Influence of large-scale climate modes on daily synoptic weather types over New Zealand

Authors


Abstract

This article examines the influence of El Niño Southern Oscillation (ENSO), Southern Annular Mode (SAM) and Interdecadal Pacific Oscillation (IPO) on the synoptic weather types over New Zealand. The effects of ENSO (indicated by the Southern Oscillation Index, SOI), SAM and IPO on the occurrence of synoptic types are estimated in a holistic framework using the maximum likelihood method via applications of generalized linear models, both annually and by seasons. The average within-class variations in the intensity and air-mass characteristics (as expressed in the Auckland Airport meteorological variables) of synoptic systems were examined for individual types at the annual level. The results show that ENSO, SAM and IPO have significant effects on the probability of occurrence and to lesser degree the intensity and air-mass characteristics of some synoptic types. The effects vary considerably with seasons, synoptic types and phases of SOI, SAM and IPO and it is the confounding effects of different large-scale modes that lead to the observed changes in the type frequencies. The findings, with respect to ENSO and SAM in all cases and for IPO at the seasonal level, are in good agreement with the literature. However, the annual changes in type frequencies associated with the 1976/1977 IPO phase shift do not support the observed changes in the strength of anomalous southwesterly windflows over New Zealand, and the influence of the recent negative IPO phase also appears different from what we expect based on the existing research. It appears that the interaction between IPO and ENSO is important for understanding the observed climatic effects of IPO on New Zealand's weather and climate. This aspect deserves further attention in future studies. The findings from this work have important methodological and practical implications for New Zealand climate research. Copyright © 2012 Royal Meteorological Society

1. Introduction

New Zealand, located in the mid-latitudes of the southwest Pacific, is influenced by a range of regional (synoptic-scale) and large-scale circulation features (Sturman and Tapper, 2008). The regional weather patterns are primarily dominated by the eastward migrating anticyclones/ridges, depressions/troughs and frontal systems, whereas weather systems of tropical origin and those from the higher latitudes can also affect this country and sometimes result in severe climatic conditions such as heavy rainfall, extreme winds, heat waves and cold outbreaks. Large-scale climate modes, in particular El Niño Southern Oscillation (ENSO), Southern Annular Mode (SAM) and Interdecadal Pacific Oscillation (IPO), are known to have significant influence on New Zealand climate variability (Salinger and Mullan, 1999; Salinger et al., 2001a; Folland et al., 2002; Kidson and Renwick, 2002; Renwick and Thompson, 2006; Lorrey et al., 2007; Marshall, 2007; Kidston et al., 2009; Purdie et al., 2011). Furthermore, it is recognized that interactions between the prevailing wind and the mountainous terrain also play a significant role in determining the weather and climate of New Zealand because of orographic effects (Salinger, 1980a, 1980b; Kidson, 2000; Salinger et al., 2004; Lorrey et al., 2008; Griffiths, 2011). Associated with such effects, regional responses to large-scale circulation drivers (e.g., ENSO and IPO) manifest very differently across homogenous climate districts (Salinger and Mullan, 1999; Salinger et al., 2001a; Lorrey et al., 2007; Ummenhofer and England, 2007) and consequently, it is often difficult to assess and manage the possible impacts of large-scale climate phenomena on the regional climate. Therefore, to improve our understanding of how the low-frequency climate modes affect the regional climate variability, it is important to investigate the influence of large-scale climate signals, i.e., ENSO, SAM and IPO, on the regional synoptic types that interact with local topography and deliver the observed contrasting climatic conditions at different locations. Such investigations are expected to be useful for improving regional climate forecasting skill on seasonal and longer scales, and feed in to the formulation of effective management and planning strategies in response to the risks imposed by future climate variability and change.

Several synoptic classifications exist for New Zealand (Kidson, 2000; Jiang, 2011; Renwick, 2011). The major classifications include the synoptic analogues derived by Kidson (1994a), Kidson (1997) and Kidson (2000) using cluster analysis (e.g., K-means) and those by Jiang (2000), Jiang et al. (2004, 2005) and Jiang (2011) based on weather-typing procedures using T-mode principal component analysis (PCA) or its variants. Whilst these classifications have been applied to a range of environmental problems in New Zealand (Kidson, 1994b; Jiang, 2000; Kidson, 2000; Gosai et al., 2003; Jiang et al., 2005; Lorrey et al., 2007; Griffiths, 2011; Jiang et al., 2011a; Purdie et al., 2011), there appear to be relatively few studies of the relationship between large-scale climate modes and daily synoptic weather types in this region (Kidson, 2000; Renwick, 2011). Jiang (2000), also Jiang et al. (2004) or Jiang (2008), derived a synoptic classification for the winter months (May–September) during 1958–1996 using T-mode PCA and subsequently examined the classification in relation to ENSO (as indicated by the Southern Oscillation Index, SOI) and the New Zealand climate shift (i.e., for the 1958–1975 and 1976–1996 subperiods which were linked to the 1976/1977 phase change of IPO; Salinger and Mullan, 1999). It was found that in winter months, the frequencies of some synoptic types varied significantly with the phases of SOI, with more pronounced changes linked to the negative SOI phase (El Niño). Associated with the 1976/1977 IPO phase change, the frequencies of some synoptic types also changed between the two subperiods, but these changes are generally statistically insignificant. It was also noted that the average meteorological characteristics of individual synoptic types, as expressed in the meteorological variables for the Auckland Airport site, were consistent across the different phases of SOI and between the two subperiods before and after 1976. Recently, Jiang (2011) established a long-term (1958–2008) daily synoptic climatological index for the New Zealand region using a new objective weather-typing procedure (obliquely rotated T-mode PCA followed by convergent K-means clustering) and further examined the correlation of SOI with the occurrence of synoptic weather types for each austral season. It was found that the frequency of synoptic weather types changed with the SOI phases in all seasons. During the negative SOI phase (El Niño), the most significant changes in type frequencies occured in spring, characterized by more frequent occurrence of synoptic types which provide prevailing southwesterly to westerly winds and reduced prominence of anticyclonic situations over the study region. During the positive SOI phase (La Niña), the most significant changes in type frequencies occured in summer, with more frequent anticyclonic types but less frequent cyclonic types.

The Kidson (2000) weather regimes (i.e., trough, zonal and blocking regimes) or equivalents (e.g., the self-similar fractal patterns that occur on climatic time scales; Lorrey et al., 2008) have also been linked to the phase changes of ENSO, SAM or IPO in some studies (Kidson, 2000; Lorrey et al., 2007; Renwick, 2011). Kidson (2000) showed that during 1958–1997, blocking regime was more frequent in positive SOI, but zonal regime was more common in negative SOI (as is consistent with Jiang, 2011). On the basis of the Kidson (2000) weather regimes, Lorrey et al. (2007) derived three basic climate regimes for New Zealand and subsequently examined the predominance of these regimes in relation to the 1976/1977 IPO phase change for the 1958–1995 period. They showed that at the annual level, no appreciable difference in climate regime occurrence was found on either side of the 1976/1977 IPO phase change. At the seasonal level, blocking regime was more frequent during spring, and zonal regime more frequent through autumn and winter after 1976. When broken down by month, they demonstrated that there were key changes in the occurrence of climate regimes throughout the year, with a statistically significant change found in late winter associated with the 1976/1977 IPO phase change (a change from more frequent blocking regime to more frequent zonal regime in August). Recently, Renwick (2011) updated the time series of the Kidson weather types to the period 1958–mid-2010 and subsequently examined the monthly occurrence of the Kidson regimes in relation to SOI and SAM. It was found that the phase changes of both SOI and SAM have significant influence on the monthly frequency of the Kidson regimes, with more zonal regimes during the negative SOI phase and more blocking regimes during the positive SOI phase (especially in summer and autumn), more trough regimes during the negative SAM (in all seasons of the year) and more zonal and blocking regimes during the positive SAM (with the largest changes found in winter and summer, respectively).

