2013 Weather Normalization Survey
Itron, Inc. 11236 El Camino Real
San Diego, CA 92130‐2650 858‐724‐2620
March 2014
Page 1
2013WeatherNormalizationSurvey Weather normalization is the process of reconstructing historical energy consumption assuming that normal weather occurred instead of actual weather. The process contains two key assumptions. First, a model is used to identify the weather response and calculate the difference between energy consumption under normal and actual weather conditions. Second, normal weather is defined and constructed to represent typical weather conditions. In November 2013, Itron conducted a survey of North American energy forecasters to understand and document the current practices in weather normalization. The survey asked three types of questions. The first set of questions was used to identify the respondents and the application of their weather normalization process. The second set of questions was asked to gain insights into their modeling assumptions. The final set of questions was asked to understand their definition of normal weather.
IdentificationQuestionsQuestions 1 through 8 The Survey includes responses from 135 companies across North America. These companies are separated into categories based on a self‐reporting question and company identification. Figure 1 and
Figure 2 s Figure 1:
how the relat
Survey Resp
20
tive size of ea
ondents
013 Weather
ach category.
r Normalizat
Page 2
tion Survey
2013 Weather Normalization Survey
Page 3
Figure 2: Survey Respondents by Size and Classification
CategoryDefinitionsThe categories used are defined as follows. Distribution. Distribution companies include both gas and electric companies that deliver
service to an end‐use customer. While these companies may include transmission and generation components, these components are not necessary for including a company into this category. Within this category, seven (7) respondents are gas only companies.
Combined Gas & Electric. These companies include both natural gas and electric distribution systems.
Retail. Retail companies are non‐regulated electric or gas companies serving either retail or wholesale customers.
ISO. Independent System Operators (ISOs) are regional organizations responsible for dispatching the electric grid and moving electricity throughout a region.
G&T. Generation and Transmission (G&T) companies maintain generation and transmission functions, but do not deliver energy to the end‐use customer. Instead, these companies deliver energy at the wholesale level.
Generation. Generation companies own power plants and do not deliver energy to end‐use customers.
Transmission. The primary business of a transmission company is to transmit energy from generators to wholesale customers.
Other. The Other category includes companies that do not fit the definitions provided in the previous categories, but still perform a weather normalization function.
The Distribution and Combined Gas & Electric categories represent final deliveries to end‐use customers. These companies account for approximately 55% of all electricity sold in the United States and Canada.
WeatherNormalizationPurposesThe 135 companies reported multiple uses for weather normalization as shown in Figure 3. While forecasting is the most common application, variance analysis, financial reporting, and rate cases are also extremely common.
Figure 3:
Category Fo V
ti Fi
sa R
re P
tr O
co
ModelQuestionsThe first aresponse questionsprocess, a
WeatheFigure 4 sresponsesweather rtotal systerespondeirrigation,
Purpose of W
Definitions.
orecasting. FVariance Analyme frame to inancial Repoales for budgeate Case Anaegulatory envlanning. Planransmission pOther. Other ompanies tha
Questions 9 through 2assumption inand calculates are used to iand the weath
erNormalhows that ths), closely follresponsive anem loads are nts normalize, wholesale, a
20
Weather Norm
The categori
Forecasting apysis. Variancunderstand dorting. Financet analysis. alysis. Rate cavironment. nning includeplanning. includes respat do not perf
ns22 n weather nore the impact oidentify the cher drivers in
izationClae most commlowed by the nd contributeweather norme the industriand farm class
013 Weather
malization
es presented
pplies normae analysis apdifferences becial reporting
ase analysis u
s applications
ponses that doform any wea
rmalization isof normal weclasses being ncluded in the
assesmon class for wcommercial cto the majormalized by 74al class. The ses.
r Normalizat
Page 4
in Figure 3 a
l weather to aplies the weaetween an org uses weathe
uses weather
s in price fore
o not fit the pather normali
s the model ueather companormalized, te model.
weather normclass (95 resprity of a syste4 and 68 respother class in
tion Survey
re defined be
a model in a father normaliriginal forecaser normalizat
normalizatio
ecasting, distr
previously defization proces
used to identired to actualthe frequency
malization is tponses). Thesm’s weather pondents, resncludes respo
elow.
