Predicting phenological events using event-history analysis

TL;DR

This paper introduces a novel event-history analysis method for predicting phenological events like bloom dates in fruit trees, emphasizing the importance of accurate temperature forecasts for improved predictions.

Contribution

The authors develop a new application of event-history analysis to phenology, providing a more accurate estimation of GDD thresholds and bloom dates.

Findings

Predicted bloom dates are quite accurate.

Predictive uncertainty is high due to crude climate modeling.

Improving temperature forecasts enhances prediction accuracy.

Abstract

This paper presents an approach to phenology, one based on the use of a method developed by the authors for event history data. Of specific interest is the prediction of the so-called "bloom--date" of fruit trees in the agriculture industry and it is this application which we consider, although the method is much more broadly applicable. Our approach provides sensible estimate for a parameter that interests phenologists -- Tbase, the thresholding parameter in the definition of the growing degree days (GDD). Our analysis supports scientists' empirical finding: the timing of a phenological event of a prenniel crop is related the cumulative sum of GDDs. Our prediction of future bloom--dates are quite accurate, but the predictive uncertainty is high, possibly due to our crude climate model for predicting future temperature, the time-dependent covariate in our regression model for phenological events. We found that if we can manage to get accurate prediction of future temperature, our prediction of bloom--date is more accurate and the predictive uncertainty is much lower.