For these parameters the model (LV0) has a fixed point at (1.236,0.382). Any trajectory that starts in the vicinity of this point will spiral inwards with an e-folding time of 0.0468. An example of such a trajectory is shown by the gray line in Fig. 1. Next,

we allow the carrying capacity of the prey to vary with time (LV1) as follows equation(8) α3=α30[1+α31sin(2πt)+α32sin(2πt/P2)].α3=α301+α31sin(2πt)+α32sin(2πt/P2).We interpret the term sin(2πt)sin(2πt) as a variation of the carrying capacity with a period of one year, and sin(2πt/P2)sin(2πt/P2) as a high frequency variation about this annual cycle with a period P2P2 years. We assume P2=0.2P2=0.2 see more years. The impact of allowing α3α3 to vary with time is shown by the black lines in Fig. 1 and Fig. 2. (Parameter values for this run are given in Table 1.) As expected, the prey and predator abundances now vary with periods of 1 and 0.2 years. The nonlinearity of the BI 2536 supplier governing equations also generates variability at other periods. This can be seen in the way the prey abundance varies with greater amplitude at about the annual cycle when the predator abundance is low (e.g., 39.5

cycle. We now perform a set of numerical experiments to compare the effectiveness of conventional and frequency dependent nudging in reducing seasonal biases in the model state. All of the model runs (see Table 1) are identical

except for the amplitude of the annual cycle of α3α3 and the form of nudging. Run LV1 includes the full time variation of carrying capacity and is not nudged. We will treat LV1 as the complete model   and sample it to generate observations   (see black lines of Fig. 1 and Fig. 2). Run LV2 is identical to LV1 except that α31=0α31=0 leading to a seasonally biased simulation. We will treat LV2 as the simplified model (see gray lines in the left panels of Fig. 2). Runs from LV3 and LV4 are identical to LV2 except that they are nudged to the mean and annual cycle of LV1 using conventional and frequency dependent nudging, respectively. We implemented the climatological bandpass filter denoted by the angle brackets in (6) using a third-order Butterworth filter defined in state space form. The cutoff frequency of the lowpass filter is 1/61/6 cycle per year and the passband of the annual filter is 0.95,1.05 cycle per year. The state space model for this filter was then combined with the predator–prey model by augmenting the predator–prey state vector, similar to the approach used by Thompson et al. (2006). The solution of (6) was then calculated numerically using an explicit Runge–Kutta scheme (ode45 routine in Matlab).