 1951), the lowering of Stevenson screen Appendix A from 2.2 m to 1.5 m (June 1961) and the transition of the artificial ventilated Stevenson screen to the KNMI round- Detected breaks and trends plated screen (June 1993). Finally, Brandsma corrected for a warming trend of 0.11 ◦C per century caused by urban warm- In this section, the available metadata and the results from ing. The correction factors Brandsma used for the changes the statistical tests described in Sect. 3 are combined. In the around 1950 and the correction factors used in this study are metadata, there are many details like routine maintenance shown in Fig. A1 and are very similar. However, the use and replacements of thermometers. These routine changes of smoothed adjustment factors in this study rather than the are made about once a year, which makes it unlikely that a original, more noisy monthly break-values, makes the adjust- step change or a discontinuous trend is related to a replaced ments for the De Bilt record associated with this break more thermometer. However, it shows that the detail of reporting conservative than those by Brandsma. The additional correc- for these stations is high, when relatively minor maintenance tion for the break around 1968 and the lack of a correction is added to the metadata records. Consistent and detailed re- of a possible warming trend makes the two records different. porting for a station is difficult to combine with vast changes The RMS difference between annual values of the De Bilt in the surroundings (leading to the break or discontinuous record obtained using the adjustments detected statistically trend) which remained unnoticed. Based on this argument, in this study and the one homogenized using Brandsma’s we were more inclined to dismiss statistical indications of method is 0.1 ◦C. a break in a record for stations that have detailed metadata records when no metadata is supportive of a break. For sta- A2 Den Helder/De Kooy tions where metadata is scarce, the absence of supporting metadata for the existence of a break was less prohibitive for No inhomogeneities have been detected. making adjustments to the record. A3 Groningen/Eelde A1 De Bilt No inhomogeneities have been detected in the annually av- The tests indicate breaks in annual averaged values near eraged values. However, evaluating the months separately 1918–1920, near 1950, near 1968–1969 and near 1976. gives possible inhomogeneities in the years 1952–1953, 1973

Clim. Past, 7, 527–542, 2011 www.clim-past.net/7/527/2011/ G. van der Schrier et al.: Central Netherlands Temperature G. van der Schrier et al.: Central Netherlands temperature13 539

Appendix 0.3 The break around 1966–1967 is most likely associated Brandsma De Bilt 1950 with two changes: the construction of a paved road on the Detected breaks and trends 0.2 De Bilt smoothed southeast side of the terrain and the related uprooting of high 0.1 trees which made the surroundings more open and the move In this section, the available metadata and the results from of the instrument field 1400 m southwards, further away from the statistical tests described in § 3 are combined. In the 0 trees in a more exposed setting in July 1966. The series is ad- metadata, there are many details like routine maintenance justed for this break. and replacements of thermometers. These routine changes -0.1 The 1971–1972 break has only a detectable break for are made about once a year, which makes it unlikely that a -0.2 March, August and September, not in the other months and step change or a discontinuous trend is related to a replaced not in the annual mean values. The only metadata from thermometer. However, it shows that the detail of reporting -0.3 around this year was the reinstallment of the thermometer for these stations is high, when even relatively minor main- screen following 1971 “new style” specifications at 5 Oc- -0.4 tenance is added to the metadata records. Consistent and tober 1970. These reinstallments were made at all stations detailed reporting for a station is difficult to combine with -0.5 around this period, making it unlikely that only this station vast changes in the surroundings (leading to the break or dis- Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec suffers from adverse effects. Furthermore, KNMI noted in continuous trend) which remained unnoticed. Based on this their logs at 14 October 1970 that a garage had been erected Fig. A1. Breaks at De Bilt for the changes around 1950. Shown are argument, we were more inclined to dismiss statistical indi- Fig. A1. Breaks at De Bilt for the changes around 1950. Shown are in the vicinity of the location. In the log the comment is thethe break break amplitudes amplitudes as as determined determined by by Brandsma Brandsma based based on on a physi- a physi- cations of a break in a record for stations that have detailed added “However, this garage has no effects on the measure- calcal methodology methodology (red) (red) and and based based on on the the current current statistical statistical approach approach ◦ ments”. It is not clear on what analysis this conclusion is metadata records when no metadata is supportive of a break. (green)(green) and and the the smoothed smoothed curve curve based based on on the the latter latter approach approach [ [C].