DENDROCHRONOLOGY 389 the year-to-year variation Autoregressive Moving Average (ARMA) models have been perhaps the most frequently used standardizing models, though the literature is full of alternative methodologies, each of which may be more suitable in particular conditions Having satisfactorily arrived at a well replicated index series representing growth, the next stage is generally to model how the climatic information influences the observed growth One of the great problems here is the availability of reliable meteorological records, and the length of those records for regions where the most environmentally sensitive trees are growing The most useful tree-ring information is often from remote areas, and even where there are local records, these are generally only available over recent decades Not surprisingly, an early study of oak growth in Britain showed that the variation in ring width attributable to mean monthly rainfall and temperature data varied according to distance from the meteorological station used Nevertheless, a process of stepwise principal component regressions generally allows a response function to be derived in which individual monthly rainfall and temperatures prior to, and during, the growth season can be recognized in order of their importance For example, in a typical temperate northern-hemisphere model, April, May and June rainfall, and May, June and July temperature may be found to have the greatest influence on growth A transfer function can then be derived so that these major elements are reconstructed throughout the length of well replicated tree-ring chronology Typically, in the more sensitive trees growing at the margins of distribution, one factor can be derived as the major influence on growth, and the reconstructions are concerned with aspects of either rainfall or temperature Given that biological responses are often not linear, and that relatively crude data in the form of monthly means are the normal tools employed, it is surprising how reproducible the results have been In 1982 a study of response functions for oaks growing at 16 sites in the British Isles was published This showed many similarities in responses, including the surprising result that temperature in the December prior to growth was significantly negatively correlated with ring width in the following growth season This was mathematically derived, without reference to the biology of the trees, and a tentative explanation was put forward This suggested that a warm December resulted in the trees metabolizing the food reserves which were subsequently not available to the tree the following spring, resulting in less growth In a cold December, the tree had ‘shut down’ allowing the reserves to be utilized for growth when conditions improved This highlighted a need for better understanding of tree physiology which is still a requirement for better modelling of the climate–growth relationship Some early reconstructions were tempted to use very short periods of meteorological data to calibrate the model – often this was all that was available, although the ring series may go back several centuries There is also a need to verify the model, so the short data series was divided into two parts, with one short series being used to calibrate the model and the other part to verify the reconstructed values over an equally short period Hundreds of years were then reconstructed on this unsatisfactory model Whilst this is tempting, and no doubt may highlight very different periods of climate within the tree-ring timeseries, it is clearly open to inappropriate interpretation Guidelines on methodology also suggest that in order to remove any effects of changes in response over time, the calibration and verification of the model should be done both ways around, i.e using the outer years to calibrate and verify using the inner years, and then reversing the periods used for calibration and verification Critics of this response function approach to dendroclimatology point out that not only are responses seldom linear, and that dramatic changes can result from crossing threshold values, but that many influences on tree growth can occur at widely varying time-scales, perhaps in a matter of hours For example, a severe frost on a single night in the late spring, when growth has not been going on for long and leaves are tender, may damage a significant proportion of the photosynthetic capacity of the tree and have a profound effect on growth for the remainder of that growth season Such an event is unlikely to show up in a crude measure such as mean monthly temperature Critics also point to the observable fact that whilst generally quite representative values are derived for much of the series, reconstructed values rarely recreate the more extreme years well, tending to underestimate the meteorological data (see Figure 1) A number of studies, particularly based in Europe, have looked at growth in ‘pointer’ years – that is, years where a large proportion of the trees show a marked growth change in the same direction – and have managed to look in greater detail, often at daily weather records, to see what factors are responsible for these growth changes These two approaches both yield very valuable information, and are not mutually exclusive It is by combining several approaches that the maximum information is likely to be gained in the future, although