Abstract. Eddy covariance has evolved as the method of choice for measurements of the ecosystem-atmosphere exchange of water vapour, sensible heat and trace gases. Under ideal conditions eddy covariance provides direct and precise flux observations, commonly approximated from single point eddy covariance measurements. While eddy covariance is appropriate over uniform terrain of infinite extent, heterogeneous land surfaces compromise the representativity of single-point measurements as a predictor for ecosystem-wide fluxes and violate assumptions of the eddy covariance method. Therefore heterogeneous land surfaces require multiple measurement units for spatially adequate sampling and representative fluxes. The complexity and cost of traditional eddy covariance instruments typically limits the feasible number of sampling units. Therefore, new low-cost eddy covariance systems are required for spatially replicated sampling not only to increase the representativity of turbulent fluxes at a single site, but also for experiments where replication is required to e.g. compare different ecosystems. The aim of this study was to test the performance of a compact low-cost pressure, temperature and relative humidity sensor for the application of evapotranspiration measurements by eddy covariance over agroforestry and conventional agriculture in Germany. We performed continuous low-cost eddy covariance measurements over agroforestry and conventional agriculture for reference, at five sites across Northern Germany over a period of two years from 2016 to 2017. We conducted side-by-side measurements using a roving enclosed-path eddy covariance set-up to assess the performance of the low-cost eddy covariance set-up. Evapotranspiration measured with low-cost eddy covariance compared well with fluxes from conventional eddy covariance. Diel cycles of evapotranspiration were well represented at a 30-min resolution. The differences between low-cost and conventional eddy covariance at 30-min resolution were small relative to the diel amplitude of the fluxes. The slopes of linear regressions for evapotranspiration comparing low-cost and conventional eddy covariance set-ups ranged from 0.86 to 1.08 for five out of ten sites, indicating a 14 % flux underestimation and a 8 % flux overestimation, respectively. Corresponding R2 values ranged from 0.71 to 0.94 across sites. This indicates that a high proportion of the flux variability of the conventional eddy covariance set-up is reproduced by the low-cost eddy covariance set-up. The spectral response characteristics of the low-cost eddy covariance set-up were inferior to the eddy covariance set-up in the inertial sub-range of the turbulent spectrum. The water vapour flux cospectrum of the low-cost eddy covariance set-up underestimated the theoretical slope of −4/3 stronger than the conventional eddy covariance set-up. That is mainly caused by the limited response time of the low-cost thermohygrometer of one second, which prevents eddies of a frequency higher than two times the response time to be adequately sampled by the thermohygrometer. We conclude that low-cost eddy covariance sensors are an alternative to conventional eddy covariance sensors when spatial replicates are required or when the scientific questions require a larger number of measurement units. An appropriately chosen high-frequency correction method is essential for the slow response sensor. The new low-cost eddy covariance set-up is a viable alternative particularly when the spatial variability of fluxes of the ecosystems of interest is larger than above reported set-up specific differences in fluxes.