future time hzation processt and actual ion to unders
on for setting
ribution plan
fined categorss.
fy the histori weather. Thy of the mode
the residentiase two classeresponse. Sypectively. Ononses for gove
horizon. ss to a historiresults. stand and pro
rates in a
ning, and
ries, as well a
cal weather he model el estimation
al class (99 s tend to be hystem peaks anly 54 ernment,
ical
oject
s
highly and
Figure 4:
DataFrData freqused in darespondeRespondeapplicatiothe mont Figure 5:
Weather No
requencyuency indicataily models annts to this quents that use ons. The neithhly or daily le
Data Freque
20
ormalization C
tes the periodnd monthly duestion. In thboth monthlyher response evel.
ency of the M
013 Weather
Classes
dicity of the wdata are used ese results, 6y and daily daincludes resp
Model
r Normalizat
Page 5
weather normin monthly m
63% use montata indicate apondents who
tion Survey
malization momodels. Figurthly data and mix of modeo do not perf
odels. Typicalre 5 shows th 7% use dailyel periodicitieform weather
lly, daily data e results fromy data. es and r normalizatio
are m 132
on at
FrequeWeather Figure 6 srespondemodels evneeded” b Figure 6:
ModelDBecause awere askecategorie
ncyofMoNormalizatiohows that 63nts refresh thvery one to fibasis.
Frequency o
Descriptioa model is useed to understs of weather
20
delParamn models are3%, or 124 resheir models mve years. On
of Model Para
onsed to obtain tand the weatvariables for
013 Weather
meterUpda periodically sponses to thmultiple timesnly 3.2% of re
ameter Updat
the weather rther variableseach class. T
r Normalizat
Page 6
atesrefreshed to is question, rs during the yespondents in
tes
response of es used in the The variable c
tion Survey
capture chanrefresh their myear, and 10.4ndicate that m
energy consummodel. The ccategories are
nges in weathmodel every y4% of respondmodels are ref
mption, a sercompiled resue defined in F
her responsesyear. 23.4% odents refreshfreshed on an
ies of questioults identify Figure 7 and
s. of h their n “as
ons
2013 Weather Normalization Survey
Page 7
Figure 8. The remainder of this section describes the models used for the system, residential, commercial, and industrial classes. Figure 7: Heating Variable Category Definitions
Heating Variable Category
Description
HDD Model includes heating degree day (HDD) and/or HDD spline variables. No other weather variables are used.
Interactions Model interacts HDD or HDD splines with another variable. Model may include HDD or HDD spline variables separately.
Other Model includes additional weather variables beyond HDD or HDD splines. However, no interactions with HDD or HDD splines are included.
HDD/Int/Oth Model includes HDD or HDD splines, interactions, and additional weather variables.
None Model is not used to normalize for cold weather.
2013 Weather Normalization Survey
Page 8
Figure 8: Cooling Variable Category Definitions
Cooling Variable Category
Description
CDD Model includes cooling degree day (CDD) and/or CDD spline variables. No other weather variables are used.
Interactions Model interacts CDD or CDD splines with another variable. Model may include CDD or CDD spline variables separately.
Other Model includes additional weather variables beyond CDD or CDD splines. However, no interactions with CDD or CDD splines are included.
CDD/Int/Oth Model includes CDD or CDD splines, interactions, and additional weather variables.
THI Model uses THI (temperature‐humidity index) instead of CDD and may include interactions and additional weather variables.
None Model is not used to normalize for hot weather.
SystemModelDescriptionThe weather variables used to capture the heating and cooling effects in a system model are shown in Figure 9. These responses are based on the definitions from Figure 7 and
Figure 8. and CDD f Figure 9:
Additionaheating efby respon
Of the 68 resfor cooling (2
System Mod
al variables arffect, and 20%ndents are sho
20
spondents no6%).
del Descriptio
re used in som% of the respown in
013 Weather
ormalizing sys
on
me system moondents use t
r Normalizat
Page 9
stem loads, m
odels. 22% othem to capt
tion Survey
most utilities u
f respondentture the cooli
use only HDD
ts use them tong effect. Th
for heating (
o capture thehe variables lis
46%)
e sted
2013 Weather Normalization Survey
Page 10
Figure 10 with the number of responses shown in parenthesis.
