ArticlePDF Available

A Reanalysis of Long-Term Surface Air Temperature Trends in New Zealand


Abstract and Figures

Detecting trends in climate is important in assessments of global change based on regional long-term data. Equally important is the reliability of the results that are widely used as a major input for a large number of societal design and planning purposes. New Zealand provides a rare long temperature time series in the Southern Hemisphere, and it is one of the longest continuous climate series available in the Southern Hemisphere Pacific. It is therefore important that this temperature dataset meets the highest quality control standards. New Zealand’s national record for the period 1909 to 2009 is analysed and the data homogenized. Current New Zealand century-long climatology based on 1981 methods produces a trend of 0.91 °C per century. Our analysis, which uses updated measurement techniques and corrects for shelter-contaminated data, produces a trend of 0.28 °C per century.
Content may be subject to copyright.
A Reanalysis of Long-Term Surface Air Temperature Trends
in New Zealand
C. R. de Freitas &M. O. Dedekind &B. E. Brill
Received: 29 December 2013 /Accepted: 7 October 2014
#Springer International Publishing Switzerland 2014
Abstract Detecting trends in climate is important in assessments
of global change based on regional long-term data. Equally im-
portant is the reliability of the results that are widely used as a
major input for a large number of societal design and planning
purposes. New Zealand provides a rare long temperature time
series in the Southern Hemisphere, and it is one of the longest
continuous climate series available in the Southern Hemisphere
Pacific. It is therefore important that this temperature dataset meets
the highest quality control standards. New Zealandsnational
record for the period 1909 to 2009 is analysed and the data
homogenized. Current New Zealand century-long climatology
based on 1981 methods produces a trend of 0.91 °C per century.
Our analysis, which uses updated measurement techniques and
corrects for shelter-contaminated data, produces a trend of 0.28 °C
per century.
Keywords Data quality control .Climate change .
Tem pe ra ture time serie s .New Zealand
1 Introduction
Although many studies have assessed the quality of national
climate datasets for use in long-term climate change detection
[25,6,7,12,19,21,24], a homogenized New Zealand national
temperature record has only once appeared in the literature [18].
This work did not set out a schedule of adjustments and was based
on a measurement technique that was significantly improved by its
author [17] over a decade later. Applying that improvement could
have a significant effect on trends, but this has never previously
been published. The aim here is to apply the method set out by
[17] (i.e. Rhoades and Salinger, 1993) exactly as they describe,
without adjusting it in any way. In our analysis, we supply the
missing schedule of adjustments, recalculated to reflect the im-
proved technique. We also correct for the contamination of raw
data identified in the refereed literature [10]. The aim is to derive a
modernized New Zealand Temperature Record (NZTR) providing
a 100-year time series of mean monthly land surface temperature
2 Background
New Zealand was one of the first countries in the Southern
Hemisphere to establish an official nationwide system of weather
records. These records provide a rare long time series for temper-
atures in the Pacific Ocean, informing the data sparse interpola-
tions required for early temperature series. Extant 1868 archives
record the national normal mean surface temperature at 13.1 °C
(when converted from degrees Fahrenheit) being the average of
10+years read at six representative weather stations. Another
major compilation, covering 35 years and based on nine stations,
was published by the Dominion Meteorologist in 1920, which
showed that the countrys average temperature has remained
remarkably stable since records began. In 2010, the National
Institute for Water and Atmospheric Research (NIWA) assessed
the current national normal at 12.74 °C being the average of
30 years read at seven stations. On the face of it, New Zealands
long-term mean temperature has remained relatively stable at
12.6 °C over the past 150 years.
In 1980, M.J. Salinger published a homogenized Com-
posite New Zealand Temperature Series[18]. Cluster analysis
C. R. de Freitas (*)
School of Environment, University of Auckland, Auckland, New
M. O. Dedekind
Research & Development, BCD Consulting, Auckland, New
B. E. Brill
P O. Box 399, Paihia, New Zealand
Environ Model Assess
DOI 10.1007/s10666-014-9429-z
was used to divide the country into six temperature response
areas, which were considered well represented by seven stations
(Auckland, Masterton, Wellington, Nelson, Hokitika, Lincoln
and Dunedin) selected for their long history, low errors and
relatively homogeneous data (see Figs. 1and 2). The degree of
temporal conformity of temperature fluctuations and trends be-
tween response areas was tested by correlating temperature data.
It was found that warming and cooling anomalies were generally
synchronous throughout New Zealand, facilitating the use of a
nationwide temperature curve. The paper notes that the raw data
were carefully adjusted for site changes and other disturbances,
but no detail is provided. The composite New Zealand Temper-
ature Record (NZTR or seven-station seriesor 7SS)thatwas
then produced is said to be highly correlated with shorter-term
area-weighted studies of New Zealand as a whole. The Salinger
[18] paper shows a warming trend of approximately 1 °C per
century during 18531975, noting that the last 25 years (1950
75) have been particularly warm.
In an annex to an unpublished 1981 doctoral thesis (New
Zealand Climate: The Instrumental Record)undertakenatVic-
toria University of Wellington, the author M.J. Salinger elabo-
rated on the adjustments he had used in homogenizing the 7SS
temperature data. They were based on the principle of neighbour
station comparisons, with an elastic definition of the term neigh-
bour, using comparisons of trend variances within the seven
station datasets. This was justified (at p.71) by noting: As there
are only a handful of stations operating for the early record,
comparisons are instead made with all available stations in
operation. The larger distances between the earlier sites mean
that comparisons are no longer neighbour station, but these
comparisons are better than none for assessing homogeneous
data series.It is also relevant that temperature fluctuations at all
Fig. 1 Location of seven climate
stations in New Zealand that have
long records, namely, Auckland,
Masterton, Wellington, Hokitika,
Nelson, Lincoln and Dunedin
C.R. de Freitas et al.
New Zealand stations are broadly synchronous. The method for
statistical measurement of adjustments was described in the
thesis annex, but calculations were omitted. We will hereafter
use the term S81to refer to the published 1980 paper [18]in
conjunction with the unpublished thesis elaborations.
James Hessell, senior climatologist at the New Zealand Me-
teorological Service, produced a contemporaneous peer
reviewed journal paper on the subject of New Zealand temper-
ature records [10]. He identified which weather stations were
reliable(i.e. unaffected by site changes) and found that those
stations showed no warming trend during the halfcentury from
1930 to 1979. The Hessell paper conducted wind-run tests and
urbanrural station contrasts to identify and exclude sites affected
by shelter or urban heat island (UHI) effects. In essence, he found
that any warming appearing from the 1930 to 1979 raw data was
attributable to non-meteorological site effects. Amongst the sta-
tions found to be unreliable were Auckland and Wellington,
which are components of Salingers 7SS. Notably, S81 had
conducted no tests for creepinginhomogeneities such as shel-
tering or UHI and consequently made no attempt to correct the
7SS for site effects.
In 1993, in collaboration with a statistician D. A. Rhoades,
Salinger published Adjustment of Temperature and Rainfall
Records for Site Changes[17]orRS93. This paper was
accepted locally as a seminal authority for the statistical tech-
niques to be used in measuring differences between the tem-
peratures of compared stations.
