Abstract

An operational weather diagnostics application for automatic generation of wind fields in near–real time from observations delivered by the high-density WegenerNet meteorological station network in the Feldbach region of Austria is introduced. The purpose of the application is to empirically provide near-surface wind fields of very high spatial and temporal resolution for evaluating convection-permitting climate models as well as investigating weather and climate variability on a local scale. The diagnostic California Meteorological Model (CALMET) is used as the core tool. This model computes 3D wind fields based on observational weather data, a digital elevation model, and land-use categories. The application first produces the required input files from the WegenerNet stations and subsequently runs the CALMET model based on this input. In a third step the modeled wind fields are stored in the WegenerNet data archives every 30 min with a spatial resolution of 100 m × 100 m, while also generating averaged weather and climate products during postprocessing. The performance of the modeling against station observations, for which wind speeds were classified into weak and strong wind speeds, is evaluated and reasonably good results were found for both wind speed classes. The statistical agreement for the vector-mean wind speed is slightly better for weak wind speeds than for strong ones while the difference between modeled and observed wind directions is smaller for strong wind speeds than for weak ones. The application is also a valuable tool for other high-density networks.

1. Introduction

Progression in computer technology and the growing power of computers in data processing leads to more complex and accurate nonhydrostatic climate models for high-resolution simulations, with horizontal resolutions at a scale of 1 km (Prein 2013; Suklitsch et al. 2011; Hohenegger et al. 2008). Because of this higher resolution, such nonhydrostatic and convection-permitting modeling (NHCM) provides more realistic simulations, especially for regions with complex terrain (Awan et al. 2011; Suklitsch et al. 2011; Prein et al. 2013).

The evaluation of regional climate models (RCMs) and further model improvements require, among other needs, meteorological observations and gridded datasets with correspondingly high spatial and temporal resolutions (Kirchengast et al. 2014). RCMs generally simulate area-averaged rather than point-scale processes (Osborn and Hulme 1998). Therefore, the most appropriate data for evaluation are gridded datasets where each grid value is a best estimate average of the grid-box observations (Haylock et al. 2008). For example, the European Climate Assessment and Data (ECA&D) activity provides the daily gridded observational dataset (E-OBS) based on station datasets and other archives, consisting of temperature, precipitation, and sea level pressure for Europe. This dataset is used on a regional scale for evaluating RCMs, monitoring climate change, and for studies regarding climate variability (Haylock et al. 2008; Klok and Klein Tank 2009; Brunner et al. 2017).

The Wegener Center at the University of Graz, Graz, Austria, established the long-term field experiment WegenerNet Feldbach region, a dense grid of more than 150 meteorological stations within an area of about 20 km × 15 km for investigating weather and climate on a local scale as well as evaluating RCMs (Kirchengast et al. 2014; O et al. 2017; Kabas 2012; Kabas et al. 2011). The processing system developed to control and manage observations from the meteorological stations is called the WegenerNet Processing System (WPS) and consists of four subsystems. The Command Receive Archiving System (CRAS) is used to transfer the raw data via general packet radio service (GPRS) to the WegenerNet Database, the quality control system (QCS) checks the quality of the data, the data product generator (DPG) produces gridded fields of weather and climate products, and the visualization and information system (VIS) offers the data to users via the WegenerNet data portal (www.wegenernet.org).

Products already implemented in the DPG are gridded weather and climate products generated by a point-specific interpolation of the main variables of temperature, precipitation, and relative humidity. In the case of the temperature, fields 2 m above the ground are produced, including calculated temperature lapse rates from the observational datasets. The gridded fields of relative humidity are constructed by an inverse-distance weighted interpolation and for precipitation the inverse-distance-squared algorithm is used. Detailed information related to the WPS and its DPG subsystem can be found in Kirchengast et al. (2014) and Kabas (2012).

The application introduced in this study fills a critical gap in the DPG. Gridded wind fields generated from the existing wind observations of the 12 WegenerNet stations that are well distributed within the 20 km × 15 km area have not yet been implemented into the DPG. Wind is often considered as one of the most difficult meteorological variables to model and depends on many different conditions, including surface properties such as topography and surface roughness. Therefore, a simple interpolation of wind station data can only be performed in cases of uniform characteristics of landscape and an additional tool is needed to determine the spatial distribution of the wind parameters (Abdel-Aal et al. 2009; Sfetsos 2002).

The key goal of this study is therefore the development of an adequately realistic and robust operational application for the automated generation of gridded high-resolution wind fields from the observational data of the WegenerNet.

In general, to determine the spatial distribution of wind speed and wind direction, different types of model simulations are used (Morales et al. 2012). Dynamic mesoscale models like the Weather Research and Forecasting (WRF) Model or the Integrated Nowcasting through Comprehensive Analysis (INCA) are the most sophisticated options, since they are capable of physically simulating synoptic processes and interactions between the earth’s surface and the atmosphere (Truhetz 2010). These complex prediction models solve prognostic equations and require extensive computational resources. Therefore, because of the considerable computing time and power needs, the spatial and temporal resolution is limited (Jancewicz 2014; Truhetz et al. 2007). An alternative model type was selected for being able to develop wind fields with high spatial and temporal resolution in near–real time without the need for extensive computing resources.

Empirical diagnostic models, used to represent the actual state of the atmosphere based on the data, are such an adequate alternative (Morales et al. 2012). They apply parameterizations to empirically take into account processes like the kinematic effects of terrain, slope flows, and terrain-blocking effects. They also include divergence minimization schemes for satisfying the incompressible mass consistency. Dynamic processes, like flow splitting, grid-resolved turbulence, etc. are not taken into account by these diagnostic models (Truhetz 2010; Wang et al. 2008; Truhetz et al. 2007). Thus, they need much less computing time for modeling.

