R.B.B. Dickison, D.R. Lavigne, E.G. Kettela


Executive Summary

This report summarizes a study of the application of the BioSIM model by Dr. Jacques Régniere and associates for prediction of spruce budworm larval development, specifically to (a) customize the model for use in New Brunswick, and (b) validate model predictions against detailed development measurements.

The customization involved the specification of eco-geographical zones and the creation of an expanded and up-dated climatological databank for New Brunswick. Zones were specified in relation to Loucks's (1962) Forest Classification of the Maritime Provinces, and the identification parameters added to a New Brunswick version of the model. The databank was created using newly available standard normal period temperatures for the period 1961-90, which required extensive adjustment of previously published normals.

Validation involved model runs to (a) test the model's capability to mimic the development phenology map for New Brunswick, (b) to assess the variation of prediction error with time, using detailed development measurements for the 1992 season, and (c) to assess the variation of prediction error for past extreme seasons -- which were adequately represented by 1990 and 1991.

The development phenology study examined normal predictions of peak III Instar at 57 grid-points in north-central New Brunswick, covering five phenological zones. A range of 1.0 days was found for the zone-average dates, or 2 1/2 days/zone; means of predictions for zones 2 and 3 were identical, but this aberration is partly due to a couple of peculiarities in site selection.

The 1992 development predictions were conducted using several combinations of input data, reflecting the availability or unavailability of antecedent larval development data and with weather data which was actually availability or which could have been available. Observed data were available from six locations. Sequential runs were made at interval dates of three days beginning May 1, and predictions made for instar levels III through VI. The results are presented in tables and graphs.

Long range predictions for the early instars were very poor, averaging at best 7-12 days too late at all but one location. (Since the model is denied a knowledge of the actual ensuing weather, the development predictions were adversely influenced by a particularly warm cool period in mid-May.) As the advance prediction period shortened the results improved noticeably: even at around two weeks in advance, predictions for instar levels IV through VI were within two days for all except two cases, i.e., in 16 of 18 cases, using the best dataset combinations.

Although the predictions improved with the incorporation of observed development data and of observations from all operating weather stations, the improvement was slight. The disadvantage, therefore, of limited data appears not to be critical. The only significant concern is the bias introduced by abnormal weather conditions occurring after the time of the model run, but this bias should be ameliorated by incorporating forecast temperature data -- which the model is designed to accept.

Predictions for past extreme years were run for 1990 and 1991, which were respectively late and early development years. The model was run for only one site in each year, and the results compared with those for a nearby 1992 station. The results were best for 1991, the warm year, and overall were comparable with those for the 1992 analyses.

In its customized version, the model provides an efficient means of predicting larval development. Unfortunately, the model gave poorest results for ITT Instar, which is close to the stage used to begin spray operations. The incorporation of weather forecast information and an intensified field survey program 'in the early stages are warranted. The report contains six recommendations, including the recommendation that BioSIM be considered for operational use within New Brunswick.