SEARCH

SEARCH BY CITATION

Keywords:

  • wind fields;
  • aircraft wind measurements;
  • mesoscale models;
  • WRF;
  • CALMET;
  • SkyArrow ERA

ABSTRACT

  1. Top of page
  2. ABSTRACT
  3. 1. Introduction
  4. 2. Materials and methods
  5. 3. Results and discussion
  6. 4. Conclusions
  7. Acknowledgements
  8. References

Long term aircraft observations of wind magnitude along an ∼250 km flight track in central Italy, performed over 1.5 years, are compared with the output of an existing mesoscale prognostic-diagnostic (WRF-CALMET) model chain aimed at assessing wind potential maps at regional scale. Aircraft measurements are used to evaluate model performance along spatial and temporal transects at moderate altitude from the ground (∼75 m), where observational frameworks are rarely available. Spatial wind analysis was capable of assessing overall model performance, while highlighting some limitations: the implemented models have better performance in inland areas with respect to coastal areas, while they are capable of representing diurnal variability in all regions correctly. Overall agreement is within 3% in the cold season and 16% in the warm season, while the greatest differences, above 30%, are obtained in coastal areas in the summer. The hypothesis supporting these results is that summer sea breeze regimes that develop consistently from the coast through the interior land are not entirely resolved from mesoscale modelling. Finally, the model performance and limitations related to complex orography are highlighted. This study demonstrates the added value that may derive from aircraft wind measurements as an additional observational framework for applied meteorology studies.

1. Introduction

  1. Top of page
  2. ABSTRACT
  3. 1. Introduction
  4. 2. Materials and methods
  5. 3. Results and discussion
  6. 4. Conclusions
  7. Acknowledgements
  8. References

Measurement and knowledge of wind spatial and temporal characteristics at the regional scale is a relevant issue for several applications in meteorology. Wind is an important variable per se in applications such as wind energy potential assessment, atmospheric transport, air pollutant dispersion and civil engineering. It is also a fundamental driver of the surface energy balance by modulating the aerodynamic resistance of the Earth's surface, therefore directly controlling energy and momentum exchange between the surface and the atmosphere (Zhang and Lemeur, 1992). Wind close to the Earth's surface is the result of a complex interaction between mesoscale circulations and the local nature of the surface terrain (Petersen et al., 1998). Changes in land properties (Wieringa, 1992; Farrugia, 2003; Bañuelos-Ruedas et al., 2010) and topography (Ayotte, 2008; Mason et al., 2010) can drive wind patterns substantially.

Wind observations are typically provided by weather stations. However, such wind measurements are affected by some constraints and limitations (Al-Yahyai et al., 2010): (1) they are costly, as typically complete meteorological stations including additional sensors are installed; (2) they have a coarse resolution, resulting in sparse networks not capable of accurately assessing wind spatial variability that is needed to derive wind maps for large regions; (3) stations are installed at specific locations such as airports, ports and areas with high population density, or where relevant phenomena need to be monitored, rather than, e.g., in high elevated remote areas which are relevant for wind energy assessment; (4) measurements are made at standard levels only (typically, 10 or 2 m), not providing information at upper levels (above 50 m). Overall, several applications such as wind energy exploitation and infrastructure construction, need a detailed knowledge of wind fields both in space and time, and at relatively high levels from the ground. For such applications, tall towers are typically installed at precise candidate locations, providing point-level information. On the other hand, Numerical Weather Prediction (NWP) models are capable of overcoming part of these constraints. These prognostic models solve the dynamic primitive equations describing the atmospheric processes numerically, consisting of simplified models of the actual physical processes of the atmosphere (Sarrat et al., 2009). Recently, many meteorological departments have started running limited area models (LAMs, i.e., where primitive equations are solved only over a limited domain), to cover the domain of a single region and the surrounding areas. Different mesoscale LAMs are available for research and operational use, such as RAMS (Pielke et al., 1992), ETA (Black, 1994), MM5 (Grell et al., 1994), WRF (Skamarock et al., 2005), COSMO (Doms and Schattler, 2008), and HRM (Majewski, 2009). NWP models are generally capable of addressing: (1) cost issues, as many of them are based on freely available open source code; (2) resolution issues, as NWP models can be run in high resolution also due to steadily increasing computational power; (3) spatial coverage, since NWP models provide gridded data for the whole model domain, both horizontally and vertically. NWP models provide flexibility to simulate relatively long periods (such as one or more years) in relatively short times, with no data gap. However, they have limitations due to simplification in physics, and uncertainty in the initial state, lateral boundary conditions and surface characteristics (Al–Yahyai et al., 2010). While NWP models are operationally applied for weather forecasts, diagnostic wind models based on mass conservation still play an important role because of their fast computation and high accuracy in local areas (Wang et al., 2008), where they are capable of considering fine scale details such as, e.g., complex topography and land–water interfaces (Hu et al., 2010). Examples of diagnostic models are AERMET (US EPA, 2004), MCSCIPUF (Sykes et al., 1998), and CALMET (Scire et al., 1999). Moreover, diagnostic models may be run in combination with prognostic mesoscale models (Bellasio et al., 2005). Recently, several wind assessment studies were carried out using NWP-alone or NWP-diagnostic coupled models, typically implemented on two or more nested domains (Beaucage et al., 2012). Hiroyuki et al. (2006) run the RAMS model through 8 and 2 km resolution simulations to investigate wind energy potential over the area of Tokyo (Japan), and found good agreement with observations, with a 4.8% prediction error on annual mean wind speed. Rather good scores were also obtained by Shimada et al. (2009) in reproducing wind data for offshore wind resource assessment over Japan after applying MM5 and WRF models with 4.5 and 1.5 km resolutions. Guilherme et al. (2009) ran the WRF model to derive wind data for Portugal using 6 and 3 km resolution, obtaining model simulation winds slightly weaker (about 5%) than the measured data. Bellasio et al. (2005) used a coupled WRF-CALMET framework obtaining good representation of actual wind patterns, even in complex topography. Mari et al. (2011) used a similar model chain to create maps of wind speed to assess the large-scale wind resource potential of the Tuscany region (central Italy). The main limitation of using prognostic-diagnostic models to derive wind fields is often the lack of validation observations at high elevation, as well as at a sufficiently large number of points that allow the assessment of spatial patterns. Therefore, their accuracy and actual capability to reproduce complex dynamic phenomena still need to be assessed against adequate observational frameworks.

