indications for a changing electricity demand pattern: the temperature dependence of electricity...

10
Indications for a changing electricity demand pattern: The temperature dependence of electricity demand in the Netherlands M. Hekkenberg , R.M.J. Benders, H.C. Moll, A.J.M. Schoot Uiterkamp University of Groningen, Center for Energy and Environmental Studies IVEM, Nijenborg 4, 9747AG Groningen, The Netherlands article info Article history: Received 16 October 2008 Accepted 15 December 2008 Available online 7 February 2009 Keywords: Electricity demand Cooling demand Climate change abstract This study assesses the electricity demand pattern in the relatively temperate climate of the Netherlands (latitude 52130 0 N). Daily electricity demand and average temperature during the period from 1970 until 2007 are investigated for possible trends in the temperature dependence of electricity demand. We hypothesize that the increased use of cooling applications has shifted the temperature dependence of electricity demand upwards in summer months. Our results show significant increases in temperature dependence of electricity demand in May, June, September, October and during the summer holidays. During the period studied, temperature dependence in these months has shifted from negative to positive, meaning that a higher temperature now leads to an increased electricity demand in these months, rather than a decreased demand as observed historically. Although electricity demand in countries with moderate summer temperatures such as the Netherlands generally peaks in winter months and shows a minimum in summer months, this trend may signal the development of an additional peak in summer, especially given the expected climatic change. As power generating capacity may be negatively influenced by higher temperatures due to decreasing process cooling possibilities, an increasing electricity demand at higher temperatures may have important consequences for power generation capacity planning and maintenance scheduling. & 2009 Elsevier Ltd. All rights reserved. 1. Introduction Daily electricity demand in countries throughout the world shows a clear seasonal pattern. Three different seasonal patterns may be observed in general; average daily electricity demand may peak during winter, during summer, or both, in which case either the winter or the summer peak is highest. Two of these patters can be found in Europe (see Fig. 1). Average daily electricity demand in most EU-15 countries, amongst which the Nether- lands, historically shows a single peak during winter months (Eurostat, 2008). Only Spain, Portugal, Italy and Greece show an additional peak during summer months. Demand patterns with only summer peaks may be found outside of Europe, e.g. as observed in Bangkok (Wangpattarapong et al., 2008) and Hong Kong (Al-Zayer and Al-Ibrahim, 1996). The seasonal pattern results from the fluctuating influx of solar radiation and the varying economic activity throughout the year in the Northern hemisphere (Pardo et al., 2002). Winter peaks in electricity demand may be attributed to increased lighting demand because of shorter daylight periods as well as to an increased heating demand and a higher average economic activity in winter than in summer due to holidays. Summer peaks are usually attributed to the use of electric cooling applications such as fans and especially air conditioners. Time series analysis of monthly electricity demand in Spain, Portugal, Italy and Greece shows that the relative importance of electricity demand during summer months has been increasing over the last decades, resulting in a summer peak average demand that topped the winter peak in Greece, Italy and Spain in 2006 (Fig. 2). This development is reflected in studies by Valor et al. (2001) and Moral-Carcedo and Vice ´ ns-Otero (2005), who correlate the Spanish daily electricity demand between 1980 and 1998 to outdoor temperature and by Bessec and Fouquau (2008) for all EU-15 countries. Their studies, like others (Nobel, 1996; Sailor and Mun ˜oz, 1997; Pardo et al., 2002; Ruth and Lin, 2006; Mirasgedis et al., 2007; Franco and Sanstad, 2008), find a ‘u’ shaped relation between outdoor temperature and electricity demand. However, they find that in recent years, the positive correlation between electricity demand and outdoor temperature for temperatures upwards of 18 1C is becoming more pronounced. This develop- ment is thought to mainly result from the increasing use of cooling applications at high temperatures. Electricity demand from cooling applications is logically expected to be positively temperature dependent: cooling uses more electricity on warmer days. Three mechanisms are princi- pally accountable for this dependence. Firstly, at higher outdoor ARTICLE IN PRESS Contents lists available at ScienceDirect journal homepage: www.elsevier.com/locate/enpol Energy Policy 0301-4215/$ - see front matter & 2009 Elsevier Ltd. All rights reserved. doi:10.1016/j.enpol.2008.12.030 Corresponding author. E-mail address: [email protected] (M. Hekkenberg). Energy Policy 37 (2009) 1542–1551

Upload: m-hekkenberg

Post on 21-Jun-2016

221 views

Category:

Documents


1 download

TRANSCRIPT

Page 1: Indications for a changing electricity demand pattern: The temperature dependence of electricity demand in the Netherlands

ARTICLE IN PRESS

Energy Policy 37 (2009) 1542–1551

Contents lists available at ScienceDirect

Energy Policy

0301-42

doi:10.1

� Corr

E-m

journal homepage: www.elsevier.com/locate/enpol

Indications for a changing electricity demand pattern: The temperaturedependence of electricity demand in the Netherlands

M. Hekkenberg �, R.M.J. Benders, H.C. Moll, A.J.M. Schoot Uiterkamp

University of Groningen, Center for Energy and Environmental Studies IVEM, Nijenborg 4, 9747AG Groningen, The Netherlands

a r t i c l e i n f o

Article history:

Received 16 October 2008

Accepted 15 December 2008Available online 7 February 2009

Keywords:

Electricity demand

Cooling demand

Climate change

15/$ - see front matter & 2009 Elsevier Ltd. A

016/j.enpol.2008.12.030

esponding author.

ail address: [email protected] (M. Hekken

a b s t r a c t

This study assesses the electricity demand pattern in the relatively temperate climate of the

Netherlands (latitude 521300N). Daily electricity demand and average temperature during the period

from 1970 until 2007 are investigated for possible trends in the temperature dependence of electricity

demand. We hypothesize that the increased use of cooling applications has shifted the temperature

dependence of electricity demand upwards in summer months. Our results show significant increases

in temperature dependence of electricity demand in May, June, September, October and during the

summer holidays. During the period studied, temperature dependence in these months has shifted from

negative to positive, meaning that a higher temperature now leads to an increased electricity demand in

these months, rather than a decreased demand as observed historically. Although electricity demand in

countries with moderate summer temperatures such as the Netherlands generally peaks in winter

months and shows a minimum in summer months, this trend may signal the development of an

additional peak in summer, especially given the expected climatic change. As power generating capacity

may be negatively influenced by higher temperatures due to decreasing process cooling possibilities, an

increasing electricity demand at higher temperatures may have important consequences for power

generation capacity planning and maintenance scheduling.