In summary, the existing studies have identified a similar correlation between the SOI (one key metric of ENSO) phases and the occurrence of synoptic weather types over New Zealand, and some preliminary investigations conducted on the linkages between SAM or IPO and the regional synoptic types (or aggregation of synoptic types). Previous studies generally focussed on examining the observed changes in the frequency of synoptic weather types or weather regimes with respect to ENSO, SAM or IPO separately. In reality, however, the observed changes in type frequencies may result from the collective/confounding effects of different climate modes (e.g., ENSO and SAM). For example, the probability of occurrence of a synoptic type during a period with both negative SOI and negative SAM may differ significantly from that in a period with negative SOI and positive SAM. Furthermore, the existing synoptic climatological studies for New Zealand, as are those for other regions in the world, often assumed that the strength and air-mass characteristics of individual synoptic types remain unchanged over time (Sheridan and Lee, 2010). However, theoretically, it is logical to expect that ENSO, SAM, IPO or the long-term trend occurring to the earth–atmosphere system like global warming (IPCC, 2007) also modulates the strength and/or air-mass features of regional synoptic systems, consequently leading to enhanced potential for regional climate anomalies (e.g., more frequent climate extremes). In light of these considerations, this article extends the work of Jiang (2011) by presenting a probabilistic analysis that quantifies the effects of ENSO, SAM and IPO on the frequency, intensity and air-mass characteristics of daily synoptic weather types over New Zealand. The analysis is unique in (1) that generalized linear models (GLMs; Dobson and Barnett, 2008) were adopted to simultaneously [with the maximum likelihood (ML) method] estimate the expected changes in the frequency of synoptic types associated with different phases of ENSO, SAM and IPO and (2) that the subtle changes in the intensity of synoptic types are examined and linked to local meteorological variables at the Auckland Airport site. The analysis was conducted on the recently established daily synoptic index defined by Jiang (2011), as detailed in Section 2.1. The synoptic index has proved useful for the interpretation and prediction of rainfall events in Auckland (Jiang et al., 2011a). This article will provide further evaluation of the utility of the Jiang (2011) classification in the New Zealand context as a whole. The data and methods are described in Section 2, whereas the results and discussions are presented in Section 3. A summary of the analysis is given in Section 4.

2. Data and methods

2.1. Jiang classification index

Jiang (2011) established a long-term (1958–2008) daily synoptic climatological index from the daily 0000 UTC (12:00 NZST) NCEP/NCAR 1000 hPa geopotential height fields for the New Zealand region using a two-stage objective classification procedure consisting of obliquely rotated T-mode PCA followed by convergent K-means clustering. The index provides a categorization of the daily weather patterns into 12 synoptic types for the period from January 1958 to December 2008. The 12 synoptic types are shown in Figure 1 and briefly described in Table I. The properties of these weather types are detailed in the study by Jiang (2011).

Figure 1.

The mean 1000 hPa geopotential height (m) maps and frequencies for 12 synoptic weather types obtained from NCEP/NCAR reanalysis for 1958–2008. The contour interval is 20 m. Source: Jiang (2011)

Table I. Description of 12 synoptic weather types
TypeFrequency (%)Description
  1. Source: Jiang (2011)

SW16.1Southwesterly flow; high to the northwest of the country
HS5.7Blocking type: high near south of South Island; shallow trough to the north of North Island
W13.7Westerly flow; shallow trough extending from the south
LN3.7Low near North Island; a blocking high in the southwest
H9.9High centred near west of the country
TSW7.3Southwesterly flow; low/trough to the southeast, tilting southeast–northwest
S6.1Southerly flow; ridge in the south and trough in the north, both tilting southeast–northwest
NE5.1Blocking type: northeasterly flow; low/trough to the northwest, high/ridge to the southeast
L4.4Low near South Island
HN13.8Zonal flow; high over North Island
HSE6.7Blocking type: high to the east of South Island
HNE7.6High to the east of North Island

It is worth noting that the main difference between the Jiang (2011) and Kidson (2000) classification schemes lies in how the seeds/prototypes were determined for starting the K-means clustering of daily weather maps. Jiang (2011) determined the prototypes for the second-stage K-means clustering using the T-mode PCA method based on the underlying correlation structure of data. Kidson (2000) began the K-means clustering with a large number of random seeds (e.g., 100 seeds), which then converged to a small number (e.g., 12 seeds) of clusters (weather types). In addition, Jiang (2011) largely suppressed the possible intensity effect of synoptic systems (e.g., seasonal and interannual variations in the intensity of weather systems) on the classification process by spatially standardizing the daily goepotential height maps prior to the convergent K-means clustering. Consequently, Jiang (2011) classified the daily weather maps from both time and phase space, focussing on the similarity in map shapes by allowing variations in the intensity of synoptic systems within each synoptic class, while Kidson (2000) has potentially more strength for comparing regions of large variance (Jiang, 2008). Hence, the Jiang (2011) classification is expected to be more suitable for comparison of both frequencies and intensities of synoptic weather types over time.

2.2. Definition of SOI, SAM and IPO phases

The SOI time series was used to indicate the phase changes of ENSO. The definition of the SOI phases is the same as that used in the study by Jiang (2011). The monthly SOI values for 1958–2008 were based on a revised calculation by the National Climate Centre of the Australian Bureau of Meteorology. The climate base period is from 1933 to 1992, and the new SOI is the actual values multiplied by ten (Allan et al., 1996). The SOI values were used to allocate each month to an SOI phase: a month was classified to the positive/high SOI phase if SOI > 7 (La Niña), to the negative/low SOI phase if SOI < − 7 (El Niño), to the normal/neutral SOI phase otherwise (neutral ENSO phase).

The monthly SAM Index (SAMI) for 1958–2008 was obtained from the British Antarctic Survey website (Marshall, 2003). The SAMI is the normalized difference in the mean zonal pressure at 40 and 65°S as derived from 12 weather stations. The monthly SAMI values were used to allocate each month to a SAM phase: a month was classified to the positive/high SAM phase if SAMI > 1, to the negative/low SAM phase if SAMI < − 1, to the normal/neutral SAM phase otherwise.

We define positive and negative IPO phases based on the IPO index obtained from the Hadley Centre, Meteorological Office, UK (Folland et al., 1999). As in the study by Jiang et al. (2011a), the period of 1977–1998 was classified as an IPO positive phase, while the 1958–1976 and 1999–2008 periods were classified as IPO negative1 and negative2 phases, respectively. The separation of two negative IPO phases was to facilitate comparison of results with the existing studies which focussed on the effects of the 1976/1977 IPO phase change on the climate variability in New Zealand.

2.3. Local meteorological data

The local meteorological data were from the Auckland Airport monitoring site, including 24-h rainfall totals and hourly measurements of wind speed, wind direction, air temperature, relative humidity, mean sea-level air pressure and global solar radiation. The length of records varies slightly for different variables, being 1962–2008 for rainfall, 1970–2008 for solar radiation and 1966–2008 for other variables. The hourly wind vector was converted into its west–east (u) and south–north (v) components. The time series of each meteorological variable was obtained by calculating the average values for the time period 11:00–13:00 NZST on each day from 1962 to 2008, except rainfall for which there was only one value per day. Seasonality was removed from the daily time series for each variable by subtracting the monthly mean climatology for the 1970–2008 period from the daily values of that variable (the normalization based on the climatology for the whole time series led to similar results). The resulting time series were further examined for the existence of long-term trends, with the linear trends identified and removed from the air temperature, relative humidity, solar radiation, u-component and wind speed data. The final daily anomalies were used to characterize the air-mass properties of individual synoptic types.

2.4. Methodology

This investigation was conducted in three steps. The first step was to quantify the effects of SOI, SAM and IPO, both annually and for each season, on the frequency of occurrence of synoptic weather types. Previous studies examined the observed changes in type frequencies separately for ENSO, SAM or IPO, without taking into account the possible confounding effects of different large-scale climate modes. In this analysis, we adopted the GLM approach, using the ML method, to simultaneously estimate the effects of SOI, SAM and IPO (detailed below). The second step was to determine whether the intensity of synoptic weather systems changes with the phases of SOI, SAM or IPO. For each synoptic type, an anomaly pattern was established by taking the difference in the mean type maps between the positive or negative SOI (SAM) phase and the neutral SOI (SAM) phase. Similarly, anomaly patterns were also obtained for the IPO positive phase (1977–1998) and IPO negative2 phase (1999–2008), compared with the IPO negative1 phase (1958–1976). The third step was to identify the significant changes occurring to the air-mass characteristics of individual synoptic types across the different phases of SOI, SAM or IPO. As in the study by Jiang et al. (2004), the air-mass characteristics of each synoptic type were expressed in terms of the meteorological variables for the Auckland Airport monitoring site (Section 2.3). In this article, for simplicity and owing to space limitations, the investigation in the second and third steps was only conducted at the annual level.

