Wind patterns in main and tributary valley systems undergo diurnal variations that are very dependent on the valley's dimensions. Characterizing the down‐valley winds is important for pollutant dispersion studies. It is also a challenge for numerical simulations. This paper uses KASCADE (KAtabatic winds and Stability over CAdarache for Dispersion of Effluents) observations to describe down‐valley flow development and characteristics in two intersecting, shallow valleys of different sizes located in the pre‐alpine region of southeast France. The Durance Valley and its tributary valley, the Cadarache Valley, are different in depth (200 and 100 m depth, respectively), slope (0.2 and 1.2°), width (5–8 and 1–2 km) and length (>50 and 6 km). Both down‐valley flows were dominant features for the period of measurements. The distinct valley scales lead to different characteristics: the smaller Cadarache down‐valley wind is primarily thermally driven and attains wind speeds of up to 1–4 m s⁻¹ with its jet nose at around 30 m in height above ground. The flow at that height initiates 2 h after sensible heat flux turnover, a time‐scale which relates to a linear wave solution for along‐valley winds. The Durance down‐valley wind involves larger scales –for both terrain features and large‐scale circulation –and is related to a mountain‐to‐plain circulation. Therefore its onset occurs with a delay of ∼6 h after sunset. It reaches speeds of 4–8 m s⁻¹ and a depth of 175–225 m above ground. Its highest occurrence is just after sunrise. Between the two valley winds, a shear layer is observed, the origin of which could be a combination of a Cadarache valley flow return current and a Durance down‐valley wind stream redirected by orography.

Near-surface winds over complex terrain generally feature a large variability at the local scale. Forecasting these winds requires high-resolution numerical weather prediction (NWP) models, which drastically increase the duration of simulations and hinder them in running on a routine basis. Nevertheless, downscaling methods can help in forecasting such wind flows at limited numerical cost. In this study, we present a statistical downscaling of WRF (Weather Research and Forecasting) wind forecasts over southeastern France (including the southwestern part of the Alps) from its original 9 km resolution onto a 1 km resolution grid (1 km NWP model outputs are used to fit our statistical models). Downscaling is performed using convolutional neural networks (CNNs), which are the most powerful machine learning tool for processing images or any kind of gridded data, as demonstrated by recent studies dealing with wind forecast downscaling. The previous studies mostly focused on testing new model architectures. In this study, we aimed to extend these works by exploring different output variables and their associated loss function. We found that there is no one approach that outperforms the others in terms of both the direction and the speed at the same time. Finally, the best overall performance is obtained by combining two CNNs, one dedicated to the direction forecast based on the calculation of the normalized wind components using a customized mean squared error (MSE) loss function and the other dedicated to the speed forecast based on the calculation of the wind components and using another customized MSE loss function. Local-scale, topography-related wind features, which were poorly forecast at 9 km, are now well reproduced, both for speed (e.g., acceleration on the ridge, leeward deceleration, sheltering in valleys) and direction (deflection, valley channeling). There is a general improvement in the forecast, especially during the nighttime stable stratification period, which is the most difficult period to forecast. The result is that, after downscaling, the wind speed bias is reduced from −0.55 to −0.01 m s⁻¹, the wind speed MAE is reduced from 1.02 to 0.69 m s⁻¹ (32 % reduction) and the wind direction MAE is reduced from 25.9 to 15.5^∘ (40 % reduction) in comparison with the 9 km resolution forecast.