In this study we employ the freely distributed diagnostic California Meteorological Model (CALMET) for the development of high-resolution wind fields. CALMET is an empirical model originally developed by the California Air Resources Board to provide wind fields for the pollutant dispersion model called CALPUFF (Scire et al. 1998; Cox et al. 2005). The model reconstructs 3D wind fields based on meteorological observations, terrain elevations, and land-use information. The quality of the modeled wind fields depends on the quality and spatial resolution of the observations from in situ meteorological station measurements as well as of the surface-related datasets, which comprise digital elevation model data and land-use-type data (Morales et al. 2012; Cox et al. 2005).

The operational requirement for our application is that the wind fields are automatically generated from the observational data of the WegenerNet in near–real time and stored to the WegenerNet archives with a spatial resolution of 100 m × 100 m and a time resolution of 30 min. Furthermore, the model performance of these produced wind fields has to be evaluated for periods with representative weather conditions.

Reporting this work, the paper is structured as follows. Section 2 provides a description of the study area, the WegenerNet Feldbach region in Austria with its over 150 meteorological stations. Section 3 presents the methodology for the empirical wind field modeling, where first the characteristics of the CALMET model and the application (developed in Python) for the automated production of the wind fields are explained. Second, a description of atmospheric weather conditions during the model evaluation periods is given. Section 4 describes the results of the wind field modeling for the selected periods in August 2008 and March 2013. Finally, section 5 provides the conclusions and prospects of the study and the next steps of follow-on work.

2. Study area and WegenerNet data

The WegenerNet Feldbach region (Fig. 1) is located in the Alpine foreland of southeastern Styria, Austria, centered near the town of Feldbach (46.93°N, 15.90°E), a region with high weather and climate variability (Kirchengast et al. 2014; Kabas et al. 2011). The terrain of this study area has many low hills and is characterized by generally small differences in altitude with maximum values of about 100 m between the valleys and crests. The highest peak is the Gleichenberger Kogel, with an elevation of 598 m, and most of the area is used for agriculture, as illustrated by Fig. 2a.

Fig. 1.

(a) Overview of the WegenerNet Feldbach region study area [white rectangle; enlarged in (b)] to the southeast of Styria, Austria. (b) The WegenerNet Feldbach region with its 153 meteorological stations, with the legend explaining map characteristics and station types.

Fig. 1.

(a) Overview of the WegenerNet Feldbach region study area [white rectangle; enlarged in (b)] to the southeast of Styria, Austria. (b) The WegenerNet Feldbach region with its 153 meteorological stations, with the legend explaining map characteristics and station types.

Fig. 2.

(a) Land use of the study area based on the CLC 2006 raster version dataset. (b) Example temperature field of the study area (10 Aug 2008; LT = 1600 UTC + 2 h).

Fig. 2.

(a) Land use of the study area based on the CLC 2006 raster version dataset. (b) Example temperature field of the study area (10 Aug 2008; LT = 1600 UTC + 2 h).

The climate of this southeastern Alpine foreland region is more Alpine at valley floors with cold winters and hot summers, and more Mediterranean along the hillsides with milder winters and hot summers (Kabas 2012; Wakonigg 1978). Heavy convective rainfall from thunderstorms, with frequent hailstorms, dominates the precipitation in summer. Strong storms can occasionally occur in the winter season (Kabas 2012; Prettenthaler et al. 2010; Wakonigg 1978). Typical for the region are thermally induced local flows and the influence of thermally-driven regional wind systems with weak wind speeds, caused by a dynamical process called Alpine pumping (Lugauer and Winkler 2005).

The hills and valleys of the region exhibit temperature contrasts, which the temperature fields produced using a modified version of CALMET for fine-resolution weather conditions are able to capture (Fig. 2b). The temperatures in the late afternoon on 10 August 2008 are shown, with higher values at lower altitudes and in the valleys and lower values along the hillsides (Fig. 2b). In this modified CALMET version, we used algorithms developed by Bellasio et al. (2005) to account for topographic shading effects and the height dependency of surface temperatures. To be consistent with the published CALMET model in this study, however, we used the original CALMET code in our wind field modeling application.

The 153 stations of the WegenerNet are used to supply in situ weather measurements as model input. These stations, with an average station distance of about 1.4 km, in an area of around 20 km × 15 km, measure meteorological parameters with a time resolution of 5 min. The stations are equipped with different sensor configurations. The 130 base stations of the network measure the main variables air temperature, relative humidity, and precipitation. Three of the stations are lacking one or two parameters (Fig. 1b; red circles). Eleven special base stations measure soil moisture and soil temperature in addition to the main variables (Fig. 1b; red triangles), 11 primary stations measure wind speed and wind direction in addition to the main variables (Fig. 1b; blue triangles), and one reference station measures wind, soil variables, air pressure, and net radiation in addition to the main variables (Fig. 1b; blue square).

All wind sensors from the primary stations and the reference station, except for stations 44, 55, and 72, are mounted on 10-m masts. Station 44 measures the wind parameters at 55-m height, station 72 at 18 m, and station 101 at 14 m (all mounted on top of silos, used in the region for storing agricultural harvests). The observations from the stations provided by the Central Institute for Meteorology and Geodynamics [Zentralanstalt für Meteorologie und Geodynamik, (ZAMG); Fig. 1b; blue stars] in Vienna, Austria, are integrated into the WPS for use as reference. Table 1 summarizes the technical characteristics of the WegenerNet stations equipped with wind sensors (Kirchengast et al. 2014; Kabas 2012).

Table 1.

Characteristics of meteorological stations with wind sensors (WN, WegenerNet station; ZAMG, National Meteorological Service station).