The main goal of the present work is to overpass such a limitation by introducing aircraft measurements to assess model performance. In fact, besides observations from anemometers installed on weather stations, winds can also be measured by aircraft. In recent years, a number of platforms have been deployed for the measurement of turbulence in the atmosphere, ranging from large research aircraft (Black et al., 2007), to small platforms such as the SkyArrow ERA (Environmental Research Aircraft) (Gioli et al., 2004), to un-manned aerial vehicles (Martin et al., 2011). Such platforms use pressure spheres to measure instantaneous angles of attack, and then retrieve actual wind components by means of calibrated motion measurements and upwash modelling (Vellinga et al., 2013). Aircraft measurements are per se sporadic, referring to specific experiments in specific conditions, thus are rarely capable of providing a comprehensive picture of wind fields across a region and across different time scales. This work is based on a previously published WRF-CALMET wind dataset calculated at 75 m altitude above the ground (Mari et al., 2011), that is assessed against an intensive regional observational dataset of aircraft wind measurements. The latter span a study area of about 100 × 120 km in Tuscany (Italy), from the coast to the inland areas, across different land use categories ranging from extensive forest to agriculture crops. Aircraft flights were scheduled as Intensive Operations Periods (IOP) in different seasons of the year, and different times of the day, aiming at maximizing the sampling of both spatial and temporal variability. Model runs were instead performed continuously for a 2 year period including all the flights. Overall, this study aims at using this innovative level of observations assessing the performance of a coupled model across spatial and temporal transects, at moderate altitude from the ground that is relevant for meteorological applications and where measurements are rarely deployed. Finally, the characteristics and the benefit of aircraft wind measurements are highlighted.

2. Materials and methods

  1. Top of page
  2. ABSTRACT
  3. 1. Introduction
  4. 2. Materials and methods
  5. 3. Results and discussion
  6. 4. Conclusions
  7. Acknowledgements
  8. References

2.1. The WRF-NMM prognostic model

WRF-NMM is a prognostic, non-hydrostatic model consisting of several modules to ingest observational data and simulate atmospheric conditions, describing the dynamics and thermodynamics of the atmospheric flow in limited areas. The numerical integrator consists of a fully compressible, Eulerian and non-hydrostatic equation set, employing a semi-staggered rotated horizontal Arakawa E grid (Janjic and Mesinger, 1984, 1989) and a terrain-following hybrid sigma-pressure vertical co-ordinate (Arakawa and Lamb, 1977). In this work WRF-NMM (v. 2.1) is implemented on a computational domain based on a medium resolution grid (0.11°, ∼10 km) covering central Europe (128 × 232 points), as shown in Figure 1. Relevant physics schemes include: Ferrier microphysics (Ferrier et al., 2002), Geophysical Fluid Dynamics Laboratory (GFDL) longwave and shortwave radiation, Noah land surface model (Ek et al., 2003), Janjic similarity surface layer and Mellor–Yamada–Janjic turbulence kinetic energy boundary layer (Janjic, 1990, 1996, 2002), Kain–Fritsch cumulus and convective parameterization (Kain, 2004). Sea Surface Temperature (SST) data are obtained from NOAA SST fields. Land use classes and soil categories are provided by the standard USGS categories (24 for land use and 16 for soil). Topography is derived from the global 30 s USGS topography data with a four-point average. Initial and boundary conditions are given by the NCEP-GFS (National Centre for Environmental Prediction-Global Forecasting System) global circulation model (T382L64, about 0.5° resolution) four times a day, at 0000, 0600, 1200 and 1800 UTC for +144 h of forecast. Boundary conditions are updated with GFS forecasts every 6 h. The GFS time step is 7.5 min for computation of dynamics and physics, while WRF-NMM time step is 30 s. The study period covers four entire years (2004–2007). The vertical configuration is made up of 35 pressure levels with the lower 10 levels mostly concentrated in the boundary layer (1000, 993, 986, 978, 970, 960, 950, 940, 930, 920 hPa). For the current application, WRF-NMM outputs have been post-processed to a regular latitude–longitude grid and vertical levels interpolated to 12 constant levels above ground (10, 75, 185, 350, 515, 705, 910, 1190, 1500, 1850, 2300, 2850 m).