& 2009 Elsevier Ltd. All rights reserved.

1. Introduction

Daily electricity demand in countries throughout the worldshows a clear seasonal pattern. Three different seasonal patternsmay be observed in general; average daily electricity demand maypeak during winter, during summer, or both, in which case eitherthe winter or the summer peak is highest. Two of these patterscan be found in Europe (see Fig. 1). Average daily electricitydemand in most EU-15 countries, amongst which the Nether-lands, historically shows a single peak during winter months(Eurostat, 2008). Only Spain, Portugal, Italy and Greece show anadditional peak during summer months. Demand patterns withonly summer peaks may be found outside of Europe, e.g. asobserved in Bangkok (Wangpattarapong et al., 2008) and HongKong (Al-Zayer and Al-Ibrahim, 1996).

The seasonal pattern results from the fluctuating influx of solarradiation and the varying economic activity throughout the yearin the Northern hemisphere (Pardo et al., 2002). Winter peaks inelectricity demand may be attributed to increased lightingdemand because of shorter daylight periods as well as to anincreased heating demand and a higher average economic activity

ll rights reserved.

berg).

in winter than in summer due to holidays. Summer peaks areusually attributed to the use of electric cooling applications suchas fans and especially air conditioners.

Time series analysis of monthly electricity demand in Spain,Portugal, Italy and Greece shows that the relative importance ofelectricity demand during summer months has been increasingover the last decades, resulting in a summer peak average demandthat topped the winter peak in Greece, Italy and Spain in 2006(Fig. 2). This development is reflected in studies by Valor et al.(2001) and Moral-Carcedo and Vicens-Otero (2005), who correlatethe Spanish daily electricity demand between 1980 and 1998 tooutdoor temperature and by Bessec and Fouquau (2008) for allEU-15 countries. Their studies, like others (Nobel, 1996; Sailor andMunoz, 1997; Pardo et al., 2002; Ruth and Lin, 2006; Mirasgediset al., 2007; Franco and Sanstad, 2008), find a ‘u’ shaped relationbetween outdoor temperature and electricity demand. However,they find that in recent years, the positive correlation betweenelectricity demand and outdoor temperature for temperaturesupwards of 18 1C is becoming more pronounced. This develop-ment is thought to mainly result from the increasing use ofcooling applications at high temperatures.

Electricity demand from cooling applications is logicallyexpected to be positively temperature dependent: cooling usesmore electricity on warmer days. Three mechanisms are princi-pally accountable for this dependence. Firstly, at higher outdoor

Page 2: Indications for a changing electricity demand pattern: The temperature dependence of electricity demand in the Netherlands

ARTICLE IN PRESS

95

100

105

110

115

120

125

1985

1987

1989

1991

1993

1995

1997

1999

2001

2003

2005

Year

Nor

mal

ized

ave

rage

dai

ly e

lect

ricity

dem

and

Winter peak Spain Winter peak Greece Winter peak ItalySummer peak Spain Summer peak Greece Summer peak Italy

Fig. 2. Development of the average daily electricity demand in the peak summer and winter months in Greece, Italy and Spain from 1986 to 2006. Annual average daily

electricity demand ¼ 100 for each year and each country. Peak months are determined for each year as the summer (April–September) and winter (October–March) month

with the highest average daily electricity demand in each given year. Data adapted from Eurostat (2008).

80

85

90

95

100

105

110

115

120

125

130

Jan

Month of the year

Nor

mal

ized

ave

rage

dai

ly e

lect

ricity

dem

and

Netherlands Spain Greece Italy Sweden

DecNovOctSepAugJulJunMayAprMarFeb

Fig. 1. Seasonal variation in daily electricity demand in several EU-15 countries over the period 1986–2006. Average daily electricity demand ¼ 100 for each country. Data

adapted from Eurostat (2008).

1 Nobel refers to normal temperatures as 2–18 1C.

M. Hekkenberg et al. / Energy Policy 37 (2009) 1542–1551 1543

temperatures, a larger temperature difference needs to be over-come, thus increasing the cooling load. Secondly, higher tempera-tures lead to less efficient heat exchange in cooling applications,therefore more vapor-compression cycles have to be made toexchange the same load of heat. Thirdly and probably mostobvious, people are more inclined to turn on air conditioningapplications on warm days than on cold days.

However, not only countries with summers generally consid-ered ‘warm’ or ‘hot’ show a societal trend towards increased use ofcooling applications. The use of cooling applications has alsoincreased in countries at higher latitudes with generally moderatesummer temperatures, such as the Netherlands (latitude521300N); supermarkets have generally increased their area ofcold products over the last decades, while most new commercialand office buildings are being equipped with air conditioningapplications (Van Arkel et al., 1999). Relatively little isknown about the effects of cooling on electricity demand in theNetherlands. An exploratory study estimated total electricity

demand from cooling applications in the Netherlands to be4–5 TWh in 2003 (Pennartz, 2005), which equals 4–5% of totalelectricity demand (CBS, 2008). Still, the Dutch monthly electricitydemand pattern does not show a summer peak such as seenin ‘warm’ countries. Nonetheless, an increasing use of coolingapplications is expected to lead to an increasingly positive relationbetween temperature and electricity demand. Apparently,this effect is obscured in the monthly figures by other (socio-economic) drivers such as the generally lower economic activityduring summer due to holidays. The only known publication thatcorrelates outdoor temperature and electricity demand for theNetherlands (Nobel, 1996) does not investigate historic develop-ment, although it does find a positive relation between averagedaily temperature and minimum electricity demand in thefollowing night for average day temperatures above normal.1

