Notice: Wiley Online Library will be unavailable on Saturday 27th February from 09:00-14:00 GMT / 04:00-09:00 EST / 17:00-22:00 SGT for essential maintenance. Apologies for the inconvenience.
1. An extensive data set describing the composition of New Zealand’s remaining indigenous forests was used to estimate the degree of correlation between measures of both soil and atmospheric water deficit and the distribution of common tree species.
2. For most species, regression models incorporating measures of air saturation deficit in early autumn, as well as an annual integral of root zone water deficit, provided the best explanation of spatial distribution. This accords strongly with the mechanistic effects of air saturation deficits on transpiration from trees, and the hydraulic risks experienced by trees under high evaporative demand.
3. Adjustment of root zone water deficits to account for reductions in rainfall in dry years substantially improved model predictions. This suggests that extreme climatic events, such as the El Niño phase of the Southern Oscillation, are likely to have strongly influenced the historic composition of forests in New Zealand’s drier eastern lowlands.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
Although considerable progress was made using this approach by Leathwick et al. (1996), their regressions consistently over-predicted the abundance of many species on sites east of New Zealand’s main mountain ranges. The problem is exemplified by the geographic range predicted for Weinmannia racemosa (Fig. 1), a broad-leaved tree that is most abundant on sites to the west and/or about the main axial ranges of central and southern New Zealand (Wardle 1966). A regression based on the environmental factors described above predicts a widespread distribution throughout the South Island, despite the virtual absence of this species east of the main mountain ranges between the Marlborough Sounds and the Catlins (41°30′ to 46° S). Similarly, our model predicts an extensive distribution in eastern areas of the North Island where it is absent or rare, including the Kaimanawa, Kaweka and northern Ruahine Ranges. Preliminary analysis to explain this discrepancy led us to suspect that high atmospheric water deficits were important. Most areas east of New Zealand’s main ranges are affected periodically by extreme föhn winds (Brinkman 1971), particularly in spring (e.g. Garnier 1958), resulting in weather conditions typified by strong winds, high temperatures and very low humidity (down to 20% or below; Ryan 1987), often with high irradiance.
Some progress was made investigating the importance of spatial variation in humidity in an analysis of the relative importance of current environment vs historic events in determining marked disjunctions in the distribution of New Zealand’s Nothofagus species (Leathwick 1998). In this study, the relative humidity at 09·00 hours NZST of the least humid month was added to the set of climatic variables described above, resulting in a substantial improvement in predictions of the distribution of many species. However, full elucidation of the role of atmospheric deficit was hindered by a lack of compositional data from the sites most affected by föhn winds, which also suffered the greatest forest clearance during early Maori occupation of New Zealand (McGlone 1983). Here, we present a more comprehensive analysis of the relative importance of root zone and atmospheric water deficits in influencing tree distributional patterns by:
•converting estimates of relative humidity to air saturation deficit to separate the independent effects of temperature and humidity on transpiration by trees;
•improving the water balance model by the use of a more robust, process-based calculation for transpiration, and allowing for discrimination between the effects of air saturation deficits and root zone water deficits as regulating variables;
•investigating the effects of reduced rainfall in drought years by using a predictive surface describing rainfall variability to adjust precipitation inputs to the water balance model;
•incorporating additional compositional data from forest fragments in eastern parts of New Zealand to gain a more comprehensive understanding of species distribution in dry, lowland regions.
In global terms, New Zealand has an atypical environment, distinguished by its extreme maritime character. Despite its small total land area, its combination of wide latitudinal range, elevated axial ranges and position athwart the westerly belt of the southern Pacific result in diverse environmental conditions. These range from warm-temperate to subantarctic, and super-humid to semiarid (Meurk 1984), with landforms consisting of youthful mountains (Whitehouse 1988), volcanic landscapes – subject to recurrent, often severe, disturbance (Froggatt & Lowe 1990) – and extensive recent alluvial and aeolian surfaces.
