Turbulent structures and concentration fluctuations above a flat forest: validation of a hybrid LES-RANS model by wind tunnel measurements

The turbulent and scalar exchange mechanisms occurring between the forest canopy and the planetary boundary layer (PBL) are of interest in numbers of environmental applications such as for forest chemistry, pollution dispersion, or forest fire propagation. Close above a forest, complex turbulent phenomena happen due to the very high shear created by the drag of the forest. In this shear layer, a number of field measurements campaigns, numerical simulations and experimental works pointed out the major role of turbulent coherent structures (CS) in the transport of momentum and scalar through and above the forest canopy. The close reproduction of these turbulent unsteady mechanisms in a numerical simulation is, therefore, of utmost importance to study interactions between the forest canopy and the PBL. In this study, an original hybrid RANS-LES simulation is performed in the purpose of reproducing the momentum and scalar fluxes above a flat forest canopy. A comparison with state-of-art numerical simulations is performed and unsteady characteristics of the flow are analysed in details with a special care on the appropriate reproduction of coherent structures. The capability of the numerical approach in reproducing complex coherent structures and scalar dispersion is validated in neutral conditions by a detailed comparison to a wind-tunnel experiment including velocity and concentration flux measurements. The link between momentum and scalar fluxes is also discussed. The simulation uses a hybrid RANS/LES approach where the RANS k- ε equations are solved in the near wall region in order to reduce the computation time compared to a classical LES. The hybrid approach is also used as a wall function. In the simulation, the forest is treated as a porous media using a drag coefficient of 0.6 and a leaf area density reproducing the experimental canopy model. An equation of passive scalar transport is added to the OpenFoam-based solver using a Schmidt number of 1/3. The simulation is performed at full scale over a h = 20 m high forest in a domain reduced to x/h = 9.6, x = 9.6 and, x = 120 with 250 000 cells. The large scales structures of the atmosphere are, therefore, not reproduced and the study focuses on the dynamic exchange near the forest canopy. Periodic boundary conditions are imposed in the horizontal directions, the top wall is a slip condition. Flow statistics are calculated on 40 000 seconds physical time period after the full development of the flow and compared to the wind tunnel results. Statistics deduced from computations, presented in the enclosed figure (top), are well reproducing the wind-tunnel experiment statistics. Values are normalised by the friction velocity (u*) calculated from the average of the Reynolds stress (u'w') in the range z/h = [1 3]. The longitudinal turbulence profile is very comparable to the experimental one whereas the transversal and vertical turbulence profiles are slightly underestimated. It is noteworthy that no discontinuity is visible at the RANS/LES transition in the hybrid approach. The Reynolds stress modeled in LES presents a slight linear decrease from the canopy top. In the wind-tunnel, the concentration flux (see enclosed figure) is decreasing faster than in the simulation. This may be due to the periodic conditions imposed in the simulation whereas experimental measurements are performed at x = 50h from the forest edge. To refine the comparison between the CFD and wind tunnel experiments, a quadrant-hole analysis (Q-H) is performed. The principle is to perform conditional statistics on the most powerful turbulent fluctuations and to estimate their effect to different quantities such as the duration, the shear stress or the concentration flux. To some extend, these turbulent fluctuations can be related to coherent turbulent structures (CS). In the method, the hole parameter, H, acts as a filter to keep the most powerful fluctuations and to remove the weakest turbulent fluctuations. As such, events are counted only if |u'w'| > (H. σ u. σ w). Here H = 1. The quadrant (Q) classifies the fluctuations such as Q1 is an “outward interaction”, Q2 an “ejection”, Q3 an “inward interaction” and Q4 a “sweep”. From this classification, the duration fraction represents the time spent in a given quadrant and conditional statistics can be performed on the contribution to the total shear stress or to the concentration flux. In the enclosed figure (bottom), the ratio of the contribution of ejections and sweeps to the total Reynolds stress (u'w') is presented and compared to various experiments and a simulation. For the preliminary results presented, Q-H analysis are performed only on a restricted number of points in the LES calculation. Altitude is here normalized by the height of the roughness sub-layer estimated from the skewness profiles. Results show a reasonable agreement compared to other simulations of the literature when the relative contribution of ejections and sweeps is compared. Looking into more details (not presented here) results show an underestimation the contribution of sweeps right above the forest canopy and an underestimation of the contribution of ejections higher. The ratio of ejections and sweeps to the total concentration flux (w'c') also compares well to the wind tunnel measurements in the region [0.5 1.5]. The validity of the RANS/LES approach for reproducing the main statistics of the flow is particularly observed close above the forest canopy. However, some mismatches appear in the modeling of coherent structures statistics. More statistics on the Q-H analysis are planned to improve the comparison, however, the development of sweeps close to the forest may be affected by the porosity settings or by the RANS layer. A study of different porosity parameters and a modification of the grid are considered to address the question.