Degree-day models in New South Wales: climatic variation in the accuracy of different algorithms and geographical bias correction procedures

Abstract

Geographical variation in the accuracy of different indirect degree-day (IDD) models may result in significant error in the estimation of developmental events in organisms whose development is temperature dependent. This study compared IDDs based on the rectangle, triangle and sine wave models to direct degree-day (DDD) estimates at 20 sites in New South Wales. The value of geographical bias correction procedures for improving the accuracy of IDDs were also examined. Temperatures were recorded at hourly or shorter intervals over periods from 6 months to 4 years depending on the site. IDDs and DDDs were calculated using 7 lower developmental thresholds between 10 and 13�C. Geographical bias correction procedures were applied at 2 resolutions: local (using data from single sites, and regional (pooling data from sites having similar climatic features). Correction equations were obtained by regressing daily IDDs on daily DDDs. Independent data were used to assess the performance of these equations at improving IDDs. Models did not perform equally well throughout the state. The triangle model performed best at sites in the east and south of the state, however, it was often inaccurate. The sine wave and rectangle models performed best at sites in the west. Discriminant analysis identified average annual rainfall, average daily January relative humidity and average daily July cloud cover as the most efficient subset of climatic variables needed to correctly classify all but 1 site on the basis of model performance. Model performance was thus dependent on atmospheric moisture indices. Local bias correction greatly improved the accuracy of IDDs at sites with humid climates but were superfluous at semi-arid sites. Within most climatically defined regions there were significant differences in the correction equations used at different sites. At some sites there were also differences between years. However, the impact of within-site variability of corrected IDDs from year to year on accumulated estimates was reduced to within acceptable limits (i.e. errors <7 days) by increasing the amount of data used to generate the equations. Regional bias correction equations gave improved IDDs but the improvement achieved was not as good as that with local correction. Model selection as a source of error in IDDs, the consequences of selecting a model for its accuracy at estimating degree-days or its precision at predicting developmental events, and the benefits and limitations of geographical bias correction are discussed.