Horsmonden, Tonbridge, Kent TN12 8LN, UK

Background and objectives
Fireblight is a complex bacterial disease affecting a range of pome fruit species. Infection can occur via open flowers, shoot tips and other young tissue. Stem invasion often follows. These features, and the sporadic nature of fireblight in time and space, make the collection of well documented disease records especially difficult. Two main aims of current systems are: the identification of likely infection risk days (for optimal timing of protective spray applications during bloom) and the judgement of times when early signs of disease will be presnt (for rapid surgical control to limit stem invasion). BIS (Billing's integrated system) [1] combines valuable features of earlier approaches to risk assessment including the Maryblight model [2]. It covers the whole growing season and considers other factors that can affect disease incidence and severity.

BIS combines several working hypotheses. During its development, early hypotheses were modified and then fine-tuned. This involved repeated critical testing before the.system was ready for use and testing by others. A statistical approach to validity testing is rarely possible with fireblight. An alternative approach (described by Popper) was used for BIS testing, with every effort being made to show that all or part of any hypothesis was false before it was deemed acceptable. Observations used in the development of hypotheses should not be used for validation tests, but with fireblight, well documented cases are difficult to gather and the system designer faces a dilemma. Should all available cases be used for system development, or should some be set aside as 'test cases'? Independent tests by others take time. Meanwhile, potential users need to evaluate systems. To overcome this difficulty, details of cases used during the development of BIS are available for evaluation purposes.

Results and conclusions
Epidemic blossom blight is associated with warm weather during bloom and flower wetting and epidemic shoot blight with damaging storms. The BIS approach is broadly similar to that in other systems and guidance on risks seems good. BIS also identifies days associated with low disease incidence. The precision of BIS assessments of disease development rates is good in most cases but there are host-related differences in critical heat sums for blossom blight which are not apparent in the Maryblight model. Assessments of stem invasion, overwintering and 'pre-bloom' risks are more tentative and difficult to test.

Lack of evaluation can mean lack of use and lack of independent testing. Where strict design and validation guides cannot easily be followed, a system can still be evaluated by other means. The poster aims to demonstrate this.

1. Billing E, 1996. Acta Horticulturae 411, 121-126. (Corrected version).
2. Steiner PW, 1990a, b. Acta Horticulturae 273, 139-48; 149-58.