3 The New Zealand Temperature Record
In 1999, the 7SS was recognized by NIWA and posted on its
website in graphical form. Since that time, its derived 1 °C per
century trend has been used constantly in official government
publications. This usage drew attention to the fact that its high
warming trend was attributable to undisclosed adjustments
insofar as there was no record of what adjustments were made
or reference to existing material that might shed light on
alterations made. The question also arises as to why data
concerns raised in the undisputed findings of the two journal
papers [10,17] were not considered. NIWA scientists re-
analysed the 7SS in 2010 [25] and derived a Schedule of
Adjustments, showing the station time and value of some 35
adjustments. This work is the revised 7SS, referred to here-
after as M10. M10 uses the period 19092008, which
showed the same trend as the original 7SS, namely, 0.91 °C
per century. The reason for omitting the years prior to 1908
was that very large gaps in the available data failed to satisfy
modern quality standards.
In this paper, we address the mean temperature data for the
7SS stations in respect of the century-long period 19092008,
agreeing with NIWA that data gaps prior to 1909 reduce
reliability below acceptable levels. Our starting point is the
Schedule of Adjustments showing the 20 data adjustments
made by S81 during the 19091975 period. The dates and
circumstances of 7SS site changes were initially identified in
Fig. 2 Location of Auckland regional stations Whenuapai, Albert Park, Mangere, Auckland Aero and Ardmore
Reanalysis of Long-Term Air Temperature Trends in New Zealand
S81 through a detailed analysis of station histories and other
available metadata. Two well-documented post-1975 Auck-
land site changes are added. We use the same broad method-
ology as S81/M10 in differencing the target station data from
those of comparison stations before and after the dates of
inhomogeneities. The selection of comparison stations is nec-
essarily confined by the dearth of high quality data pre-World
War II, and we have used the same comparison stations as
M10 for all adjustments. In the result, our methodology and
data inputs wholly coincide with M10 except in two respects:
(a) the use of RS93 statistical techniques to measure differ-
ences, as opposed to S81/M1 measurement techniques
(Table 1), and (b) acceptance of the findings of Hessells
[10] paper regarding the contamination of Auckland and
Well in gto n raw d at a.
4 Rhoades and SalingerRS93
The Rhoades and Salinger [17] paper (RS93) deals with the
detection and elimination of temperature data movements that
are of non-meteorological origin. With respect to changes in the
environment adjacent to observation sites, the authors advise
that for studies of climate change, it is best to choose stations
that are unlikely to be affected by gradual changes in shading or
urbanisation([17], p. 899). However, sudden changes, such as
those caused by shifts in observation sites or replacement of
measuring devices, can be identified, measured and corrected
by the application of appropriate statistical methodology.
The work Rhoades and Salinger [17] are concerned with is
the measurement of site change effects when the nature and
times of changes are known a priori from station records or
other metadata. It describes a method for target stations where
neighbouring stations are available for comparisondefining
neighbouringas subject to similar local weather condi-
tions([17], p. 900). A different method is provided for
isolated stations, noting that it is more difficult to distinguish
site change effects from regional meteorological effects. The
authors warn that statistical tests for change usually
assume that successive observations are independent
identically distributed random variables, whereas the
complexity of weather systems may mean that those
assumptions are violated.
In dealing with a site change known a priori by way of
comparison with neighbouringstations, RS93 faces the
same measurement issues as did S81. Both papers consider
statistical differences between before-and-after temperature
averages at the target station and those at comparison stations.
In RS93, the measurement techniques for those differences
contrast with those applied in S81 in the following respects:
1. RS93 uses monthly data whilst S81 used annual data.
2. RS93 takes short before-and-after periods (e.g.
24 months), whilst S81 used long periods (e.g. 10 years
or more).
3. RS93 weights the averages of data from comparison
stations based on relative correlation coefficients, whist
S81 used unweighted averages.
4. RS93 excludes proposed adjustments that are not statisti-
cally significant at the 95 % level, whilst S81 did not
measure confidence intervals for adjustments.
In summary, the RS93 method is a pairwise station com-
parison technique to quantify the magnitude of a site move by
differencing data from two similar stations before and after a
known site change. To minimize the potential impact of
unrecognized long-term gradual inhomogeneities (such as
shelter growth or UHI) at one or both of the stations, RS93
reduces the length of the comparison period using monthly
data. Seasonal effects are eliminated by differencing each
series by corresponding months only, either 1 or 2 years
removed. A strong (fourth power) weighting is applied to
reduce the effects of poorly correlating stations while enhanc-
ing the influence of well-correlated stations. The RS93 meth-
od has been applied as exactly as possible, without attempting
to improve or adjust it, to determine what effect its application
has on the S81 seven-station series.
The use of monthly (if not daily) data and significance
testing [15] has become standard international practices,
reflected in recent papers such as [23]and[22]. The use of
monthly data offers more degrees of freedom, while signifi-
cance testing balances the risks between type I and type II
errors. Weighting the average outcome of comparisons is also
standard, although the drivers of that weighting may vary with
the circumstances. In the case of the 7SS, the reference sta-
tions are not neighbouringand are sometimes geographical-
ly distant. Using the heavy correlation weightings of RS93
Tabl e 1 Statistical techniques
employed in the S81 and RS93
S81 RS93
Data Annual means Monthly means
Length of comparison (±) Maximum period available Up to 36 months
Weighting Unweighted Correlation (power of 4)
Significance testing Untested 95 % confidence tests
Station selection Any No known site changes
C.R. de Freitas et al.
self-selects those reference stations with the highest relevance
for the strongest role.
The limited comparison period circumscribes the ever-
present risk of distortion by undocumented data inhomogenei-
ties, whether sudden or gradual, at a reference station. This risk
is greater when the raw data have not been screened or tested
for data irregularities or suspected site effects, as is the case with
the S81/M10 datasets. RS93 ([17], at p. 900) notes: The use of
monthly differences means that the t-statistic has relatively high
degrees of freedom even when computed from a short time
interval of only 1 or 2 years before and after the site change. The
period of comparison is kept relatively short in order to avoid
contamination by gradual effects, or sudden but unrecognized
effects, at one or more of the neighbouring stationsThe usual
concern to maximise the power of the test is balanced by an
opposing concern that the modelling assumptions are likely to
be more seriously invalidated as the period of comparison is
lengthened.We note that [14] (p. 1206) also used ±24-month
comparison periods by default for their algorithm based on
pairwise comparisons. RS93 illustrates its overall method in a
detailed example of adjustments for a known site change in
Christchurch using ±2-year periods ([17], p. 104108).
We regarded the RS93 method as superior to that of its
predecessor S81 for several reasons. Foremost among them is
that RS93 itself makes a compelling case for each of the
methods four characteristics. Secondly, the author common
to both papers, Salinger, clearly regarded his published 1993
version as an improvement upon his earlier unpublished at-
tempts. Thirdly, RS93 was published in the peer-reviewed
literature, chosen as one of 21 exemplars in the omnibus paper
by Peterson et al. [16] and cited with approval in the WMO
AdjustmentsManual of 2003. It has been the adjustment
method of choice in New Zealand climatology for two de-
cades, and we are unaware of any serious criticism or dispute
regarding it. All this is not to say that improvements could not
be found, particularly amongst modern automated homoge-
nizing methods. Rather, this seminal local paper provides an
obvious foundation for homogenizing any New Zealand tem-
perature series.
5.1 Description
Our method follows the RS93 neighbour comparison tech-
niques for estimating the effect of known site changes but
extends that approach to comparisons between well-correlated
distant stations. We de-trend the inhomogeneous section
where necessary by using the slope calculated from a refer-
ence time series. Note that we have not modified the RS93
method at all, and we follow the same process as laid out in the
worked example (section 2.4) in that paper.