Characteristics of meteorological stations with wind sensors (WN, WegenerNet station; ZAMG, National Meteorological Service station).
Characteristics of meteorological stations with wind sensors (WN, WegenerNet station; ZAMG, National Meteorological Service station).

The observations of the meteorological stations and the gridded data products produced by the DPG are available to users in near–real time since 2007, generated with a latency of about 1–2 h. The same requirements apply to the new wind field modeling application introduced here, to reproduce the wind fields since 2007 and to provide the ongoing data with a maximum latency of 2 h.

3. Methods and evaluation periods

The core part of the new operational wind field application is the CALMET model briefly introduced in section 1 above (Scire et al. 1998).

The wind fields are computed by CALMET using a two-step approach. The first step includes the adjustment of an initial guess wind field in regard to kinematic effects of terrain, slope flows, and terrain-blocking effects. Based on the settings in the CALMET control file, a user has different options to generate the initial guess wind field. In the current study we use the so-called observation-only approach in order to ensure genuinely empirical wind fields, which means that the initial guess wind field is produced by an interpolation based solely on observational data. We enhanced the original CALMET code related to the interpolation for the initial guess wind field to enable less weight (influence radius) for stations that are influenced by local terrain (the original CALMET foresees a fixed influence radius).

In a second step, the observational data are used again and blended into the step 1 wind field by an inverse distance weighting interpolation to produce the consolidated step 2 wind field. Each station site with wind observations affects the step 2 wind field within a user-defined radius of influence. In addition, relative weighting parameters are used to weight the observations and the wind fields previously produced in step 1. To derive mass-consistent wind fields, the horizontal winds were adjusted by a divergence minimization procedure. In the observation-only approach the user has to define several critical parameters, which can affect the results of the model runs significantly.

The chosen settings for the WegenerNet Feldbach region, based on extensive sensitivity tests, and the explanation of relevant model parameters are shown in Table 2. Details related to model settings and options can be found in the CALMET user’s manual (Scire et al. 1998). Adjustment of our application to other regions needs repetition of the sensitivity test in those regions, as is unavoidable with this type of empirical modeling.

Table 2.

Settings of critical area-specific model parameters in CALMET, used in this study.

Settings of critical area-specific model parameters in CALMET, used in this study.
Settings of critical area-specific model parameters in CALMET, used in this study.

Figure 3 shows the flowchart of our application, implemented in Python, that automatically creates wind fields with a time resolution of 30 min and a spatial gridding of 100 m × 100 m. This application reads the meteorological data for each time step from the WegenerNet database and creates the needed surface meteorological data files and the upper-air data files in a CALMET-compliant format. This preparatory processing step of the application includes the calculation of vector-mean values from the 5-min observational wind data. The vector mean of the observed wind speed and wind direction is estimated by

 
formula

and

 
formula

where the mean values of the north component υ and the east component u are calculated from six observations for each CALMET time step (N = 6 for a temporal CALMET resolution of 30 min) by

 
formula

and

 
formula

CALMET requires upper-air data consisting of vertical profiles of wind speed, wind direction, temperature, pressure, and elevation, usually obtained from radiosonde stations. The existing radiosonde station location in southeastern Austria is not within or nearby the study area and therefore is not representative of the WegenerNet Feldbach region. Because of the distance between radiosonde stations, and to keep the key operational input independent from data external to the WegenerNet, 12 upper-air data files are created automatically for each time step from the observed temperature, pressure, and wind measurements at the locations of the WegenerNet primary stations and the reference station.

Fig. 3.

Work flow for the automatic generation of high-resolution wind fields from the WegenerNet dataset.

Fig. 3.

Work flow for the automatic generation of high-resolution wind fields from the WegenerNet dataset.

For this purpose, temperature lapse rates are estimated based on a linear and bilinear least squares adjustment for each time step from the 153 WegenerNet temperature observations at different elevations in this hilly terrain. The bilinear fits that are composed of two lines are performed in 4-m increments of altitude z and are estimated by

 
formula

where T(z) is the temperature at altitude z. The temperature lapse rates are defined by and , and the two lines intersect at the altitude with the corresponding temperature estimate . For the linear fit we set to zero and fit just one line using the equation above. This linear fit is compared with the best-guess approximation of all bilinear fits, based on the performance parameter (fit residuals). The regression parameters of the fit with the smallest are adapted as a result for , , , and . These estimates are then used to compute the temperature at the defined vertical levels (Table 2) for the WegenerNet primary stations and the reference station. The CALMET model uses a terrain-following vertical coordinate system, which means that its vertical levels represent the height above ground.

The upper-air pressure variables are produced for each level and time step by using the barometric law, starting from the surface pressure observed by the reference station. The air pressure P at the altitude z is calculated by

 
formula

where is the air pressure at sea level (z = 0 m), which is calculated from the observations of the reference station (z = 302 m) through the inverse of the equation above. The sea level temperature is computed for the WegenerNet primary stations and the reference station by the use of the regression parameters. Afterward, we use the equation to estimate the pressure for the terrain-following vertical levels. The regression parameters , , , and are those determined by the least squares adjustment explained above. The gas constant R, mean molar mass of dry air , and the gravity acceleration g are taken to be constant with values, respectively, equal to 8.3145 J mol−1 K−1, 0.028966 kg mol−1, and 9.80795 m s−2.

In the case of upper-air wind variables the measurements from the primary stations and the reference station are used for the 10-m vertical level. The upper-air wind variables for the highest defined vertical level (Table 2) are set to the observed values from the highest WegenerNet station with valid wind observations at the current time step in each file.

