Modelling Plant Development Using Phenological Data:

a report on studies conducted at Royal Botanic Garden Edinburgh from 2002 to 2014

G.H.Harper

Most plant phenological studies are based on populations, comprising large numbers of plants sometimes spread over large areas. This means that there is little genetic or environmental uniformity, and standard statistical methods are used to establish significance of results, which is usually easy enough, given the large data sets that are often available. The innovative approach used in this RBGE study, however, brings with it problems in the statistical handling of the results. In the RBGE herbarium the taxonomists study individual plants on herbarium sheets in order to decide which can be grouped together in sets to be included in the scientific classification of plants; by analogy with this approach, it is assumed that there may be just as much variation between individual plants in their physiological properties as in their physical characteristics, and the attempt was accordingly made to study individual plants at ‘high resolution’ – meaning that observations of first-flower date (FFD) and in some cases first-leaf, leaf-fall, and last-flower dates, would be recorded for many years on a daily basis. Each data set then would be as large as the number of years the observations on the individual accession (an individual woody plant, or small patch in the case of herbs) were continued. The analysis was abandoned after 13 years, when I moved to Hereford, but observations have been continued at RBGE on some plants three times per week, so that it is hoped that some of the hypotheses developed in the study up to 2014 may be tested on the new data. Alternatively some hypotheses could be testable on independent data sets. The first diagram illustrates some interesting points: it shows the analysis of FFD in a small patch of Cuckoo-pint or Lord-&-Ladies Arum maculatum in the north-west corner of the garden. The equation calculates the FFD in terms of a chilling period (days in November of the previous year, PY, when mean temperature was below 5degC) and a thermal-acceleration period (average daily minimum temperature in March of the current year, CY, and average daily maximum temperature in April); the flowering period is shown by the heavy-outline box on the lower time-line. The equation was calculated on the basis of 2002-09 data, and the two curves, of observed and calculated dates, for 2010-14 show how well the model performed. To begin with the model made good predictions, but the condition of the plants was deteriorating, and by 2014 was very poor; not surprisingly the model no longer worked, and the plants died soon after. This demonstrates that models cannot be expected to function well when the physiological condition, or the health, or the developmental stage, or the environment, of an accession changes. So, in many plants, large high-quality data sets of, say, 30 years, will not be available, even if the project continues that long, although prospects are better for trees. Since adequate data sets were not available for standard statistical tests of significance, a hypothetico-deductive approach to testing was adopted, with each model being used to predict the FFD in the following year(s). Usually the model was recalculated each year on the basis of the enlarged data set. While Arumac70BEF illustrates an ordinary case of a plant with a single chilling period & single thermal acceleration period (although with different temperature sensitivities in the early & late stages), many other accessions in the study point to more interesting & complicated patterns of temperature-sensitive stages of development. If these developmental patterns can be demonstrated with adequate data, they could be useful for improving our understanding of plants’ developmental physiology and their reactions to weather & climate change.

Out of 57 models passing a prediction test in 2014, 28 (about half the total) involved a single chilling and a single thermal-acceleration period. About a quarter of the total (15) showed more than one chilling period, and these were in various combinations of PY, 2PY (year before PY), 3PY (year before 2PY) and 4PY. Colspe63A, for example, seems to have chilling periods in three years.

Of particular interest are various species of apple (Malus), in which PY + 3PY (M.baccata 20A, floribunda 29A) and 2PY + 4PY (M.sieboldii 41A) suggest 24-month cycles, which are already known in fruiting behaviour. (Where something like 24-month cycles are involved, and not all members in a population are in sync, it is necessary to study individual plants or physiologically synchronized groups if the  phenomenon is to be isolated for study.)

If data sets are not large enough for orthodox statistical tests of significance, a more innovative approach may be necessary. It is to be hoped that this study can be continued, either at RBGE, using observations made since 2014, or by exploiting other, independent data sets, such as those at Kew Gardens in London. For fruit trees it is possible that there may be good data sets at research stations such as East Malling Research Station in Kent, England. Botanic gardens & other research institutions are encouraged to set up projects like those at RBGE, since a better understanding of development in all plants, including those of commercial significance, and especially in relation to climate change, could be of great importance as an over-populated world tries to feed itself. A more detailed discussion of possibilities for future studies is in the report.