◦C]. For stations where metadata is scarce, the absence of support- based. No corrections are made for this break. ing metadata for the existence of a break was less prohibitive The metadata for the years around 1984 only indicate rou- for making adjustments to the record. and 1996. The 1952–1953 break has an amplitude of a tine maintenance and replacements of the instruments, the 2.2 m to 1.5 m (June 1961) and the transition of artificial ven- most profound being a replacement of the thermograph on tilatedroughly Stevenson equal size screen for to July KNMI and round-plated December, but screen of opposite (June A1 De Bilt sign. The metadata indicate that on 28 February 1952 the 19 March 1983 due to a bended pen arm and adjustments to 1993). Finally, Brandsma corrected◦ for a warming trend of the thermograph on 6 October 1983 and 29 November 1984 0.11thermograph◦C per century was corrected caused by with urban 1 C warming. and that it The was correc- replaced ◦ ◦ The tests indicate breaks in annual averaged values near by 1.0 C and 0.5 C respectively. Thermographs were re- tionon factors 2 June Brandsma 1953. The used evidence for the from changes the metadata around 1950 is judged and 1918–20, near 1950, near 1968–69 and near 1976. There placed approximately once a year, which makes it improb- thetoo correction scanty to factors justify used a correction in this study for this are break. shown in fig. A1 is no information from the metadata which could explain able that these defects relate to the observed breaks in the and areOn very1 May similar. 1973 the However, measurement the use field of was smoothed relocated adjust- to the the breaks near 1918–20 and 1976. The break at 1950 is record. However, this latter break was large and robust mentwest factors side of in the this runway. study rather This thanmove the coincides original, with more changing noisy clearly related with a string of modifications which happened enough to justify adjustment in the absence of metadata. monthlymeasurement break-values, practice makes from that manual the adjustments to the use of for electronic the De in the early 1950s. On May 17th 1950, the large thermometer Biltequipment. record associated The Groningen/Eelde with this break series are more is adjusted conservativ for thise screen (the so-called ‘Pagoda’) was replaced by a Stevenson A5 Winterswijk/Hupsel thanbreak. those by Brandsma. The additional correction for the screen. On September 16th 1950, a relocation westwards had breakNo around metadata 1968 indicating and the a lack possible of a cause correction for the of 1996 a pos- break Large breaks in the Winterswijk/Hupsel record are detected taken place, subsequently followed by a 300 m relocation siblecould warming be found. trend Apparently, makes thesome two records suspicion different. at the KNMI The in 1940 and 1950. Much smaller breaks are detected near southward on August 27th 1951. RMSstaff difference of the time between must have annual existed, values since of the on De4 October Bilt record 1996, 1960 and 1970–1972. The break in 1940 is possibly related The break at 1968–69 is most probably related to the a comparison is made between the temperature sensors and obtained using the adjustments detected statistically in this to a relocation in 12 March 1940 to a more ideally located uprooting of the nearest part of the neighbouring orchard a calibrated sensor. No deviations were found though. The study and the one homogenized using Brandsma’s method is site. The new site is open, facing the observer’s house to (which was turned into a parking lot) at the end of March series is not corrected for this detected break. 0.1◦C. the north and a meadow to the south, but the thermometer 1967 and/or the construction of a new 25 m high KNMI of- screen is placed between two shrubs. On 27 February 1950, A4 Oudenbosch/Gilze Rijen fice which started in November 1967 and had reached its the thermometer screen was relocated 5 m eastwards away highest point on January 2nd 1969. A2 Den Helder/De Kooy Breaks in the Oudenbosch/Gilze Rijen series are detected from the shrubs. The growth of the shrubs might have intro- The only metadata which could possible relate to the de- duced an artificial trend in the data. The annually averaged Noaround inhomogeneities 1946–1948, are 1966–1967, detected. 