In addition, the CALMET model requires a preprocessed gridded geophysical data file consisting of terrain elevations and land-use categories (CALMET input data files; orange fields in Fig. 3). As our digital elevation model (DEM), a dataset derived from airborne laser scanning point clouds (provided online by gis.steiermark.at) is used, illustrated by the elevations in Fig. 1b. Before putting the data into the CALMET model, the original spatial resolution of 10 m is resampled and averaged to 100 m. Figure 2a shows the land-use map of the study area based on the Corine Land Cover 2006 dataset (CLC 2006; see EEA 2007) converted into a CALMET-compliant format. The definitions of the CLC land-use categories differ from the CALMET classes. In the entire CLC dataset the third and most detailed level contains 44 classes, while the default classification scheme of the CALMET model distinguishes up to 14 land-use types only (Oleniacz and Rzeszutek 2014). We therefore reclassified 13 CLC 2006 land-use categories of the study area into six CALMET-compliant classes, and assigned the appropriate CALMET grid code to each category. For each of these land-use categories the parameters shown in Table 3 are applied to the CALMET geophysical dataset.

Table 3.

Geophysical parameters based on CLC 2006 and used in this study.

Geophysical parameters based on CLC 2006 and used in this study.
Geophysical parameters based on CLC 2006 and used in this study.

The model performance is evaluated using periods with mainly two types of wind events: the thermally induced wind events and the strong wind events, which are representative of this study area. We chose the periods August 2008 and March 2013. In August 2008 the study area was mainly controlled by autochthonous weather conditions, which mainly led to thermally induced wind systems. Such weather conditions are characterized by low wind speeds, cloudless skies, low relative humidity, and increased radiation fluxes between the earth surface and the atmosphere (Prettenthaler et al. 2010). The synoptic influences are small, and the wind field is dominated by small-scale temperature and pressure gradients. In March 2013 several episodes of strong wind occurred. The wind speed was classified as weak () or strong (), to be able to estimate the performance for different conditions. Wind speeds < 0.5 m s−1 were classified as calm. In August 2008, 56% of the observed wind speeds from the WegenerNet stations were classified as weak and 8% as strong (36% being calm). In March 2013 as well 56% were weak but 19% were strong (25% being calm). The limit of 2.5 m s−1 was chosen because it is the typical maximum wind speed in the study area during autochthonous days, caused by Alpine pumping (Lugauer and Winkler 2005).

For statistically evaluating the modeling skill, we use the statistical performance parameters summarized in Table 4. As an evaluation method, we use the leave-one-out cross validation for the WegenerNet and ZAMG stations with wind data as listed in Table 5 (more details in section 4 below). This evaluation methodology means that observations at one wind station are removed from the model input and generated wind fields are evaluated against the wind data from this station.

Table 4.

Statistical performance parameters used for the evaluation of the wind field modeling results.

Statistical performance parameters used for the evaluation of the wind field modeling results.
Statistical performance parameters used for the evaluation of the wind field modeling results.
Table 5.

Statistical performance measures calculated for representative meteorological stations for weak and strong wind speeds (top half for August 2008; bottom half for March 2013). See Table 4 for more information on the calculations for the performance parameters. The results for ZAMG FB and WN 132 are illustrated as examples in Figs. 69.

Statistical performance measures calculated for representative meteorological stations for weak and strong wind speeds (top half for August 2008; bottom half for March 2013). See Table 4 for more information on the calculations for the performance parameters. The results for ZAMG FB and WN 132 are illustrated as examples in Figs. 6–9.
Statistical performance measures calculated for representative meteorological stations for weak and strong wind speeds (top half for August 2008; bottom half for March 2013). See Table 4 for more information on the calculations for the performance parameters. The results for ZAMG FB and WN 132 are illustrated as examples in Figs. 6–9.

Measurements from the Feldbach and Bad Gleichenberg ZAMG stations are only used as reference for evaluation and not as model input. We compared the modeled outputs with the reference station datasets for each time step of the validation period. For each reference station location, the statistical performance parameters (Table 4) are calculated by comparing the nearest-neighbor gridpoint values to the observations at the corresponding location.

4. Results

We show typical examples of modeled wind fields for thermally driven circulations on 10 August 2008 (Fig. 4a) and strong wind speeds on 15 March 2013 (Fig. 4b), both at a height of 10 m. The top panel on the left-hand side of Fig. 4 illustrates nighttime winds with a down-valley direction caused by temperature and pressure gradients on a local scale. The middle-left panel of Fig. 4 shows the thermally induced wind field in the afternoon, caused by the Alpine pumping. This typical thermally induced regional wind of the study area is called Antirandgebirgswind and arises usually in the afternoon as a southerly wind (Wakonigg 1978). This Antirandgebirgswind, with maximum near-surface wind speeds of around 2.5 m s−1, is a characteristic regional flow between the bordering mountains of the eastern Alps (Koralpe, Gleinalpe, Fischbacher Alps, and the mountainous region to the north of Graz) and the so-called Riedelland, which is the hilly country region in southeastern Styria comprising the study area and part of the Alpine Foreland (Wakonigg 1978). The bottom-left panel of Fig. 4 shows the early evening situation, where valley winds arise, generally from an up-valley direction.

Fig. 4.

Modeled wind fields typical of the study area. (a) Thermally induced wind fields (10 Aug 2008) and (b) strong region-scale winds (15 Mar 2013). Times shown are UTC (corresponding to LT − 2 h on 10 Aug 2008 and LT − 1 h on 15 Mar 2013).

Fig. 4.

Modeled wind fields typical of the study area. (a) Thermally induced wind fields (10 Aug 2008) and (b) strong region-scale winds (15 Mar 2013). Times shown are UTC (corresponding to LT − 2 h on 10 Aug 2008 and LT − 1 h on 15 Mar 2013).