Through the GLM approach, the frequency (probability) of occurrence of the 12 synoptic types by Jiang (2011) may be modelled together as a multinominal quantity (with 12 categories), or, modelled separately as individual binary outcomes (with 2 categories for each type). For easy interpretation of the results, we adopted the latter approach in the current analysis. Logistic regression models (a member of the GLM family) were developed to account for the effects of the phase changes in SOI, SAM and IPO (as covariates) on the occurrence of each synoptic type (as the binary response variable). That is, letting Yit indicate an occurrence of some weather type t on the ith day,

equation image

Yit has a Bernoulli (pit) distribution, where pit is the probability of having type t on day i, with t indexing one of the 12 synoptic types as shown in Figure 1 and i = 1, …, N (total number of days). For each synoptic type, the annual effects of SOI, SAM and IPO can be modelled as follows:

equation image(1)

The inclusion of the seasonality effect term in Model (1) is to suppress the mean seasonal variability in the occurrence of synoptic types (Kidson, 2000; Jiang, 2011). In other words, the main effects of SOI, SAM and IPO are modelled after taking into account the natural seasonal cycles associated with the mean large-scale circulations over the study region.

Studies have shown that the effects of SOI and SAM on the type frequencies vary with seasons (Jiang, 2011; Renwick, 2011). Hence, logistic regressions were also performed for each season separately. For each synoptic type, Model (1) is then modified as follows:

equation image(2)

where pits is the probability of occurrence of type t in season s (summer = December–February, autumn = March–May, winter = June–August, spring = September–November) on day i.

To estimate the parameters in Model (1) or (2), we need to specify a base/reference category for each categorical covariate. For convenience, we defined summer as the base category for the seasonal effect (only applies to Equation (1)), the normal phase as the base category for the SOI and SAM effect, and the IPO negative1 phase (1958–1976) as the base category for the IPO effect. Hence, the effects of the covariates were actually modelled relative to the base categories (Jiang et al., 2011a). The model parameters were estimated using the ML method. The ML method technically facilitates the estimation of the collective effects of SOI, SAM and IPO (covariates on the right-hand side of the model) on the occurrence of individual weather types (as expressed in the logit form on the left-hand side of the model) in a holistic framework. It is known that, for a logistic regression model, the exponential transformation of a parameter estimate can be interpreted as an odds ratio (here, the odds ratio for the occurrence of a synoptic type compared with the base category). Hence, in this text, the modelled effects of the covariates are described as percentage of changes in the odds of occurrence of individual synoptic types. The use of odds facilitates the comparison of results across weather types, seasons and covariates (i.e., SOI, SAM and IPO).

A few considerations should be noted in the development of the GLM models. First, no time-dependent terms (linear or nonlinear function of time) were included in the regression models, as the trend in type frequencies was regarded insignificant in this analysis (Jiang, 2011). Second, for simplicity, the possible interactions between covariates (e.g., SOI and IPO) were not considered in the modelling process in this article. Third, the predictive power of the regression models will not be discussed in this text, as the purpose here was to simultaneously estimate the changes in the frequency of synoptic types using the ML method. Last, also for simplicity, the present investigation was limited to modelling the simultaneous effects of the covariates, with the effects of weather type persistence (autocorrelation, e.g., type lifetime or synoptic sequence) not included in Equations (1) and (2). These aspects will be discussed further elsewhere.

3. Results and discussion

3.1. Expected changes in the frequency of synoptic weather types

This section describes the modelled changes (expected changes, hereafter) in the frequency of occurrence of synoptic weather types in relation to the phase changes of SOI, SAM and IPO. Associated with the logistic regression models, the exponential transformation of the parameter estimates for each covariate may be interpreted as the odds ratio of a synoptic type compared with the base category. An odds ratio < 1 indicates a decrease in the odds of a weather type, whereas an odds ratio > 1 suggests an increase in the odds of the weather type. The changes in odds (as percentages) were calculated both annually and for each season, for each synoptic type and by the phases of SOI, SAM and IPO, respectively. To focus the discussion on the significant changes in odds, only values that are significant at the 0.05 level for a chi-square test are shown in Table II for the annual effect and in Table III by seasons. A plus (+) or minus (−) sign in the table indicates an increase or decrease in odds that is statistically significant at the 0.1 level but not at the 0.05 level.

Table II. Modelled annual changes (%) in the odds of individual synoptic types.a
TypeSOISAMIPO
 Negative versus normalPositive versus normalNegative versus normalPositive versus normalPositive versus negative1Negative2 versus negative1
  • a

    Shown are values significant at the 0.05 level for a chi-square test. A positive or negative sign indicates that the increase or decrease is significant at the 0.1 level but not at the 0.05 level. IPO positive phase: 1977–1998; IPO negative1 phase: 1958–1976; IPO negative2 phase: 1999–2008

SW42.3  − 11.3  
HS− 23.623.0− 36.5   
W18.8− 12.454.0− 24.9  
LN− 22.4− 22.821.122.5 − 21.3
H− 30.917.5− 32.6+  
TSW17.0− 17.018.8− 20.0 
S23.6  20.3  
NE− 45.8   
L− 28.6 50.3 + 
HN13.2 − 14.9   
HSE− 23.017.1− 15.230.6 21.9
HNE     
Table III. Modelled changes (%) in the odds of individual synoptic types by seasons.a
 Modelled phase effectsSeasonSWHSWLNHTSWSNELHNHSEHNE
  • a

    Shown are values significant at the 0.05 level for a chi-square test. A positive or negative sign indicates that the change is significant at the 0.1 level but not at the 0.05 level. IPO positive phase: 1977–1998; IPO negative1 phase: 1958–1976; IPO negative2 phase: 1999–2008

ENSO (SOI)NegativeAutumn67.7− 42.4  − 32.5+43.8− 39.3  − 29.0
 versusWinter+65.031.9 − 42.2  − 39.0    
 normalSpring59.5− 76.1+− 41.2 31.2 − 66.8− 39.8 − 57.7
  Summer35.1− 30.0 − 30.140.0+− 45.8  +
 PositiveAutumn    38.6− 54.3   26.8  
 versusWinter    − 35.2    − 30.1  
 normalSpring   +       
  Summer 45.7− 45.5− 48.828.9− 47.3 53.8 66.0
SAMNegativeAutumn − 45.687.579.6+  77.2− 19.2  
 versusWinter 26.9 − 36.3   41.6   
 normalSpring  69.2 − 49.4 − 38.7  − 30.3 
  Summer − 32.333.4+− 35.243.335.9 104.7   
 PositiveAutumn  +  +   37.9
 versusWinter  − 28.4 45.7− 39.7  − 43.854.849.145.1
 normalSpring− 17.461.6+        
  Summer− 24.4 − 32.492.3   84.7 + 
IPOPositiveAutumn− 28.0   29.7− 38.7   24.4  
 versusWinter − 33.0 + − 21.5      
 negative1Spring    − 39.8+      
  Summer  44.5     67.1 +
 Negative2Autumn− 35.6  + + 30.045.6 
 versusWinter+ 27.8         
 negative1Spring    47.9− 62.5    
  Summer    − 24.4110.4     

3.1.1. Expected changes in the positive and negative SOI phases

At the annual level (Table II), the negative SOI is associated with 13–42% increases in the odds of the SW, W, TSW, S and HN types (which provide prevailing westerly to southwesterly winds over New Zealand) and 13–46% decreases in the odds of synoptic situations with an anticyclone centred near the South Island (H, HS, and LN) or to the east of New Zealand (NE, HSE, and HNE) (which provide easterly or northerly airflows over the study region), with a 29% decrease also found in the odds of the L type. The largest increase in odds (42%) occurs to the SW type, whereas the largest decrease in odds (46%) corresponds to the NE type. On the other hand, during the positive SOI phase, there are around 17–23% increases in the odds of the anticyclonic types HS, H and HSE (which provide easterly to northeasterly flows over New Zealand) and about 12–17% decreases in the odds of cyclonic types (TSW and W) (which are associated with prevailing westerly or southwesterly airflows over New Zealand). It is interesting (yet not readily understood) that the odds of the LN type decreases (by 22–23%) during both the positive and negative SOI phases compared with the normal SOI phase. Overall, the negative SOI (El Niño) has broader effects on the occurrence of synoptic types than the positive SOI (La Niña).

The effects of SOI on the occurrence of synoptic weather types differ significantly across seasons (Table III). Compared with the normal SOI phase, the most significant changes in type frequencies occur in spring during the negative SOI phase, but in summer during the positive SOI phase. It is notable that the change in the odds of the SW and NE types is consistent across different seasons during the negative SOI phase. Such consistency manifests as the largest annual changes in the odds of these two types in Table II. Across four seasons, however, during the same SOI phase (either positive or negative), the odds of some weather types can change in opposite directions. For example, during the negative SOI, the HS type is more prominent in winter (positive change) but less frequent in spring and secondarily autumn and summer (negative changes). Hence, because of the cancellation of effects, the changes in odds for HS appear smaller in amplitudes at the annual level.