The evergreen forests occupying these landscapes are dominated by varying mixtures of conifers, mostly from the Podocarpaceae; broad-leaved trees, many from families with subtropical to tropical affinities; and southern beech (Nothofagus) species (Wardle 1991). Broad-leaved trees are most numerous on warm, moist, fertile sites, but are generally replaced by Nothofagus as conditions become cooler, drier and/or less fertile (Wardle, Bulfin & Dugdale 1983; Wardle 1991; Leathwick 1995). All four Nothofagus species show pronounced distributional disjunctions that are not explicable in terms of environmental factors currently known to influence tree distribution at the landscape scale (Leathwick 1998). Conifers are widespread throughout New Zealand’s forests, often as scattered emergents, but they may become dominant on sites either recently disturbed or with harsh edaphic conditions (Leathwick 1995).
Materials and Methods
Data describing forest composition were drawn from a range of historic and contemporary plot-based forest surveys, and included 14 684 plots. Of these, 134 plots were additional to the data set described earlier by Leathwick (1995, 1998), and were collected specifically for this study to better describe forest composition in dry environments. Thirty-three entities were selected for analysis. Because of ambiguities in much of the original field data, three of these consisted of closely related pairs or triplets of species of similar appearance, i.e. Podocarpus totara and P. hallii;Quintinia acutifolia, Q. elliptica and Q. serrata;Nestegis cunninghamii and N. lanceolata (nomenclature follows Allan 1961; Connor & Edgar 1987). Widespread hybridism is observed between the two Podocarpus species (McKelvey 1963). Because the timber inventory that constituted the majority of the dataset largely disregarded trees below millable size (< 305 mm or 12 inches in diameter), and provided only counts of larger trees that were considered nonmerchantable, data were summarized to show numbers of trees by species with diameters > 305 mm per unit area for each plot.
Environmental variables consisted of a range of climatic and landform factors similar to those used by Leathwick (1998) (Table 1). All climate parameters were estimated from thin-plate spline surfaces that allowed the spatial interpolation of data collected at irregularly distributed meteorological station networks (Hutchinson & Gessler 1994). Surfaces for temperature and rainfall were fitted to normals for 1951–80 (New Zealand Meteorological Service 1983, 1984), the period that most closely matches the time over which most of the compositional data were collected. Improved surfaces for humidity and solar radiation were fitted to long-run average data collected up to 1980.
Table 1. Environmental variables used in the analysis
Mean annual temperature
where J= July minimum temperature; T= annual temperature, J = mean of J; σJ= standard deviation of J, etc.
Mean annual solar radiation
14·9 MJ m−2 day−1
11·8–15·3 MJ m−2 day−1
0·0 MJ m−2 day−1
0·78–1·3 MJ m−2 day−1
where J= June solar radiation; S = annual solar radiation, etc. as for Tw
Annual integral of root zone water potentials below field capacity; see text for description
The daily water balance model of Leathwick et al. (1996), which estimates root zone water deficit (W) by summing (over one year) the daily difference between the water potential at field capacity (assumed to be −0·02 MPa) and the actual water potential, was modified following Whitehead, Leathwick, & Walcroft (2001). Firstly, the evaporation component of the model was altered to allow a diurnal trend in canopy transpiration and soil evaporation using Gaussian integration (Goudriaan & van Laar 1994). Transpiration from the canopy (Ec, mm d−1) was calculated using the simple diffusion equation appropriate for aerodynamically rough canopies:
where D (kPa) is the air saturation deficit above the canopy, gs (mm s−1) is the average stomatal conductance, L (m2 m−2) is the leaf area index, cp (J kg−1 °C−1) is the specific heat of air, ρ (kg m−3) is the density of air, λ (J kg−1) is the latent heat of vaporization and γ (kPa °C−1) is the psychrometric constant. Values of gs were calculated from a process-based model coupling conductance with photosynthesis, as described in Whitehead et al. (2001), and were adjusted downwards as soil water content decreased, declining to zero at an assumed permanent wilting point (−1·5 MPa). Evaporation from the understorey and soil (Eu, mm d−1) was calculated from the available energy flux density below the canopy, Ru (MJ m−2 d−1), such that:
where τ is a dimensionless coefficient describing the degree of coupling of the understorey with the air above the canopy, and s is the slope of the relationship between saturated vapour pressure and temperature at a given temperature (kPa °C−1). Ru was calculated from the solar radiation above the canopy using Beer’s Law (see Whitehead et al. 2000). Soil rooting depths and textures, estimated by overlaying plot co-ordinates onto a digital map of New Zealand soils, were used to calculate soil water holding capacities and water potentials as described in Leathwick et al. (1996). The cumulative root zone water deficit (MPa days) was then calculated for each site, assuming, in the absence of explicit spatial data, a constant leaf area index of 5.
