2.3 Turbulent Heat Fluxes
Sensible (H ) and latent heat fluxes (LE ) (both in W m−2) were calculated using the covariance between the vertical wind speed w (m s−1) and air temperature T (K), as well as between w and the specific humidityq (kg kg−1), such that:
\(H=\ \rho_{a}\ c_{\text{pa}}\ \overset{\overline{}}{w^{{}^{\prime}}\ T^{{}^{\prime}}}\)(1)
\(\text{LE}=\ \lambda\rho_{a}\ \overset{\overline{}}{w^{{}^{\prime}}\ q^{{}^{\prime}}}\)(2)
where ρa is the humid air density (kg m−3), cpa is the specific heat of humid air (J kg−1 K−1) andλ is the latent heat of vaporization (J kg−1). Here, primes denote fluctuations from the 30-min average, indicated by an overbar.
Raw wind velocity data from the raft were first corrected for wave-induced motion, using the accelerometer data, following the method proposed by Miller et al. (2008). In short, the apparent wind measured by the sonic anemometer was corrected to account for Euler angles, angular velocities and linear accelerations monitored by the accelerometer. Then, for the shore and raft flux towers, we processed the turbulence data using the EddyPro® software, version 7.0 (LI-COR Biosciences, USA). In doing so, we applied time-lag compensation, linear detrending, double rotation approach (Baldocchi et al., 1988; Wilczak et al., 2001), density fluctuation compensation (Webb et al., 1980), spike removal (Papale et al., 2006), and other statistical tests (Vickers & Mahrt, 1997). Poor-quality data were flagged (Mauder & Foken, 2011) and removed. Data from the raft and shore stations were aggregated into one dataset by favoring data with the best quality criteria (Mauder et al., 2013). Note that shore data were retained only when winds originated from the reservoir. To complete the final dataset, gap-filling was implemented based on the method developed by Reichstein et al. (2005).
Over the whole study period (1447 days), 57 % of the turbulent flux data had to be gap-filled due to the fact that the raft was only deployed from June to October and the shore flux tower was frequently exposed to winds from the surrounding land. We assessed flux errors by applying the Finkelstein and Sims (2001) random uncertainty method.