In general, the modelled changes in the odds of synoptic types are in good agreement (in a qualitative sense) with the observed changes in type frequencies reported in the study by Jiang (2011) for the same classification, and also are consistent with Kidson (2000) and Renwick (2011) who examined the correlation between SOI and the Kidson weather regimes (details in Section 1). However, as expected, there exist quantitative discrepancies between the present analysis and Jiang (2011), in particular in terms of the relative importance of the changes in type frequencies across seasons. For example, associated with the positive SOI phase, Jiang (2011) reported that the NE type has a significantly increased (decreased) occurrence in summer (spring), but the present analysis (modelling) suggests that a significant frequency change only occurs in summer. Moreover, Table III shows that in the positive SOI phase, the modelled change in the odds of the NE type (53.8) is greater than that of the HS type (45.7) in summer. In contrast, Jiang (2011) reported that during positive SOI, the (observed) rate of frequency change for the HS type is greater than that for the NE type in summer. While the parameter estimates (odds) in the present analysis are subject to modelling errors, the discrepancies between two studies are associated with the confounding effects from other climate modes (e.g., SAM and IPO) on the probability of occurrence of synoptic types over the study region, as will be further discussed in Sections 3.1.4 and 3.1.5.

3.1.2. Expected changes in the positive and negative SAM phases

The odds of synoptic weather types also changes significantly over the different phases of SAM. At the annual level (Table II), compared with the normal SAM, the negative SAM is associated with up to 54% increases in the odds of the cyclonic types (W, L and TSW) and up to 37% decreases in the odds of the anticyclonic types (HS, H, HSE, HN and NE). On the other hand, the positive SAM is associated with up to 31% increases in odds for the anticyclonic types (HSE, S and H) which provide relatively settled situations over New Zealand, and up to 25% decreases in the odds of the W, TSW and SW types which provide westerly to southwesterly winds over New Zealand. It is noteworthy that the LN type, characterized by a low centred near the North Island and a high to the south of the South Island, is associated with 21–23% increases in odds during both the positive and negative phases of SAM compared with the normal phase of SAM (as is opposed to the changes in relation to SOI; the mechanisms for these increases are unclear and deserve attention in future studies). Overall, the changes in odds are relatively more significant during the negative SAM than that during the positive SAM, but with the asymmetry not as significant as for SOI. Furthermore, the effects of SAM also vary significantly with seasons (Table III). The positive SAM has the strongest influence in summer (with up to 92% changes in odds) and winter (with up to 55% changes in odds), whereas the negative SAM has the most significant effect in summer (with up to 105% changes in odds) and secondarily in autumn (with up to 88% changes in odds).

Although not directly comparable for individual weather types, the modelled effects of SAM shown in Tables II and III are generally consistent with the results reported by Renwick (2011), where the mean monthly frequencies of the Kidson (2000) regimes were examined for the positive and negative SAM. For example, Renwick (2011) noted that there are more trough regime during the negative SAM (in all seasons of the year) and more zonal and blocking regime during the positive SAM (with the largest changes found in winter and summer, respectively). Furthermore, the present analysis also agrees with Kidston et al. (2009) in the major aspects. For instance, while Kidston et al. (2009) found that the mean position of the Southern Hemisphere storm track shifts equatorward during the negative SAM phase, this study identified a significant increase in the odds of cyclonic types (L and W) over the New Zealand region. These findings agree with the observation that the negative SAM is often associated with enhanced westerly winds and unsettled weather over New Zealand, whereas the positive SAM corresponds to relatively light winds and more settled conditions in this region (Renwick and Thompson, 2006).

3.1.3. Expected effects of positive and negative2 IPO phases

The changes in the odds of synoptic types with respect to IPO phases are less extensive, especially, at the annual level, than those associated with SOI or SAM (Tables II and III). A few previous studies examined the impacts of the 1976/1977 IPO phase change on New Zealand regional climate variability (Salinger and Mullan, 1999; Salinger et al., 2001a; Salinger et al., 2008). Hence, for clarity, we report the modelled effects of IPO on the frequency of synoptic types separately for the IPO positive phase (1977–1998) and the IPO negative2 phase (1999–2008) compared with the IPO negative1 phase (1958–1976). Changes in the IPO positive phase (1977–1998) compared with the IPO negative1 phase (1958–1976) At the annual level, there are only two synoptic types exhibiting significant changes in odds at the 0.1 level for a chi-square test. Compared with the IPO negative1 phase (1958–1976), the IPO positive phase (1977–1998) is associated with a significant increase (17%) in odds for the L type (with a low centred to the west of the South Island) and a significant decrease (11%) in odds for the TSW type (with a low centred to the southeast of New Zealand). The effects of the 1976/1977 IPO phase change vary significantly with seasons (Table III). In autumn, increases in odds occur to the zonal types H (30%) and HN (24%), which is accompanied by decreases in odds for the trough type TSW (39%; with prevailing westerly wind) and the southwesterly type SW (28%). An opposite change tends to occur in spring, with an increased odds for the trough type TSW and deceased odds for the anticyclonic type H. In winter, there is a decrease for the TSW and HS types but an increase for the LN type (which seems counterintuitive, as is consistent with Jiang, 2008). In contrast, summer is associated with significant increases in odds for the cyclonic types W and L types, yet with the blocking states HSE and HNE exhibiting divergent changes in odds.

The modelled results are consistent (in the major aspects) with Lorrey et al. (2007), who examined the effects of the 1976/1977 IPO phase change on the monthly occurrence of New Zealand climate regimes (summation of the Kidson type frequencies) during 1958–1995. That study noted that at the annual level, no appreciable difference in the climate regime predominance was found on either side of the 1976/1977 IPO phase change. At the monthly level, however, considerable changes occurred to the predominance of climate regimes for 1958–1995. After the 1976/1977 IPO phase change, trough regime was more frequent from early winter through early summer, but less frequent in late autumn. In contrast, blocking regime for the same period was more frequent for late springtime (with peak predominance in late summer), but decreased significantly in winter (particularly in August). Moreover, the present analysis agrees well with the anomalous precipitation and sea-level pressure (SLP) patterns identified for New Zealand in relation to the 1976/1977 IPO phase change (Salinger and Mullan, 1999; Salinger et al., 2001a, 2001b, 2008). For example, the annual changes in the odds of the TSW and L types (Table II) imply a significant change in the frequency of occurrence of precipitation events over the South Island between the two IPO phases. Consistently, Salinger and Mullan (1999) showed that, in terms of New Zealand annual rainfall, the region that was most influenced by the 1976/1977 IPO phase change was the west coast of the South Island. At the seasonal level, the frequency changes indentified in Table III are very consistent with Salinger et al. (2001a). They showed that, compared with the prior-1976 negative phase, the SLP fields in the IPO positive phase showed a strong increase over the entire New Zealand region in autumn, but a decrease over and to the east of the country in spring. In contrast, higher pressures were found near New Zealand in winter, whereas lower pressures occurred to the southeast of the country in summer. They also showed that the largest changes in precipitation include an increase in the west and south of the South Island in summer, but a decrease over the northern North Island in autumn.