1971 and near 1984. tected break at 1919–20 are routine replacements of ther- difference record does show a trend over this 10-year period, mometers. The detected break near 1918–20 is left uncor- The metadata indicates that corrections on the minimum ◦ and maximum recording thermometers changed frequently but it is small (approx. 0.18 C in 10 years) and has been left rected due to the absence of metadata indicating possible A3 Groningen/Eelde uncorrected. causes for this jump. and significantly around the 1946–1947 period. Breaks in the minimum and maximum temperatures could result in breaks The breaks near 1960 and 1970–1972 have been left un- After correcting for the 1950 and 1968 jumps, a new run No inhomogeneities are detected in the annually averaged in the daily averaged temperature since Oudenbosch is one corrected due to absence of metadata which could be related through the program did detect the possible inhomogeneity values. However, evaluating the months separately gives pos- of the stations where Eq. (2) is used. However, the changes to the break. at 1976 but it failed to exceed the 95% significance level. sible inhomogeneities in the years 1952-53, 1973 and 1996. in corrections to the min. and max. temperatures affected Interestingly, a break is detected near 1984–1985. Based Brandsma (KNMI, personal communication) homoge- The 1952-53 break has an amplitude of roughly equal size temperatures below freezing mostly and amounted to ap- on the F -statistic, this break is not significant and is not nized the De Bilt record using more physical methods rather for July and December, but of opposite sign. The metadata prox. 0.1 ◦C to 0.2 ◦C. Application of models Eqs. (3) and corrected for. However, it seems to be related to a mod- than the statistical approach used here. He corrected for the indicate that on February 28 1952 the thermograph was cor- est relocation of the station some 50 m in SW direction on (5) indicate◦ statistically significant breaks in summer and au- replacement of the screen to the Stevenson screen and its rectedtumn. with Based 1 C on and incongruous that it has evidence been replaced from the on metadata June 2rd and 27 March 1985. transition (September 1950), a relocation of the Stevenson 1953.the tests, Theevidence we leave thisfrom break the metadata uncorrected. is judged too scanty screen (August 1951), the lowering of Stevenson screen from to justify a correction for this break.

www.clim-past.net/7/527/2011/ Clim. Past, 7, 527–542, 2011 540G. van G. der van Schrier der Schrier et al.: et Central al.: Central Netherlands Netherlands Temperature temperature 15

A6 Gemert/Volkel 0.4 data exists which might substantiate this break. No adjust- ments are made to this record. 0.3 A large break and a discontinuous trend are detected around 1950. The break and discontinuous trend are obvious from 0.2 A9 Hoorn a visual inspection of the difference record (Fig. A2). This 0.1 is clearly associated with the reinstallment of the station on Hoorn shows a break which is barely significant in 1948. Ini- 0 27 September 1949. Preceding this reinstallment, reports had tially, the observation site was located on a terrain for agri- been made (in June and July of 1949) indicating that the site -0.1 cultural use. On November 1st 1946, the observation terrain was relocated to the gardens of the local slaughterhouse, fac- did not meet regulations regarding the surroundings. The -0.2 clearing was too small for proper ventilation. Additionally, ing nearby buildings in southeast to southwest directions. On the height of the thermometer screen was not according to -0.3 November 21st 1947, measurements ceased and on 28 April regulations. -0.4 1948, the station was relocated back to its original terrain. Having identified the combination of a step and a discon- No adjustments are made for this break. -0.5 tinuous trend for 1950, estimates of the adjustments from the 1900 1920 1940 1960 1980 2000 2020 Another break is detected around 1970–73, but only for the other model, Eq. (7), are to be used. It turns out that the es- month March. The metadata indicate that a school was build timates of the size of the trend for the 1950 break, calculated Fig.Fig. A2. A2.TrendsTrends and and breaks breaks in the in the annual annual mean mean temperature temperature dif- dif- at a distance of approximately 20 m from the observation site for each sliding window, show much variability. This is prob- ferenceference between between the the Gemert/Volkel Gemert/Volkel series series and and the the reference reference series series in the early 1970s and a relocation in SW direction of 15 m [◦C].