An inverse Lagrangian framework is used to derive the vertical distribution of sources and sinks for water vapor, CO₂ and sensible heat within a triticale canopy. Measurements of concentration profiles and u* were performed from middle growing season to late senescent stage of the field crop. To evaluate the uptake and emission rates, simultaneous porometer measurements of assimilation and transpiration at the leaf level were scaled up. Since the position of leaves from grass plants are very regular, a reliable upscaling procedure could be performed. Measurements from subsequent leaves at each crop were assigned to three different canopy layers and multiplied with active leaf area for each layer. The Lagrangians calculation showed a good energy budget closure. Comparison of model calculations with the upscaled leaf level transpiration rates agreed well, although CO₂ uptake in the middle layer was overestimated. The exact propagation of errors from input parameters was determined by a detailed sensitivity study. Some general cases were found, were the common inverse Lagrangian approach failed: (1) Under instationary conditions, especially during the morning and evening hours, first order closure scheme is not appropriate because the profiles are changing rapidly and are leading to small gradients with very high uncertainties in the predicted source/sink distributions. (2) Under weak turbulence, the uncertainty increases due to the propagation of the error in u*. It goes to infinity as u* goes to zero. (3) Stability conditions have to be taken into account, especially during calm nights, when unstable conditions near the ground, caused by the heated soil and cooling in the upper canopy, drive a weak but efficient free convective mixing. In the latter case, a modified scaling scheme to derive the dispersion matrix in the Lagrangian model is proposed. It combines (common) friction velocity scaling from u* in the upper canopy and free convection scaling from w* at the bottom part. By comparison with soil heat flux measurements, the applicability of the modified scaling scheme could be demonstrated.