In summary, while the modelled changes in the frequency of synoptic types are generally consistent with previous studies in terms of the mean anomalies in the large-scale SLP field and the precipitation records over New Zealand, these results, especially at the annual level (Table II), do not conform to our general contention that the change to the IPO positive phase is associated with increased prevalence of southwesterly circulations over the study region (Salinger and Mullan, 1999; Salinger et al., 2001a). This discrepancy may be due to the coupling effect of the relatively weak strength of the IPO signal compared with ENSO or SAM and the complex interactions between these large-scale climate modes. Technically, when climate modes of different time scales are considered together in the GLM models (Equations (1) and (2)), it is expected that the effects of climate modes with higher frequencies (larger variance) (i.e., ENSO and SAM) will dominate the analysis. In other words, the GLMs have potentially less power for detecting the IPO signal from the data. Lorrey et al. (2007) demonstrated that associated with the 1976/1977 IPO phase change, no statistically significant changes were observed in the climate regime predominance until the assessment was conducted at the monthly level (as opposed to the annual level). It is possible that the above discrepancy may be related to the changes in the frequency (odds) of other synoptic types which were not identified at the annual level. The collective effects of those frequency changes, although statistically insignificant individually, may to some degree account for the observed southwesterly anomalies over New Zealand during the IPO positive phase (compared with the IPO negative1 phase). However, based on the present analysis, it appears that the previously identified anomalous southwesterlies during the IPO positive phase are associated with the confounding effects of ENSO on type frequencies as well as the possible changes in the intensity of regional circulations accompanying the state changes of both ENSO and IPO. The more frequent occurrence of El Niño events during 1977–1998 might have contributed to the enhanced southwesterlies in the IPO positive phase over New Zealand (Salinger and Mullan, 1999; Jiang et al., 2004; Lorrey et al., 2007). As will be shown in Section 3.2, the intensity of synoptic systems varies with the phases of IPO. It was found that the strongest anomalous westerly gradient flows are associated with the SW, W and S types (associated with prevailing westerlies or southwesterlies) and mainly located over southern New Zealand or the Southern Ocean. Consistently, the Auckland Airport meteorological data (Section 3.3) also suggest the association of stronger westerly winds with the IPO positive phase under the influence of weather types that provide prevailing southwesterly wind regimes over New Zealand (e.g., SW, S and TSW). Hence, the interactions between ENSO and IPO (Folland et al., 2002; Schneider and Cornuelle, 2005; Folland, 2008) and the intensity effects of large-scale climate modes, which are not yet considered in the GLM model development for this study, may deserve more in-depth investigations in future research. Changes in the IPO negative2 phase (1999–2008) compared with the IPO negative1 phase (1958–1976) There is limited knowledge available in the literature on the influence of the recent IPO phase (1999-present) on the synoptic types over New Zealand. In addition, based on the current understanding of IPO phase duration of 20–30 years (Folland et al., 2002), the present IPO negative phase is far from complete. As such, the synoptic types that have occurred in the current IPO negative phase may inadequately represent the total range of traits normally exhibited within this climate mode. Therefore, results from the negative2 versus negative1 phase intercomparison should be viewed cautiously and revisited in the future. This section focusses on describing the modelled changes in the odds of synoptic types, with further evaluations of the results to be reported elsewhere.

Annually, when compared with the IPO negative1 phase (1958–1976), the IPO negative2 phase is related to significant increases in mid-latitude blocking situations, with decreased odds for the LN and NE types but increased odds for the HSE type. These changes imply that the recent negative IPO phase tends to provide higher chance of relatively drier conditions over the North Island and possibly to the east of New Zealand, compared with the IPO negative1 phase. At the seasonal level, autumn has decreased odds for the SW, W and TSW types (which contribute to prevailing southwesterlies) and increased odds for both zonal (HN and H) and blocking states (HSE and NE). This change is similar to that occurring in the IPO positive phase for the same season, possibly implying less frequent precipitation events over the western or southern New Zealand because of reduced occurrence of southwesterly winds, and more frequent rainfall over the eastern part the country because of increased occurrence of northerly or northeasterly flows. Spring has more frequent southerly wind regimes (S) and less frequent northeasterly type (NE), accompanied by less frequent blocking situation LN and zonal type H. It is expected that the frequency of precipitation may decrease over the North Island, but increase over the southern South Island. Summer has significantly increased odds for the TSW type but decreased odds for the H type, indicating more frequent southwesterly flow and possibly increased chance for precipitation over the western and southern South Island. The main changes in winter are characterized by increased occurrence of the W and SW types that provide prevailing southwesterly winds over New Zealand, implying an increased chance of more frequent precipitation in the western part of the country.

Overall, the effects of the recent negative IPO phase appear different from that of the (1958–1976) IPO negative1 phase, with the changes in type frequencies not consistent with the general contention that New Zealand will receive more precipitation in the northeast of the country (Salinger and Mullan, 1999). This finding possibly reflects a mechanism change in how IPO influences the weather and climate of New Zealand. It is possible that the climate anomalies occurring within the IPO negative2 phase (compared to other phases) may result from the interactions between IPO and ENSO (Folland, 2008), as will be supported by the changes in the intensity of synoptic systems described later in Section 3.3. However, since the present analysis was based on an incomplete negative IPO phase (i.e., 1999–2008), more in-depth research is needed for any further conclusions in this aspect.

3.1.4. Comparison of the effects of SOI, SAM and IPO on type frequencies

Previous studies examined the observed changes in the type frequencies with respect to ENSO, SAM or IPO separately, without taking into account the possible confounding effects of these climate modes. This study shows that there exist confounding effects of different large-scale climate phenomena (Tables II and III). For instance, the SW type has increased odds during the negative SOI, but decreased odds during the positive SAM. It is interesting that while the changes in the type frequencies are generally in phase for SAM and SOI, the changes in odds for the L, LN and HN types are completely out of phase between the two climate modes. Logically, it is such confounding effects that determine the net/observed changes in the frequency of individual synoptic types. Therefore, the use of the GLM approach and the ML estimation method in the present analysis has facilitated a holistic assessment of the effects of SOI, SAM and IPO on the occurrence of synoptic types.

The changes in the odds of synoptic types differ significantly between SOI, SAM and IPO across seasons (Table III). For example, the negative SOI has strongest effects on the occurrence of the SW and NE types, while the phase changes of SAM have more significant influences on the occurrence of the W and L types. The occurrence of the HS type is affected by the negative SOI and secondarily negative SAM, but the H type responds to the phase changes of SOI, SAM and IPO. Overall, the phase changes of IPO have less significant influences (than SOI or SAM) on the occurrence of daily synoptic weather types, that is, the signal of IPO is relatively weak in the time space (in the time series) of synoptic-scale circulations over the study region. This finding is supported by Jiang et al. (2011a), who noted that the effect of IPO on the occurrence of wet days in Auckland is statistically insignificant.

These results are in line with studies conducted elsewhere (Chan and Zhou, 2005; Hope et al., 2006; Wang et al., 2007; Zhou et al., 2007a, 2007b; Verdon and Kiem, 2009; Jiang et al., 2011). For example, Verdon and Kiem (2009), based on monthly mean SLP data, found that the phase changes in ENSO, SAM, IPO and Indian Ocean Dipole were associated with changes in the frequency of monthly mean synoptic types over the eastern Australian region. The consistent findings across Australia–New Zealand indicate that the common signals of large-scale climate modes are indeed influencing regional weather and climate by significantly modulating the frequency of synoptic-scale weather systems in the study region (Jiang et al., 2004; Jiang, 2011).

3.1.5. Seasonal large-scale circulation linkages to regional climate drivers and synoptic type changes

For New Zealand, the preponderance of specific weather types have been previously linked to the operation of underpinning regional climate drivers that impinge on the mid-latitudes of the southwest Pacific (Kidson, 2000; Jiang et al., 2004; Jiang, 2011; Lorrey et al., 2011; Renwick, 2011). Synoptic types clearly change in frequency and importance by season, and combine to cause significant impacts on seasonal climate anomalies and the occurrence of extreme weather events in New Zealand (Fitzharris et al., 1997; Kidson, 2000; McKerchar and Henderson, 2003; Renwick, 2011). Some of the differences between summer and winter mean circulation states are important to note, and the fundamental interplays of the relative strengths of meridional and zonal anomalies caused by the ENSO, SAM and IPO phenomena working in unison all contribute to synoptic type frequency changes. Therefore, some discussion about regional circulation processes and how the synoptic types are nested therein is warranted, along with characterization of how ENSO, IPO and SAM climate drivers influence hemispheric circulation during different seasons to promote the synoptic type anomalies.

During negative SOI (El Niños), the subtropical high pressure belt to the north of the country (in part driven and contributed to by the descending limb of the Hadley circulation) and centres of west–east travelling anticyclones typically reside further north than normal. In addition, the South Pacific Convergence Zone (SPCZ) (positioned by the interplay of the prevailing westerly flow and equatorial and subequatorial easterlies arising from the northern and western flank of the Southeast Pacific Anticyclone) usually resides to the north and east of normal during negative SOI phases as a result of weaker subtropical easterlies and stronger westerlies that are displaced northward of normal. La Niñas are typically characterized by opposite responses of all the aforementioned features relative to El Niños. Positive SAM typically is associated with a more southerly position of the westerly jet, whereas negative SAM sees a more northerly position for the jet. The repositioning of the major circulation features from these two phenomena clearly could influence the seasonal occurrence of some synoptic types, in particular those that are related to extropical cyclone transitions into the area east of New Zealand and the Tasman Sea region, and low-pressure systems of mid-to-high latitude origin.