Each of the nstations is denoted using the convention i=
0,1,2,,nwhere i=0is the candidatestation with the site
change. The other stations are the reference stations. We
denote x
as the average temperature measured in month
number tat station iwhere t=1,2, Assume a station site
change occurred at time τand that τfalls on the first day of the
month. This is a valid assumption for most planned site
changes in New Zealand. First, the difference series y
calculated, for 12-month (k=1) and 24-month (k=2) cases;
that is:
τþt12kwhere ¼1;2;;12k
In other words, if the station change occurred at the end of
December 1975, y
for any station (when k=1) is the January
1976 temperature minus the January 1975 temperature. The
term y
is February 1976 minus February 1975, up to Decem-
ber 1976 minus December 1975. For k=2, y
is January 1976
minus January 1974, y
is February 1976 minus February
1974, up to December 1976 minus December 1974. So, when
k=1, there are just 12 y
values, and when k=2, there are 24.
The differencing is intended to remove any seasonal effect,
and, in the absence of a trend or a real effect due to the site
change, y
would be a random variable with zero mean[9]
(RS93, p. 904).
Once all the y
series have been assembled, the correla-
tions ρ
are calculated (using k=1 in this case, although other
values of k are permissible [13] (RS93, p.906)) between each
differenced series y
as follows:
rwhere t¼1to12k
The overbar in the preceding equation indicates an average
over all t. The weights (w) are computed using the 4th power
of the correlations, where:
The weighted differences (z) between the y
series and the
base series y
Finally, the mean of the differences is calculated:
The preceding calculations are performed for both cases k=
1andk= 2. The 95 % confidence intervals are computed in the
standard manner: z±2.201 s for k=1 and z±2.069 s for k=2,
Reanalysis of Long-Term Air Temperature Trends in New Zealand
where sis the standard error of z, with no adjustment for
autocorrelation of the time series. Throughout this document,
the 95 % confidence interval will be quoted unless explicitly
stated otherwise. If the 95 % confidence interval does not
contain zero, the adjustment is valid. The adjustment is made
by subtracting zfrom the base x
series for all values pre-
change (i.e. replace x
by x0ðÞ
zfor t<τ). This has the effect
of raising pre-change values if the value of zis negative and
lowering them if zis positive.
The convention has been to use the mean of both the
significant results, for k=1 and k=2,whenmakinganadjust-
ment. If, on the other hand, the 95 % confidence interval
contains zero (i.e. the 95 % confidence limit is greater than
the shift itself), no adjustment is made. In some cases, where
there are conflicting results between k=1 and 2, k=3 has been
used to break the deadlock, but these cases are rare. RS93
advocates taking a conservative approach: adjustments should
not be made unless there is clear and unambiguous evidence
of a genuine shift in temperatures as the result of a site change.
5.2 Gradual Inhomogeneities
Many archivists exclude stations in metropolitan areas from
the calculation of national long-term temperature trends, a
clear example being the Australian Bureau of Meteorology,
as explained in [6]and[20]. McAneney et al. [13] found that
sheltering by nearby trees can increase daily maximum tem-
peratures by 1 °C per 10 m of shelter growth over a 6-year
period. After examining a range of New Zealand stations, [10]
determined that two of the stations used in the 7SS were
climatically unrepresentative and assessed them to have in-
creased sheltering from trees and/or significant urbanisa-
tion and/or screen changes[[10], p. 5]. The sites are Albert
Park in Auckland and Kelburn in Wellington. These sites,
within the central business districts of two of New Zealands
largest cities, form the major portion of the Auckland and
Wellington temperature series.
Hessell ([10], p. 1) describes the Albert Park site (1909
1976) as follows:
Visitors to this central city park today cannot fail to be
impressed by the many large exotic trees, most of which
were planted about the turn of the century and some of
which are still growing. The instrument enclosure is
surrounded on all sides by trees and buildings which
shelter the site to a great degree.
The metadata refers to a considerable effect on
windflowas well as sunshine. The daily mean wind
run data 19161975 graphed by [10] (p. 4) displays a
continuing dramatic reduction (except at the time of an
anemometer change) not experienced elsewhere in the
region. Hessell [10] also finds a strong temperature trend
bias in the data unlikely to be due to a broadscale
climatic effect. A 12-year comparison suggested that
Albert Park had warmed 0.6 °C more than a rural coun-
terpart 10 km distant ([10], p. 8, Table 6).
There are no nearby long series to compare with Albert
Park. The closest, which is described as reliablein [10], is at
Te Aroha, some 114 km to the southeast. A Te Aroha com-
parison suggests that steady relative mean warming at Albert
Park averaging 0.09 °C per decade during the period data from
this station is available. If that distortion is not subtracted, any
overlap comparison performed on this series will compound
theerror.ThisisillustratedinFig.3using a diagram from [9].
For the full 19161975 period of Albert Park contamina-
tion, the only available reference series comprises the homog-
enized datasets for Masterton, Nelson, Hokitika, Lincoln, and
Dunedin. The averaged trend of these five stations is com-
pared to Albert Park in Fig. 4. This averaged trend was used to
reduce the Albert Park trend over 19161975.
From 1976, Auckland data are drawn from the Mangere
treatment plant, sited amongst newly commissioned settling
ponds (Fig. 2). In 1998, the site moved a short distance to
Auckland Aero, the countrys principal international/domestic
Fig. 3 Schematic illustration of a temperature record at a site experienc-
ing urban warming. The full caption for this figure from [9]reads: a
Schematic illustration of a temperature record at a site experienc-
ing urban warming and a station moved from the urban center to
the urban outskirts. bThe temperature record adjusted for the
discontinuity has a stronger warming trend than that in the un-
disturbed environment
C.R. de Freitas et al.
airport. Apparent warming trends in the dataset for Auckland
Aero (southwest of Auckland) were de-trended [1]byrefer-
ence to two other less urban airports on the metropolitan
fringes, namely, Whenuapai (27 km to the northwest) and
Ardmore (17 km to the southeast). The station histories for
these two stations reveal no significant site changes during
19762009, and screening disclosed no abrupt shifts. For the
period 19621993, Auckland Aero warmed 0.96 °C/century
faster than Whenuapai, and over 19692011 it warmed
0.97 °C/century faster than Ardmore (Table 2). The Mangere
station was compared with the same two airports and showed
a similar relative warming, implying that the overall Mangere
region was greatly affected by UHI, as the population grew by
1,200 % from 15,700 in 1957 to 190,000 in 1981. The data
from the Mangere sites during 19762009 are detrended by
reference to the average slope of Whenuapai and Ardmore.
Mangere station was decreased by 0.0093 °C/year and Auck-
land Aero by 0.00965 °C per year.
Hessell [10] also found downward trends since 1930 in
mean daily wind-run at the Kelburn site in Wellington, which
has a high mean wind speed of about 12 knots, calculating that
the shelter effect during 19451970 was about one half that of
Albert Park. The metadata discloses that the encroaching
vegetation was cut back in 1949, 1959 and 1969. In 1986, it
was reported in The Dominiondaily newspaper that the
New Zealand Meteorological Service had relocated the
anemometer and was considering relocation of the other in-
struments as wind speeds were being permanently reduced by
about 25 % by the Pohutukawa factor(in reference to a
common native tree species). Kelburns contaminated period,
19282005, is detrended as proposed by [1] by reference to
the average slope of the five homogenized non-contaminated
stations, as with Albert Park (see Fig. 4).