image

Figure 1. The computational domains of WRF-NMM (10 km spatial resolution, contour map) and CALMET (2 km spatial resolution, black rectangle). Contours represent terrain elevation. This figure is available in colour online at wileyonlinelibrary.com/journal/met

Download figure to PowerPoint

2.2. The CALMET diagnostic model

CALMET is a meteorological model including a diagnostic wind field module, that generates the wind fields used in this study, and two micrometeorological modules for overwater and overland boundary layers (Scire et al., 1999). A multi-step approach is used for the computation of wind fields. In the current configuration, an initial-guess wind field is obtained from the gridded wind fields generated by the WRF model at 10 km horizontal resolution over a 25 × 21 grid covering the whole Tuscany region (Figure 1). The wind field is then used to compute the terrain-forced vertical velocity and the kinematic effects of terrain on horizontal wind components, by applying an iterative divergence-minimization scheme (Liu and Yocke, 1980). Slope flows are computed based on the Mahrt (1982) parameterization in terms of terrain slope, distance to the crest, and local sensible heat flux. The thermodynamic blocking effects of terrain on wind flow are parameterized in terms of the local Froude number (Allwine and Whiteman, 1985). The micrometeorological module over land computes the surface energy balance (Holtslag and van Ulden, 1983), surface friction velocity, Monin–Obukhov length, and convective velocity scale. Over water, a profile technique, based on air–sea temperature differences, is used to compute the micrometeorological parameters in the marine boundary layers (Scire et al., 1999). The final three dimensional wind field at 2 km horizontal resolution is obtained by interpolation based on the inverse-distance method, obtaining a 120 × 107 horizontal grid, while the same WRF 12 vertical levels (from 10 to 2850 m above ground) are used, thus not requiring vertical interpolation. CALMET was run as WRF with a 1 h time step through a 4 year period (2004–2007).

2.3. Study area and flight tracks

The study region object of this study is a sub-portion of the CALMET computational domain represented in Figure 1, located in Tuscany, and extending from the coastline until about 100 km inland, including all the areas sampled with aircraft measurements (Figure 2). The climate of this region is typically Mediterranean, with mild winters and hot and dry long summers. The mean annual rainfall is between 700 and 1000 mm, while average annual temperature is around 14–16 °C. A marked dry season characterizes all this area, and its length mainly depends on the distance from the sea and the altitude. Aircraft flight operations were originally designed in the frame of an extensive campaign, aimed at measuring regional carbon fluxes for the CARBIUS project (Maselli et al., 2010). Tracks were designed along paths that are representative of the region (Figure 2), in terms of land use, orography, and meteorological conditions.

image

Figure 2. Flight tracks overimposed to (a) orography and (b) land cover. Numbers indicate along-track distance in kilometres from the south limit. Track is divided into the three sections S1, S2, S3. This figure is available in colour online at wileyonlinelibrary.com/journal/met

Download figure to PowerPoint

Flights were made across the entire track or across sub-portions of it. A unique geographical reference system is adopted for all the flights, by defining an along-track spatial co-ordinate, which originates at the most southerly point of the track, and terminates at the most northerly and more inland point (Figure 2). In this way, all aircraft observations may be easily located along the resulting track, about 240 km long.

The area is dominated by two main broad land cover types, according to the Corine Land Cover 2006 classification (ISPRA, 2010), i.e., forest and agricultural areas, since urban areas were specifically avoided for flights. The terrain is relatively flat on the coast portion of the track, then the orography is more variable moving from southwest through northeast inner area, reaching an altitude of about 600 m ASL, while it is moderately hilly on the final inland part. The land cover is characterized by the alternation of agricultural fields, pastures and wood-lands. Forests prevail on the upper hilly areas (Figure 2(b)).

The experimental plan was based on performing Intensive Observations Periods (IOPs) in different seasons of the year, over a repetitive path, and in different times of the day with the aim of encompassing both spatial and temporal variability. Flights were performed from July 2004 to December 2005, during 16 IOPs, each consisting of 4–12 flights covering daily variability in illumination conditions and wind regimes (Table 1). The study period is divided into two broad classes, representing spring to late summer season (P2, from April to September) and a cold period encompassing the rest of the year (P1, from October to March).

Table 1. Classification of the study period into intensive operation periods (IOP)
IOPDateTimeNPeriodTaPARRH
  1. For each IOP, some basic informations are reported: starting and ending dates (Date), starting and ending standard time(LST) (Time), number of flights (N), period P1 or P2 (Period), average air temperature in °C (Ta), average incoming PAR radiation (PAR) in µmol m−2 s−1, average relative humidity (RH) in percentage. Variabilities represent standard deviations.