Figure 4b displays a strong all-day northerly wind from 15 March 2013, with maximum 30-min wind speeds of around 8 m s−1 at 1000 UTC (1100 LT) (middle-right panel). It can be seen that in the area around WegenerNet station 135 the wind direction is forced into a NW-to-W component, resulting from the observations at this station. The influence of observations from a given station, weighted by the influence radius parameter (Table 2) in step 2 of CALMET interpolation scheme (as explained in section 3 above) becomes obvious in this particular example.

The enlarged sections in Fig. 5 with an area of about 3 km × 3 km around the “Steinberg” hill display wind fields (top two rows) and vertical cross sections (bottom row) for different atmospheric stratification (for Fig. 5a the same time slice as in Fig. 4a, bottom left). In Fig. 5a the wind is again dominated by the thermally driven Antirandgebirgswind under unstable conditions, as diagnosed by CALMET based on a small value of an internal stability factor (ratio of Brunt–Väisälä frequency to wind speed) and a negative Monin–Obukhov length. Higher vertical wind speeds can be observed, especially at heights of 50 m above ground. In CALMET such terrain-forced vertical wind speeds are estimated based on a stability-dependent decay function weighted by the stability factor (Scire et al. 1998). The horizontal direction of flow remains mainly unchanged and is not modified across the local terrain. The temperature contour of the vertical cross section (Fig. 5, bottom left) shows temperature lapse rates with maximum temperatures of about 21°C in the Raab valley north of the Steinberg hill and minimum temperatures of about 20.3°C at the top of the hill.

Fig. 5.

Enlarged view of the subregion around the Steinberg hill east of reference station 77. (a) Thermally induced wind fields under unstable conditions (10 Aug 2008) and (b) weak winds under stable conditions (15 Mar 2013); horizontal (υ, black) and vertical (w, blue) wind components at (top) 10 and (middle) 50 m above ground are indicated. (bottom) North-to-south vertical cross sections of wind vectors over the hill at the 10- and 50-m levels as well as temperature contours (color shading). Times shown are UTC.

Fig. 5.

Enlarged view of the subregion around the Steinberg hill east of reference station 77. (a) Thermally induced wind fields under unstable conditions (10 Aug 2008) and (b) weak winds under stable conditions (15 Mar 2013); horizontal (υ, black) and vertical (w, blue) wind components at (top) 10 and (middle) 50 m above ground are indicated. (bottom) North-to-south vertical cross sections of wind vectors over the hill at the 10- and 50-m levels as well as temperature contours (color shading). Times shown are UTC.

In winter, temperature inversions occur frequently in relation to high pressure weather conditions with weak wind speeds in the study area (Fig. 5b). In Fig. 5, the top and middle panels on the right-hand side indicate the modification of the wind field in a direction around the Steinberg hill with generally low horizontal and vertical wind speeds. The temperature contour of the vertical cross section (Fig. 5, bottom right) illustrates the strong temperature inversion with maximum temperatures of about 5°C on top of the hill and minimum temperatures in the Raab valley north of the Steinberg hill of about −2°C.

Periods with rapidly changing weather conditions, such as fast-developing thunderstorms, cannot be resolved in detail by the CALMET diagnostic model because of its limited time resolution of not shorter than 30 min.

Figures 6 and 7 illustrate the model performance for the Feldbach ZAMG station and WegenerNet station 132 as typical examples of the performance of a single station. The ZAMG station is located at 330-m height near the town of Feldbach in the Raab valley, the main valley of the study area (Fig. 1b). WegenerNet station 132 is located near the Poppendorfer stream at a height of 295 m. The environment of the latter station is characterized by low-density residential areas that influence the representativity of wind observations for the 1-km scale somewhat. In the scatterplots shown, the 30-min vector mean values from the observed wind speeds are compared with the modeled 30-min nearest-neighbor grid points.

Fig. 6.

Modeled vs observed wind speeds for the ZAMG station at Feldbach: (a) August 2008 and (b) March 2013 for weak (; gray dots) and strong (; black dots) wind speeds.

Fig. 6.

Modeled vs observed wind speeds for the ZAMG station at Feldbach: (a) August 2008 and (b) March 2013 for weak (; gray dots) and strong (; black dots) wind speeds.

Fig. 7.

As in Fig. 6, but for WegenerNet station 132.

Fig. 7.

As in Fig. 6, but for WegenerNet station 132.

For the Feldbach station, the comparison shows a high degree of similarity between modeled and observed wind speeds for both weak and strong conditions, with correlation coefficients ranging from 0.75 to 0.85 (Fig. 6). This indicates good representativeness for the 1-km scale. Because of the influence of local terrain on WegenerNet station 132, the results are slightly worse for this station, especially for strong wind speeds (Fig. 7). For example, the R value in August 2008 for strong conditions is 0.46 (Fig. 7a) compared with the Feldbach station with a value of 0.75 (Fig. 6a); the corresponding R value in March 2013 is also 0.46 (Fig. 7b) compared with a value of 0.84 for the Feldbach station (Fig. 6b). The mean absolute error () of the wind direction is for both stations higher in the case of weak wind speeds, as one might expect from the challenging effects that weaker wind speeds have on wind direction estimations.

Figure 8 shows the relative frequency of wind directions from the model compared with the observed wind directions for the ZAMG Feldbach station, again for the same periods and wind speed classes. It can be seen that the distribution of wind directions is similar among the observations and modeled values, which is a satisfying result, indicating the applicability of the wind fields. The largest difference between the modeled and observed wind directions for this station can be seen in Fig. 8c in August 2008 for strong wind speeds, with a shift between the SSE and S sectors. In this case the model calculates about 40% of the wind directions for the SSE sector, while the observations show about 40% in the S sector. This shift can be explained by the influence the environment of the station has on the wind field.