To illustrate the homogenization process used, the following
details the procedure for the Dunedin temperature record for
Fig. 4 Albert Park (19161976) and Kelburn (19282005) mean temperature anomalies relative to five homogenized reference stations
Tabl e 2 Summary of the difference in trends between two independent
Mangere sites (Mangere and Auckland Aero) and more rural Auckland
airports Whenuapai and Ardmore. The results show that there is a
consistent and significant difference between the Mangere sites and more
rural sites. Also shown are 95 % confidence intervals
Station Period Trend difference
Mangere minus Whenuapai 19591993 0.92± 0.28
Mangere minus Ardmore 19691998 0.94± 0.39
Auckland Aero minus Whenuapai 19621993 0.96± 0.36
Auckland Aero minus Ardmore 19692011 0.97±0.25
Reanalysis of Long-Term Air Temperature Trends in New Zealand
the period 19092009 using the RS93 [17]method(Table3).
Reported changes for the Dunedin climate station occurred in
1997, 1960, 1947, 1942 and 1913. The first instrument change
occurred at the end of August 1997. The reference stations
chosen are other South Island stations, namely, Ashburton
Council, Timaru 2, Palmerston and Invercargill Aero. Corre-
lations are 0.97, 0.95, 0.97 and 0.96 respectively, while the
corresponding weighting factors are 0.26, 0.24, 0.25 and 0.24.
The RS93 shifts are calculated to be 0.04± 0.20 °C for k=1
and 0.04±0.20 °C for k=2. The calculated shift is not signif-
icant at the 95 % confidence level, so no adjustment is made.
At the end of October 1960, the site was moved a few
hundred yards. We examine the effect of this move on the
temperature record using the following stations: Kelburn,
Table 3 Summary of homogenization adjustments to the Auckland,
Masterton, Nelson, Hokitika, Lincoln and Dunedin temperature series
using the RS93 method. Included are the various site names, the period of
operation, the RS93-calculated adjustment for each site change and the
accumulated adjustment sum (relative to the reference site) as applied to
each unadjusted series
Station Site name From To Adjustment (°C) Sum (°C)
Dunedin Leith Valley Jan 1900 Dec 1912 0.00 +0.16
Botanical Gardens Jan 1913 Nov 1942 0.69 +0.16
Beta Street Dec 1942 May 1947 +0.85 +0.85
Musselburgh Jun 1947 Oct 1960 0.00 0.00
Musselburgh Nov 1960 Aug 1997 0.00 0.00
Musselburgh EWS (reference) Sep 1997 present 0.00 0.00
Auckland Albert Park Sep 1909 Mar 1976 0.12 0.10
Mangere Apr 1976 Jul 1998 +0.02 +0.02
Auckland Aero (reference) Aug 1998 Present 0.00 0.00
Masterton Waingawa Feb 1912 Apr 1920 0.00 0.00
Waingawa Jun 1920 Sep 1942 0.00 0.00
Waingawa Oct 1942 Dec 1990 0.00 0.00
East Taratahi (reference) Jan 1991 Oct 2009 0.00 0.00
Wellington Buckle Street Jun 1906 Jun 1912 +0.21 0.48
Thorndon Jul 1912 Dec 1927 1.00 0.69
Kelburn (reference) Jan 1928 Aug 2005 0.00 0.00
Kelburn AWS Sep 2005 Present 0.06 0.06
Nelson Nelson Oct 1907 Nov 1920 0.40 0.35
Nelson Dec 1920 Dec 1931 0.00 +0.05
Appleby Jan 1932 Nov 1996 0.23 +0.05
Nelson Aero Dec 1996 May 1997 +0.28 +0.28
Nelson Aero (reference) Jun 1997 present 0.00 0.00
Hokitika Hokitika Town Jan 1900 Aug 1912 0.50 0.27
Hokitika Town Sep 1912 Oct 1928 0.00 +0.23
Hokitika Town Nov 1928 Jul 1943 +0.57 +0.23
Hokitika Town Aug 1943 Dec 1944 0.68 0.34
Hokitika Southside Jan 1945 Dec 1963 +0.29 +0.34
Hokitika Aero Jan 1964 Oct 1967 +0.05 +0.05
Hokitika Aero (reference) Nov 1967 present 0.00 0.00
Lincoln Lincoln Jan 1905 Nov 1915 0.45 0.42
Lincoln Dec 1915 Oct 1923 +0.59 +0.03
Lincoln Nov 1923 Dec 1925 0.51 0.56
Lincoln Jan 1926 Dec 1943 0.60 0.05
Lincoln Jan 1944 Apr 1964 +0.55 +0.55
Lincoln May 1964 Dec 1975 0.00 0.00
Lincoln Jan 1976 May 1987 0.00 0.00
Lincoln Broadfield EDL Jun 1987 Dec 1999 0.00 0.00
Lincoln Broadfield EWS (reference) Jan 2000 present 0.00 0.00
C.R. de Freitas et al.
Adair, Waimate and Invergargill Aero. The correlations for
1 year on either side of the shift (k=1) were 0.41, 0.82, 0.85
and 0.55, respectively. The weightings were therefore 0.02,
0.42, 0.47 and 0.08. Note that the two lesser correlated sites
Kelburn and Invercargill Aero have been weighted very low.
The RS93 shifts are calculated to be 0.23±0.27 °C for k=1
and 0.24±0.24 °C for k=2. Because this shift is not signif-
icant at the 95 % confidence level, no adjustment is made.
At the end of May 1947, the site was moved. To check the
effect of this move, we use the RS93 method with reference
stations Kelburn, Alexandra and East Gore. The correlations
(k=1) are 0.84, 0.90 and 0.87, respectively, and the weightings
are therefore 0.29, 0.38 and 0.33. The shifts for k=1 and k=2
are +0.86±0.36 °C and +0.84±0.26 °C. As these are both
clearly significant at the 95 % confidence level, the shift is
valid. So, the adjustment is to raise all the pre-June 1947
values by (0.86+0.84)/2= +0.85 °C.
At the end of November 1942, the site was moved. The
reference stations used here are once again Kelburn,
Alexandra and East Gore, with correlations 0.71, 0.96 and
0.89, respectively, giving weightings of 0.15, 0.50 and 0.35.
The shift is calculated using the RS93 method to be 0.72±
0.26 °C for k=1 and 0.65± 0.35 °C for k=2. Because this
result is significant at the 95 % confidence level, the shift is
Fig. 5 Annual meantemperatures for Dunedin from 1909 to 2009showing the unadjusted and RS93-adjusted series. The dashed lines indicate the linear
trends of each series
Tabl e 4 Summary of trends before and after homogenization for the
seven sites Auckland, Masterton, Wellington, Nelson, Hokitika, Lincoln
and Dunedin. Included are the station names, time period and unadjusted
and adjusted trends of each. Also given are 95 % confidence intervals
Station Period Trend (°C/century)
Unadjusted Adjusted
Auckland 19092009 0.69±0.35 0.24±0.32
Masterton 19122009 0.36±0.36 0.36±0.36
Wellington 19092009 0.01±0.36 0.43±0.30
Nelson 19092009 0.07±0.35 0.27± 0.32
Hokitika 19092009 0.44±0.38 0.21± 0.35
Lincoln 19092009 0.04±0.35 0.19± 0.30
Dunedin 19092009 0.57±0.33 0.30± 0.30
7SS average 19092009 0.27± 0.31 0.28±0.29
Tabl e 5 Outcomes of the different adjustment methods (°C/century)
Unadjusted data 7SS (S81) 7SS (RS93) Difference
Auckland 0.69 1.34 0.27 1.07
Masterton 0.36 0.80 0.36 0.44
Wellington 0.01 0.79 0.43 0.36
Nelson 0.0 0.81 0.27 0.54
Hokitika 0.44 1.07 0.21 0.86
Lincoln 0.08 0.99 0.19 0.80
Dunedin 0.53 0.58 0.24 0.34
7SS average 0.27 0.92 0.28 0.64
This column is the result of subtracting column 2 (S81) from column 3
Reanalysis of Long-Term Air Temperature Trends in New Zealand
valid. In light of this, the adjustment undertaken is to lower the
pre-December 1942 values by (0.720.65)/2, giving an ad-
justment of 0.69 °C.