The increased occurrences of synoptic types that are largely promoted by blocking over the south Pacific to the east of New Zealand (Trenberth and Mo, 1985; Renwick, 1998; Renwick and Revell, 1999) (LN, HSE, NE and HS) produce more frequent northerly or easterly flow and/or settled (calm) conditions across the country. A commonality of these ‘blocking’ types during summer is an association with positive SOI (La Niña) and positive SAM conditions. The conjoint SOI and SAM states (which often occur as positive SOI and positive SAM during summer; L'Heureux and Thompson, 2006) would induce a southerly displaced subtropical high pressure belt (and westerly jet) and a southwest-displaced SPCZ (with stronger easterlies and highs to the north and east of New Zealand) in the southwest Pacific (note that the frequency of the LN type decreases under positive SOI, but attains more significant increases under positive SAM). Commonality of the synoptic type changes in the ‘blocking’ group during autumn suggests that the more southerly position of westerly jet (weaker westerly flow across New Zealand) generates increases in the frequency of ‘blocking’ types, while negative SOI (El Niño) and positive IPO conditions (El Niño-like) serve to diminish types that are more characteristic of northerly and easterly flow. Large decreases in the ‘blocking’ group also occur during El Niños and when low-frequency climate state is El Niño-like (positive IPO), as opposed to increases when the westerly jet is further south during winter. These interpretations are consistent with a Southern Hemisphere blocking study conducted by Sáez de Adana and Colluci (2005), which showed linkages between ENSO variability and late autumn/winter blocking event occurrences. The most significant agreement between synoptic types in the ‘blocking’ group indicate decreases occur during El Niño springtime, and reductions are associated with a westerly jet further north than normal and the positive IPO (El Niño-like) climate state in the south west Pacific.

Changes in the frequency of synoptic types that produce zonal flow across the country (SW, W, HNE and HN) are much clearer for summer than winter, whereby negative SOI (El Niño), negative SAM (a northerly displaced westerly jet position) and positive IPO conditions (low-frequency climate patterns in the south west Pacific more akin to El Niño) occur in conjunction (although the changes in HNE and HN in summer are not significant). The ‘zonal’ synoptic type changes exhibit conflicting results for SOI and SAM modulation in autumn and winter, but are associated with increases during positive IPO conditions in the late austral warm season. ‘Zonal’ type increases also coincide with El Niño springtime coincident with reductions (increases) when the westerly jet is more southerly (northerly), with exception of the HN type that decreases when SAM is negative (more northerly jet position).

The TSW, S and L types are typified by disturbed circulation over New Zealand or southerly quadrant flow that can be collectively termed ‘trough’ types. These patterns are known to produce some of the largest positive (negative) rainfall (temperature) anomalies across the country on an interseasonal basis (Renwick, 2011). Kidson (2000) established similar synoptic types to our study, and also illustrated that there were preferential swapping of ‘trough’ and ‘blocking’ types for opposite ENSO phases for some seasons. This study indicates that during summer, the ‘trough’ types appear significantly driven by a SAM negative state (more northerly westerly jet) with reinforcements by negative SOI (El Niño) and positive IPO (El Niño-like) conditions in the equatorial Pacific. In part, one could infer weaker easterly flow in the subtropics (and a northeast-displaced SPCZ) north of New Zealand during these times from the association with ENSO and IPO, which would also reduce the frequency of low-pressure patterns associated with extropical cyclone transition into the New Zealand sector in summer (Diamond et al., in preparation). Moreover, the synoptic patterns observed in the ‘trough’ group during summer appear to be nested within a hemispheric-scale circulation mode akin to a PSA wave pattern number three (PSA WP3). This observation suggests that large-scale PSA WP3 existence in summer can be driven by SAM with influences on mid-latitude wave pattern spatial expression (i.e., longitudinal and latitudinal breadth of the three low-pressure centres that radiate outward from higher to lower latitudes) generated from equatorial climate drivers such as ENSO and IPO (at least in the southwest Pacific sector). Relationships between TSW, S and L types during autumn can be reconciled by purported strengthening of ENSO teleconnections to New Zealand during the positive IPO phase (Salinger et al., 2004), but because of the shortness of documentation for that multidecadal climate oscillation further work is needed to confirm or refute this hypothesis. The decrease in the frequency of trough types in autumn under positive IPO may reflect their replacement by blocking types during the austral warm season (summer/autumn) when New Zealand is connected more strongly with ENSO variability. In addition, the increase in the relative importance of trough types during autumn occurs when polar drivers (like SAM) operate in conjunction with El Niño events (negative SOI), which could be a response to a subtly repositioned mid-to-high-latitude storm track further north than usual. The frequency of ‘trough’ types decreases when the background state is more El Niño-like (positive IPO), increases (decreases) when the jet is further north (south) during winter, while changes in the group as a whole during spring are less coherent.

3.2. Spatial variations in mean type maps—anomaly patterns

Given the significant changes that occur to the frequencies of some synoptic types, the mean maps for each weather type were compared and the spatial variations (anomaly patterns, hereafter) calculated between the different phases of SOI, SAM and IPO respectively, to determine whether the large-scale climate modes also affect the intensity of synoptic systems over New Zealand. It was found that, on the one hand, both locations and shapes of synoptic high (ridge) and low (trough) systems in the mean maps of individual weather types are very similar across the different phases of SOI, SAM or IPO, indicating the temporal consistency (stability) of the synoptic classification under examination. On the other hand, small but statistically significant (at the 0.05 level for a two-tailed t-test; e.g., Figure 2) spatial anomalies do exist, and can be linked to the existing studies in the literature, indicating that the large-scale circulation modes modulate regional synoptic system intensity over New Zealand. The main findings are described in the following three subsections.

Figure 2.

Anomalous 1000 hPa geopotential height (m) maps of the SW, NE, LN and HS types for the negative and positive SOI phases compared with the normal/neutral SOI phase. The contour interval is 3 m. Triangles indicate that anomalies are significant at the 0.05 level for a two-tailed t-test

3.2.1. Anomaly patterns by phases of SOI

There exist systematic spatial adjustments in the intensity of synoptic systems associated with the phase changes of SOI. The geopotential height anomaly patterns for the negative and positive SOI phases, as expressed by the mean difference in the goepotential height fields compared with the normal SOI phase, are exemplified in Figure 2 for the SW, NE, LN and HS types. During the negative SOI phase, the height anomalies over New Zealand are generally positive across all synoptic types (except for HS). In contrast, associated with the positive SOI phase, the height anomalies are less significant and tend to be negative over the South Island (three types) or the whole country (eight types), with only one exception (LN). The height anomalies indicate that the negative SOI contributes to enhanced (weakened) synoptic-scale anticyclonic (cyclonic) systems over the study region, but the positive SOI imposes the reverse influence on the regional circulations. Although small in amplitudes, the anomaly patterns for individual types parallel the documented modulation of ENSO to the strength of the prevailing southwesterly wind over New Zealand (Sturman and Tapper, 2008). During the negative SOI phase, the anomalous gradient flows are mostly southwesterly over the North Island (e.g., blocking situations NE, LN and HS) with exceptions in S, L and HN, but westerly or southerly over the southern South Island and near the Southern Ocean (e.g., SW and W that provide prevailing southwesterly winds) with exceptions in L and HNE. In contrast, during the positive SOI phase, the anomalous gradient flows of the synoptic types appear less systematic over New Zealand and mostly from the northwest and northeast quadrants (except for H, NE and LN) for the North Island and the northern South Island, but with the anomalous westerlies or southwesterlies (except for HS and S) mainly located further south, over the southern South Island or south of New Zealand (not shown). Overall, the anomalous gradient flows appear stronger in the negative SOI phase compared with the positive phase, consistent with the observation that El Niño episodes have stronger influence on New Zealand's weather and climate compared with La Niña events (Mullan, 1995).