Land cover (LC) data incorporation, for weather modeling purposes, highlights many problems. The straightforward Single Level Mode (SLM) aggregation is not adapted for high‐resolution LC maps, with a high number of classes, because it could generate false classifications. We propose a Multi‐Level Mode (MLM) aggregation method that includes a hierarchical structure. This study focuses on the Corine Land Cover (CLC) data set. Differences between MLM and SLM methods are small at the finest horizontal resolution and increase to a value of around 16% at 9‐km horizontal resolution. To further integrate CLC data into WRF (Weather Research Forecasting model), we included a dedicated table of physical parameters next to using the classical conversion toward the USGS one. To evaluate the LC impact on the modeled boundary layer, we used WRF at 1 km to simulate a 4‐day period with diurnal cycles of valley winds in the heterogeneous western pre‐Alps area. We tested three LCs (USGS, MODIS, and CLC), where CLC has two physical parameter tables and two aggregation methods. CLC performs better than other tested LCs. The CLC aggregation methods revealed limited differences between the simulated variables, although the MLM method gives better results. Since these comparisons are restricted to a single location with the same LC type, the differences between various simulations regarding atmospheric parameters probably result from horizontal advection from upwind areas where surface conditions differ. Extended time series with multiple locations and diverse meteorological conditions need to be considered to reinforce the results presented here.

In regions of complex topography, local flows are difficult to forecast on a routine basis, especially in stable conditions, due to the coarse resolution of operational models. The Cadarache valley (southeastern France) features this sort of complex topography. The Weather Research and Forecasting (WRF) model is run daily to forecast the weather in this region with a horizontal resolution of 3 km. Such a resolution cannot resolve all topography details of the small Cadarache valley, and therefore its local wind patterns. Other variables, however, that are less dependent on the subgrid topography, are satisfactorily forecasted, and used as inputs to an artificial neural network (ANN) designed to reproduce wind observations inside the valley from WRF forecasts. A variable selection procedure identified 5 key input variables that best drive the ANN. With respect to the WRF output, the ANN significantly improves forecasted low-level winds, both for speed and direction. This study demonstrates the potential for the ANN technique to be used as a correcting tool to forecast weather conditions at the local scale when numerical modeling is performed at a resolution too coarse to take into account the effect of local topography.

In an overall approach aiming at the development and qualification of various tools designed to diagnose and/or forecast the flows at the local scale in complex terrain, we qualified a numerical model based on the WRF platform and operated in a two-way nested domain mode, down to a horizontal resolution of 111 m for the smallest domain. The area in question is the Cadarache valley (CV), in southeast France, which is surrounded by hills and valleys of various sizes. The CV dimensions (1 km wide and 100 m deep) favor the development of local flows greatly influenced by the diurnal cycle and are prone to thermal stratification, especially during stable conditions. This cycle was well documented due to permanent observations and dedicated field campaigns. These observations were used to evaluate the performance of the model on a specific day among the intensive observation periods carried out during the KASCADE-2017 campaign. The model reproduced the wind flow and its diurnal cycle well, notably at the local CV scale, which constitutes considerable progress with respect to the performances of previous WRF simulations conducted in this area with kilometric resolution, be it operational weather forecasts or dedicated studies conducted on specific days. The diurnal temperature range is underestimated however, together with the stratification intensity of the cold pool observed at night. Consequently, the slope drainage flows along the CV sidewalls are higher in the simulation than in the observations, and the resulting scalar fields (such as specific humidity) are less heterogeneous in the model than in the observations.

We hereby present a new method with which to nowcast a thermally driven, downvalley wind using an artificial neural network (ANN) based on remote observations. The method allows the retrieval of wind speed and direction. The ANN was trained and evaluated using a 3-month winter-period dataset of routine weather observations made in and above the valley. The targeted valley winds feature two main directions (91% of the total dataset) that are aligned with the valley axis. They result from downward momentum transport, channeling mechanisms, and thermally driven flows. A selection procedure of the most pertinent ANN input variables, among the routine observations, highlighted three key variables: a potential temperature difference between the top and the bottom of the valley and the two wind components above the valley. These variables are directly related to the mechanisms that generate the valley winds. The performance of the ANN method improves on an earlier-proposed nowcasting method, based solely on a vertical temperature difference, as well as a multilinear regression model. The assessment of the wind speed and direction indicates good performance (i.e., wind speed bias of −0.28 m s⁻¹ and 84% of calculated directions stray from observations by less than 45°). Major sources of error are due to the misrepresentation of cross-valley winds and very light winds. The validated method was then successfully applied to a 1-yr period with a similar performance. Potentially, this method could be used to downscale valley wind characteristics for unresolved valleys in mesoscale simulations.