121 July to 22 July 20040900–18098P227.9 ± 2.11392 ± 51442.1 ± 14.8
29 August to 11 August 20040841–154610P225.4 ± 2.11306 ± 28957.4 ± 12.3
324 August to 25 August 20040925–16146P224.4 ± 1.71127 ± 40458.5 ± 15.4
416 November to 18 November 20040910–16178P111.9 ± 2.3594 ± 27356.2 ± 15.0
57 December to 8 December 20041000–16204P113.5 ± 1.8280 ± 23871.5 ± 7.0
621 December to 23 December 20040933–16228P14.4 ± 2.7443 ± 22258.4 ± 8.6
712 January to 14 January 20051014–15368P17.9 ± 1.5259 ± 21184.1 ± 10.2
816 March to 18 March 20050932–161410P111.3 ± 7.11160 ± 22152.6 ± 14.8
95 April to 14 April 20050840–16118P212.6 ± 2.21063 ± 27458.4 ± 17.7
103 May to 10 May 20050826–172112P216.8 ± 2.11159 ± 43560.7 ± 18.5
1112 July to 13 July 20050841–18006P222.4 ± 2.01187 ± 52451.5 ± 7.1
1227 July to 29 July 20050849–18449P228.6 ± 2.61324 ± 47845.0 ± 14.7
1331 August to 2 September 20050842–181810P224.2 ± 2.41121 ± 38956.2 ± 10.0
1421 September to 23 September 20050859–162310P216.8 ± 2.5853 ± 36765.7 ± 7.8
1524 October to 26 October 20051129–15374P118.7 ± 1.3801 ± 32581.1 ± 10.9
1614 December to 16 December 20051008–16348P16.4 ± 2.3325 ± 23768.2 ± 11.5

The flight track is divided for the analysis presented here into three continuous and contiguous main sections: (1) S1, characterized by being close to the coast-line, with flat orography and mostly agricultural and grassland land use; two exceptions are represented by some small mountains in the vicinity of the track, at 37 and 92 km, respectively; (2) S2, a transect going from the coast to the interior areas, characterized by a mostly forest land cover, and more complex orography; (3) S3, an inland transect, with mixed agriculture and forest land covers and gentle hilly orography (Figure 2 and Table 2). Such spatial classification is made consistently with the flight tracks to isolate three self-contained segments that present overall different characteristics at the landscape scale (Table 2), and are subject to different meteorological patterns and to a different influence of the sea. At the same time, the two sub-periods (P1 and P2) are identified as an optimum compromise to assess separately time periods in which different wind regimes may develop, especially sea breeze regimes on the coast, while maintaining a sufficient number of flights to derive reliable statistics.

Table 2. Land use and orographic characteristics of the three track segments S1, S2, S3
SectionsAgriculture (%)Forest (%)Terrain height (m)
  1. Percentages of agricultural and forest land cover classes, and average terrain height with associated standard deviation.

S1722482.3 ± 104.8
S2793332.3 ± 172.2
S36135281.0 ± 71.4

2.4. Aircraft platform

Aircraft measurements were made with the Sky Arrow ERA, a small certified platform equipped with sensors to measure three-dimensional wind and turbulence together with gas concentrations and other atmospheric parameters at high frequency (Gioli et al., 2006). Wind measurements are accomplished by the so-called Mobile Flux Platform (MFP). In brief, the velocity of air with respect to aircraft is measured using a hemispheric nine-hole pressure sphere (Crawford and Dobosy, 1992). The actual wind components (horizontal U, V and vertical W) relative to the Earth are calculated afterwards by removing aircraft motion, that is sampled through multi-antenna GPS coupled to accelerometers, measuring three-dimensional velocity, pitch, roll and heading of the aircraft. The probe is equipped with a fast-response thermocouple that measures air temperature with a response time of 0.02 s. Additional measurements include carbon dioxide and water vapour densities measured with an open path infrared gas analyser (Licor 7500, LiCor, Lincoln, Nebraska), incoming and reflected Photosynthetically Active Radiation (PAR) measured with quantum sensors (Li190, LiCor, Lincoln, Nebraska). A recent detailed description of the MFP concept and its calibration for precise wind measurements is provided in Vellinga et al. (2013).

2.5. Aircraft and model data comparison

Aircraft and WRF-CALMET model data originate at different spatial, both horizontal and vertical, and temporal resolutions. Aircraft measurements are designed to resolve turbulent eddies, thus are high frequency measurements obtained at 50 Hz. This corresponds, at the average aircraft speed of 40 m s−1, to 80 cm spatial horizontal resolution. On the other hand, WRF-CALMET outputs are natively at 2 km and 1 h spatial horizontal and temporal resolution, respectively. Aircraft data have been spatially averaged into 2 km length segments, each segment having a space (the segment centre) and a time (the average time of the segment data) stamp. Model data have then been linearly interpolated in horizontal space and in time at such stamps, in order to derive a model output associated to each aircraft segment. Vertically, the model output layer normally placed at 75 m above ground level (AGL), is directly compared with aircraft data whose altitude ranged from 63 to 120 m during the entire campaign, depending on orography and safety constraints. Flight segments with an average altitude higher than 100 m AGL have been therefore discarded. The complete aircraft dataset consists of 9048 segments; some are excluded because of the altitude limit and because some aircraft flights extended outside the Tuscany region and model domain. Overall, 7308 data constitute the coupled dataset of observations and model outputs that are analysed here (Table 3).