A site move occurred at the end of January 1913. The
reference stations used are Albert Park, Nelson, Christchurch
Gardens and Lincoln, with correlations of 0.45, 0.68, 0.81 and
0.86 and weightings of 0.03, 0.18, 0.36 and 0.44, respectively.
The calculated shift is +0.28±0.46 °C for k=1 and +0.37±
0.37 °C for k=2. The shift is not significant at the 95 %
confidence level and is therefore not valid. Consequently, no
adjustment is made for the 1913 site move.
Tab le 3summarizes the adjustments for Dunedin, and
Fig. 5shows the adjusted and unadjusted time series for the
site from 1909 to 2009. The effect of the adjustments is to
reduce the trend from 0.57± 0.33 °C/century to 0.30±0.30 °C/
century. Note that in the 7SS, NIWA have excluded pre-
1913 years due to poor data quality. If these early years are
excluded, then the trend drops to 0.24±0.30 °C/century. The
results for the six remaining stations are also summarized in
Tab le s 4and 5.
Combining the seven homogenized station histories into
one series, we find that over the past century the warming
trend for 19092009 is 0.28±0.29 °C per century (Fig. 6).
This is similar to the raw data trend for the same period (0.27±
0.31 °C/century). These results are reassuring as New Zealand
site changes were wholly random, displaying no apparent
pattern, and there were no systematic changes made to all
stations at the same time. It is to be expected, therefore, that
the random changes would tend to balance over time. If the
contaminated stations at Auckland and Wellington are exclud-
ed from the series, the warming trend drops slightly to 0.26±
0.30 °C per century. A comparison with the S81/M10 trend is
discussed in the Conclusionsection.
Fig. 6 Unadjusted and RS93-adjusted mean annual temperature anomalies 19092009 for New Zealand (baseline 19712000) using the seven
homogenized series from Auckland, Masterton, Wellington, Nelson, Hokitika, Lincoln, and Dunedin. The linear trends are 0.27±0.31 °C/century for
unadjusted and 0.28±0.29 °C/century for RS93-adjusted
Tabl e 6 Summary of the pre-1960 adjustments using the methods of
Salinger (1981) and that of the current work
Station Year Salinger RS93
Albert Park 1950 +0.1 0.0
Waingawa 1920 0.7 0.0
Waingawa 1942 0.4 0.0
Buckle St 1912 0.9 +0.21
Thorndon 1927 1.2 1.0
Nile St 1920 0.7 0.4
Cawthron 1931 0.2 0.0
Hokitika Tn 19431944 0.6 0.11
Lincoln 1915 1.0 0.45
Lincoln 1923 +0.6 +0.59
Lincoln 1929 0.8 0.51
Lincoln 1943 0.3 0.6
Dunedin 1912 +0.3 0.0
Dunedin 1942 0.1 0.69
Dunedin 1947 +0.6 +0.85
Ave ra ge 0.35 0.14
C.R. de Freitas et al.
7 Discussion
Hessell [10] examined apparent continuous warming over
New Zealand since 1930 but concluded ([10], p. 1): “… no
important change in annual mean temperature since 1930 has
been found in stations where these factors [changes in shelter,
screenage, and/or urbanization] are negligible.Our study
similarly concludes that no importantchange in mean tem-
perature occurred over the period 19092009 once the known
contamination is corrected.
Tab le 5highlights the disparity of our results with those of
S81. The major disagreements occur as a result of the appli-
cation of significance testing and the correction of Aucklands
distorted trend. S81 did not attempt to quantify or correct for
UHI or sheltering effects. The two rural stations (Hokitika and
Lincoln) show average warming of 0.2 °C/century in contrast
to the 0.32 °C/century of the five city stations. A concern
highlighted by RS93 [17] is that all seven datasets have not yet
been thoroughly screened for possible UHI effects or other
undocumented changes. There are few adjustments in the
second half of the series and no material differences between
analyses in that period. The adjustments for 190959 are set
out in Table 6. Folland and Salinger [8] estimated 18711993
SST variations for an area including New Zealand at about
0.6 °C/century but acknowledged that there is low confidence
in the data in the crucial pre-1949 period.
8 Conclusion
Detecting trends in climate variables is important in assess-
ments of global and regional change based on long-term
observations as they are widely used as inputs for societal
design and planning purposes [11]. The information is also
extensively used in hindcast verifications for regional and
local models. The ongoing controversy regarding the contri-
bution of human versus natural causes of climate change has
increased the importance of scrutinizing the instrumental re-
cord of climate data for the longest possible time periods. For
all these reasons, it is important that we have the best temper-
ature trend estimates possible.
Using well-accepted homogenization methods, we have
derived a mean land surface air temperature trend for New
Zealand over the past century of 0.28±0.29 °C per century,
which is considerably less than the S81/M10 value of 0.91±
0.30 °C per century. By excluding weighted averages and
including adjustments which are not statistically significant,
S81 may have allowed too many false positivesto occur. In
addition, using long comparison time periods may have
allowed creeping inhomogeneities and undocumented shifts
at reference sites to skew the individual adjustments. This is
borne out by Table 5, which demonstrates that in every case
the S81 station adjustments greatly increased the individual
trends, while the RS93 method resulted in equal numbers of
increases and decreases. As noted previously, S81 did not
account for gradual effects such as sheltering or UHI. The
detrending of Albert Park in Auckland and Kelburn in Wel-
lington also contributed significantly to the mean trend result.
We have also shown that a very similar outcome would follow
if those two stations were not corrected but simply omitted
from the series.
1. Aguilar,E.,Auer,I.,Brune,M.,Peterson,T.C.,&Wieringa,J.(2003).
Guidelines on climate metadata and homogenization. Geneva: World
Meteorological Organization. WCDMP No. 53, WMD/TD No. 1186.
2. Auer, I., Böhm, R., Jurkovic, A., Orlik, A., Potzmann, R., Schöner, W.,
P., Efthymiadis, D., Mestre, O., Moisselin, J. M., Begert, M., Brazdil, R.,
Bochnicek, O., Cegnar, T., Gajic-Capka, M., Zaninovic, K., Majstorovic,
Z.,Szalai,S.,Szentimrey,T.,&Mercalli, L. (2005). A new instrumental
precipitation dataset for the Greater Alpine Region for the period 1800
2002. Int J Climatol, 25,139166.
3. Begert, M., Schlegel, T., & Kirchhofer, W. (2005). Homogeneous
temperature and precipitation series of Switzerland from 18642000.
Int J Climatol, 25,6580.
4. Böhm, R., Auer, I., Brunetti, M., Maugeri, M., Nanni, T., & Schöner, W.
(2001). Regional temperature variability in the European Alps 1760
1998 from homogenized instrumental time series. Int J Climatol, 21,
5. Camuffo, D. and Jones P. (eds) (2002). Improved understanding of past
climatic variability from early daily European instrumental sources.
Climatic Change,53,1392.
6. Della-Marta, P., Collins, D., & Braganza, K. (2004). Updating
Australias high-quality annual temperature dataset. Aust Meteorol
Mag, 53(2), 7593.
