Title: Predicting variations in methane emissions from boreal peatlands through regression models.

Authors: A. Kettunen, V. Kaitala, J. Alm, J. Silvola, H. Nykänen and P.J. Martikainen

Status: Boreal Environment Research Vol. 5 (nro 2), 2000. pp. 115-131

Abstract: We used frequently measured data on methane emissions, water tables and temperature profiles from different microsites in a boreal low-sedge Sphagnum papillosum pine fen to detect the modes of spatial and temporal variations in methane emissions and to test how well regression model approaches predict the variations. Methane fluxes showed strong spatial and seasonal variations, but diurnal variations were less evident. Diurnal pattern with maxima during the afternoon from 12.00 to 20.00 and minima from 00.00 to 8.00 were detected only in the microsites where plant transport apparently was negligible and only during the early season. In the microsites where segdes dominated no diurnal pattern could be found and the diurnal pattern in the microsites where segdes did not dominate was no longer evident by July. The number of episodic emissions and their contribution to the seasonal cumulative estimate were found to increase from the wet microsites where episodic fluxes accounted for 1 - 4 % of the seasonal estimate to hummocks where the contribution of episodic fluxes rose to 10 %. In the spatial microscale, the average water table and abundance of sedges apparently affected the level of emissions, as both average and cumulative emissions were lowest from high hummocks (from 10 to 13 g CH₄ m^-2 a^-1), intermediate from flarks and lawn-low hummock (from 26 to 29 g CH₄ m^-2 a^-1), and highest (50 g CH₄ m^-2 a^-1) from the intermediately moist lawn surface with a high sedge cover. The temporal pattern of methane emissions was related to peat temperatures via regression models, but not to water tables as water tables remained high throughout the season. The regression models of methane emissions vs. peat temperatures explained the methane emissions quite satisfactorily (r²=0.61-0.78, N = 302 to 583) for flarks and low hummocks, but less satisfactorily for the high hummocks (r²=0.32-0.37, N = 302 to 564). For the independent data sets that were not used in estimation the goodness of fit values were remarkably lower (in most cases r² < 0.30). High episodic pulses and diurnal variations were not captured by the models and the models overestimated the spring emissions and underestimated midsummer high emissions.