◦ ably related to the relatively high year-to-year variability of [ C]. was effectuated on 2 July 1971, but it is unclear if this rather the difference series in relation to the modest trend. Because modest relocation could explain the break to warmer condi- of this, the few estimates (<10%) of a negative trend were tions. No correction for this break are made to the Hoorn sationreasons procedure for having detects the a largest break adjustment and discontinues factors trend associated at not used in the final estimate of the trend, nor were estimates record. thiswith year, the with move a negative to a more trend ideally in the located difference setting series (Fig. aft2).er used which showed trends of >0.5 ◦C/10 year (7 instances). 1980. No break nor discontinuous trend is detected when the A10 Schiphol The record is corrected for the break and the discontinuous transitionA8 Twenthe between the series is set at 1990. The poor quality trend. of the measurements from the 1980s of the Volkel air base The Schiphol (Amsterdam International Airport) record There is some discussion on the validity of the measure- (notTwenthe shown) shows may be a break related in to 1969 the spurious in the annually trend. Theaveraged intro- val- shows a significant break in 1981. The Schiphol metadata ments from Volkel airbase. When the transition between the ductionues, but of significant automated values measurements fail to show in the in earlyan analysis 1990s for will each is not very clear about the cause for this break. Presumably Volkel and Gemert records is set at 1980, then the homogeni- havemonth improved separately. the quality Twenthe of the is a data. military With air the base nearest and no tree meta- it is related to a relocation of the measurement field from sation procedure detects a break and discontinues trend at linedata at 245 exists m ofwhich the thermometer might substantiate screen inthis N-NW break. direction, No adjust- the vicinity of the main buildings to the outer edge of the this year, with a negative trend in the difference series after withments trees have of heights been made between to this 18–20 record. m, the situation at the Schiphol area, near a runway. The KNMI archives holds a 1980. No break nor discontinuous trend is detected when the observation site is in line with WMO regulations for a mete- map for the situation around 1960 and one for 1986 from transition between the series is set at 1990. The poor quality orologicalA9 Hoorn observation site. which a change in position of the measurement field is ev- No corrections are made to the Volkel record which is of the measurements from the 1980s of the Volkel air base ident but a more precise timing of the relocation cannot be (not shown) may be related to the spurious trend. The intro- blendedHoorn to shows the Gemert a break record which from is 1990 barely onwards. significant in 1948. Initially, the observation site was located on a terrain for given. Given the rapid growth of the airport, this period is duction of automated measurements in the early 1990s will likely to have seen more than one relocation of the instru- A7agricultural Maastricht/Beek use. On 1 November 1946, the observation have improved the quality of the data. With the nearest tree mentation. The record is adjusted for this break. line at 245 m of the thermometer screen in N–NW direction, terrain was relocated to the gardens of the local slaughter- with trees of heights between 18–20 m, the situation at the Thehouse, Maastricht/Beek facing nearby record buildings shows in a break southeast in 1931 to southwest in the an- di- A11 Deelen observation site is in line with the WMO regulations for a nuallyrections. averaged On 21values, November but fails 1947, to show measurements significant values ceasedin and meteorological observation site. anon analysis 28 April for 1948, each month the station separately. was relocated The Maastricht back to obser its origi-- vations were made on top of a tower (at approx. 