Recent developments and needs in wind farm design and optimization require improvements in the modelling of the atmospheric boundary layer (ABL), from the surface to the free atmosphere. Such developments point at progressively incorporating more realistic atmospheric physics in order to improve the simulation of wind farm design procedures. The traditional approach to wind resource assessment and evaluation of turbine site suitability is to model the microscale flow around and within a wind farm based on site measurements and microscale flow models that rely on the similarity theory of Monin–Obukhov. For this purpose, steady-state is often assumed and neutral atmospheric conditions are typically considered.
Several downscaling approaches have been attempted to bridge the gap between the large scale dynamics and the local and microscales providing relevant information about the wind field. A seamless transition from the mesoscale to the microscale through setting up a unified model chain is one of the present challenges to develop in the next generation of wind assessment tools and also from the perspective on increasing Earth System Model resolution and the realism of model physics. From the perspective of microscale model developments, models have to extend their range to simulate the entire boundary layer and to include the relevant physics involved (e.g. Coriolis) as well as a realistic large scale forcing, dependent on thermal stratification from the surface layer to the free atmosphere. The dynamics of these mechanisms determine the relationships between the large scale and the meso and microscale wind climatology and wind conditions relevant for wind resource assessment or for wind turbine siting.
Thus, a higher level of complexity can be provided by retaining the nonlinearity of the Navier-Stokes equations and simulate momentum, turbulence and energy with computational fluid dynamic (CFD) codes. Nevertheless, the application of CFD in wind assessment is still largely based on steady-state Reynolds-averaged Navier Stokes (RANS) turbulence models based on neutral and surface-layer approximations. More complex approaches include thermal stratification in RANS or use Large Eddy Simulation (LES) methods that can simulate terrain gravity waves. Nevertheless, regardless of the complexity of the turbulence model, they all assume idealized microscale conditions where mesoscale tendencies are not incorporated as additional forcing in the computational domain.
Some of the analysis carried out within Enerxico considers three types of simulated data that are used to describe tendencies or to offer complementary information to them. ERA5 reanalysis data are considered under the hypothesis that their approximately 26 km lat. x lon. resolution is sufficient to capture the spatial and temporal representative features of gradient and advection tendencies. WRF mesoscale model output including a specific code for the calculation of tendencies is also used to compare with ERA5 estimates. Also, tendency estimates obtained with a postprocessing tool developed to feed the Alya microscale model are also considered
One of the codes exploited to perform these calculi is WRF. As part of WP1, there has been a work on improving the computational efficiency of the code by reducing the MPI calls that were implemented in order to reduce the communication time that limit the code scalability.
This is, the observed loose of efficiency is due to the serialization and temporal umbalances of the code that increase with the scale. The analysis identified different regions with temporal umbalances as well as how the imbalance of the largest computing region is absorbed by the point to point communications. In this sense, work has focused on two specific WRF modules designed and implemented by CIEMAT: PBL (YSU) and SL (Revised MM5 surface layer scheme), obtaining a ~8% improvement in the strong scalability tests carried out once the umbalances were overcome. The results of this work are compiled in the project deliverable “D1.2 Report on intra-node and multi-node optimizations for HPC codes”.
With this new version, several studies within Enerxico WP2 have been carried out. Specifically speaking, the information provided by stability indices, vertical temperature, humidity, and horizontal wind profiles, as well as advection and gradient changes with height was obtained from ERA5 reanalysis at model levels for an area of interest around the Alaiz site at the NE of the Iberian Peninsula. Results have shown that thermal and dynamic profiles as well as stability indices present small variations within nearby grid points of the ERA5 reanalysis (~27 km lat. x lon. resolution), thus reflecting mesoscale variability at intra and interdaily timescales.
Advection and gradient tendencies were calculated from ERA5 output and from WRF simulations of comparable resolution, including both a direct implementation within the WRF code and using a post-processing script that considers WRF output and estimates values averaged in time and space over a different mesh as assimilated by the Alya microscale model.
Advection and gradient tendencies as represented by ERA5 do not show clear spatial links to stability, thermal, and dynamic variables; except for tendencies being larger for areas of increased horizontal wind components. Advection and gradient tendencies obtained from ERA5 and WRF are spatially consistent in their large scale structure and the timing of changes, albeit with small scale differences. WRF tends to produce larger advection values close to the surface. Postprocessed and space averaged WRF outputs tend to produce larger tendencies than direct WRF outputs. Differences between both can be considered of the same magnitude as differences with ERA5 output (Figs. 8 and 9).
Increasing WRF resolution produces an increase of spatial gradients in the resulting advection and gradient tendencies. Reducing temporal sampling of ERA5 output from hourly to daily resolution preserves the vertical and temporal (synoptic) structure of tendencies.
Gradient (shaded) and advection (contour) tendencies from 2018-09-27 00:00 to 2018-10-04 00:00 in the nearest point to Alaiz wind farm calculated with WRF output using the domain at 27 km of horizontal resolution (Fig. 2 right). Top: by using the code to estimate tendencies; bottom: obtained by applying Lehner (2012) and Sanz Rodrigo et al (2017) b but without spatial and temporal averaging