AbstractInland freshwater bodies form the largest natural source of carbon to the atmosphere. To study this contribution to the atmospheric carbon cycle, eddy-covariance flux measurements at lake sites have become increasingly popular. The eddy-covariance method is derived for solely local processes from the surface (lake). Non-local processes, such as entrainment or advection, would add erroneous contributions to the eddy-covariance flux estimations. Here, we use four years of eddy-covariance measurements of carbon dioxide from Lake Erken, a freshwater lake in mid-Sweden. When the lake is covered with ice, unexpected lake fluxes were still observed. A statistical approach using only surface-layer data reveals that non-local processes produce these erroneous fluxes. The occurrence and strength of non-local processes depend on a combination of wind speed and distance between the instrumented tower and upwind shore (fetch), which we here define as the time over water. The greater the wind speed and the shorter the fetch, the higher the contribution of non-local processes to the eddy-covariance fluxes. A correction approach for the measured scalar fluxes due to the non-local processes is proposed and also applied to open-water time periods. The gas transfer velocity determined from the corrected fluxes is close to commonly used wind-speed based parametrizations.