Table 3. Wind magnitude statistics calculated for the WRF-CALMET model and aircraft data and relative percentage difference
 Sections
PeriodsS1S2S3All sections
  1. Values are reported for two study periods P1 and P2, and three track segments S1, S2, S3 (Figure 2). Bold: aircraft measurements (obs); normal: WRF−CALMET estimations (mod); brackets: percentage error (PE). Variabilities represent 95% confidence intervals of the averages. Percentage errors (PE) are computed as n= 100 (mod − obs)/obs.

Data sample    
P178810947442626
P21176211413924682
All periods1964320821367308
Mean wind speed (m s−1)    
P14.71 ± 0.204.35 ± 0.143.43 ± 0.164.20 ± 0.10
 3.75 ± 0.184.43 ± 0.183.96 ± 0.224.09 ± 0.12
 (−20.4%)(+1.8%)(+15.5%)(−2.6%)
P25.20 ± 0.144.23 ± 0.083.28 ± 0.104.19 ± 0.06
 3.42 ± 0.083.06 ± 0.083.12 ± 0.123.17 ± 0.04
 (−34.2%)(−27.7%)(−4.2%)(−24.3%)
All periods5.00 ± 0.124.27 ± 0.063.34 ± 0.084.19 ± 0.06
 3.56 ± 0.083.52 ± 0.083.41 ± 0.103.50 ± 0.06
 (−30.8%)(−17.6%)(+2.0%)(−16.5%)

2.6. Tall tower wind measurements at Cabauw (NL)

A measurement flight was performed on 2 February 2002 with the SkyArrow ERA in the vicinity (within 300 m horizontal separation) of the Cabauw meteorological mast, a 220 m tall tower located in The Netherlands (51° 58′ 12″ N, 4° 55′ 34″ E). Vertical profiles of wind speed and direction were measured at Cabauw at 10, 20, 40, 80, 140 and 200 m levels, with 15 min temporal resolution. Aircraft descending and ascending profiles were performed as large spirals, down to 47 m above ground and compared with tower data as 1 Hz averages.

3. Results and discussion

  1. Top of page
  2. ABSTRACT
  3. 1. Introduction
  4. 2. Materials and methods
  5. 3. Results and discussion
  6. 4. Conclusions
  7. Acknowledgements
  8. References

3.1. Tall tower wind measurements at Cabauw (NL)

The comparison with the Cabauw mast profile measurements is reported for validation of the aircraft's wind measurement capability. Wind speed vertical variability is correctly reproduced within 5%, over a span from 12.2 to 16.6 m s−1 (Figure 3(a)). Wind direction vertical pattern is also correctly reproduced, with a slight clockwise direction turn from ∼190 to ∼200° within the 50–200 m altitude range (Figure 3(b)). Aircraft wind data observed variability is larger than tower; this is likely related to the much smaller averaging time of aircraft observations (1 s) with respect to tower observations (15 min), needed to resolve vertical variability along aircraft profiles adequately.

image

Figure 3. Vertical profiles of wind speed (a) and direction (b) measured by aircraft (empty circles) and at the Cabauw tower (diamonds).

Download figure to PowerPoint

3.2. Overall wind fields comparison

Central Italy data presentation is organized starting from aggregated overall comparisons, to more spatially and temporally resolved analysis. Examples from single flights that are representative of interesting behaviours are finally reported. Flight operations were designed to cover both spatial and temporal variability, through flights performed across the same pre-defined tracks at various times of the day and periods of the year (Table 1). Figure 4 reports temporal distribution of individual measurements across the course of the day and across different seasons of the year. Overall, hourly distribution exhibits two peaks at 1000 h and 1500 LST (= UTC + 1). Those are mostly related to the logistic organization of flights: since typically two flights per day were made, a gap between first and second flight is present at mid-day hours. Nevertheless, effort was put into sampling in all possible conditions, and all hours from 0800 h to 1900 h were sampled. During the course of the 1.5 years study period, all seasons of the year were sampled with about the same intensity, with a minimum in winter related to the lower amount of daytime that was available for flights (Figure 4(b)). Overall, this dataset cannot be strictly considered an unbiased temporal average, since it is likely biased towards some hours of the day (Figure 4(a)) and towards fair weather with respect to unfavourable weather conditions. Nevertheless, it constitutes a rare example of long-term aircraft observations of micro-meteorological variables in day time, and allows the comparison of model and data in a large variety of different locations and meteorological conditions.

image

Figure 4. Standard time hourly temporal distribution (a) and seasonal distribution (b) of aircraft observation. Left axis indicate absolute number of samples; right axis indicate percentage amounts.