7. Della-Marta, P. M., & Wanner, H. (2006). A method of homogeniz-
ing the extremes and mean of daily temperature measurements. J
Clim, 19,41794197. doi:10.1175/JCLI3855.1.
8. Folland, C. K., & Salinger, M. J. (1995). Surface temperature trends
and variations in New Zealand and the surrounding ocean 1871
1993. Int J Climatol, 15, 11951218.
Easterling, D., Peterson, T., & Karl, T. (2001). A closer look at
United States and global surface temperature change. JGeophys
Res, 106, 23. 94723 963.
10. Hessell, J. W. D. (1980). Apparent trends of mean temperature in
New Zealand since 1930. NZJSci,23,19.
11. Livezey,R.E.,Vinnikov,K.Y.,Timofeyeva,M.M.,Tinker,R.,&anden
Dool, H. M. (2007). Estimation and extrapolation of climate normals and
climate trends. J Appl Meteorol Climatol, 46, 17591776
12. Manton, M. J., & Coauthors. (2001). Trend in extreme daily rain-fall
and temperature in Southeast Asia and the South Pacific: 19611998.
Int J Climatol, 21,269284.
13. McAneney, K. J., Salinger, M. J., Porteus, A. S., & Barber, R. F.
(1990). Modification of an orchard climate with increasing shelter-
belt height. Agric For Meteorol, 49,177189.
14. Menne, M. J., & Williams, C. N. (2009). Homogenization of tem-
perature series via pairwise comparisons. JClim,22, 17001717.
15. Neyman, J., & Pearson, E. S. (1933). The testing of statistical
hypotheses in relation to probabilities a priori. Math Proc Camb
Philos Soc, 29,492510. doi:10.1017/S030500410001152X.
Reanalysis of Long-Term Air Temperature Trends in New Zealand
16. Peterson, T. C., Easterling, D. R., Karl, T. R., Groisman, P.,
Nicholls, N., Plummer, N., Torok, S., Auer, I., Boehm, R.,
O., Szentimrey, T., Salinger, J., Forland, E. J., Hanssen-Bauer,
I., Alexandersson, H., Jones, P., & Parker, D. (1998).
Homogeneity adjustments of in situ atmosphericclimatedata:
areview.Int J Climatol, 18,14931517.
17. Rhoades, D. A., & Salinger, M. J. (1993). Adjustment of temperature
and rainfall records for site changes. Int J Climatol, 13,899913.
18. Salinger, M. J. (1980). The New Zealand temperature series. Climate
Monitor, 9(4), 112118.
19. Serra, Y. L., AHearn, P., Freitag, H. P., & McPhaden, M. J. (2001).
ATLAS self-siphoning rain gauge error estimates. Journal of
Atmospheric and Oceanic Technolology, 18,19892002.
20. Torok, S. J., & Nicholls, N. (1996). A historical annual temperature
data set for Australia. Aust Meteorol Mag, 45,251294.
21. Tuomenvirta, H. (2001). Homogeneity adjustments of temperature
and precipitation series? Finnish and Nordic data. Int J Climatol,
21(4), 495506.
22. Venema, V. K. C., Mestre, O., Aguilar, E., Auer, I., Guijarro, J. A.,
Domonkos, P., Vertacnik, G., Szentimrey, T., Stepanek, P.,
Zahradnicek, P., Viarre, J., Muller-Westermeier, G., Lakatos, M.,
Williams, C. N., Menne, M. J., Lindau, R., Rasol, D., Rustemeier,
E., Kolokythas, K., Marinova, T., Andresen, L., Acquaotta, F.,
Fratianni, S., Cheval, S., Klancar, M., Brunetti, M., Gruber, C.,
Prohom, D. M., Likso, T., Esteban, P., & Brandsma, T. (2012).
Benchmarking homogenization algorithms for monthly data. Clim
Past, 8,89115.
Benchmarking the performance of pairwise homogenization of sur-
face temperatures in the United States. J Geophys Res, 117,D05116.
24. Wijngaard, J. B., Klein Tank, A. M. G., & Können, G. P.
(2003). Homogeneity of 20th century European daily temper-
ature and precipitation series. Int J Climatol, 23,679692.
Report on the review of NIWAsseven-stationtemperature
series. NIWA Information Series No. 78. 175 p.
C.R. de Freitas et al.
... de Freitas et al. (2015) (henceforth dFDB) report a trend of 0.28°C per century over the period 1909-2009 for New Zealand land surface temperatures, from their reanalysis of a composite of seven long-term records. This is much lower than the warming trend of about 0.9°C per century reported previously by other researchers and much smaller than trends estimated from independent sea surface temperature data from the surrounding region. ...
... A paper by de Freitas et al. ( [1], hereafter referred to as dFDB) claims to find a very small warming rate for New Zealand temperatures over the past century, a result very much at odds with all other published evidence (e.g., [2], Fig. SPM.1). The temperature records analyzed are collectively known as the New Zealand Bseven-station series^(7SS), built up from temperature measurements made at seven locations across the country, which go back to the early years of the twentieth century or even before. ...
... Finally, while there are many numbers reported in the dFDB paper [1], it does not include sufficient specific information, in general, for us to reproduce the underlying calculations. To examine these calculations, we have referred back to unpublished reports produced in 2011 by the same authors, on behalf of the New Zealand Climate Science Coalition. 1 These reports, referred to here collectively as CSC11, challenged NIWA's calculations in M10. ...
Full-text available
de Freitas et al. (2015) (henceforth dFDB) report a trend of 0.28 °C per century over the period 1909–2009 for New Zealand land surface temperatures, from their reanalysis of a composite of seven long-term records. This is much lower than the warming trend of about 0.9 °C per century reported previously by other researchers and much smaller than trends estimated from independent sea surface temperature data from the surrounding region. We show these differences result primarily from the way inhomogeneities in temperature time series at individual stations due to site or instrument changes are identified and adjusted for in the dFDB paper. The adjustments reported in that paper are based on a method designed by one of us (Salinger), but use only a short (1–2-year) overlap period with comparison stations and consider only inhomogeneities in monthly mean (rather than monthly maximum and minimum) temperatures. This leads to underestimates of the statistical significance of individual temperature discontinuities and hence rejection of many valid adjustments. Since there was a systematic tendency for the seven-station sites to be relocated to colder locations as the early half of the twentieth century progressed, this rejection of valid adjustments produces an artificially low rate of warming. We therefore disagree with the trend calculations in the dFDB paper and consider there is no reason to reject the previous estimates of around 0.9 °C warming per century.
... While it is generally agreed that homogenized time series better represent actual historical weather and climate trends (Datsenko et al., 2002;Peterson et al., 1998), the method of homogenization is under increasing scrutiny along with the different temperature trends generated (Boretti, 2013;De Freitas et al., 2014;Stockwell and Stewart, 2012). For example, Zhang et al. (2014) showed that data homogenization for weather station moves in Beijing municipality, China, could lead to a significant overestimate of warming trends of surface air temperature. ...
... A large number of different techniques and philosophies are employed in the development of homogenized temperature series (Datsenko et al., 2002;De Freitas et al., 2014;Easterling and Peterson, 1995). Furthermore, subjective judgment by experienced climatologists has been an important tool in many adjustment methodologies, and it is acknowledged that this can modify the weight given to various inputs based on a myriad of factors, which according to Peterson et al. (1998) can be too laborious to program. ...