Page 3: Indications for a changing electricity demand pattern: The temperature dependence of electricity demand in the Netherlands

ARTICLE IN PRESS

M. Hekkenberg et al. / Energy Policy 37 (2009) 1542–15511544

However, temperature dependence patterns are important forassessments of future electricity demand, especially in the contextof climate change (Sailor and Pavlova, 2003; Amato et al., 2005;Ruth and Lin, 2006; Isaac and Van Vuuren, 2009). Possible trendsin such patterns may thus improve such assessments. Moreover, achanging temperature dependence may be a first indication of thedevelopment of an (initially small) summer peak, after all, such atrend may result in an increasing importance of summer monthsin the yearly total. Furthermore, a changing temperature depen-dence may have implications for maintenance scheduling. Thisstudy therefore aims to disclose the effect of an increased use ofcooling applications on the Dutch electricity demand.

2. Methodology

We have collected daily data on temperature and electricityuse, and analyzed the resulting data points with temperature asthe independent variable and electricity use as the dependentvariable. Unfortunately, there is a high degree of noise in such databecause of daily variation in electricity consumption unrelated totemperature differences. Such noise in the data may obscure anyexisting trend in small data samples.

Therefore, studies investigating temperature dependence ofelectricity demand usually analyze time series of a year or longer,and distill long-term trends such as economic growth orpopulation dynamics using extensive data mining techniques.Such approaches generally require formal assumptions about thevarious trends that may exist in the time series. In case theappropriate parameters are included in such an analysis, suchapproaches could possibly find a sophisticated description of theobserved temperature dependence, perhaps even throughout theyears. However, in our view the underlying socio-economicvariation is too complex to adequately capture in formalassumptions, and our data sample is not sufficiently consistent(see ‘data’ section) to allow a comprehensive analysis. Thereforewe have not followed a complex data mining approach.

Instead, we have deliberately chosen a relatively simpleapproach that uses the raw data available to establish a possibletrend in temperature dependence regardless of all other trends.After all, our primary aim is to find evidence that the temperaturedependence may be subject to change, not to establish asophisticated formula to describe temperature dependence.Instead of aggregating available data into yearly or larger datasets,we have separated the available data into monthly sets to describethe temperature dependence of electricity demand in the monthsfrom 1970 until 2007 for which suitable data was available.

Box 1–‘Non-working days’

The following days are excluded from our study.

(1) Weekends (all Saturdays and Sundays)

(2) National holidays (New Years Day, Good Friday, Easter day and E

Ascension Day, Whit Sunday and Whit Monday, Christmas Day

(3) Christmas holiday (all dates between and including 24 Decembe

(4) Single days that are preceded and followed by weekends or hol

Two other days have been excluded (September 5th 1973 and Dereported electricity use from the monthly average exceeded 20% fordata points.

Following these exclusion rules leads to the exclusion of 3972 (3been included.

For each working day in each month we have calculated thedeviation of daily electricity use relative to the average dailyelectricity use on working days in the same month. We havesubsequently plotted the relative deviation of electricity use foreach day against the average temperature on that day. For each ofthe resulting scatter plots we have calculated an ordinary leastsquared linear regression coefficient. Each regression coefficientrepresents the temperature dependence of electricity demand fora particular month in a particular year. Negative coefficients meanthat a higher temperature leads to lower electricity demand,positive coefficients mean that a higher temperature leads tohigher electricity demand. Through this approach, we havecollected a time series of temperature dependence coefficientsfor each month of the year. In this way, observed trendsthroughout the time series represent changes in the temperaturedependence of electricity demand. Finally, observed trends areanalyzed statistically to test the robustness of the findings.

The main advantage of this approach is that it compares thedeviations within one month, which result in dependencecoefficients that are representative for that particular month.Such an approach diminishes the possible distortive influence ofinconsistent seasonal parameters such as relative economicactivity and solar radiation of which the combined effects arehard to estimate, and of possible methodological changes inmeasuring electricity demand or temperature. Provided that thedata within the month are (more) consistent, this possibly resultsin clearer temperature dependence signals. A disadvantage is thatthe results are harder to interpret because it will not result in a‘general’ temperature dependence of electricity demand.

Trends found within these short time series can be highlydependent on single observations. For example, the chanceoccurrence of a relatively low electricity demand on a day withrelatively high temperature may pull the observed trend down-wards, whereas if a low demand occurred on a day with arelatively low temperature the trend would be pushed upwards.Because of the noise in the available data, each separate dataset isrelatively sensitive to such chance occurrences.

Part of the temperature independent noise may be anticipated,such as general economic activity differences between workingdays and non-working days. The difference between working daysand non-working days shows clear patterns that can accountfor a large share of day to day variation in electricity demand(Moral-Carcedo and Vicens-Otero, 2005). Eliminating this varia-bility can contribute to extracting the signal we are looking for.We have therefore focused our research on working days only.Box 1 describes the ‘non-working days’ which have been excludedfrom our study. The average deviation of electricity demand onexcluded data compared with the corresponding monthly average

aster Monday, Queens Day (April 30th, except when on Sunday),

and Boxing Day, New Years Eve).

r and 2 January)

idays (e.g. Friday after Ascension Day).

cember 17th 2003), because on these days the deviation of theno apparent reason. It is assumed that these are ‘‘contaminated’’

2.0%) of 12,410 available data points. All other data points have

Page 4: Indications for a changing electricity demand pattern: The temperature dependence of electricity demand in the Netherlands

ARTICLE IN PRESS

M. Hekkenberg et al. / Energy Policy 37 (2009) 1542–1551 1545

electricity demand on included data is �20.7%, suggesting thatelectricity demand on excluded days differs from that on includeddays indeed. By excluding non-working day observations we loseinformation on the temperature dependence trend on non-working days. However, non-working days cannot be regardedas a homogenous group. Saturdays differ from Sundays, and eachholiday may again have its own characteristics. Thereforeresearching a trend in non-working days would require furtherspecification of the type of non-working day. Since we havechosen to analyze the data on a monthly data for the reasonsdescribed above, there are too few observations of each non-working day type per month to make a meaningful analysis.

