Isoprene emission potentials from European oak forests derived from canopy flux measurements : An assessment of uncertainties and inter-algorithm variability

Biogenic emission algorithms predict that oak forests account for ~ 70 % of the total European isoprene budget. Yet the isoprene emission potentials that underpin these model estimates are calculated from a very limited number of leaf-level observations and hence are highly uncertain. Increasingly, micrometeorological techniques such as eddy covariance are used to measure whole-canopy fluxes directly, from which isoprene emission potentials can be calculated. Here, we review five observational datasets of isoprene fluxes from a range of oak forests in the UK, Italy and France. We outline procedures to correct the measured net fluxes for losses from deposition and chemical flux divergence, which were found to be on the order of 5–8 % and 4–5 %, respectively. The corrected observational data were used to derive isoprene emission potentials at each site in a two-step process. Firstly, six commonly used emission algorithms were inverted to back out time series of isoprene emission potential, and then an average isoprene emission potential was calculated for each site with an associated uncertainty. We used these data to assess how the derived emission potentials change depending upon the specific emission algorithm used and importantly, on the particular approach adopted to derive an average site specific emission potential. Our results show that isoprene emission potentials can vary by up to a factor of four depending on the specific algorithm used and whether or not it is used in a big-leaf or canopy environment model format. When using the same algorithm, the calculated average isoprene emission potential was found to vary by as much as 34 % depending on how the average was derived. In order to best replicate the observed fluxes we propose a new weighted average method whereby the isoprene emission potential is calculated as the average of all flux observations divided by the average activity factor (γ) of the emission algorithm. This approach ensures that modelled fluxes always have the same average as the measurements. Using this new approach, with version 2.1 of the Model for Emissions of Gases and Aerosols from Nature (MEGAN), we derive new ecosystem-scale isoprene emission potentials for the five measurement sites, Alice Holt, UK (10 500 ± 2500 µg m−2 h−1), Bosco Fontana, Italy (1610 ± 420 µg m−2 h−1), Castelporziano, Italy (43 ± 10 µg m−2 h−1), Ispra, Italy (7590 ± 1070 µg m−2 h−1) and the Observatoire de Haute Provence, France (7990 ± 1010 µg m−2 h−1). Ecosystem-scale isoprene emission potentials were then extrapolated to the leaf-level and compared to previous leaf-level measurements for Quercus robur and Quercus pubescens, two species thought to account for 50 % of the total European isoprene budget. The literature values agreed closely with emission potentials calculated using the G93 algorithm, which were 85 ± 75 µg g−1 h−1 and 78 ± 25 µg g−1 h−1 for Q. robur and Q. pubescens respectively. By contrast, emission potentials calculated using the G06 algorithm, the same algorithm used in a previous study to derive the European budget, were significantly lower, which we attribute to the influence of past light and temperature conditions. Adopting these new G06 specific emission potentials for Q. robur (55 ± 24 µg g−1 h−1) and Q. pubescens (47 ± 16 µg g−1 h−1) reduced the projected European budget by ~ 17 %. Our findings demonstrate that calculated isoprene emission potentials vary considerably depending upon the specific approach used in their calculation. Therefore, it is our recommendation that the community now adopt a standardised approach to the way in which micrometeorological flux measurements are corrected and used to derive isoprene, and other biogenic VOC, emission potentials. Modellers who use derived emission potentials should pay particular attention to the way in which an emission potential was derived and ensure that the algorithm they are using, and the implementation thereof, is consistent with that used to derive the emission potential. Our results show that, in the worst cases, failure to account for this may result in modelled fluxes that differ from observations by up to a factor of four