Secondly, to test the comparative effects of reduced rainfall expected in dry years on species distribution vs the long-run average rainfall, we calculated measures of root zone water deficit using both average monthly rainfall (P, mm), and reduced monthly rainfall (P′) inputs such that:
where α is a dimensionless coefficient and σp is the site-by-site standard deviation in monthly rainfall (mm). Monthly standard deviations were calculated from monthly estimates of the ratio of the variance to the mean for rainfall derived from a thin-plate spline surface (Leathwick, unpubl. data) fitted to data for 307 sites throughout New Zealand (New Zealand Meteorological Service 1979).
Separate estimates of mean monthly air saturation deficit (D) were calculated from monthly estimates of average minimum and maximum daily temperature, and daily relative humidity measured at 09·00 hours NZST (Jones 1994). Temperatures at 09·00 hours NZST were estimated using a sine-based interpolation method (Goudriaan & van Laar 1994).
A substantial increase in both the spatial resolution and consistency of landform variables was obtained by overlaying plot co-ordinates onto a digital copy of the New Zealand Land Resource Inventory (NZLRI) database, a polygon-based spatial database describing a range of land attributes (Ministry of Works 1974). Lithology was expanded from the six categories used in Leathwick (1998) to 15 classes derived by manipulation of the NZLRI descriptions of surface rock (G, Table 1). Use of NZLRI drainage data overcame marked inconsistencies in drainage description between the differing vegetation surveys making up the total data set – this was a five-step scale (F, Table 1) as opposed to the three-step scale used by Leathwick (1995, 1998). Finally, NZLRI data describing the predominant slope or relief (S, °) were added. Use of the slope data recorded at each site at the time of original plot measurement had been considered for inclusion in earlier work, but was rejected because of measurement inconsistencies.
Correlations between tree numbers per unit area and environmental variables were estimated using multiple regression. Before regression fitting, data for each species were constrained by excluding plots more than 100 observations beyond the upper and lower limits of occurrence in relation to each continuous predictor, or from levels of factor variables (G, F) for which there were less than three positive occurrences. This eliminated large numbers of zero observations from sites clearly outside the environmental range of each species (Austin & Meyers 1996). All regressions were fitted in S-Plus (v. 4·5; MathSoft Inc., Seattle, WA, USA) using generalized additive models (GAMs: Bio, Alkemade, & Barendregt 1998; Hastie & Tibshirani 1990; Yee & Mitchell 1991) with a Poisson error term and logarithmic link function. Problems with over-dispersion were accommodated by fitting a quasi-likelihood model, in which the dispersion parameter was estimated from the data, rather than being fixed at a value of 1 (MathSoft 1997).
An initial set of models was fitted using a backward stepwise procedure, to allow the best predictors of root zone and atmospheric water deficit to be identified. Candidate variables were the temperature (Ta, Tw), solar radiation (Sa, Sw), root zone water deficit (W0, i.e. based on average rainfall), and landform variables (S, G, and F) described above. The significance of removing each variable was tested in turn using the scaled change in deviance, which was distributed approximately as for an F-statistic (Venables & Ripley 1994). The significance of fitting continuous variables as smooth or linear terms was also tested, with smooth terms reduced to linear terms where contributions of smooth components were nonsignificant. Model fitting proceeded until no fitted terms could be removed or modified, or previously removed terms added, at P < 0·01.
We then tested whether any further improvements in model fit could be obtained by replacing the annual integral of root zone water deficit computed using long-run average rainfall (W0), with integrals computed to represent increasing severity of drought. These were derived by running the water balance model with values of α (equation 3) equal to 0·3, 0·5, and 0·7 (W0·3…W0·7). Having fitted the measure of W producing the greatest overall reductions in deviance, we then assessed which of the 12 monthly estimates of air saturation deficit (DJan …DDec) brought about the greatest additional improvements in model fit. This was achieved by separately adding each month’s estimate, and comparing scaled reductions in deviance averaged for all species. Finally, we re-tested the effects of adding estimates of root zone water deficit with varying degrees of drought-correction to confirm that the addition of D estimates to the regression models had not altered the outcome from earlier testing. Having identified the set of variables with the highest overall correlations with species abundances, a final set of regressions was fitted and tested using the stepwise procedure described above.
