To the Editor

Hewitt et al.1 reported the detection of circadian control of isoprene emissions from two tropical rainforests in Malaysia, based on their finding that a model without a circadian control cannot reproduce the observations. By including circadian-controlled isoprene emissions into models of atmospheric chemistry and transport, they suggested that plant circadian rhythms indirectly affect the global concentration of surface-level ozone. Here, we argue that the circadian rhythm postulated by Hewitt et al.1 is not robust, and depends on untested assumptions regarding both the temperature and light response of isoprene emissions, and the unaccounted-for effects of canopy structure. The apparent circadian control disappears if different, biologically realistic, model parameters are used.

Hewitt et al. used the Guenther et al. algorithms of the MEGAN model2 to detect the circadian control. We base our hypothesis on the notion that non-random deviations of the model parameters from their unknown true value can lead to an apparent circadian rhythm. We tested our hypothesis using the same algorithms2 used by Hewitt et al., applying a Bayesian model inversion to synthetic data (Fig. 1).

Figure 1: The three lines represent the three steps in the test of our hypothesis.
figure 1

First, we generated a synthetic emissions time series using the standard MEGAN model with a constant basal emission rate (blue). We then generated a 'circadian' time series using MEGAN, by assuming a circadian basal emission rate (red; see Supplementary Eq. S7 of Hewitt et al.1). Finally, we optimized the parameters of the MEGAN model using Markov-chain Monte Carlo with simulated annealing, and tested whether the 'circadian' time series can be modelled without circadian control (grey; a fixed basal emission rate, but with slightly different model parameters). Parameter values were allowed to vary by 30%, which is well within observed and biologically realistic variability11. See Supplementary Fig. S1b of Hewitt et al.1 for comparison.

Full size image

The optimized MEGAN model proved flexible enough to reproduce the synthetic circadian emissions with parameter changes within parameter uncertainty, and no circadian control (Fig. 1). The net effect of the parameter changes was a shift in the relative importance of radiation in comparison to temperature in the control of isoprene emissions. So why would the Malaysian forests have a different light and temperature response curve to that included in MEGAN?

There are various isoprene emissions models available, and the shape of the light and temperature response curves of each is decidedly different3. This reflects significant variations in light and temperature responses of isoprene emissions in and across species, due to inter-specific variations in isoprene synthase expression and differences in dynamic substrate pools4. It may, therefore, not be reasonable to expect the dependencies applied in MEGAN, estimated from leaves of temperate forests2, to apply to Malaysian rainforest canopies.

Furthermore, all isoprene models are leaf-level models that are later scaled to the canopy, thereby being highly sensitive to assumptions regarding canopy structure5,6. The extraction of basal emissions at a canopy scale is complicated both by strong diurnal gradients in canopy micrometeorology and the high variability of basal emissions rates in the canopy itself7. To adhere to the Hewitt et al. study, we have used their isoprene model without detailing canopy structure, albeit being another potential factor that could change the expected diurnal pattern of whole-canopy emissions, for example, by altering the contributions of different foliage layers to total canopy emissions.

We agree with Hewitt et al. that isoprene emission models should be improved, but we show here that the diurnal response of isoprene emissions in their study cannot be conclusively attributed to a circadian control. Although leaf-level circadian and ultradian controls have been previously reported8,9, we argue that the extent to which this affects canopy-scale emissions has yet to be rigorously assessed. Clearly, more work is needed to gain insight into variations of light and temperature responses of isoprene emissions across the globe. Model optimization techniques10, such as those used here, could aid in quantifying the extent of natural variability, and the associated implications for modelling global isoprene emissions.