2013 Weather Normalization Survey
Page 11
Figure 10: System Other Variables
Other Heating Variables Other Cooling Variables
Wind (6) Cloud Cover (5) Lag Weather (3) Dew Point/Humidity (2) Effective Temperature (1) High/Low Temperature Spread(1) Precipitation (1)
Dew Point/Humidity (8) Wind (5) Cloud Cover (4) High Temperature (3) Precipitation (3) High/Low Temperature Spread (1) Lag Weather (1)
Interactive variables allow for the heating and cooling response to change under specific conditions. 16% of the responses use interactions in the heating effect, and 18% of the responses use interactions for the cooing effect. The interacted variables listed by respondents are shown in Figure 11 with the number of responses shown in parenthesis. The primary interaction is daytypes, which includes daily, monthly, and seasonal binary variables. Figure 11: System Interactive Variables
Heating Interactions Cooling Interactions
Daytypes (9) End Use Trend (2) Economic Trend (1) Lag Temperatures (1) Deviations from Normal (1) Peak Temperature (1)
Daytypes (11) End Use Trend (3) Economic Trend (1) Hours of Light (1) Peak Temperature (1)
ResidentialModelDescriptionThe weather variables used to capture the heating and cooling effects in a residential model are shown in Figure 12. These responses are based on the definitions from Figure 7 and
Figure 8. heating (5 Figure 12
Other varvariables parenthescommon. Figure 13
Wind (5) Cloud CovHeating DLag WeatDew PoinHigh/LowPrecipitat
Interactivdominant Figure 14
Daytypes End Use TEconomicCustomerDaylight H
CommerThe weathshown in
Of the 99 res55%) and CDD
: Residential
riables are uselisted by respsis. Among o
: Residential
Other He
ver (5) Degree Hour (her (2) t/Humidity (2
w Temperaturetion (1)
ve variables art interaction i
: Residential
Heatin
(12) Trend (6) c Trend (2) r Counts (2) Hours (1)
rcialModelDher variables Figure 15. Th
20
spondents noD for cooling (
l Model Desc
ed by 12% of pondents are ther variable
l Other Varia
eating Variab
3)
2) e Spread(1)
re used by 21s with daytyp
l Interactive V
ng Interaction
Descriptionused to capthese response
013 Weather
ormalizing res(43%).
ription
respondentsshown in Figs used, wind,
bles
bles
1% of respondpe binary vari
Variables
ns
ure the heaties are based
r Normalizat
Page 12
sidential cons
to capture bure 13 with t, cloud cover
Dew PWind (CoolinLag WPrecipCloud High/L
dents for heatables as show
DaytypEnd UsEconoDayligCustom
ng and coolinon the defini
tion Survey
umption, mo
oth heating ahe number oand dew poin
Other
Point/Humidit(4) ng Degree Houeather (2) pitation (2) Cover (1) Low Tempera
ting and 18%wn in Figure 1
Coo
pes (14) se Trend (4)mic Trend (3)ht Hours (1)mer Counts (1
ng effects in ttions from Fi
ost utilities us
and cooling ref responses snt/humidity a
r Cooling Vari
ty (5)
ur (2)
ature Spread (
% for cooling e14.
ling Interacti
)
1)
the commercgure 7 and
e only HDD fo
esponses. Thshown in are the most
iables
(1)
effects. The
ons
ial model are
or
he
e
Figure 8. heating (5 Figure 15
Some resplisted by tAmong ot Figure 16
Wind (7) Cloud CovDew PoinHeating DHigh/LowLag WeatPrecipitat
The interaand syste Figure 17
Daytypes EconomicEnd Use TCustomerDaylight H
IndustriaThe weathin Figure 1
Of the 95 res56%) and CDD
: Commercia
pondents usethese respondther variables
: Commercia
Other He
ver (5) t/Humidity (3Degree Hour (w Temperatureher (1) tion (1)
active variablm models, th
: Commercia
Heatin
(11) c Trend (2) Trend (1) r Counts (1) Hours (1)
alModelDeher variables 18. The respo
20
spondents noD for cooling (
al Model Desc
e other variabdents are shos used, wind,
al Other Varia
eating Variab
3) 1) e Spread(1)
es used in thehe main categ
al Interactive
ng Interaction
escriptionused to captonses are bas
013 Weather
ormalizing com(44%).