Still, our generic exclusion rules do not totally suppress thedata noise. Because differences in electricity demand resultingfrom the temperature dependence of electricity demand may stillbe small relative to remaining noise, the findings within separatedatasets have little individual meaning in themselves. Therefore,conclusions about the possible trends in temperature dependenceshould not be based on individual datasets, but only on theircombination.

3. Data

We analyzed the relation between outdoor temperature andelectricity demand on the Dutch high voltage network during theperiod 1970–2007. We therefore collected data on daily electricityload and daily outdoor temperature during this period.

3.1. Electricity data

Our data on electricity load consist of two separate datasets.We obtained historic half-hourly electricity load figures for theperiod January 1970 to May 1999 from the current DutchTransport System Operator (TSO) TenneT. This dataset (‘SEPdataset’) consists of the historical measurement data from thecooperation of electricity producers (SEP), which measuredaverage electricity load on the high voltage network during5 min periods for each half hour. TenneT was formed from SEP in1998 as the independent TSO in the Netherlands conform the newpolicy of a liberalized electricity market. Unfortunately, electricityload data during the first years after the reform are not available.From 2003 onwards, electricity load data for 15 min intervals isavailable on the internet (TenneT, 2008). We have used these data(‘TenneT dataset’) from March 2003 up until December 2007.Because of the missing data, the combined dataset therefore

0

20000

40000

60000

80000

100000

120000

1970

1972

1974

1976

1978

1980

1982

1984

Ann

ual e

lect

ricity

dem

and

(GW

h) CBS dataset "electricity availaTenneT datasetSEP dataset

Fig. 3. Comparison of reported electricity demand in the Netherlands in the SEP and

covers a total of 411 months in the period 1970 to 2007; 34 or 35datasets are available for each month of the year.

Comparison of the two datasets thus obtained shows that theSEP dataset does not fit the TenneT dataset. The difference isthought to originate from a changed measurement methodbetween both datasets. The SEP-dataset consists of the measuredload at the connections of the regional grids with the high voltagegrid and centrally operated production facilities (operated by SEP).The TenneT dataset is the sum of data provided by each gridadministrator, which reports on each connected party with agenerating capacity of more than 10 MW (TenneT, 2008). There-fore, the TenneT dataset may include some decentralized sourcesthat were excluded in the SEP dataset. Furthermore, comparisonof both datasets with total Dutch electricity available through thepublic grid as calculated by Statistics Netherlands (CBS), showsthat the SEP dataset progressively deviates from the data reportedby CBS. The SEP dataset covers 99% of the total electricity availablethrough the public grid in 1976–1986 but the coverage declines toonly 83% in 1998 (Fig. 3). This trend may be partly explained by anincreasing share of decentralized electricity production in localgrids that was not transported through the high voltage gridlinesmeasured by SEP. The TenneT dataset seems to be closing this‘gap’, possibly because the grid administrators currently reportingelectricity demand may include such data in their reports.

The inconsistency between the reported values in the SEPdataset and the TenneT dataset and their discrepancy comparedwith CBS total electricity consumption data complicates thecomparison of electricity load over the years. However, thedescribed changes in measurement methodology and alteredeconomic structure that result in this discrepancy are expected tohave resulted in long term rather than short-term data variance.Our analysis analyzes absolute values only within a one monthtimeframe. It therefore does not compare absolute values that arefar apart in time, nor values coming from different datasets. Onlyafter this initial processing of the data, the data are comparedwith data from other months, thus reducing the distortion arisingfrom using separate datasets.

3.2. Temperature data

Daily temperatures may vary throughout the country, mainlyrelated to the proximity of the North Sea and the prevailing winddirection. To account properly for these variations, our tempera-ture indicator would ideally average temperatures throughout thecountry, weighted by relative share in national electricityconsumption, which may be influenced by e.g. local population

1986

1988

1990

1992

1994

1996

1998

2000

2002

2004

2006

year

ble through the public grid"

TenneT dataset with CBS data for electricity available through the public grid.

Page 5: Indications for a changing electricity demand pattern: The temperature dependence of electricity demand in the Netherlands

ARTICLE IN PRESS

M. Hekkenberg et al. / Energy Policy 37 (2009) 1542–15511546

density and local economic structure. The Royal Dutch Meteor-ologic Institute (KNMI) publishes historical daily weather data forten different measurement stations on their website (KNMI,2008), but publishes no average value nor a methodology tocalculate this. Considering the required effort of determining eachmeasurement station’s relative weight compared to the expectedadded accuracy of the temperature indicator we decided to use

R2 = 0.3578

-6%

-4%

-2%

0%

2%

4%

6%

8%

10Daily average temperature °C

Dev

iatio

n fr

om m

onth

ly a

vera

geel

ectr

icity

dem

and

30252015

Fig. 4. Percentage deviation from average electricity demand on each working day

in June 1976 plotted against the average temperature on that day.

R2 = 0.8367

-6%

-4%

-2%

0%

2%

4%

6%

8%

10Daily average temperature °C

Dev

iatio

n fr

om m

onth

ly a

vera

geel

ectr

icity

dem

and

30252015

Fig. 5. Percentage deviation from average electricity demand on each working day

in June 1996 plotted against the average temperature on that day.