... Furthermore, subjective judgment by experienced climatologists has been an important tool in many adjustment methodologies, and it is acknowledged that this can modify the weight given to various inputs based on a myriad of factors, which according to Peterson et al. (1998) can be too laborious to program. The resulting continuous temperature series are used to report climatic trends (De Freitas et al., 2014;Trewin, 2013). Until now, however, there has been no objective test of the assumed improvement in the temperature series resulting from homogenization. ...
... While it is generally agreed that homogenized time series better represent actual historical weather and climate trends (Datsenko, et al. 2002;Peterson et al., 1998), the method of homogenization is under increasing scrutiny along with the different temperature trends generated (Boretti, 2013;De Freitas, et al. 2014;Stockwell and Stewart, 2012). For example, Zhang et al. (2014) showed that data homogenization for weather station moves in Beijing A C C E P T E D M A N U S C R I P T ...
... A large number of different techniques and philosophies are employed in the development of homogenized temperature series (Datsenko et al., 2002;De Freitas et al., 2014;Easterling and Peterson, 1995). Furthermore, subjective judgement by experienced climatologists has been an important tool in many adjustment methodologies, and it is acknowledged that this can modify the weight given to various inputs based on a myriad of factors, which according to Peterson et al. (1998) can be too laborious to program. ...
... Furthermore, subjective judgement by experienced climatologists has been an important tool in many adjustment methodologies, and it is acknowledged that this can modify the weight given to various inputs based on a myriad of factors, which according to Peterson et al. (1998) can be too laborious to program. The resulting continuous temperature series are used to report climatic trends (de Freitas et al., 2014;Trewin, 2013). Until now, however, there has been no objective test of the assumed improvement in the temperature series resulting from homogenization. ...
Hurricane Sandy - a tropical cyclone that transitioned into an extratropical cyclone near the time of landfall along the east coast of the United States - caused historic damage in many regions which rarely receive such a direct hit from a storm of this magnitude, including many of the large metropolitan areas along the U.S. eastern seaboard. Specifically, Sandy generated record low-pressure, a large wind field with corresponding storm surge and copious amounts of precipitation in some areas, including record snowfall in mountainous regions. Sandy presented several forecast challenges to the National Oceanic and Atmospheric Administration’s (NOAA) National Weather Service (NWS). Satellites played an integral role in the analysis and forecast of Sandy’s track and intensity. The NOAA National Hurricane Center, Ocean Prediction Center, and Weather Prediction Center all relied on information from satellites to make critical warning decisions using various satellite products that assist with diagnosing tropical cyclone intensity, surface winds over the ocean, and heavy precipitation. All of the skillful global forecast models used satellite data for initiation to better forecast the track and intensity of Sandy. As part of the Geostationary Operational Environmental Satellite - R-series (GOES-R) and Joint Polar Satellite System (JPSS) Proving Ground activities, new satellite products were available to forecasters at these national centers in experimental form to assist with observing this unique, high impact event. This paper will demonstrate how current satellite products assisted NOAA forecasters during Sandy and introduce some new satellite products that could be used to analyze and predict future high impact weather systems.
... Since 1900, New Zealand's mean annual temperatures have risen at a rate similar to the global mean, with most of the warming occurring after 1946 (0.12-0.14 C decade À1 , 50-year trend ending in 2006, Dean and Stott, 2009). However, irregularities in observations and poor station coverage in alpine areas result in uncertainties in warming rates for high elevation areas (de Freitas et al., 2014). What is known, however, is that warming has reduced New Zealand's glacier volume since 1850 (Mark et al., 2006). ...
Alpine ecosystems in the Pacific Islands are isolated and unique, characterized by high levels of endemism. Only Hawai‘i and New Zealand have elevations high enough to contain substantial alpine climates, and about 11% of the land area of both island groups is located above treeline. Both of these volcanically active archipelagos are characterized by complex topography, with peaks over 3700 m. These alpine ecosystems have significant cultural, social, and economic value; however, they are threatened by invasion of exotic species, climate change, and human impacts. Nonnative ungulates reduce native shrubland and grassland cover, and threaten populations of endangered birds. Exotic plants alter water yields and increase fire risk, and increased recreational visitation to these remote areas facilitates the introduction of exotic plant seeds, pests, and pathogens. Both New Zealand and Hawai‘i have experienced strong warming at higher elevations, and future projections indicate that these robust warming trends will continue. Glacial retreat has been noted in the Southern Alps, with 34% ice volume lost since 1977, and New Zealand may lose 88% of its ice volume by 2100. Snowfall on Hawai‘i's mountain peaks is projected to almost entirely disappear by 2100. Changes are occurring rapidly, and additional monitoring and research are needed to conserve these uniquely sensitive, remote regions.
... The most extreme case, however, is that in which the data are made to conform to their expected behavior with appropriate adjustments to the homogenization procedure (Marohasy, 2016) ) (Parker, The Artefacts of Data Biases in Surface Temperature, 2015) (Davison, 2016) (Kennel, 2016) (Rose, 2017). In the case of reanalysis, station data are homogenized to correct for differences in instrumentation, measurement methods, and the time of day that the measurements were made so that the data may be "assimilated" into a well behaved time series that meets climate model requirements (Kennel, 2016) (DeFreitas, 2015 (Davison, 2016) (Mulvey, 2015) (Bengtsson, 2004) (Vose, 2012) (Wang, 2012) (Edwards, 1999) (Hulme, 2010). ...
Full-text available
A month by month trend analysis of daily maximum and daily minimum temperature data from Fort Collins, Boulder, and Cheesman USHCN weather stations in northeastern Colorado for the period 1893-2014 is presented. The results show anomalous and incompatible patterns in long term temperature trends among the three stations that serve to question the use of the Fort Collins data as empirical evidence of global warming. The findings are consistent with questions raised by others with respect USHCN data integrity, the homogenization of incompatible station data into regional climatology, and the greater validity and information content of observational data over derived synthetic data 1 .
... On the other hand, the impact of changes in environmental variables changes, as the temperature or the precipitation, on environmental issues have been assessed and analyzed in order to explore the implications under a future climate [3]. The most applied statistical models in order to analyze climate time series, in particular of temperatures data sets, are the linear models and the usual time series models as ARIMA models [1,2,4]. These models are largely known as simple and they have well-known optimal properties under the usual assumptions (normality of errors, etc.). ...
Conference Paper
Full-text available
This work presents a periodic state space model to model monthly temperature data. Additionally, some issues are discussed, as the parameter estimation or the Kalman filter recursions adapted to a periodic model. This framework is applied to monthly long-term temperature time series of Lisbon.
... Withers et al. (2009) found largely different temperature trend behavior, which was probably due to diverse geographic factors as well as the length of the time series. De Freitas et al. (2015) derived a mean land surface air temperature trend for New Zealand over the past century of about 0.28°C per century. ...
Full-text available
Among several variables affecting climate change and climate variability, temperature plays a crucial role in the process because its variations in monthly and extreme values can impact on the global hydrologic cycle and energy balance through thermal forcing. In this study, an analysis of temperature data has been performed over 22 series observed in New Zealand. In particular, to detect possible trends in the time series, the Mann-Kendall non-parametric test was first applied at monthly scale and then to several indices of extreme daily temperatures computed since 1951. The results showed a positive trend in both the maximum and the minimum temperatures, in particular, in the autumn-winter period. This increase has been evaluated faster in maximum temperature than in minimum one. The trend analysis of the temperature indices suggests that there has been an increase in the frequency and intensity of hot extremes, while most of the cold extremes showed a downward tendency.