3.2.2. Anomalous patterns by phases of SAM

Following the approach used above, significant spatial variations in the mean type maps were also identified for the different phases of SAM. The anomaly patterns for the positive and negative SAM phases relative to the normal SAM phase are given in Figure 3 for the SW, W, HN and HNE types. It was noticed that, for the whole region and across all weather types, negative geopotential height anomalies occur in the negative SAM phase, but positive anomalies observed in the positive SAM phase. In particular, the relatively large anomalies tend to be south of 40°S, except for HNE, HS, NE and LN (blocking types) which have large anomalies extending further north. The anomalies indicate that the positive SAM contributes to enhanced (weakened) anticyclonic (cyclonic) systems, but the negative SAM imposes opposite effects on the synoptic-scale weather systems. The geopotential height anomalies and related gradient flows vary significantly between different synoptic types. For example, the relatively larger anomalies are related to synoptic situations with a depression/trough over or near the country, e.g., TSW, W, L, LN and NE, while the anomalous centres are relatively shallow for the H, HN and HSE types compared with those for other synoptic patterns. During the positive SAM phase, the anomalous gradient dictating surface flows over New Zealand tend to be zonally orientated (i.e., easterly) for most synoptic types (with exceptions in L, LN and HNE). In contrast, the negative SAM tends to be associated with both zonal and meridional gradients over the country. These results are consistent with Kidston et al. (2009) who found that the storm track in the Southern Hemisphere shifts equatorward from its mean position during the negative SAM, also in good agreement with Marshall (2007) who noted that the positive SAM corresponds to higher-than-average pressure and weaker-than-average westerlies in mid-latitudes, but the negative SAM is related to the reverse trend in pressure and windier conditions in the New Zealand region.

Figure 3.

Anomalous 1000 hPa geopotential height (m) maps of the SW, W, HN and HNE types for the negative and positive SAM phases compared with the normal/neutral SAM phase. The contour interval is 3 m. Triangles indicate that anomalies are significant at the 0.05 level for a two-tailed t-test

3.2.3. Anomalous patterns by phases of IPO

The spatial variations in the mean type maps were examined between the different phases of IPO. For example, Figure 4 shows the geopotential height anomaly patterns for the IPO positive phase (1977–1998) and the IPO negative2 phase (1999–2008) compared with the IPO negative1 phase (1958–1976) for the SW, NE, LN and TSW types. In general, the anomaly types somehow mimic the spatial features of the mean type maps in Figure 1, exhibiting some localized variability in the intensity or location of synoptic systems associated with the phase changes of IPO. For all weather types except SW, H and HSE, high/ridge systems tend to be stronger (larger positive height anomalies) in the IPO positive phase, but the low/trough systems appear stronger (larger negative height anomalies) in the IPO negative2 phase (e.g., NE). In contrast, a reversal occurs for the SW, H and HSE types, with the positive anomaly centres being larger in the recent negative IPO phase than the IPO positive phase. Despite the exceptions, the results are consistent with previous studies that examined the SLP changes associated with the 1976/1977 IPO phase change. For instance, Salinger et al. (2001a) showed a strong increase in SLP over the entire New Zealand region in autumn, but a decrease over and to the east of New Zealand in spring; the winter and summer anomalies follow a muted version of these patterns, with winter tending towards higher pressure near New Zealand and summer lower pressure to the southeast of New Zealand.

Figure 4.

Anomalous 1000 hPa geopotential height (m) maps of the SW, NE, LN and TSW types for the positive (1977–1998) and negative2 (1999–2008) IPO phases compared with the negative1 (1958–1976) IPO phase. The contour interval is 3 m. Triangles indicate that anomalies are significant at the 0.05 level for a two-tailed t-test

The anomalous circulation over New Zealand generally propagates in similar directions to that of the prevailing winds associated with individual synoptic types (Figure 1), except for that the HS, LN and TSW types that exhibit very different anomalous flow patterns between the positive phase and negative phase2 of IPO. Strong anomalous westerly flows, mainly associated with the SW, W, HN and secondarily S types, are located over the southern New Zealand or to the south of New Zealand (over the Southern Ocean), with the gradients (and by inference the windflow) stronger in the recent IPO phase than the IPO positive phase. This finding is consistent with the contention that the recent IPO positive phase is typical of stronger westerlies and increased precipitation over the western South Island (Salinger and Mullan, 1999).

3.3. Average air-mass characteristics of synoptic types

An analysis was undertaken to examine whether the air-mass characteristics of individual synoptic types also change over the different phases of SOI, SAM or IPO. As a case study, the air-mass characteristics of synoptic weather types are expressed in mean daily anomalies of the Auckland Airport meteorological variables. The daily anomalies for each meteorological variable are deviations of daily values from the monthly mean climatology (for the common period 1970–2008), with the seasonality largely removed, so that the analysis was conducted for the full dataset as a whole (rather than by seasons). It was found that the synoptic type–local meteorology relationships are consistent across the different phases of SOI, SAM and IPO, conforming well to general meteorological principles (Sturman and Tapper, 2008) and in agreement with Jiang et al. (2004) and Jiang (2008). However, there are distinguishable within-type variations in local meteorological conditions over the different phases of SOI, SAM and IPO. Although much smaller in magnitudes compared with the across-type discriminations, these variations parallel the anomaly patterns of the mean type maps described in Section 3.2 and are very consistent with the existing studies with respect to the linkages of the large-scale climate phenomena with New Zealand climate variability. The main findings are described, respectively, for SOI, SAM and IPO in this section.

3.3.1. Average air-mass characteristics by SOI phases

Figure 5 shows the mean air-mass characteristics and their 95% confidence intervals (CI) by synoptic types for the different phases of SOI. The level of separation between a pair of CIs indicates whether the related mean values are significantly different: the larger the separation, the more different the mean values; the two overlapping CIs indicate that the mean values are not significantly different. It can be seen that, the between-type discriminations of local meteorological conditions are very similar over the different phases of SOI. The most significant within-type variations in local meteorology occur to the anomalies of v-component (e.g., SW and HNE), temperature (all types except LN) and relative humidity (e.g., SW, W, LN and HNE), with the across-phase changes less systematic in solar radiation and 24-h rainfall. These variations over the SOI phases conform well to the general perception of the influence of ENSO on New Zealand climate variability (Kidson, 2000). During the negative SOI phase, the synoptic types typical of prevailing southwesterlies over New Zealand (e.g., SW, W, TSW and S) correspond to stronger southwesterly (weaker northeasterly) wind anomalies, below-normal air temperature and relative humidity, possibly above-normal solar radiation, and below-normal rainfall in Auckland. During the positive SOI phase, the zonal types (e.g., H and HN) and the blocking states (e.g., HS, HSE and HNE) provide local conditions characterized by stronger than normal northeasterly (weaker southwesterly) wind anomalies, above-normal temperature, similar to or above-normal relative humidity and rainfall, and similar to or below-normal solar radiation. In summary, these within-type variations are consistent with the height anomaly patterns in the mean type maps (Figure 2), thus indicating that the state of ENSO also influences the intensity and thus air-mass characteristics of synoptic types over the New Zealand region.

Figure 5.

Means (circles) and 95% confidence intervals (bars) of the Auckland Airport meteorological variables by synoptic types and phases of SOI. This figure is available in colour online at wileyonlinelibrary.com/journal/joc

3.3.2. Average air-mass characteristics by SAM phases

The average air-mass characteristics were also examined in relation to the phases of SAM, with examples given in Figure 6. Again, the local meteorology–synoptic type relationship is very consistent over the different phases of SAM. The most distinguishable variations across the SAM phases were found in the anomalies of mean SLP (all types except LN, S and L), secondarily, u-component (e.g., SW, W, HN and HNE), wind speed (e.g., SW and W) and relative humidity (e.g., H, HN, HSE and HNE). In general, if compared to the normal SAM phase, the positive SAM contributes to more settled weather conditions in Auckland (higher air pressure, lower wind speed, weaker southwesterlies, lower relative humidity and higher temperature), but the negative SAM appears to be associated with the reverse conditions. Overall, these results are consistent with the anomaly patterns identified in the synoptic types (Figure 3) and in good agreement with the existing studies, e.g., by Renwick and Thompson (2006), Marshall (2007) and Kidston et al. (2009). Therefore, we infer that SAM also modulates the intensity of synoptic systems and thus affects the air-mass characteristics of synoptic types.

Figure 6.