Identification of species optima and ranges from these final regressions was made difficult by both the irregular shape of the environmental space defined by the sample points and the uneven distribution of plots within it. Distributions of species in relation to root zone and atmospheric water deficits were therefore determined by examining the fitted values from the regressions, with the core environmental range defined as that over which fitted tree numbers exceeded one third of the maximum fitted tree number. Spatial distributions were predicted from these regressions using a set of points on a 1 km grid across New Zealand with environmental estimates equivalent to those used to fit the regressions.
IDENTIFICATION OF THE BEST PREDICTORS OF WATER DEFICIT
Modifying estimates of root zone water deficit to take account of rainfall in drought years resulted in substantial reductions in residual deviance in the initial models (Table 2), and this increased progressively with increasing drought correction of rainfall. Testing with values of α > 0·7 was considered, but rejected because this would have reduced rainfall estimates to an unrealistic degree. Use of α = 0·7 typically resulted in a 30–40% reduction in rainfall, equating on most sites to the rainfall received in a 1-in-20-year drought.
Table 2. Summary of scaled changes in deviance and number of species regressions showing significant change in deviance (P < 0·01) when adding annual integrals of root zone water deficit, calculated with increasing degrees of correction (α) for reduced rainfall in drought years
Value of α
Average scaled reduction
Number of significant models
Comparison of the effects of adding estimates of air saturation deficit for each month indicated that DMar produced the largest average reductions in residual deviance, while DFeb and DMar were retained as significant terms in regressions for the most species (Table 3). Addition of DAug produced the smallest changes in residual deviance, but was still statistically significant for the majority of species. A reassessment of the relative significance of W with varying values of α confirmed that a value of α = 0·7 still brought about the greatest reductions in residual deviance, despite the addition of DMar to the suite of candidate variables.
Table 3. Summary of scaled changes in deviance and numbers of species regressions showing significant change in deviance (P < 0·01) when separately adding estimates of air saturation deficit for each month
Average scaled reduction
Number of significant models
Final models for 11 of the 33 species contained all nine environmental variables, with two or more variables dropped for 13 species. Annual temperature (Ta) had the greatest contribution, as assessed by average scaled changes in deviance (Table 4), followed by annual and winter solar radiation (Sa and Sw). Root zone water deficit was the next most important predictor, followed by DMar. The latter was retained as a significant term in all but four regressions, but W0·7 was removed for only two species. Slope and drainage were the variables least frequently retained, with the latter kept in the final models for only 17 species.
Table 4. Summary of final stepwise regressions. Table entries indicate the average scaled increases in residual deviance resulting from dropping each term, where included, from the final models, and the number of species for which each term was retained as significant
Average scaled reduction
Number of models in which significant
Examination of the fitted ranges and optima along the two water deficit gradients reveals strong sorting (Fig. 2, Table 5). Although most species occur over a wide range of the root zone water deficits sampled in the data set (0·5–398 MPa days), and 11 occur throughout this range, core distributions for species extend over approximately 210 Mpa d on average. Species optima are frequently skewed within this core range; e.g. species such as Dacrydium cupressinum, Prumnopitys ferruginea, Podocarpus totara/hallii and Metrosideros umbellata, although capable of growing on sites with annual root zone water deficits of 100 MPa days or greater, are predicted to reach their maximum abundance on sites with minimal deficits. Species predicted to reach their maximum abundances in drought-prone climates include Dacrycarpus dacrydioides, Alectryon excelsus, Laurelia novae-zealandiae, Pittosporum eugenioides, Prumnopitys taxifolia and Griselinia littoralis. The first three of these species, although occurring in few to moderate numbers on well-drained hill slopes, generally reach their maximum abundances on flat sites with moderate to poor drainage.
Table 5. Optima and core ranges for common New Zealand tree species in relation to root zone and atmospheric water deficits. Table entries in bold indicate the environmental conditions at which fitted values from the regressions reach their maxima, while the remaining values indicate the environmental ranges over which fitted values exceed one third of the maximum fitted values
W0·7 (MPa days)
Similar patterns are evident in relation to air saturation deficits, with 12 species occurring throughout the range sampled by the data set (0·05–0·59 kPa); core ranges for species had an average span of 0·29 kPa. Species reaching maximum abundance at low atmospheric deficits included Metrosideros umbellata, Dacrydium cupressinum, Prumnopitys ferruginea and Podocarpus totara/hallii, all of which are most abundant on sites with root zone water deficits of < 5 MPa days. Similarly, several species predicted to reach maximum abundance on sites with large root zone deficit, Pittosporum eugenioides, Alectryon excelsus, Dacrycarpus dacrydioides and Prumnopitys taxifolia, are also predicted to reach maximum abundance on sites with large air saturation deficits.