Fig. 8.

Relative frequency of wind directions for observed (blue line) and modeled (red line) values for the ZAMG Feldbach station: (left) August 2008 and (right) March 2013 for (a),(b) weak () and (c),(d) strong () wind speeds.

Fig. 8.

Relative frequency of wind directions for observed (blue line) and modeled (red line) values for the ZAMG Feldbach station: (left) August 2008 and (right) March 2013 for (a),(b) weak () and (c),(d) strong () wind speeds.

Figure 9 shows wind roses divided by wind speed categories for the Feldbach ZAMG station. Again, similar patterns between observed and modeled values for weak conditions are visible. For both periods, the model properly calculates values below 2.5 m s−1 (Figs. 9a,b). In Fig. 9c, the shift between observed and modeled values from the S to the SSE sector for strong wind speeds can be seen in detail; for example, the 40% model wind directions in the SSE sector all have weak values below 2.5 m s−1, while the observations only show around 15% in this sector, with wind speeds that are all strong up to 3.5 m s−1. In March 2013 the strong wind case (Fig. 9) aligns very well, with the observations inclined to somewhat stronger southward wind speeds than the modeled ones.

Fig. 9.

Relative frequency of wind directions based on wind speed categories for (first and third rows) observed and (second and fourth rows) modeled values for the ZAMG Feldbach station: (left) August 2008 and (right) March 2013 for (a),(b) weak () and (c),(d) strong () wind speeds.

Fig. 9.

Relative frequency of wind directions based on wind speed categories for (first and third rows) observed and (second and fourth rows) modeled values for the ZAMG Feldbach station: (left) August 2008 and (right) March 2013 for (a),(b) weak () and (c),(d) strong () wind speeds.

Figures 10 and 11 show the analogous results for WegenerNet station 132. The distribution of the observed wind directions from station 132 has narrower wind corridors for weak wind speeds compared with the modeled wind directions (Figs. 10a,b). From Figs. 11a and 11b the same results become obvious: the corridor of the observed wind directions with weak wind speeds is narrower, with prevailing wind directions from the NNW to the N or the S sector. For the strong wind speed category the pattern of the modeled wind directions is more similar to the observed wind directions (Figs. 10c,d and 11c,d), with the modeled wind speeds somewhat underestimated (Figs. 11c,d). Because of the quite good accordance between modeled and observed wind parameters for WegenerNet station 132, the overall reasonableness of the wind field results is also underscored by this station.

Fig. 10.

As in Fig. 8, but for WegenerNet station 132.

Fig. 10.

As in Fig. 8, but for WegenerNet station 132.

Fig. 11.

As in Fig. 9, but for WegenerNet station 132.

Fig. 11.

As in Fig. 9, but for WegenerNet station 132.

Table 5 summarizes the statistical results for all of meteorological stations that were used for the performance evaluation. In fact for internal extensive evaluation, all stations were used, but the ones examined in detail and summarized in Table 5 are well representative. The results of the relative statistical parameters applied to the vector mean of the wind speed [R and index of agreement (IOA)] generally show better values for the weak wind speed class than for the strong wind speed class. On the other hand, the statistical measure applied to evaluate wind directions () shows better results for strong wind speeds compared to weak wind speeds. In general, the bias B is somewhat negative, except for the Bad Gleichenberg ZAMG station and for WegenerNet station 132 in March 2013 during weak conditions.

The resulting RMSE values range from 0.37 to 0.67 m s−1 for weak wind speeds and from 0.43 to 1.77 m s−1 for strong wind speeds. For the weak wind speed class the correlation coefficient is higher than 0.75 for all stations expect for the Bad Gleichenberg ZAMG station and WegenerNet station 135. For the strong wind speed class the correlation coefficient is generally worse, especially for WegenerNet station 11.

Regarding the IOA, we note that in this study the IOA redefined by Willmott et al. (2012) is used with a lower limit of −1 and an upper limit of +1 with values approaching +1 representing a better degree of model performance. For example, an IOA of 0.5 means that the sum of the difference magnitudes between modeled and observed values is one-half of the sum of the observed deviation magnitudes. Conversely, an IOA of −0.5 implies that the sum of the difference magnitudes is twice the sum of the observed deviation magnitudes. Values of IOA near −1.0 mean either that the model-estimated deviations about are poor estimates of the observed deviations or that there is in fact little observed variability (Willmott et al. 2012). The higher values of the IOA for weak wind speeds compared with strong wind speeds therefore indicate that the sum of the difference magnitudes compared with the observed deviation magnitudes is lower for weak wind speeds (somewhat better performance) than for strong wind speeds.

5. Conclusions and prospects

This work has introduced an operational weather diagnostics application for the automatic generation of high-resolution wind fields from the dense WegenerNet Feldbach region network of meteorological stations operated by the Wegener Center in southeastern Styria, Austria. The wind fields are computed in near–real time and stored in the WegenerNet data archives, available at 30-min temporal resolution and with a spatial resolution of 100 m × 100 m. The core part of the new application is the freely available empirical model CALMET, which we employ to simulate the wind fields based on the WegenerNet meteorological observations, a digital elevation model, and land-use categories. The generated half-hourly high-resolution wind fields are also averaged to further hourly and daily weather data products as well as monthly, seasonal, and annual climate data products. These data products can be used for investigating weather and climate variability on a local scale, as well as for the evaluation of convection-permitting climate models.