Examples of means (circles) and 95% confidence intervals (bars) of the Auckland Airport meteorological variables by synoptic types and phases of SAM. This figure is available in colour online at wileyonlinelibrary.com/journal/joc

3.3.3. Average air-mass characteristics by IPO phases

In a similar fashion, the air-mass characteristics of synoptic weather types were also examined with respect to the positive and negative phases of IPO, with examples given in Figure 7 (the linear trends were removed from the air temperature, relative humidity, u-component and wind speed anomalies). The solar radiation data were not used in this analysis, because the records cover only 5 years of the IPO negative1 (1958–1976) phase. Again, the local meteorology–weather type relationships (i.e., the atmospheric and terrestrial climate responses to an imposed circulation pattern) are generally stationary across the phases of IPO, consistent with Jiang (2008). Significant changes occur to the SW, HNE and HS types for local wind conditions (especially wind speed), the HS, NE and HN types for temperature, the TSW type for relative humidity and the HS type for mean SLP. The variations in local wind anomalies are accompanied by changes in temperature and relative humidity across the IPO phases. Overall, the SW, S, W, TSW and L types, which are associated with prevailing westerly or southwesterly air flows over New Zealand, correspond locally to stronger winds, lower relative humidity and temperature in the positive IPO phase than the negative IPO phases. In contrast, the anticyclonic types H, HN, HS, HNE, NE and LN (mostly blocking states) are associated with lighter winds, lower temperature, and higher/similar relative humidity in Auckland during the IPO positive phase than the negative1 IPO phase, with the mean wind speed and temperature (except for NE and LN) lower than the IPO negative2 phase. These results are consistent with the observation that New Zealand experiences stronger westerlies in the positive IPO than the negative IPO (Salinger and Mullan, 1999; Salinger et al., 2000, 2001b).

Figure 7.

Examples of means (circles) and 95% confidence intervals (bars) of the Auckland Airport meteorological variables by synoptic types and phases of IPO. IPO positive phase: 1977–1998; IPO negative1 phase: 1958–1976; IPO negative2 phase: 1999–2008. This figure is available in colour online at wileyonlinelibrary.com/journal/joc

4. Summary and conclusion

This article documents an analysis that holistically examines the effects of SOI (thus ENSO), SAM and IPO on the frequency, intensity and air-mass properties of synoptic weather types over the New Zealand region. The analysis was based on the Jiang (2011) classification and the application of the GLM approach (i.e., logistic regressions). The GLM approach provides a tool to simultaneously estimate the expected effects of ENSO, SAM and IPO, so that results are comparable across seasons, synoptic types and different large-scale climate modes. The results are consistent with previous studies (Jiang et al., 2004; Jiang, 2008; Jiang, 2011; Renwick, 2011), with the main findings summarized below:

  • 1)The phase changes of SOI, SAM and IPO have significant effects on the frequency of occurrence of certain synoptic types over the New Zealand region. Such effects are asymmetric between the negative and positive phases and can vary significantly with seasons. Overall, the effects of SOI and SAM appear to be more significant than that of IPO, implying that, for short- to medium-term climate predictions in the New Zealand context, it is necessary to include both SOI and SAM as predictors.
  • 2)The influence of SOI, SAM and IPO on individual synoptic types is selective and can counter-effect one another. For example, the negative SOI is associated with decreased occurrence of the L type, while the negative SAM corresponds to increased occurrence of this synoptic state. In order words, the observed changes in the frequency of synoptic weather types usually result from the effects of more than one influential factor. In climate studies, therefore, it is meaningful to consider the effects of these large-scale climate drivers in a holistic framework, as is demonstrated in the present analysis. This also echoes the sentiment provided through the Regional Climate Regime Classification hypothesis (Lorrey et al., 2007, 2008), where the conjoint operation of several climate drivers working in synch contributes to synoptic type changes, which in turn dictate the overall circulation patterns that contribute to regional climate anomalies over seasonal and longer (centennial scale and beyond) time scales.
  • 3)While the Jiang (2011) synoptic classification is generally consistent over the different phases of the large-scale climate modes, there exist considerable within-class variations in both intensities and air-mass characteristics of individual synoptic types. Although small in amplitudes, such variations can be meaningfully related to the existing studies in the literature, indicating that the signal of these large-scale climate modes also manifest in the intensity of synoptic weather systems in this region.
  • 4)At the annual level, the significant changes in type frequencies do not conform to our general contention that the positive IPO phase was associated with enhanced anomalous southwesterly circulations over New Zealand. It is possible that the previously identified anomalous southwesterlies in the positive IPO phase (compared to the negative IPO phase) were because of the modulation to the intensity of the regional circulations by IPO and the confounding effects of ENSO. Given the correlation between IPO and ENSO, inclusion of interaction terms in the GLM models may improve the interpretability of results.
  • 5)The effect of the recent negative IPO phase appears different from that of the previous negative IPO phase (i.e., 1958–1976). This finding possibly reflects a change in the mechanism of how the IPO affects the weather and climate of New Zealand. However, bearing in mind that the present analysis was based on an incomplete negative IPO phase (i.e., 1999–2008), more in-depth research is needed in this aspect.

In conclusion, this study has demonstrated the differential effects of ENSO, SAM and IPO on New Zealand synoptic weather types in a holistic framework. Jiang et al. (2004) and Jiang (2011) noted that it is mainly the changes in the frequency of synoptic weather types that lead to the significant differences in New Zealand's climate between La Niña and El Niño events. This study further demonstrates that ENSO, SAM and IPO influence the climatic conditions in this region mainly by modulating the frequencies and to a lesser degree the intensity of synoptic systems and consequently the air-mass characteristics of certain synoptic types. The findings from this study may be useful for regional weather or climate prediction. Given that the continued climate warming will probably result in changes in large-scale atmospheric circulation patterns (Folland et al., 2003; IPCC, 2007), improved understanding of how synoptic weather systems manifest will assist in prediction of regional climate conditions associated with such changes. The findings also have important methodological implications for applications of synoptic climatology in the New Zealand or wider context. For example, Kidson (1997, 2000) noted that the synoptic climatological approach is largely of qualitative value for climate studies in New Zealand, mainly because of the very large within-type variability in climate elements. This analysis implies that the modulation by the large-scale climate phenomena might have added to the within-type variability in the existing synoptic classifications for New Zealand. Given that the existing classifications were essentially based on the discrete-cluster approach (Jiang, 2010) and each defined a small set of (often around 12) synoptic types (clusters) to represent the full range of regional circulation patterns, this study implies that the within-type variability in climate elements may be to some degree reduced by obtaining a classification with a larger (but manageable) set of synoptic types which take into account the subtle, systematic variations in the intensity of regional circulations associated with the phase changes of large-scale climate modes (e.g., ENSO and SAM). Hence, an extension of this study is to apply the batch self-organizing map weather-typing method described in the study by Jiang et al. (2011b) to the New Zealand context (ongoing).

Application of the findings in this study is also relevant to emerging palaeoclimate reconstructions based on synoptic approaches for New Zealand and Australasia (Lorrey et al., 2008, 2011; Ackerley et al., 2011). This study indicates consistency for a site-specific response to synoptic type occurrences (Auckland) for multiple climate drivers over multiple time scales. This suggests that the incident direction of the flow and orographic influences is key determinants (and most critically important) for generating anomalous patterns, and that the responses are stable through time. Proxy signals seen in palaeoclimate archives can also be considered in the same way (in terms of a circulation response). The holistic approach suggested here, in terms of the integration of multiple omnipresent climate phenomenon (ENSO, IPO, and SAM) for generating climate and weather anomalies, is in line with previous palaeoclimate work (Lorrey et al., 2007, 2008, 2011) that suggested interpreting proxy signals in terms of one driver should be cast aside in favour of multiproxy networks integrated under past circulation patterns and geopotential height reconstructions. That approach would also potentially help to reveal the relative importance of multiple drivers responsible for those patterns with greater certainty, and surmount issues of phenomenon teleconnection stability through time.

Future studies also include the development of more comprehensive prediction models for the probability of occurrence of individual synoptic types, for instance, by inclusion of the autocorrelation effects of synoptic types and the interactions between different large-scale climate modes. Furthermore, the results can be applied to other environmental problems in New Zealand, e.g., for evaluating the impacts of both large-scale modes and regional synoptic types on the air quality in Auckland.

Acknowledgements

This work was initiated when the first author studied statistics in Macquarie University, Sydney, Australia. The second and third authors acknowledge the Ministry of Science and Innovation (MSI) for supporting this work through core funding to National Institute of Water and Atmospheric Research (NIWA), and contribute to Climate Present and Past (formerly a part of ‘Adaptation to Climate Variability and Change’ contract COX0701). The authors are grateful to the National Climate Centre of the Australian Bureau of Meteorology for accessing the monthly SOI values, the British Antarctic Survey for accessing the monthly SAM Index, the Hadley Centre, Meteorological Office, UK for accessing the IPO index, and NIWA, New Zealand for accessing the Auckland Airport meteorological data from the National Climate Database. Special thanks should go to Dr James Renwick, NIWA, Wellington, New Zealand and the unknown referees for constructive comments on the text.

Ancillary