-1

-0.5

0

0.5

1

1.5

2

1970

Tem

pera

ture

dep

ende

nce

of e

lect

ricity

dem

and

(% p

er °C

)

May JunTrend May Trend Jun

1980

Fig. 6. Development of temperature dependence coefficients in May, Ju

only the data for the measurement station in De Bilt (station 260).De Bilt is a town centrally located in the Netherlands, and close tothe highly populated Randstad region; it is therefore believed tobe suitable as a proxy for population based average temperaturein the Netherlands.

4. Results

Figs. 4 and 5 show examples of two of the scatter plots we haveplotted for the 411 monthly datasets. They show the percentagedeviation from average monthly electricity demand on eachworking day in June 1976 and June 1996 plotted against theaverage day temperature on each of those working days. Eachgraph also shows the regression line found to best fit the observeddata, based on an ordinary least square linear regression. Theregression coefficient for the predictor (temperature) can beregarded as an indicator for the temperature dependence ofelectricity demand during that month. The coefficient measuresthe relative deviation of the average monthly electricity demandper degree of temperature change. Negative coefficients, visible asdownward sloping regression lines, indicate that in the observedmonth, a higher temperature correlates with a lower electricitydemand. Positive coefficients indicate that a higher temperaturecorrelates with higher electricity demand. As mentioned earlier,little value should be assigned to each individual graph, since eachseparate dataset is relatively vulnerable to random occurrence ofdistorting noise.

Figs. 4 and 5 return in miniature in Fig. A1, which shows all 411scatter plots for the 411 datasets. The combination of graphs inthis figure shows the full time series of the 411 months fromJanuary 1970 to December 2007 for which data was available. Italso shows the linear regression functions that best fit each of the411 separate data sets. However, Fig. A1 reveals a clear trend fromdownward sloping regression functions (negative coefficients) toupward sloping regression functions (positive coefficients) forseveral months throughout the time series. This trend is mostclearly visible for the months May and June, but is also found forSeptember and October. The months from November to Aprilshow mainly downwards sloping regression functions. Thesummer months July and August show a rather chaotic pattern,of alternating upwards and downwards sloping functions.

Further analysis of the temperature dependence coefficientsreveals that coefficients in May, June, September and October didnot only turn from negative to positive over the 37 year long

Year

Sep OctTrend Sep Trend Oct

201020001990

ne, September and October in the Netherlands from 1970 to 2007.

Page 6: Indications for a changing electricity demand pattern: The temperature dependence of electricity demand in the Netherlands

ARTICLE IN PRESS

Table 1Regression statistics for temperature dependence coefficient change from 1970 to 2007.

Month Coefficient change

(Pct/deg.)/yr

R2 Level of

significance

Month Coefficient change

(Pct/deg.)/yr

R2 Level of

significance

January �3.24�10�3 0.036 0.283 July 9.35�10�3 0.031 0.317

February �3.49�10�3 0.018 0.444 August 16.7�10�3 0.050 0.202

March 7.14�10�3 0.110 0.051 September 21.9�10�3 0.477 0.000

April 7.42�10�3 0.096 0.070 October 16.8�10�3 0.388 0.000

May 31.3�10�3 0.724 0.000 November �3.47�10�3 0.044 0.236

June 34.4�10�3 0.772 0.000 December �3.74�10�3 0.042 0.245

M. Hekkenberg et al. / Energy Policy 37 (2009) 1542–1551 1547

period, but show a clear progression over time (Fig. 6). Thisfinding means that electricity demand in May, June, Septemberand October is increasingly depending on temperature differ-ences, whereby higher temperatures lead to increased electricitydemand. The observed trend suggests that a temperaturedifference of 1 1C in May, June, or September would currentlyresult in a deviation of total Dutch electricity demand in thesemonths of over 0.5%.

Table 1 shows the statistics of the regression of the time seriesfor all months. Notice that the coefficient changes for May andJune are the largest, whereas the changes in September andOctober are more modest. Although the correlation in othermonths is relatively minor and not found to be significant atpo0.05, it is noteworthy that all months from March to Octobershow an increasing coefficient (positive coefficient change),whereas November to February show a decreasing coefficient(negative coefficient change).

5. Discussion

In this study we find significant changes in temperaturedependence in May, June, September and October. Our resultsindicate positive and increasing temperature dependence in thewarmer months of the year. This finding is consistent with ourhypothesis that increasing use of cooling applications would leadto an increasingly positive relation between temperature andelectricity demand, especially during the summer months. Severalpoints of discussion regarding these findings and their signifi-cance should be addressed.

Firstly, although our results suggest increasing temperaturedependence for several months, the lack of a significant trend inthe real summer months July and August is remarkable. In fact, ifin any month, the effect of increased use of cooling applicationsand air conditioning would be expected to be visible in thewarmest months first. This lack of a trend may be due to the factthat summer holidays distort the electricity demand pattern inthese months.

As noted in the methodology section, we have chosen not touse data correction methods in our study, except for the exclusionof non-working days. The chosen approach may be expected tolead to a high degree of variance in the data in the summermonths. On most days of the summer months, a certain butfluctuating percentage of the population is on holidays, whichgenerally leads to lower electricity demand because oflower economic activity. However, excluding holidays is difficultin these months because the timing of summer holidays isstaggered over the country and varies per year. Genericallyexcluding holidays would leave little data included for thesemonths. Since days with relatively high economic activity (onwhich a relatively large part of the population is working) may fallon relatively warm days one year and on relatively cool daysanother year, the variation in economic activity may obscure any

temperature dependence trend over the years. Thus, the effect offluctuating economic activity on electricity demand in July andAugust may be expected to be larger than the effect of fluctuatingtemperature. Consequently, although the temperature depen-dence trend is found to be positive for both July and August, thevariance of the deviation of this trend is too high to provestatistically significant.