Download figure to PowerPoint

Mean observed wind is 4.19 ± 0.06 m s−1 (95% confidence interval of the mean), while modelled wind is lower at 3.50 ± 0.06 m s−1 (Table 3). The frequency distribution of such overall results (Figure 5(a)) reveals that the aircraft reports more frequent high winds, especially in the range 3–10 m s−1, while model reports more frequent low winds in the range 1–3 m s−1. Wind direction distribution reveals a peak in northeast (45°) direction both in data and model, then a large amount of southerly to westerly winds with a peak at south (180°) for aircraft data, and at west (270°) for model data (Figure 5(b)).

image

Figure 5. Frequency histograms of aircraft (grey bars) and WRF-CALMET (white bars) wind speed (a) and wind direction (b).

Download figure to PowerPoint

3.3. Sub-sections and sub-periods wind fields comparison

To investigate whether these differences change both spatially and temporally, data are partitioned in the different sections (S1, S2, S3) of the study area (Figure 2), and different sub-periods P1 spanning fall and winter, and P2 spanning spring and summer. In cold season (P1), modelled wind magnitude (4.09 ± 0.12 m s−1) is overall underestimated by only 2.6% with respect observations (4.20 ± 0.10 m s−1), which is well within the respective confidence intervals (Table 3). However, such agreement is not spatially consistent: in coastal section S1 modelled wind is negatively biased by a factor of 20%, in section S2 it is in very good agreement within 2%, while in section S3 it is overestimated by 15% (Table 3). In the warm season (P2), modelled winds are systematically lower than observed winds, with an average bias of 24% (Table 3). The largest differences, above 30%, are computed in coastal areas S1, then in forested areas S2, and finally in inland areas S3. To investigate whether the observed bias is related to specific temporal patterns, the daily course of wind speed across the three sub-tracks (S1, S2, S3) and seasons (P1, P2) was computed, and reported (Figure 6). The associated r2 of the scatter plots reported in Figure 6 are computed between 0.46 P2-S3 and 0.98 for P1-S3. The combination P2-S1 has the highest absolute bias (Table 3), initially small in the morning and then larger in the central part of the day. Nevertheless, temporal variability is well reproduced by the model in all conditions.

image

Figure 6. Standard time hourly averages of aircraft (solid line) and WRF-CALMET (dashed line) wind speed, computed for any combination of the two study periods (P1 and P2) and the three track segments (S1, S2 and S3). Error bars represent 95% confidence intervals of the averages. The corresponding r2 is reported at lower-right corner of each panel.

Download figure to PowerPoint

Overall, these results highlight that the WRF-CALMET model is capable of describing the temporal dynamic of wind regimes during the course of the day, while the magnitude of the computed wind has a larger negative bias (underestimation) in the coast than in the inland areas, in the central part of the day and in the warm rather than cold season. The working hypothesis is that this behaviour is related to sea breeze regimes and associated circulations. In summer, the sea breeze develops consistently in the morning in coastal areas, and penetrates inland interacting with the mesoscale circulation and with orography. This is a local phenomenon that interacts with and affects regional scale circulations. It is likely that the native horizontal and vertical resolutions of WRF are insufficient to simulate the sea breeze development completely. To investigate this hypothesis further, spatial wind transects along the whole flight track encompassing the three sub-regions and the two sub-periods, are compared. By averaging all passages over a single pixel, short term temporal variability is ruled out focusing instead on spatial variability of wind magnitude. Spatial correlation (rs) and percentage errors are computed as statistical indicators. Figure 7 reports wind spatial transects over imposed terrain orography, along the entire flight track. In period P1 (Figure 7(a)), wind measurements are very well reproduced in S2 (spatial correlation rs = 0.68, percentage error PE = 1.5%), and overestimated (PE = +15%) in S3 (Table 3) maintaining a high spatial correlation value of 0.58 (Figure 7(a)). Section S1 exhibits a larger variability in observations, that is also driven by orography with a negative forcing, i.e., local small orographic structures act as obstacles reducing wind speed while larger flat coastal areas exhibit higher winds, likely related to moderate sea breeze developments (Figure 7(a)). The model overall did not entirely reproduce such spatial features (rs = 0.42), and underestimates wind by 20% (Table 3). A clear influence of orography on wind speed is observed in sections S2, when observations of wind magnitude are positively correlated with terrain height, and the same pattern is reproduced by the model. Such an outcome is also in agreement with Prieto et al. (2007), who applied a MM5 (3 km resolution) nested scheme to feed two CALMET nested domains (1 km and 100 m) over complex terrain in northern Spain, and found that the coupled system accounted for orography effects better than the mesoscale model only.

image

Figure 7. Spatial transects of aircraft (black circles) and model (white circles) wind magnitude (left axis). Black line (right axis) indicates terrain height at individual points on the flight path while grey area (right axis) indicates 5 km2 average terrain height centred on the flight points. Aircraft wind is obtained at 63–100 m above ground, model wind at 75 m above ground. Top panel (a) refers to period P1 (October to March), bottom panel (b) to period P2 (April to September).