cription
bles to capturown in Figure cloud cover a
ables
bles
e commerciagory of interac
Variables
ns
ure the heatised on the de
r Normalizat
Page 13
mmercial con
re both heatin16 with the nand dew poin
Wind (Dew PPrecipCloud CoolinDayligHigh/LLag W
l models are sctions is the d
DaytypEconoCustomEnd UsDay Lig
ng and coolinefinitions from
tion Survey
nsumption, m
ng and coolinnumber of rent/humidity a
Other
(4) Point/Humiditpitation (2) Cover (1)
ng Degree Houht Hours (1)Low Temperaeather (1)
shown in Figudaytype varia
Coo
pes (12) mic Trend (2)mer Counts (2se Trend (1)ght Hours (1)
ng effects in tm Figure 7 an
ost utilities u
g responses. sponses showre the most c
r Cooling Vari
ty (4)
ur (1)
ature Spread (
ure 17. As wiable.
ling Interacti
) 2)
)
the industrial d
use only HDD
The variablewn in parenthcommon.
iables
(1)
ith the reside
ons
model are sh
for
es hesis.
ential
hown
Figure 8. heating (4 Figure 18
The otherresponsesreported Figure 19
Wind (3) Cloud CovDew Poin
Figure 20
Daytypes Economic
TemperHDD and referenceresponse.referencethe tempemultiple c Figure 21
Of the 54 res46%) and CDD
: Industrial M
r interactive vs are shown ispecific other
: Industrial O
Other He
ver (3) t/Humidity (1
: Industrial I
Heatin
(4) c Trend (1)
ratureCutCDD are calcue point. Regre. A single cute point. Multierature referecut points to c
: Heating an
20
spondents noD for cooling (
Model Descri
variables usedn Figure 19 ar and interact
Other Variabl
eating Variab
1)
nteractive Va
ng Interaction
tPointsulated as the ession model point variabliple cut pointence point. Fcapture the h
d Cooling De
013 Weather
ormalizing Ind(41%).
ption
d by some resnd Figure 20.tive variables
les
bles
ariables
ns
difference bes use these vle is used whet variables areFigure 21 shoheating and co
gree Day Cut
r Normalizat
Page 14
dustrial consu
spondents to In both cates.
Wind (Dew PPrecipCloud
DaytypEcono
etween the aariables to caen assuming e used when aws the perceooling respon
t Points
tion Survey
umption, mos
capture bothegories, a low
Other
(2) Point/Humiditpitation (1) Cover (1)
Coo
pes (4) mic Trend (1)
ctual temperapture the noa linear respoassuming a cntage of respnse.
st utilities use
h heating andw number of r
r Cooling Vari
ty (2)
ling Interacti
)
rature and a ton‐linear heatonse from thehanging lineapondents that
e only HDD fo
d cooling respondents
iables
ons
temperature ting and coolie temperaturar response frt use single ve
r
ing re rom ersus
20
013 Weatherr Normalizat
Page 15
tion Survey
2013 Weather Normalization Survey
Page 16
TemperatureHumidityIndexCalculationA Temperature Humidity Index (THI) is used to combine temperature and humidity into a single numerical value that captures the effects of moisture in the air. Recently, utilities have reported a wide variety of mathematical calculations to capture this effect. This survey allowed for respondents to define their index calculations. Of the 13 responses to this question, four distinct equations were provided. These four equations capture the interaction between dry bulb temperatures (T) and moisture in the form of dew point (DP) or relative humidity (RH). The equations are shown below.