However, notwithstanding the staggered timing of holidays,the end of July and the beginning of August may generally beconsidered holidays throughout the country. Therefore the work-ing days during this summer holiday period possibly can betreated as a set of days with a reasonably stable activity pattern,although markedly different from that in non-holiday-months.We have therefore re-analyzed the data in sets that cover thetimeframe from July 16 to August 15. We have excluded weekendssimilarly as in the other datasets. The analysis of thesedatasets shows a significant trend of 25.0�10�3 Pct/degree/year(p ¼ 0.009, R2

¼ 0.194) for the coefficient change over theinvestigated time period from 1970 to 2007. Thus, a significanttrend is found after all in the holiday summer months. However,the low R2 shows that outdoor temperature cannot explain muchof the volatility in electricity demand in this period, whichcorresponds with the hypothesis of a highly variable economicactivity described above.

Secondly, we may wonder why we find a significantly positivetrend in October, but not in April, even though days are shorter inOctober and both months are generally considered to have similarweather. A closer look at the data reveals that the averagetemperature in October during the studied period is 1.61 higherthan in April (Fig. 7), although the difference has decreased overthe decades we investigated. Thus the perception of similarweather is not correct. The temperature difference may explainthe difference in the observed trend, since a higher temperaturewould lead to higher cooling and lower heating activity inOctober. An anonymous reviewer suggested that behavioraldifferences in air conditioning use between the start and theend of the summer season could also contribute to the observeddifference. Furthermore, although the October trend is found to besignificant and the April trend is not, both trends’ low R2 (0.388for October, 0.096 for April) indicate a low predictive power. Thisobservation and the fact that the temperature dependencecoefficients in both months are relatively small, indicates thatthe temperature dependence in both April and October is arelatively inconclusive predictor for electricity demand. A repeti-tion of the analysis using maximum temperatures instead ofaverage temperatures supports this statement. In this additionalanalysis we generally found slightly different values for coeffi-cients, R2 and level of significance for all months, although thegeneral conclusions remain the same. In this analysis, the trendfor April and March becomes significant, although both still have avery low R2, thus indicating that little difference in fact existsbetween April and October. Although this analysis generates thesetwo additional significant results, we chose to describe our results

Page 7: Indications for a changing electricity demand pattern: The temperature dependence of electricity demand in the Netherlands

ARTICLE IN PRESS

0.0

2.0

4.0

6.0

8.0

10.0

12.0

14.0

16.0

18.0

20.0

Jan

Month of the year

Ave

rage

wor

king

day

tem

pera

ture

(°C

)

1970-1979 1980-1989 1990-1999 2000-2007

DecNovOctSepAugJulJunMayAprMarFeb

Fig. 7. Average daily temperature on the working days included in the study, divided per decade. Adapted from KNMI (2008).

Ele

ctric

ity D

eman

d

T1 T2

1

2

Fig. 8. The effect of a changing average monthly temperature on temperature

dependence. Since we define temperature dependence as the relation between

electricity demand and temperature, temperature dependence at any temperature

is represented in this graph by the tangent of the ‘u’ curve at that temperature. If

average monthly temperature changes from T1 to T2, temperature dependence is

thus expected to change from 1 to 2.

M. Hekkenberg et al. / Energy Policy 37 (2009) 1542–15511548

based on average temperatures, because most relevant literatureis based on average temperature. The additional significant resultsfor March and April were judged to be of lesser importance thanfollowing the existing literature.

Thirdly, the value of the dependence coefficient found in thisstudy suggests a current electricity demand variation of 0.5% oftotal demand per degree temperature change in May, June,September and the summer holidays. Since the value of thiscoefficient is defined by the observed variation and the absolutemonthly electricity demand, the weight of this percentagedepends on the absolute value of total Dutch electricity demand.Since total Dutch electricity demand is growing annually, agrowing percentage signals a bigger share of a bigger pie.However, because of this dependence on total demand, someremarks on the developments in electricity generation should bekept in mind. The electricity load measurements used in thisstudy are not synonymous to the total Dutch electricity demand.Increasingly, part of the demand is met by decentralized powergeneration, such as combined heat and power. Therefore thetiming of operation of these decentralized systems influences theimplications of our results. Decentralized systems that areoperated on the basis of electricity prices, may have a dimmingeffect on the results from our study. After all, whenelectricity demand rises due to temperature increase, electricityprices go up. This price hike would lead to more decentralizedsystems being operated, which decreases the need for centralizedproduction which is measured in our data. However, if decen-tralized systems are operated continuously, to reduce the priceper kWh (as suggested by Boonekamp and Van Hilten (1990)before the liberalization of the electricity market), our findingsmay slightly overestimate the actual effect on total demand.Unfortunately, we do not have enough insight in decentralizedelectricity production data to conclusively assess its impact on ourresults.

Fourthly, given the (relatively small) 4–5% estimated share ofcooling applications in the total electricity demand throughoutthe year, the finding of a 0.5%/degree temperature dependence ontotal electricity demand suggests a large temperature dependenceof cooling demand itself. However, the 0.5% degree is stillrelatively small compared to the temperature dependenceobserved in summer months in Spain, which we estimate in

the order of 1.5–2%/degree based on Spanish data (Pardo et al.,2002).

