Fitting Population Dynamic Models to Time-Series Data by Gradient Matching

Abstract

We describe and test a method for fitting noisy differential equation models to a time series of population counts, motivated by stage-structured models of insect and zooplankton populations. We consider semimechanistic models, in which the model structure is derived from knowledge of the life cycle, but the rate equations are estimated nonparametrically from the time-series data. The method involves smoothing the population time series x(t) in order to estimate the gradient dx/dt, and then fitting rate equations using penalized regression splines. Computer-intensive methods are used to estimate and remove the biases that result from the data being discrete time samples with sampling errors from a continuous time process. Semimechanistic modeling makes it possible to test assumptions about the mechanisms behind population fluctuations without the results being confounded by possibly arbitrary choices of parametric forms for process-rate equations. To illustrate this application, we analyze time-series data on laboratory populations of blowflies Lucilia cuprina and Lucilia sericata. The models assume that the populations are limited by competition among adults affecting their current birth and death rates. The results correspond to the actual experimental conditions. For L. cuprina (where the model's structure is appropriate) a good fit can be obtained, while for L. sericata (where the model is inappropriate), the fitted model does not reproduce some major features of the observed cycles. A documented set of R functions for all steps in the model-fitting process is provided as a supplement to this article.

Forcing Versus Feedback: Epidemic Malaria and Monsoon Rains in Northwest India

“…Alternatives to our likelihood-based approach for inference on nonlinear dynamic systems include spectral matching [43], gradient matching [44] and Bayesian methodology [45], [46]. Our choice of likelihood-based methods was influenced by their statistical efficiency (even in the face of poor estimability of some parameters [47]), the availability of computationally efficient numerical algorithms [21], and the lack of scientifically supported prior distributions for a Bayesian analysis.…”

Census error and the detection of density dependence

“…Ellner et al . (2002) apply a technique that attempts to reduce or remove bias in parameters or functional forms of density relationships resulting from measurement errors.…”

“…Ellner et al . (2002) show that this method is effective in some cases, although less effective in others.…”

A two-stage estimation method for random coefficient differential equation models with application to longitudinal HIV dynamic data

“…The spirit of two-stage method was initiated by Varah (1982) who used a cubic spline to smooth the data in the first stage, and employed the least squares method for parameter estimation in the second stage. Ellner, Seifu, and Smith (2002) fitted the dynamic models to time series data using the local polynomial regression. Liang and Wu (2008) established the theoretical properties of the method for ODE models by applying the local polynomial smoothing in the first stage.…”