Index = 0.55 * T + 0.20 * DP + 17.50 Index = T ‐ (0.55 ‐ 0.55*RH/100) * (T ‐ 58) Index = ‐42.379 + ((2.04901523*T) + (10.14333127*RH)) ‐ (0.22475541*T*RH) ‐ (0.00683783 * (T2)) ‐ (0.05481717 * (RH2)) + (0.00122874 * (T2) * RH) + (0.00085282 * T * (RH2)) ‐ (0.00000199 * (T2) * (RH2)) Index = 16.923 + ((1.85212 * 10‐1) * T) + (5.37941 * RH) ‐ ((1.00254 * 10‐1) * T * RH)
+ ((9.41695 * 10‐3) * T2) + ((7.28898 * 10‐3) * RH2) + ((3.45372 x 10‐4) * T2 * RH) ‐ ((8.14971 * 10‐4) * T * RH2) + ((1.02102 * 10‐5) * T2 * RH2) ‐ ((3.8646 * 10‐5) * T3) + ((2.91583 * 10‐5) * RH3) + ((1.42721 * 10‐6) * T3 * RH) + ((1.97483 * 10‐7) * T * RH3) ‐ ((2.18429 * 10‐8) * T3 * RH2) + ((8.43296 * 10‐10) * T2 * RH3) ‐ ((4.81975 * 10‐11) * T3 * RH3)
NormaQuestionsThe seconrepresentcan impacof questioa similar wsurvey.
NumbeFigure 22 responsesdata. Thissurvey. Tincrease i Figure 22
ChanginChanging handling cnumber oThese res Figure 23
Upd
Response
Changed
Has Not C
alWeathes 23 through nd assumptiots an expectect the averageons to undersweather norm
rofYearsshows the nus. In 2013, 33s response cohe largest chn percentage
: Number of
ngtheNumthe number climate changof years recenults are show
: Recent Cha
date Frequen
s
Recently
Changed
20
erQuestio30 n in weather d weather coe calculation stand the commalization sur
intheNorumber of yea3% of the 126ompares to 43anges betweee using 10 yea
Years in the
mberofYeof years usedge. In 2006, 2ntly. In 2013, wn in Figure 2
anges to the N
ncy 2
013 Weather
ons
normalizatioondition and iincluding the
mmon practicervey. Several
rmalCalcurs used to ca6 respondents3% using 30 yen 2006 and ars.
Normal Calcu
earsd in the norma25% of surveythe same qu3.
Number of Ye
2013 Survey
125
32%
68%
r Normalizat
Page 17
n is the defins typically repe number andes in calculatof the topics
ulationlculate normas define weatyear averages2013 are red
ulation
al weather cay respondentestion was as
ears
2006 Sur
115
25%
75%
tion Survey
nition of normpresented byd range of yeaing the averas show compa
al weather cother based ons from the 10uction in the
alculation hasts indicated thsked with 32%
rvey
mal weather. y an average. ars. This survages. In 2006arative result
ompared to thn 30 years of 6 responses ipercent usin
s been a commhat they had % indicating a
Normal weat Multiple factvey asked a se, Itron condus with the 20
he 2006 survehistorical wein the 2006 g 30 years an
mon techniquchanged the a recent chan
ther tors eries cted 006
ey ather
nd the
ue for
ge.
Figure 24 respondepreviouslydependin Figure 24
FrequeEach yearFigure 25 from the
shows the rents who havey used. The og on the purp
: How the Nu
ncyofNorr, the availabishows that 82006 Survey.
20
esults of a folle changed recther responspose of the w
umber of Yea
rmalCalculity of new w81% of respon
013 Weather
low‐up questcently, 58% ues indicated c
weather norm
ars Has Chang
ulationsUpeather data cndents update
r Normalizat
Page 18
ion asking hose fewer yearchanges that alization proc
ged
pdatecreates the ope their norma
tion Survey
ow the numbers while 32% use multiple cess.
pportunity toal weather ea
er of years hause more yeadefinitions fo
o recalculate nach year comp
as changed. Oars than or normal we
normal weathpared to 69%
Of the
eather
her. %
2013 Weather Normalization Survey
Page 19
Figure 26 displays the last year of data included in the normal calculation. In this figure, 83% of the respondents include data from 2011, 2012, and 2013 in their calculation. Figure 25: Update Normal Weather Annually
Update Frequency 2013 Survey 2006 Survey
Responses 124 114
Update Annually 81% 69%
Do Not Update Annually 19% 31%
Figure 26
OversigBecause ninvolved iwhose noresponses Figure 27
Upd
Response
Regulator
No Regula
ClimateWhile maweather cchanging
: Last Year o
ghtofRegunormal weathn overseeingormal weathes obtained in
: Normal We
date Frequen
s
ry Oversight
atory Oversig
eChangeny utilities mcalculation, ththe number o
20
f Normal Calc
ulatorsher can impac the normal wr calculation the 2006 sur
eather Calcula
ncy 2
ht
anage climathe survey reqof years.