Download figure to PowerPoint

In period P2 (Figure 7(b), spatial correlations are computed at 0.16, −0.22 and 0.57 for sections S1, S2 and S3, respectively. In section S1, where wind is severely underestimated (PE = −51%) as previously reported (Table 3), the bias results not to be associated to specific spatial features (Figure 7(b)). In section S2, observed and modelled spatial patterns are negatively correlated, where the observations report a decreasing magnitude proceeding from the coast to the inland, while model values are largely driven by local orography (Figure 7(b)). In section S3, the agreement between the two is really good both in terms of spatial correlation and absolute values (Figure 7(b) and Table 3).

These results support the hypothesis that sea breeze regime simulation is the main limiting factor of the WRF-CALMET model chain in describing wind spatial patterns. In the inland areas (section S3), not likely affected by any sea breeze influence, the agreement is excellent (within 5%) in either season and spatial signals are highly correlated (r = 0.58 and 0.57 in P1 and P2). The S2 area is likely not affected as well by sea breeze development in non-summer (period P1), while it experiences a transition from sea breeze circulations to mesoscale circulations in summer time. This acts as a descending trend of wind magnitude reflecting progressive penetration and attenuation of sea breeze regimes proceeding from the coast inland. This trend variability masks completely the effect of orography that is instead observed in other seasons. Models not reflecting sea breeze penetration show a clear positive correlation with ground elevation in all seasons (Figure 7(b)). On the coastal area, where sea breeze development is more pronounced, the bias is maximum, with the model underestimating magnitudes by 34%. These results are basically consistent with Bellasio et al. (2005), who applied CALMET over a 4 km resolution domain in northern Italy after ingestion of a number of surface stations and one tall tower station, and proved the model was not capable of resolving microclimate effects such as lake breezes. It has been well recognized in the literature that CALMET is capable of estimating horizontal wind fields under significant spatial and temporal variability, when ingested observations are enough to resolve characteristic local flows (Wang et al., 2008). Yim et al. (2007) pointed out that CALMET capability in reproducing local wind fields strictly depends on density, frequency and accuracy of the observations used as input. Similar recommendations were provided by Morales et al. (2012), who run CALMET at 1 km resolution over southern Chile using as input NCEP reanalysis data and four surface stations. Gilliam et al. (2005) applying CALMET over a 50 × 50 km domain in New York City with a 0.5 km resolution, found that it is unable to adequately develop sea breeze circulations over complex terrain, even when ingested by observations from an extensive network; these effects are instead properly reproduced by MM5, applied through a triple nesting with the innermost domain having a 1 km resolution. Hence, use of gridded fields from a full-physics mesoscale model rather than observations is recommended to improve CALMET fields in the sea breeze simulation (Surapipith et al., 2011). While the modelling setup adopted in this study meets such recommendation, the 10 km horizontal WRF resolution, that is necessarily a compromise between accuracy and computational cost for a regional scale application such as this, is likely not adequate to resolve completely local effects related to sea breeze development at grid points across the coastline.

3.4. Single flights wind spatial patterns

Finally, some illustrative snapshots are reported from specific individual flights that highlight important differences in measured and modelled wind fields in complex orography. Two out of the 129 flights are analysed with the aim of providing more insight on local effect of orography, illustrating an example of the level of information that aircraft wind data provide. The two flights are chosen because they exhibit very similar conditions, with a general circulation from the north in winter, and northeasterly spatially homogenous winds close to the coast. Such stationary conditions can better highlight spatial variability and specific influences of orography. The flight sub portion is comprised between 15 and 70 km of the flight track. Aircraft data sampled at relatively high spatial resolution (∼250 m) are over-imposed at the entire model grid at 2 km resolution, both to reveal modelling limitations related to insufficient resolution, and to provide a comprehensive vision of mesoscale circulations at the time of flights (Figure 8). At both southern and northern sides of this section, the same similar spatially stable wind field is present in observations and is in very good agreement with the model (Figure 8). Several orographic structures are present in the surroundings of the track at the central portion. Here, actual observed winds result greatly influenced by such structures, with wind magnitudes dropping about 70% and wind direction turning to the northwest for about 20 km. WRF-CALMET is not actually capable of resolving such spatial patterns, and exhibits a more constant wind field over the study area (Figure 8). The reasons for this are likely related to an insufficient grid resolution (2 km), that is also reflected in the DTM (Digital Terrain Model) on which the model computes atmosphere dynamics. Soares et al. (2011) similarly reported increasing bias between observed and mesoscale modelled wind fields as terrain complexity increases. It is worth noticing that the WRF-CALMET model chain on which this analysis is based (Mari et al., 2011) is a regional scale tool, thus neither designed nor adequate to resolve relatively complex orography structures at fine resolution, that lead to sharp variations of atmospheric properties within short distance. Also, as pointed out by Wang et al. (2008), CALMET is a strictly area-dependent model, i.e. its setting parameters need to be carefully tuned for each area where the model is applied, because their selection may depend on local wind characteristics. These examples are instead reported to highlight the quality and the potential of aircraft derived winds that may constitute a useful observational framework in addition to existing weather station networks in any area of the globe.