We evaluated the results by identifying representative monthly periods that included both frequently occurring thermally induced weak wind speeds (August 2008) and strong wind speeds (March 2013). Thanks to the dense station network, the statistics show reasonably good results for both periods and confirm the utility of the new wind fields. The statistical performance measures applied to the vector-mean wind speed show better results for weak wind speeds than for strong wind speeds. The results related to wind direction are found to be more accurate for strong wind speeds than for weak wind speeds. The application has been running operationally in the Feldbach region since mid-2016 and the earlier data since 2007 have been reprocessed.

Ongoing future work deals with applying the developed application to automatically produce wind fields for a second study area, the WegenerNet Johnsbachtal (Fuchsberger et al. 2016; Strasser et al. 2013), located to the north of Styria in an Alpine mountainous region. This second study area is characterized through a region of complex terrain with high relief energy. The original CALMET model calculates the energy balance without considering topographic shading through relief. This is a challenge in complex terrain since such shading significantly affects the energy balance and, subsequently, the wind field. Furthermore, surface temperature fields are produced by a simple inverse distance interpolation without taking the vertical temperature gradient into account.

To improve the modeling of these physical effects, for the second study area we implemented algorithms developed by Bellasio et al. (2005) into CALMET (version 6.5.0). These algorithms take into account the topographic effects for the calculation of solar radiation as well as the terrain elevation for estimating the temperature field close to the surface; detailed information on the implemented algorithms can be found in Bellasio et al. (2005). We will use the original and the enhanced CALMET model to identify and adapt the best method of generating wind fields for the WegenerNet Johnsbachtal network.

The wind fields produced based on the enhanced CALMET version will as well be cross validated for the WegenerNet Feldbach region. As a scientific application, we will then use the empirical wind fields for the evaluation of nonhydrostatic climate model simulations in the two WegenerNet regions for selected challenging weather situations. Beyond the two regions the new application can also serve, after appropriate tuning, as a valuable tool for high-resolution wind field modeling for other networks.

Acknowledgments

The authors thank Roberto Bellasio (Enviroware), Italy, for providing the modified CALMET 5.2 code, including algorithms to account for topographic shading effects and vertical temperature gradients. We also thank Heimo Truhetz (Wegener Center, University of Graz) for valuable discussions about scientific issues and the model setup, and three anonymous reviewers for valuable comments that helped to significantly improve the paper. The CALMET 6.5.0 model code was available online (www.src.com/calpuff/). CORINE Land Cover data for the study area (www.eea.europa.eu), digital elevation model data (www.gis.steiermark.at), and WegenerNet Feldbach region data (www.wegenernet.org) were available online. WegenerNet funding is provided by the Austrian Ministry for Science and Research, the University of Graz, the state of Styria (which also included European Union regional development funds), and the city of Graz; detailed information can be found online (www.wegcenter.at/wegenernet).

