A critical review of the use and performance of different function types for modeling temperature-dependent development of arthropod larvae

Abstract

Temperature-dependent development influences production rates of arthropods, including crustaceans important to fisheries and agricultural pests. Numerous candidate equation types (development functions) exist to describe the effect of temperature on development time, yet most studies use only a single type of equation and there is no consensus as to which, if any model predicts development rates better than the others, nor what the consequences of selecting a potentially incorrect model equation are on predicted development times. In this study, a literature search was performed of studies fitting development functions to development data of arthropod larvae (99 species). The published data of most (79) of these species were then fit with 33 commonly-used development functions. Overall performance of each function type and consequences of using a function other than the best one to model data were assessed. Performance was also related to taxonomy and the range of temperatures examined. The majority (91.1%) of studies were found to not use the best function out of those tested. Using the incorrect model lead to significantly less accurate (e.g., mean difference±SE 85.9±27.4%, range: -1.7 to 1725.5%) predictions of development times than the best function. Overall, more complex functions performed poorly relative to simpler ones. However, performance of some complex functions improved when wide temperature ranges were tested, which tended to be confined to studies of insects or arachnids compared with those of crustaceans. Results indicate the biological significance of choosing the best-fitting model to describe temperature-dependent development time data.

Modeling temperature‐dependent development rate and phenology in insects: review of major developments, challenges, and future directions

“…Thus, efforts were made to develop models reducing the number of parameters (e.g., Briere model; Figure 2C), while keeping key model characteristics: (1) the asymmetry in development rate with a sharp decline above optimal temperature of development, and (2) the characterization of a lower and upper threshold of development (Briere et al, 1999;Shi et al, 2015). Despite these advances and Lamb's evidence on nonlinearity and the development of simplified models (above), to date the straight-line equation is still the most used in papers (Quinn, 2017;Rebaudo et al, 2017), even if it is biased at extreme temperatures.…”

“…The characterization of the relationship between temperature and development rate is still a challenge, notably because of mortality at extreme temperatures and interindividual variation within the same species (R egni ere et al, 2012), which makes model choice constrained by uncertainty. In the review of the various performance functions modeling temperature-dependent development of arthropods, Quinn (2017) revealed that model choice is of critical importance because of the significant differences between model predictions (see also Shi et al, 2015;and Figure 1B-D). In Quinn's (2017) study, although no function type was described as standing out from the rest, considering and comparing various functions in regard to the available datasets should lead to better predictions.…”

“…In the review of the various performance functions modeling temperature-dependent development of arthropods, Quinn (2017) revealed that model choice is of critical importance because of the significant differences between model predictions (see also Shi et al, 2015;and Figure 1B-D). In Quinn's (2017) study, although no function type was described as standing out from the rest, considering and comparing various functions in regard to the available datasets should lead to better predictions. The thermal performance curve approach was built out of uncertainties regarding the underlying mechanisms in insects, and the relationship between temperature and development in these models should be considered in regard to their associated assumptions (Sinclair et al, 2016).…”

“…The thermal performance curve approach was built out of uncertainties regarding the underlying mechanisms in insects, and the relationship between temperature and development in these models should be considered in regard to their associated assumptions (Sinclair et al, 2016). Some initiatives, such as the ILCYM (Insect Life Cycle Modeling) software for insect pests , the ArthroThermoModel software (Mirhosseini et al, 2017), or the R package dev-Rate for arthropods R Core Team, 2018;Rebaudo & Struelens, 2018) selecting the previously fitted model in the literature on a similar species without considering alternatives, a practice that should be avoided (Quinn, 2017). Choosing between models on the basis of criteria such as Akaike's Information Criterion (AIC) may be an intuitive choice, but must be used with caution as each criterion has its advantages and disadvantages (e.g., AIC favors more complex models, which could lead to an overfitting of the data), and combination of various methods should be preferred (Angilletta, 2006).…”

“…Recent advances have highlighted the importance of thermal performance curves in ecology and entomology, as long as their inherent assumptions are acknowledged (Sinclair et al, 2016). The scope of this review is not to present and detail the various models available for understanding and predicting temperature-dependent development ratesee Damos & Savopoulou-Soultani (2012), Shi et al (2015), Mirhosseini et al (2017), and Quinn (2017) for that purposebut to document how the different approaches have evolved over time, in order to explain why so many models are available, and to identify current and future challenges.…”