20m above In the period 1954–57, measurements were taken during No corrections are made to the Volkel record which is nal terrain. No adjustments are made for this break. street level) on a building in the centre of the city. From 1 weekdays only at Deelen airbase, reliable monthly means blended to the Gemert record from 1990 onwards. Another break is detected around 1970–1973, but only for Julythe 1951, month parallel of March. measurements The metadata were indicate made at that the a outskirts school was could only be constructed for January 1958 onwards. Breaks ofbuilt the city at a in distance a garden of area approximately which show 20 considerable m from the differ- observa- are detected near 1962 and around 1984–85. However, the A7 Maastricht/Beek encestion with site inthe the Maastricht early 1970s observations. and a relocation The suboptimal in SW direction set- metadata from this military airport provided no leads to what tingof of 15 the m was Maastricht effectuated station on and2 July the 1971, relocation but it toisunclear a new sit ife this might have caused these breaks. The breaks are large enough to convincingly exceed the critical significance levels and are The Maastricht/Beek record shows a break in 1931 in the an- somerather 65 modest m higher relocation in altitude could may explain be the theprincipal break reasons to warmer adjusted. nually averaged values, but fails to show significant values in forconditions. having the Nolargest correction adjustment for thisfactors break associated have been with made the to an analysis for each month separately. The Maastricht obser- movethe Hoornto a more record. ideally located setting (fig. 2). In the annual averaged values for Deelen and to a lesser vations were made on top of a tower (at approx. 20 m above extent in the February monthly means, a discontinuous trend street level) on a building in the centre of the city. From A8A10 Twenthe Schiphol is detected with a break in 1977. Before 1977, a distinct 1 July 1951 onwards, parallel measurements were made at upwards trend is detected indicating that Deelen warms faster the outskirts of the city in a garden area which show consid- TwentheThe Schiphol shows a break (Amsterdam in 1969 in International the annually averaged Airport) val- record than the reference record, after that year the warming trends erable differences to the Maastricht observations. The sub- ues,shows but significant a significant values break fail in to show1981. in The an analysis Schiphol for metadata each are similar. Again, there is no indication what might be the optimal setting of the Maastricht station and the relocation to monthis not separately. very clear Twenthe about the is a cause military for air this base break. and no Presumably meta- reason for the discontinuous trend and it is left unadjusted. a new site some 65 m higher in altitude may be the principal it is related to a relocation of the measurement field from

Clim. Past, 7, 527–542, 2011 www.clim-past.net/7/527/2011/ G. van der Schrier et al.: Central Netherlands temperature 541 the vicinity of the main buildings to the outer edge of the After corrections for the 1969–1970 and 1986–1988 Schiphol area, near a runway. The KNMI archives holds a breaks, the Eindhoven series was put through the break de- map for the situation around 1960 and one for 1986 from tection script again and this yielded a break in 1980–1981 which a change in position of the measurement field is ev- and a newly detected break in 1958, which was apparently ident but a more precise timing of the relocation cannot be not large enough (in terms of the F -statistic) to be reported given. Given the rapid growth of the airport, this period is in the uncorrected series. This break must be related to a sta- likely to have seen more than one relocation of the instru- tion relocation at 17 October 1958. Before this date, the site mentation. The record is adjusted for this break. did not meet KNMI specifications. This break is corrected for. A11 Deelen A13 Rotterdam In the period 1954–1957, measurements were taken during weekdays only at Deelen airbase, reliable monthly means No breaks have been detected at Rotterdam. could only be constructed for January 1958 onwards. Breaks have been detected near 1962 and around 1984–1985. How- Acknowledgements. The authors would like to thank Theo ever, the metadata from this military airport provided no Brandsma and the Royal Netherlands Meteorological Institute for leads as to what might have caused these breaks. The breaks making available the homogenized De Bilt record from Brandsma (http://www.knmi.nl/klimatologie/onderzoeks-ge-ge-vens/ are large enough to convincingly exceed the critical signifi- homogeen 260/tg hom mnd260.txt). We thank Jan Huizinga cance levels and have been adjusted. for his assistance in uncovering the metadata from the KNMI In the annual averaged values for Deelen and to a lesser archives. The two anonymous referees are thanked for their extent in the February monthly means, a discontinuous trend constructive comments. was detected with a break in 1977. Before 1977, a distinct upwards trend was detected indicating that Deelen warms Edited by: J. Guiot faster than the reference record, after that year the warming trends are similar. Again, there is no indication as to what might be the reason for