013 Weather
culation Peri
ct forecasts, pweather calcuis overseen avey.
ation Specifie
2013 Survey
123
16%
84%
e change effequested inform
r Normalizat
Page 20
od
planning studulation. Figura regulatory e
ed by Regulat
2006 Sur
166
13%
87%
ects by changmation about
tion Survey
ies, and ratesre 27 shows tentity. The pe
tors
rvey
ging the numbt climate chan
s, regulatory the number oercentage is s
ber of years unge adjustme
entities may of respondentsimilar to the
used in the noent beyond
be ts
ormal
2013 Weather Normalization Survey
Page 21
Figure 28 shows that 9% of respondents use a method for climate change beyond controlling the number of years
2013 Weather Normalization Survey
Page 22
Figure 28: Account for Climate Change
Update Frequency 2013 Survey
Responses 124
Account for Climate Change 9%
Do Not Account for Climate Change 91%
NormalPeakWeatherNormal peak weather is used to normalize peak weather events. Two types of normal calculations are typically used in the normal peak weather calculation. These calculations are defined below.
Peak Day Weather. Peak day weather is defined as the weather conditions on the peak day only. After identifying these days, the temperatures (or HDD and CDD values) are averaged across these historical events.
High or Low. High or low weather is defined by identifying the highest and lowest historic temperatures in a month and averaging across these events regardless of when the monthly peak event occurred. The High and Low weather may have occurred on a weekend and did not cause the highest load event in the month.
2013 Weather Normalization Survey
Page 23
Figure 29 shows the results from 96 responses to this question. In this figure, 61% of respondents use the peak day weather approach. The other responses include different methods reported by respondents. These methods are listed below with the number of respondents include in parenthesis.
Temperature on Peak Hour (4) High Temperature Variations such as THI or a heat index (3) Rank and Average (3) Load Factor Method (2) Current and Preceding Day (1) Probability Distribution (1) Cold Snap Duration (1) Other (6)
Figure 29
SummaIn Novemdocumenttwo key acurrent m While thecommon C
to
Wvasudva
HC
Nav
: Normal Pea
aryber 2013, Itrot their currenssumptions –
models and no
e process of echaracteristic
lasses. Mosto be the most
Weather Variaariables. Howuch as wind saytype variabariations base
DD and CDD DD cut point
Normal Weathverages, but t
20
ak Weather
on conductednt weather no– a model andormal weathe
ach companycs are observe
companies nt weather sen
ables. When wever, severapeed, cloud cbles are also ced on weekda
Definition. Wto capture th
her Calculatiothere is a tran
013 Weather
d this survey oormalization pd normal weaer definitions
y contains vared through th
normalize thensitive and re
normalizing aal responses scover, dew pocommon becaays, months,
When defininhe non‐linear
on. The normnsition to usin
r Normalizat
Page 24
of North Amepractices. Thather. This suused by 135 c
riations basedhis survey. Th
e residential apresent the m
a class, most show a signifioint, and humause they capand seasons.
ng HDD and Cweather‐con
mal weather cng shorter av
tion Survey
erican energyhe weather nourvey capturecompanies.
d on their cushese characte
and commercmajority of im
models are dcant interest midity. Additipture heating
DD, most comnsumption res
calculation is sverages.
y forecasters tormalization es the charact
stomer base aeristics are su
ial classes. Tmpacts due to
driven by HDDin other weaonally, interag and cooling
mpanies use asponses.
still dominate
to understanprocess incluteristics of the
and needs, a ummarized be
hese classes to weather.
D and CDD ather variableactions with response
a single HDD
ed by 30 year
d and udes e
few elow.
tend
es
and
r
2013 Weather Normalization Survey
Page 25
Normal Weather Updates. Most companies update the normal weather calculation each year to remain current with the latest weather information
Weather normalization continues to be a major task for companies as seen by the strong response to the well‐defined applications in forecasting, variance analysis, financial reporting, and rate cases.