Full-text available
In recent decades, the world has been confronted with the consequences of global warming; however, this phenomenon is not reflected equally in every part of the globe. Thus, the warming phenomenon must be monitored in a more regional or local scale. This paper analyzes monthly long-term time series of air temperatures in three Portuguese cities: Lisbon, Oporto and Coimbra. We propose a periodic state space framework, associated with a suitable version of the Kalman filter; which allows for the estimation of monthly warming rates taking into account the seasonal behavior and serial correlation. Results about the monthly mean of the daily mid-range temperature time series show that there are different monthly warming rates. The greatest annual mean rise was found in Oporto with 2.17ºC whereas, in Lisbon and Coimbra, it was 16 respectively, 0.62ºC and 0.55ºC, per century.
Full-text available
This report describes sampling and error characteristics of self-siphoning rain gauges used on moored buoys designed and assembled at NOAA's Pacific Marine Environmental Laboratory (PMEL) for deployment in the tropical Pacific and Atlantic Oceans in support of climate studies. Self-siphoning rain gauges were chosen for use on these buoys because they can be calibrated at PMEL before and after deployment. The rainfall data are recorded at 1-min intervals, from which daily mean rate, standard deviation, and percent time raining are calculated and telemetered to PMEL in real time. At the end of the deployment, the 1-min, internally recorded data are recovered and processed to produce 10-min rain rates. Field data from a subset of these rain gauges are analyzed to determine data quality and noise levels. In addition, laboratory experiments are performed to assess gauge performance. The field data indicate that the noise level during periods of no rain is 0.3 mm h 21 for 1-min data, and 0.1 mm h21 for 10-min data. The estimated error in the derived rain rates, based on the laboratory data, is 1.3 mm h 21 for 1-min data, and 0.4 mm h21 for 10-min data. The error in the real-time daily rain rates is estimated to be at most 0.03 mm h 21. These error estimates do not take into account underestimates in accumulations due to effects of wind speed on catchment efficiency, which, though substantial, may be correctable. Estimated errors due to evaporation and sea spray, on the other hand, are found to be insignificant.
The evidence of apparent continous warming over New Zealand since 1940 is examined from both physical and statistical standpoints. It is found that the exposures of most of the thermometers have been affected by changes in shelter, screenage, and/or urbanisation, all of which tend to increase the observed mean temperature. -from Author
A high-quality historical surface air temperature dataset, for mean annual temperatures, has been prepared for Australia by adjusting data for inhomogeneities caused by station relocations, changes in exposure and other discontinuities. An objective procedure was developed for determining the necessary adjustments. Station history documentation was also used for this purpose. Time-series of annual mean maximum and mean minimum temperatures have been produced for 224 stations. Trends in annual mean maximum, minimum, the mean of the maximum and minimum, and the range between maximum and minimum, have been calculated at each site. The dataset provides adequate spatial coverage of Australia back to 1910 for the production of all-Australia average temperatures. Maximum and minimum temperatures have increased since about 1950, with minimum temperatures increasing faster than the maximum temperatures.
An updated and improved version of the Australian high-quality annual mean temperature dataset of Torok and Nicholls (1996) has been produced. This was achieved by undertaking a thorough post-1993 homogeneity assessment using a number of objective and semi-objective techniques, by matching closed records onto continuing records, and by adding some shorter duration records in data-sparse regions. Each record has been re-assessed for quality on the basis of recent metadata, resulting in many records being rejected from the dataset. In addition, records have been re-examined for possible urban contamination using some new approaches. This update has highlighted the need for accurate and complete station metadata. It has also demonstrated the value of at least two years of overlapping observations for major site changes to ensure the homogeneity of the climate record. A total of 133 good-quality, homogenised records have been produced. A non-urban subset of 99 stations provides reliable calculations of Australia's annual mean temperature anomalies with observation error variances between 15 and 25 per cent of the total variance and decorrelation length scales greater than the average inter-station separation.
Changes in the circumstances behind in situ temperature measurements often lead to biases in individual station records that, collectively, can also bias regional temperature trends. Since these biases are comparable in magnitude to climate change signals, homogeneity "corrections" are necessary to make the records suitable for climate analysis. To quantify the effectiveness of U.S. surface temperature homogenization, a randomized perturbed ensemble of the USHCN pairwise homogenization algorithm was run against a suite of benchmark analogs to real monthly temperature data. Results indicate that all randomized versions of the algorithm consistently produce homogenized data closer to the true climate signal in the presence of widespread systematic errors. When applied to the real-world observations, the randomized ensemble reinforces previous understanding that the two dominant sources of bias in the U.S. temperature records are caused by changes to time of observation (spurious cooling in minimum and maximum) and conversion to electronic resistance thermometers (spurious cooling in maximum and warming in minimum). Error bounds defined by the ensemble output indicate that maximum temperature trends are positive for the past 30, 50 and 100 years, and that these maximums contain pervasive negative biases that cause the unhomogenized (raw) trends to fall below the lower limits of uncertainty. Moreover, because residual bias in the homogenized analogs is one-tailed under biased errors, it is likely that maximum temperature trends have been underestimated in the USHCN. Trends for minimum temperature are also positive over the three periods, but the ensemble error bounds encompass trends from the unhomogenized data.
WMO-recommended 30-yr normals are no longer generally useful for the design, planning, and decision-making purposes for which they were intended. They not only have little relevance to the future climate, but are often unrepresentative of the current climate. The reason for this is rapid global climate change over the last 30 yr that is likely to continue into the future. It is demonstrated that simple empirical alternatives already are available that not only produce reasonably accurate normals for the current climate but also often justify their extrapolation to several years into the future. This result is tied to the condition that recent trends in the climate are approximately linear or have a substantial linear component. This condition is generally satisfied for the U.S. climate-division data. One alternative [the optimal climate normal (OCN)] is multiyear averages that are not fixed at 30 yr like WMO normals are but rather are adapted climate record by climate record based on easily estimated characteristics of the records. The OCN works well except with very strong trends or longer extrapolations with more moderate trends. In these cases least squares linear trend fits to the period since the mid-1970s are viable alternatives. An even better alternative is the use of "hinge fit" normals, based on modeling the time dependence of large-scale climate change. Here, longer records can be exploited to stabilize estimates of modern trends. Related issues are the need to avoid arbitrary trend fitting and to account for trends in studies of ENSO impacts. Given these results, the authors recommend that (a) the WMO and national climate services address new policies for changing climate normals using the results here as a starting point and (b) NOAA initiate a program for improved estimates and forecasts of official U.S. normals, including operational implementation of a simple hybrid system that combines the advantages of both the OCN and the hinge fit.
To be confident in the analyses of long-term changes in daily climate extremes, it is necessary for the data to be homogenized because of nonclimatic influences. Here a new method of homogenizing daily tempera- ture data is presented that is capable of adjusting not only the mean of a daily temperature series but also the higher-order moments. This method uses a nonlinear model to estimate the relationship between a candidate station and a highly correlated reference station. The model is built in a homogeneous subperiod before an inhomogeneity and is then used to estimate the observations at the candidate station after the inhomogeneity using observations from the reference series. The differences between the predicted and observed values are binned according to which decile the predicted values fit in the candidate station's observed cumulative distribution function defined using homogeneous daily temperatures before the in- homogeneity. In this way, adjustments for each decile were produced. This method is demonstrated using February daily maximum temperatures measured in Graz, Austria, and an artificial dataset with known inhomogeneities introduced. Results show that given a suitably reliable reference station, this method produces reliable adjustments to the mean, variance, and skewness.