2.1.9
EMPIRICAL MODELS FOR PREDICTION OF REGIONAL LIGHT LEAF SPOT (PYRENOPEZIZA BRASSICAE) INCIDENCE ON WINTER OILSEED RAPE IN THE UK

SJ WELHAMl, JA TURNER2, BDL FITTl, P GLADDERS3 and KG SUTHERLAND4

1 IACR-Rothamsted, Harpenden, Herts AL5 2JQ, UK; 2 Central Science Laboratory, MAFF, Sand Hutton, York Y04 ILZ, UK; 3 ADAS Boxworth, Cambridge CB3 8NN, UK; 4 SAC, Ferguson Building, Craibstone Estate, Bucksburn, Aberdeen AB21 9YA, UK

Background and objectives
Empirical models have been used to forecast plant disease epidemics and guide spray decisions. Light leaf spot (Pyrenopeziza brassicae) is a serious disease of winter oilseed rape in the UK, showing seasonal and regional variation in epidemics, as measured by percentage of plants with infected leaves in March or with infected pods in July. Work has started on a forecasting system for light leaf spot, to identify regions and crops at risk in each season [1]. This paper describes the empirical modelling used to determine seasonal risk.

Materials and methods
Data collected in England and Wales and stored on the CSL/ADAS winter oilseed rape disease and pest survey database were used to calculate regional incidence of light leaf spot on leaves in the spring (March) and on pods in the summer (July) for commercial crops over the period 1987-97. Regions were defined as the climatic regions used by the UK Meterological Office. Weather variables used were regional 30-year monthly means of winter temperature and rainfall, and deviations about these regional means. Light leaf spot incidence in the previous summer was used as a measure of available inoculum in each autumn. Multiple regression models were used to predict spring disease from the previous autumn and to predict summer incidence from spring incidence; in both cases only data from unsprayed crops was used. The models were developed in two stages. The first stage used available inoculum and 30-year mean weather variables to predict forward from autumn for an 'average' season. The second stage also included deviations about the 30-year means to take account of current weather patterns and modify the predictions accordingly. This scheme produces an initial regional forecast in autumn which is updated, taking account of actual weather patterns. Patterns in variation for individual crops were used to estimate the risk of severe infection for unsprayed crops.

Results and conclusions
A response surface model was developed to predict spring disease in an average season, incorporating available inoculum, average winter temperature and rainfall, and the interaction between temperature and rainfall. This first-stage model separated out regions: given similar available inoculum, more light leaf spot would be expected in colder, wetter regions than in warmer, dryer regions, with the predictions being more sensitive to changes in rainfall than in temperature. When deviations from average temperature were included, it was found that for every additional rain-day in winter (above the 30-year mean) an additional 2% incidence was predicted in spring. Also, for every 0. 1C decrease in autumn temperature (below the 30-year mean), an additional 0.5% plants infected in spring was predicted. The spring survey incidence was the best predictor of summer disease in an average season. This prediction could be improved by including the number of rain-days in May: each extra rain-day was equivalent to an extra 2% pod infection in July on unsprayed crops. These results reflect factors influencing the epidemics: light leaf spot is a polycyclic disease which requires a certain period of leaf wetness to infect the plant and can continue to grow at low temperatures. The empirical models described here are intended to give initial predictions of risk for regions of the UK, which can be updated as the season progresses. The predictions can also be modified for individual crops to take account of disease in the crop (based on sampling), sowing date and NIAB cultivar resistance rating [1]. Further work requires the incorporation of fungicide use into the predictions. These empirical models will be complemented by more mechanistic models being developed [2] and are intended to contribute to decision support systems.

References
1. Fitt BDL, Gladders P, Turner JA, Welham SJ, Davies JMLI, 1997. Aspects of Applied Biology 48, 135-142 2. Papastamati K, Welham SJ, Fitt BDL, Gladders P, 1998. Proceedings 7th International Congress of Plant Pathology.