However, there is much less correlation between species optima at intermediate values along these two gradients (Fig. 2). For example, at moderately large W (50–100 MPa days), Weinmannia silvicola and Pseudopanax edgerleyi are predicted to reach their maximum abundance on sites with D < 0·3 kPa, while Phyllocladus trichomanoides and Dysoxylum spectabile are predicted to reach their maximum abundance where D = 0·5 kPa. Similar contrasts can be found at intermediate positions along the air saturation deficit gradient, where Fuchsia excorticata and Phyllocladus alpinus are most abundant on sites with minimal W (2–3 MPa days), but Beilschmiedia tarairi and Nestegis spp. are predicted to be abundant at W c. 200 MPa days.
Improvements in the spatial prediction of abundance produced from these regressions are typified by the distribution predicted for Weinmannia racemosa (Fig. 3). This shows a substantial improvement in recovery of the regional-scale distribution patterns (cf. Figure 1). Although low abundances are still erroneously predicted for sites east of the main axial ranges of both islands, amounts are substantially less than from our previous regressions, particularly in south-eastern parts of the South Island, and in the Kaimanawa, Kaweka and northern Ruahine Ranges. Greater abundance is also predicted for western sites, reflecting the improved model fit in environments with low air saturation deficit where Weinmannia racemosa is most abundant. An added advantage is that the new prediction exhibits much greater fine-scale differentiation, reflecting the improved resolution of the landform variables used in this analysis.
Although the approach used in this study is broadly similar to that used in earlier predictions of tree distribution (Leathwick 1995, 1998), it makes use of more refined measures of water stress, developed within a conceptual framework based on results from studies of tree physiology. The most striking feature of our results is the demonstration of strong sorting of species optima in relation to both root zone water deficit and air saturation deficit (Fig. 3), with regressions including these variables producing markedly improved predictions of species distribution compared to earlier regressions. Although measures of soil water deficit have been used in a number of other studies, demonstration of strong correlations between air saturation deficit and species abundance is a significant advance that has not been tested elsewhere at a landscape scale.
The physiological basis for the response of species abundance and distribution to air saturation deficit stems from the sensitivity of stomatal conductance to environment and the cumulative effect of canopy conductance on photosynthesis and productivity. Studies with individual plants of many woody species have shown that stomatal conductance is very sensitive to air saturation deficit (Schulze & Hall 1982). A large fraction of its diurnal and seasonal variability can be explained by variance in ambient air saturation deficit and irradiance (e.g. Jarvis 1976). Although few data are available to demonstrate this effect for species native to New Zealand, the marked sensitivity of stomatal conductance to air saturation deficit has been shown for Nothofagus species (Benecke & Havranek 1980; Hollinger 1987). The strong link between the rate of photosynthesis and stomatal conductance for individual leaves and canopies is recognized by the inclusion of a coupled term in process-based models of canopy carbon uptake (Collatz et al. 1991; Leuning et al. 1995). Experimental evidence, obtained by manipulating stomatal conductance or the cross-sectional area of the conducting pathway in stems, has shown that good relationships exist between stomatal conductance, transpiration and the hydraulic conductivity of stems to water transfer (Whitehead 1998). Decreases in canopy conductance during periods when root zone water deficit increases, or when air saturation deficit is high, are interpreted as a mechanism for the avoidance of cavitation of xylem vessels in trees, as demonstrated using models (Tyree & Sperry 1988; Williams et al. 1996; Bond & Kavanagh 1999).
At broader spatial scales, Mencuccini & Grace (1995) have demonstrated long-term adjustment of canopy conductance and leaf area to accommodate differences in prevailing conditions of air saturation deficit in Pinus sylvestris growing at different sites. Similarly, measurements of carbon isotope (13C/12C) discrimination for 22 species of Nothofagus growing in common conditions were consistent with adaptation of stem hydraulic behaviour by species growing in drier climates (Read & Farquhar 1991). Although all the mechanistic links need to be conclusively demonstrated, we argue that there is sufficient evidence to argue plausibly that species abundances are related to prevailing air saturation deficit which, along with rainfall, strongly influences tree water balance, productivity and long-term survival.