Fifthly, although the hypothesis of increasing use of coolingapplication would explain our findings, there is also an alternativeexplanation of our findings. As mentioned before, the relationbetween electricity demand and temperature is usually found tobe ‘u’ shaped (Nobel, 1996; Valor et al., 2001; Pardo et al., 2002;Moral-Carcedo and Vicens-Otero, 2005; Mirasgedis et al., 2007).This means that increasing average temperatures, would also leadto an increasing coefficient of temperature dependence in acertain month (see Fig. 8). Fig. 7 shows that indeed, averagemonthly temperatures in the last two decades were higher than inthe first two decades under investigation. However, regressionanalysis between observed average monthly temperature andobserved temperature dependence shows very low explanatorypower (R2 between 0.00 and 0.279). Even so, higher temperatures,especially uncomfortably high temperatures, may be a driver for

Page 8: Indications for a changing electricity demand pattern: The temperature dependence of electricity demand in the Netherlands

ARTICLE IN PRESS

M. Hekkenberg et al. / Energy Policy 37 (2009) 1542–1551 1549

the purchase of air conditioning and other cooling applications(Sailor and Pavlova, 2003). In this regard, higher temperaturesmay drive a socio-economic trend which may have a long lastingeffect on future electricity demand. Thus, although the increasingaverage monthly temperature may partly explain the observedfindings, it cannot account for all the observed development on itsown. Furthermore, average October temperatures in the lastdecade (11.4 1C) are still much lower than average June tempera-tures in the 1970s (15.3 1C), but the observed temperaturedependence coefficient is currently larger in October than in Junein the 1970s. Therefore, the observed trend in temperaturedependence requires an alternative explanation.

Lastly the dependence of total electricity demand on tempera-ture results from the total underlying socio-economic structure.Therefore, a decrease in the importance of negatively temperaturedependent processes, for instance a decreased importance ofheating due to better insulation or higher heating efficiency,may lead to an increasingly positive temperature dependence justas an increase of positively temperature dependent processeswould do. However, the electricity demand for heating in theNetherlands is limited mainly to the operation of electricalpumps, because Dutch heating systems are generally based onnatural gas. Moreover, if changes in heating demand would beresponsible for the increases in temperature dependence, weshould find these changes mainly in winter instead of in summermonths.

6. Implications of a changing temperature dependence

Given the dominance of winter electricity demand in Dutchelectricity pattern, an increasing temperature dependence insummer months may not be thought to lead to general capacityissues in the Netherlands. However, the 2003 ‘code red’ situationfor Dutch electricity supply has shown that high summertemperatures in combination with extreme drought may lead tocapacity problems (ECN, 2004). Because cooling water availabilitywas limited due to low water levels in rivers, and because surfacewater temperature limited the amount of cooling water allowedto be dumped, reserve capacity in the Netherlands dropped below700 MW, which resulted in extreme price spikes. An increasingtemperature dependence may lead to such situations to occurmore frequently. The expected temperature increase due to globalwarming may add to this increased frequency.

Apart from the effect of global warming on peak demand, itmay also have effects on the future monthly electricity pattern.Since the temperature dependence coefficients for May, June,September, October and the summer holidays have becomepositive, it may be expected that a temperature increase due toglobal warming will lead to higher electricity demand in thesemonths, thus possibly resulting in a (small) summer peak.Historically, the period of summer minimum was used asthe main time to service and revise electricity plants in theNetherlands (Dijk and Geerts, 1988). Since the liberalization of theelectricity market however, maintenance planning should beregarded at a larger geographical scale, as expected demand andavailable production capacity in the whole of Western Europe setsthe price for electricity, which may be expected to ultimately drivedecisions to schedule servicing and maintenance. Nonetheless,since similar developments may occur in the rest of Europe,increasing summer demand may lead to plant servicing in springand autumn, when prices are less volatile.

Moreover, additional capacity may be required to meet theadditional summer demand. It has not escaped our attention thatcooling demand in summer runs parallel with solar radiation andthus the capacity for solar electricity generation. Since conven-

tional power stations have the temperature related capacity issuesdescribed above, sometimes leading to price spikes, this renew-able electricity source may be a suitable alternative to meet thesummer cooling demand.

Furthermore, our findings may also have consequences for theexpected effects of global warming on future energy demand. Theincreasing temperature dependence found in this study may shiftthe balance between decreasing energy demand for heatingpurposes and increasing energy demand for cooling purposes asa result of global warming. Most studies investigating this relationassume a fixed balance temperature (Sailor and Munoz, 1997;Sailor, 2001; Amato et al., 2005; Crowley and Joutz, 2005; Hadleyet al., 2006; Ruth and Lin, 2006), mostly at 18 1C. The observedshift from negative to positive temperature dependence in thisstudy in months with average temperatures below this tempera-ture, suggests that this balance point may instead be variable intime. The assumptions on temperature dependence in thesestudies investigating this balance therefore need to be criticallyevaluated.

Lastly, many European countries have higher average summertemperatures than the Netherlands. The developments in thetemperature dependence of electricity demand that we found forthe Netherlands may be illustrative for other countries, in Europeand elsewhere. Interactions in the increasingly internationallyoriented electricity market make that these findings should alsobe regarded at a pan-European level.

7. Conclusion

In this study we have investigated the electricity demandpattern in the relatively temperate climate of the Netherlands forpossible changes related to the increased use of cooling applica-tions. We have found significant increases in temperaturedependence of electricity demand in May, June, September,October and during the summer holidays during the period1970–2007. This trend has resulted in an increasingly positivetemperature dependence during summer months, currentlyestimated around 0.5% of total electricity demand per degreetemperature difference. If this trend continues, it may result in asummer electricity demand peak in the Netherlands, as seen insouthern European countries, which may have important con-sequences for electricity generation capacity, maintenance sche-duling and electricity prices. This development may also alter theexpected effects of global warming on electricity demand,possibly shifting the balance between decreasing electricitydemand for heating and increasing electricity demand for coolingresulting from global warming.

Acknowledgements

The authors would like to acknowledge TSO TenneT b.v. forproviding the detailed electricity load datasets and StatisticsNetherlands for providing the historic data for total Dutchelectricity use. We would also like to thank Mr. Veldkamp(TenneT), Mr. Nobel (TenneT) and Mr. Kloots (Statistics Netherlands)for their interpretation of and discussion on the observed data.Finally, we wish to acknowledge the two anonymous reviewers fortheir constructive comments.

Appendix A

See Fig. A1.