the discontinuous trend and it is left References unadjusted. Alexandersson, H. and Moberg, A.: Homogenization of Swedish temperature data, part I: homogeneity test for linear trends, Int. A12 Eindhoven J. Climatol., 17, 25–34, doi:10.1002/joc.1097, 1997. Brandsma, T. and van der Meulen, J. P.: Thermometer screen in- A curiously low value for annual averaged temperature for ◦ tercomparison in De Bilt (the Netherlands) – Part II: Description Eindhoven airport was observed for 1952, which is 0.95 C and modeling of mean temperature differences and extremes, Int. lower than the reference. A comparison with adjusted J. Climatol., 28, 389–400, doi:10.1002/joc.1524, 2008. records for Oudenbosch/Gilze Rijen and Gemert/Volkel in- Easterling, D. R. and Peterson, T. C.: A new method for detecting dicated that the monthly averages for the month May to undocumented discontinuities in climatological time series, Int. July were up to 3 ◦C lower than surrounding stations. The J. Climatol., 15, 369–377, doi:10.1002/joc.3370150403, 1995. monthly averages of 1952 for these months were replaced Hansen, J., Ruedy, R., Sato, M., and Lo, K.: Global by an average of the corresponding months of the adjusted surface temperature change, Rev. Geophys., 48, RG4004, records of Oudenbosch/Gilze Rijen and Gemert/Volkel. The doi:10.1029/2010RG000345, 2010. metadata indicated that observations from 1 May 1952 on- Haylock, M. R., Hofstra, N., Klein Tank, A. M. G., Klok, wards were made by airforce personnel rather than civil ser- E. J., Jones, P. D., and New, M.: A European daily high- resolution gridded daat set of surface temperature and precipi- vants from the aviation authority, and reports of KNMI in- tation for 1950–2006, J. Geophys. Res.-Atmos., 113, D20119, spectors of the mid-1950s complained of many false read- doi:10.1029/2008JD010201, 2008. ings. Jaruskova,´ D.: Change-Point Detection in Meteorological Measure- Breaks were detected at 1969–1970 and 1986–1988. The ment, Mon. Weather Rev., 124, 1535–1543, 1996. exact timing of the latter break is vague, non-significant Labrijn, A.: The climate of the Netherlands during the last two and breaks were also reported for 1985, but strangely enough, a half centuries, Scientific Rep. KNMI No. 102, Royal Nether- none for 1987. The metadata provided no information on the lands Meteorological Institute, mededelingen en Verhandelin- possible origin of the first break. The relocation to a new gen, 1945. terrain on 3 July 1984 may be related to the latter break. Lund, R. and Reeves, J.: Detection of Undocumented Change- There is some indication of a break in the month of May points: A Revision of the Two-Phase Regression Model, J. Cli- only, around 1980–1981; only 3 sliding windows indicated mate, 15, 2547–2554, 2002. Peterson, T. C. and Easterling, D. R.: Creation of homogeneous this break. No metadata had been found which might account composite climatological reference series, Int. J. Climatol., 14, for this break and it is left uncorrected. 671–680, 1994.

www.clim-past.net/7/527/2011/ Clim. Past, 7, 527–542, 2011 542 G. van der Schrier et al.: Central Netherlands temperature

Peterson, T. C., Easterling, D. R., Karl, T. R., Groisman, P., van Ulden, A.: The Construction of a Central Netherlands tem- Nicholls, N., Plummer, N., Torok, S., Auer, I., Boehm, R., perature, EMS8/ECAC7 Abstracts, Vol. 5, EMS2008-A-00421, Gullett, D., Vincent, L., Heino, R., Toumenvirta, H., Mestre, 8th Annual Meeting of the EMS/7th ECAC, 2008. O., Szentimrey, T., Salinger, J., Forland, E., Hanssen-Bauer, I., van Ulden, A., van Oldenborgh, G. J., and van der Schrier, G., Alexandersson, H., Jones, P. D., and Parker, D.: Homogeneity The Construction of a Central Netherlands temperature, Scien- adjustments of in situ atmospheric climate data: A review, Int. J. tific Rep. WR 2009-03, KNMI, 2009. Climatol., 18, 1493–1517, 1998. Vincent, L. A.: A Technique for the Identification of Inhomo- Potter, K. W., Illustration of a New Test for Detecting a Shift in geneities in Canadian Temperature Series, J. Climate, 11, 1094– Mean in Precipitation Series, Mon. Wea. Rev., 109, 2040–2045, 1104, 1998. 1981. Wang, X. L.: Comments on “Detection of Undocumented Change- Reeves, J., Chen, J., Wang, X. L., Lund, R., and Lu, Q.: A Re- points: A Revision of the Two-Phase Regression Model”, J. Cli- view and Comparison of Changepoint Detection Techniques for mate, 16, 3383–3385, 2003. Climate Data, J. Appl. Meteorol. Clim., 46, 900–915, 2007. Wilks, D. S.: Statistical methods in the Atmospheric Sciences, Aca- van der Hoeven, P. C. T.: Etmaaltemperatuur en Dagextremen uit demic Press, 1995. Termijnwaarnemingen, memorandum KD 92-08, KNMI, 1992.

Clim. Past, 7, 527–542, 2011 www.clim-past.net/7/527/2011/