New and improved degree-days models for predicting when fish eggs will hatch in the field

Background/Question/Methods

Lab and hatchery data show a clear relationship between egg development and incubation temperature in fish. Although this relationship is useful for predicting hatch date (D_H) when incubation temperature is relatively constant, it is of little use when incubation temperature is variable (e.g., in flow-through hatcheries or the wild). In this study, we used published data to develop and evaluate degree-day models for predicting D_H when incubation temperature is variable. These models rely on an estimate of the cumulative degree-days (CDD; ^oC*days) to hatch above some base temperature (T_i; ^oC). Our study was in two parts. In part 1, we used published relationships between days to hatch and incubation temperature to develop species-specific degree-day models. We then used published data to evaluate the ability of these models to predict D_H in the field. In part 2, we developed alternative degree-day models for the species in part 1 by deriving empirical relationships between CDD, T_i, and the optimal temperature for egg development. We then determined the relative ability of these alternative models to predict D_Hin the field.

Results/Conclusions

We identified 64 fish species for which published data were available to develop species-specific degree-day models for predicting D_H in the field. Published data were also available to evaluate model accuracy for 17 (27%) of these species. Across all species, our degree-day models were very accurate: ratio of predicted D_H to observed D_H 1.02±0.06 95% CI). Predicted D_H was within 10% of observed D_H for 14 of 17 species (82%). The largest errors in predicted D_H occurred when observed D_H was <20 days (i.e., for spring spawning species). In part 2 of our study, we found strong relationships between T_i estimate in part 1 and published estimates of optimal temperature for egg development (n = 55 species, r² = 0.76), and between T_i and CDD estimated in part 1 (n = 63 species, r² = 0.86). Degree-days models developed using these equations were only slightly less accurate than degree-day models from part 1, which suggests that this approach is a viable alternative. Overall, our results suggest that a degree-day approach that does not simply assume that T_i is zero can accurately predict D_H the field.