Page 9: Indications for a changing electricity demand pattern: The temperature dependence of electricity demand in the Netherlands

ARTICLE IN PRESS

Fig. A1. Percentage deviation from average monthly electricity demand on each working day plotted against the average temperature on that day as in Figs. 4 and 5, for all

411 months from January 1970 to December 2007 for which data was available. Months are plotted from left (1 ¼ January) to right (12 ¼ December), years are plotted from

bottom to top. Following the months May and June throughout the time series reveals a clearly shifting regression line. Note that regression lines in each individual graph

are not tested for significance and generally have a low R2.

M. Hekkenberg et al. / Energy Policy 37 (2009) 1542–15511550

Page 10: Indications for a changing electricity demand pattern: The temperature dependence of electricity demand in the Netherlands

ARTICLE IN PRESS

M. Hekkenberg et al. / Energy Policy 37 (2009) 1542–1551 1551

References

Al-Zayer, J., Al-Ibrahim, A., 1996. Modelling the impact of temperature onelectricity consumption in the eastern province of Saudi Arabia. Journal ofForecasting 1597-106.

Amato, A., Ruth, M., Kirshen, P., Horwitz, J., 2005. Regional energy demandresponses to climate change: methodology and application to the common-wealth of Massachusetts. Climatic Change 71 (1), 175–201.

Bessec, M., Fouquau, J., 2008. The non-linear link between electricity consumptionand temperature in Europe: a threshold panel approach. Energy Economics 30(5), 2705–2721.

Boonekamp, P.G.M., Van Hilten, O., 1990. Toekomstig verbruikspatroon vanelektriciteit-mogelijke veranderingen en effecten (Future electricity consump-tion pattern—possible changes and effects). ECN, Petten.

CBS, 2008. Elektriciteit afzet binnenland 2003 (Domestic electricity supply 2003)./http://statline.cbs.nl/StatWeb/publication/?DM=SLNL&PA=37985&D1=33&D2=l&VW=TS.

Crowley, C., Joutz, F.L., 2005. Weather effects on electricity loads: modeling andforecasting. USEPA, Washington.

Dijk, D., Geerts, M., 1988. Report on the Power System Simulation Model SCELEC.Center for Energy and Environmental Sciences IVEM, Groningen.

ECN, 2004. Energie Verslag Nederland 2003 elektriciteits- en gasmarkt (Energyreport Netherlands 2003—electricity and gas market). ECN, Petten.

Eurostat, 2008. Monthly gross electricity consumption in Europe 1985–2006./http://epp.eurostat.ec.europa.eu/portal/page?_pageid=1073,46870091&_dad=portal&_schema=PORTAL&p_product_code=NRG_IND_342MS.

Franco, G., Sanstad, A., 2008. Climate change and electricity demand in California.Climatic Change 87 (0), 139–151.

Hadley, S.W., Erickson III, D.J., Luis Hernandez, J., Broniak, C.T., Blasing, T.J., 2006.Responses of energy use to climate change: a climate modeling study.Geophysical Research Letters 33 (L17703).

Isaac, M., Van Vuuren, D., 2009. Modeling global residential sector energy demandfor heating and air conditioning in the context of climate change. Energy Policy(forthcoming).

KNMI, 2008. Daily weather data of the Netherlands. /http://www.knmi.nl/klimatologie/daggegevens/download.cgi?language=engS.

Mirasgedis, S., Sarafidis, Y., Georgopoulou, E., Kotroni, V., Lagouvardos, K., Lalas,D.P., 2007. Modeling framework for estimating impacts of climate change on

electricity demand at regional level: case of Greece. Energy Conversion andManagement 48 (5), 1737–1750.

Moral-Carcedo, J., Vicens-Otero, J., 2005. Modelling the non-linear response ofSpanish electricity demand to temperature variations. Energy Economics 27(3), 477–494.

Nobel, F., 1996. Koude en zwoele nachten laten minimale belasting stijgen(Cold and sultry nights make minimal load rise). INspanning (April 1996)22-23.

Pardo, A., Meneu, V., Valor, E., 2002. Temperature and seasonality influences onSpanish electricity load. Energy Economics 24 (1), 55–70.

Pennartz, A.M.G., 2005. Onderzoek naar het elektrisch energieverbruik vankoelinstallaties in Nederland-concept (Study of the electric energy consump-tion of cooling applications in the Netherlands). KWA bedrijfsadviseurs,Amersfoort.

Ruth, M., Lin, A.C., 2006. Regional energy demand and adaptations to climatechange: methodology and application to the state of Maryland, USA. EnergyPolicy 34 (17), 2820–2833.

Sailor, D.J., Pavlova, A.A., 2003. Air conditioning market saturation and long-termresponse of residential cooling energy demand to climate change. Energy 28(9), 941–951.

Sailor, D.J., 2001. Relating residential and commercial sector electricity loads toclimate—evaluating state level sensitivities and vulnerabilities. Energy 26 (7),645–657.

Sailor, D.J., Munoz, J.R., 1997. Sensitivity of electricity and natural gas consumptionto climate in the USA—Methodology and results for eight states. Energy 22(10), 987–998.

TenneT, 2008. Measured feed to the Dutch national grid. /http://www.tennet.org/english/operational_management/System_data_relating_processing/measured_feed.aspxS.

Valor, E., Meneu, V., Caselles, V., 2001. Daily Air temperature and electricity load inSpain. Journal of Applied Meteorology 40 (8), 1413–1421.

Van Arkel, W.G., Jeeninga, H., Menkveld, M., Ruig, R.J., 1999. Energieverbruik vanebouwgebonden energiefuncties in woningen en utiliteitsgebouwen (Energyuse of energy functions in houses and buildings). ECN, Petten.

Wangpattarapong, K., Maneewan, S., Ketjoy, N., Rakwichian, W., 2008. Theimpacts of climatic and economic factors on residential electricity consump-tion of Bangkok Metropolis. Energy and Buildings 40 (8), 1419–1425.