THE IMPORTANCE OF EXTREME CLIMATIC EVENTS
Another significant feature of this analysis is its confirmation of the findings of a range of other studies that identified the importance of droughts in determining the growth and survival of forest trees in New Zealand (Bannister 1986; Coulter 1966; Grant 1984; Innes & Kelly 1992; Jane & Green 1984). We suspect that further improvements could be made, including the correction of the residual over-prediction of abundance for species such as Weinmannia racemosa in eastern sites by replacing our long-term average-based estimates of D with data describing extreme events. In particular, evaporative demand is likely to be very much greater than average during föhn events, which are typified by both high D and temperature. Therefore, it appears that El Niño events, which result in increased frequencies of westerly föhn winds and reduced rainfall in eastern parts of New Zealand (Mullan 1996), are likely to have played a major role in shaping the composition of the forests that once occupied these landscapes. In addition, given that föhn winds, which play a large role in determining the spatial distribution of extreme air saturation deficits in New Zealand, also occur throughout many other parts of the globe (e.g. Barry 1981), similar effects are highly likely to occur elsewhere.
LIKELY PREHUMAN FOREST COMPOSITION IN DRY ENVIRONMENTS
Inclusion of additional plots in the data set to better represent dry, lowland environments gives some interesting insight into the likely composition of prehuman forests in parts of New Zealand where pre-European deforestation was most extensive (McGlone 1983). It must be acknowledged that this assumes that the fragments of forest, on which predictions of composition in dry environments are based, are representative of historic forest composition. A strong possibility exists that forests within dry environments may have been selectively removed (McGlone 1989), leaving only remnants of more fire-resistant composition and/or those on the most topographically dissected or poorly drained sites.
Lowland species able to be analysed here for the first time include Alectryon excelsus, Pittosporum eugenioides, and Vitex lucens, all of which are indicated as having considerable tolerance to high root zone and atmospheric water deficits. Other species, many of which reach maximum abundance or are at least common in upland and/or moist forests (Fig. 2), are also shown to be capable of tolerating warm, dry lowland conditions, e.g. Prumnopitys taxifolia, Melicytus ramiflorus, Nothofagus solandri, Griselinia littoralis, Fuchsia excorticata, Podocarpus totara/hallii, Elaeocarpus dentatus, Dacrycarpus dacrydioides and Laurelia novae-zealandiae (Table 5). Although the last two species, along with Vitex lucens, are most common on poorly drained sites, they also occur in fewer numbers on well-drained hill slopes, suggesting tolerance of both drought and water-logging. This is consistent with Jane & Green (1985), who showed that species growing on sites with normally poor drainage can be exposed to surprising drought stress, due in part to difficulties in developing adequate root systems in the anaerobic conditions that normally prevail. Similar tolerance of both drought and water-logging is exhibited by Nothofagus solandri (Sun et al. 1995).
This analysis provides strong evidence that both root zone water deficit and air saturation deficit are likely to be important factors determining the distribution of many of New Zealand’s indigenous tree species. Such a result is consistent with evidence from physiological studies of a range of tree species, demonstrating not only the sensitivity of transpiration to atmospheric water deficits but also the potential for physical damage to result from excessive evaporative demands. Furthermore, our results confirm that extreme climatic events are likely to be as important in shaping species distribution as long-run averages.
This analysis would not have been possible without the wealth of forest composition data contained in the National Vegetation Survey database, a collection funded by New Zealand’s Foundation of Research, Science and Technology. Staff of the former New Zealand Forest Service collected most of the data used – our debt is gratefully acknowledged. Graeme Hall made a major contribution to the formation of this database. Gary Barker, Jake Overton and Joshua Baré assisted in the collection of additional field data. Development of our ideas benefited greatly from discussions with Gary Barker, Peter Bellingham, Bruce Burns, Bruce Clarkson, Alan Green, John Hunt, Anthony Lehmann, Matt McGlone, Jake Overton, Geoff Rogers, Peter Wardle, Richard Waring, Adrian Walcroft and Bastow Wilson.