REFERENCES

REFERENCES
Abdel-Aal
,
R.
,
M.
Elhadidy
, and
S.
Shaahid
,
2009
:
Modeling and forecasting the mean hourly wind speed time series using GMDH-based abductive networks
.
Renewable Energy
,
34
,
1686
1699
, doi:.
Awan
,
N. K.
,
H.
Truhetz
, and
A.
Gobiet
,
2011
:
Parameterization-induced error characteristics of MM5 and WRF operated in climate mode over the Alpine region: An ensemble-based analysis
.
J. Climate
,
24
,
3107
3123
, doi:.
Bellasio
,
R.
,
G.
Maffeis
,
J. S.
Scire
,
M. G.
Longoni
,
R.
Bianconi
, and
N.
Quaranta
,
2005
:
Algorithms to account for topographic shading effects and surface temperature dependence on terrain elevation in diagnostic meteorological models
.
Bound.-Layer Meteor.
,
114
,
595
614
, doi:.
Brunner
,
L.
,
G. C.
Hegerl
, and
A. K.
Steiner
,
2017
:
Connecting atmospheric blocking to European temperature extremes in spring
.
J. Climate
,
30
,
585
594
, doi:.
Cox
,
R. M.
,
J.
Sontowski
, and
C. M.
Dougherty
,
2005
:
An evaluation of three diagnostic wind models (CALMET, MCSCIPUF, and SWIFT) with wind data from the Dipole Pride 26 field experiments
.
Meteor. Appl.
,
12
,
329
341
, doi:.
EEA
,
2007
: CLC2006 technical guidelines. European Environment Agency Tech. Rep., 66 pp. [Available online at http://www.eea.europa.eu/publications/technical_report_2007_17.]
Fuchsberger
,
J.
,
G.
Kirchengast
,
T.
Kabas
, and
C.
Bichler
,
2016
: Wegenernet climate station networks: Overview and examples. LTER-Austria Conf. 2016, Vienna, Austria, Austrian Long-Term Ecosystem Research Network. [Available online at https://wegenernet.org/downloads/poster_LTER_2016_wegnet_v2.pdf.]
Haylock
,
M. R.
,
N.
Hofstra
,
A. M. G.
Klein Tank
,
E. J.
Klok
,
P. D.
Jones
, and
M.
New
,
2008
:
A European daily high-resolution gridded data set of surface temperature and precipitation for 1950–2006
.
J. Geophys. Res
,
113
,
D20119
, doi:.
Hohenegger
,
C.
,
P.
Brockhaus
, and
C.
Schär
,
2008
:
Towards climate simulations at cloud-resolving scales
.
Meteor. Z.
,
17
,
383
394
, doi:.
Jancewicz
,
K.
,
2014
:
Remote sensing data in wind velocity field modelling: A case study from the Sudetes (SW Poland)
.
Pure Appl. Geophys.
,
171
,
941
964
, doi:.
Kabas
,
T.
,
2012
: WegenerNet climate station network region Feldbach: Experimental setup and high resolution data for weather and climate research (in German). Wegener Center Scientific Rep. 47-2012, 177 pp. [Available online at http://wegcwww.uni-graz.at/publ/wegcreports/2012/WCV-WissBer-No47-TKabas-Jan2012.pdf.]
Kabas
,
T.
,
U.
Foelsche
, and
G.
Kirchengast
,
2011
:
Seasonal and annual trends of temperature and precipitation within 1951/1971-2007 in south-eastern Styria, Austria
.
Meteor. Z.
,
20
,
277
289
, doi:.
Kirchengast
,
G.
,
T.
Kabas
,
A.
Leuprecht
,
C.
Bichler
, and
H.
Truhetz
,
2014
:
Wegenernet: A pioneering high-resolution network for monitoring weather and climate
.
Bull. Amer. Meteor. Soc.
,
95
,
227
242
, doi:.
Klok
,
E. J.
, and
A. M. G.
Klein Tank
,
2009
:
Updated and extended European dataset of daily climate observations
.
Int. J. Climatol.
,
29
,
1182
1191
, doi:.
Lugauer
,
M.
, and
P.
Winkler
,
2005
:
Thermal circulation in south Bavaria climatology and synoptic aspects
.
Meteor. Z.
,
14
,
15
30
, doi:.
Morales
,
L.
,
F.
Lang
, and
C.
Mattar
,
2012
:
Mesoscale wind speed simulation using CALMET model and reanalysis information: An application to wind potential
.
Renewable Energy
,
48
,
57
71
, doi:.
O
,
S.
,
U.
Foelsche
,
G.
Kirchengast
, and
J.
Fuchsberger
,
2017
:
Validation and correction of rainfall data from the WegenerNet high density network in southeast Austria
.
J. Hydrol.
, doi:, in press.
Oleniacz
,
R.
, and
M.
Rzeszutek
,
2014
:
Determination of optimal spatial databases for the area of Poland to the calculation of air pollutant dispersion using the CALMET/CALPUFF model
.
Geomatics Environ. Eng.
,
8
(
2
),
57
69
, doi:.
Osborn
,
T. J.
, and
M.
Hulme
,
1998
:
Evaluation of the European daily precipitation characteristics from the Atmospheric Model Intercomparison Project
.
Int. J. Climatol.
,
18
,
505
522
, doi:.
Prein
,
A. F.
,
2013
: Added value of convection resolving climate simulations. Wegener Center Scientific Rep. 53-2013, 164 pp. [Available online at http://wegcwww.uni-graz.at/publ/wegcreports/2013/WCV-SciRep-No53-APrein-Jul2013.pdf.]
Prein
,
A. F.
,
G. J.
Holland
,
R. M.
Rasmussen
,
J.
Done
,
K.
Ikeda
,
M. P.
Clark
, and
C. H.
Liu
,
2013
:
Importance of regional climate model grid spacing for the simulation of heavy precipitation in the Colorado headwaters
.
J. Climate
,
26
,
4848
4857
, doi:.
Prettenthaler
,
F.
,
A.
Podesser
, and
H.
Pilger
,
2010
: Climate Atlas Styria, Period 1971–2000: An Application-Oriented Climatology (in German). Vol. 4. Oesterreichischen Akademie der Wissenschaften, 358 pp.
Scire
,
J. S.
,
F. R.
Robe
,
M. E.
Fernau
, and
R. J.
Yamartino
,
1998
: A user’s guide for the CALMET meteorological model (version 5). Earth Tech, Inc., 332 pp. [Available online at http://www.src.com/calpuff/download/CALMET_UsersGuide.pdf.]
Sfetsos
,
A.
,
2002
:
A novel approach for the forecasting of mean hourly wind speed time series
.
Renewable Energy
,
27
,
163
174
, doi:.
Strasser
,
U.
,
T.
Marke
,
O.
Sass
,
S.
Birk
, and
G.
Winkler
,
2013
:
John’s creek valley: A mountainous catchment for long-term interdisciplinary human-environment system research in upper Styria (Austria)
.
Environ. Earth Sci.
,
69
,
695
705
, doi:.
Suklitsch
,
M.
,
A.
Gobiet
,
H.
Truhetz
,
N. K.
Awan
,
H.
Göttel
, and
D.
Jacob
,
2011
:
Error characteristics of high resolution regional climate models over the Alpine area
.
Climate Dyn.
,
37
,
377
390
, doi:.
Truhetz
,
H.
,
2010
: High resolution wind field modelling over complex topography: Analysis and future scenarios. Wegener Center Scientific Rep. 32-2010, 170 pp. [Available online at http://wegcwww.uni-graz.at/publ/wegcreports/2010/WCV-SciRep-No32-HTruhetz-Apr2010.pdf.]
Truhetz
,
H.
,
A.
Gobiet
, and
G.
Kirchengast
,
2007
:
Evaluation of a dynamic-diagnostic modelling approach to generate highly resolved wind fields in the alpine region
.
Meteor. Z.
,
16
,
191
201
, doi:.
Wakonigg
,
H.
,
1978
: Weather and Climate in Styria (in German). Technische Universitaet Graz, 473 pp.
Wang
,
W.
,
W. J.
Shaw
,
T. E.
Seiple
,
J. P.
Rishel
, and
Y.
Xie
,
2008
:
An evaluation of a diagnostic wind model (CALMET)
.
J. Appl. Meteor. Climatol.
,
47
,
1739
1756
, doi:.
Willmott
,
C. J.
,
S. M.
Robeson
, and
K.
Matsuura
,
2012
:
A refined index of model performance
.
Int. J. Climatol.
,
32
,
2088
2094
, doi:.

Footnotes

© 2017 American Meteorological Society. For information regarding reuse of this content and general copyright information, consult the AMS Copyright Policy (www.ametsoc.org/PUBSReuseLicenses).