image

Figure 8. Snapshots from two flights, representing aircraft wind vector (thin red arrows) along the track at 250 m spatial resolution, and WRF-CALMET wind vector (thick black arrows) on the 2 km resolution grid at the time of the flight. Panel (a) flight segment was performed on 16 November 2004 at 1345 h (local standard time); panel (b) flight segment was performed on 21 September 2005 at 1015 h. This figure is available in colour online at wileyonlinelibrary.com/journal/met

Download figure to PowerPoint

4. Conclusions

  1. Top of page
  2. ABSTRACT
  3. 1. Introduction
  4. 2. Materials and methods
  5. 3. Results and discussion
  6. 4. Conclusions
  7. Acknowledgements
  8. References

Long term aircraft observations of wind speed across an ∼250 km flight track in central Italy, performed during a 1.5 year study period, are reported and compared with the output of an existing WRF-CALMET model chain aimed at assessing wind potential maps at regional scale. Aircraft measurements are used to evaluate model performance along spatial transects and at moderate altitude from the ground (∼75 m AGL). Spatial wind analysis was capable of assessing overall model performance, while highlighting some limitations: the implemented mesoscale and prognostic models have better performance in inland areas with respect to coastal areas, while they are capable of representing diurnal variability in all regions correctly. The main reason for wind underestimation in coastal areas is related to models not being fully capable of resolving sea breeze regimes that consistently develop especially in the warm seasons. In addition, when assessed against high resolution aircraft wind, some model limitations in orographically complex areas were highlighted (Figure 8) revealing how orographic structures may generate local wind patterns that are not reproduced by models. Barantiev et al. (2011) also pointed out that the variety of physical, geographical and climate conditions related to sea breeze circulations, as well as weather patterns in coastal regions is huge, requiring mesoscale models to be constantly and vastly evaluated. Overall, the presented dataset allowed an extensive assessment of a modelling framework that is innovative, due to the typical lack of these types of observations, and was capable of providing a level of information that may be directly used to improve future models. Differences in model outputs may arise from different spatial resolutions, or only from different surface schemes (Draxl et al., 2012). Spatial resolutions, both horizontal and vertical, need to be improved to increase the model's capacity of representing orography driven patterns and better capture land-sea circulations. In fact, very recent implementations are today maturing to provide information at very high resolution for large regions, up to ∼250 m (Tammelin et al., 2013). Since mesoscale simulations may be re-run for past periods as new models become available, the dataset deployed in this work may constitute a benchmark to assess and evaluate future model implementations. Improvements may in fact be achieved both in the long-range forcing used to set initial and boundary conditions, and in the prognostic models and the computational capability needed to run them at higher resolutions.

Aircraft wind observations are not potential tools to replace the installation of tall towers on specific candidate locations for installations of infrastructures such as wind generators, since they provide a sporadic sampling in time. For this, UAV small aircraft might be better candidates, due to low cost and high capacity to fly almost continuously in day time. Instead, aircraft data, especially if combined with mesoscale and prognostic models, represent an additional tool on top of those actually used, that is specifically capable of assessing wind spatial variability. In orographically complex and/or remote locations, where spatial variability is extremely high and dense observation networks do not exist, aircraft data may fill an observational gap at relatively low cost. To assess wind patterns over more extensive areas, rather than along linear tracks, different experimental strategies may be adopted. Aircraft flight patterns composed of regular equally spaced transects in both north–south and east–west directions, flown across a small study area, would be a candidate experiment to build an observational ‘grid’ that may support a deeper analysis of the wind spatial behaviour in topographically complex areas.

Acknowledgements

  1. Top of page
  2. ABSTRACT
  3. 1. Introduction
  4. 2. Materials and methods
  5. 3. Results and discussion
  6. 4. Conclusions
  7. Acknowledgements
  8. References

Aircraft measurements were made within the project CARBIUS (Carbon Regional Balance Italy-USA). Modelling work was supported by the Tuscany Region Authority in the framework of the project: ‘WIND-GIS: a project to develop a web service to assess the wind potential of Tuscany region’. Fred Bosveld (KNMI) provided the Cabauw tower wind data within the RECAB EC project. We wish to thank the SkyArrow pilot Paolo Amico, and Alessandro Zaldei for the technical support.

References

  1. Top of page
  2. ABSTRACT
  3. 1. Introduction
  4. 2. Materials and methods
  5. 3. Results and discussion
  6. 4. Conclusions
  7. Acknowledgements
  8. References
  • Allwine KJ, Whiteman CD. 1985. MELSAR: A Mesoscale Air Quality Model for Complex Terrain, Vol. 1: Overview, Technical Description and User's Guide. Pacific Northwest Laboratory: Richland, WA, Washington, DC.
  • Al–Yahyai S, Charabi Y, Gastli A. 2010. Review of the use of Numerical Weather Prediction (NWP) models for wind energy assessment. Renew. Sustain. Energy Rev. 14: 31923198.
  • Arakawa A, Lamb VR. 1977. Computational design of the basic dynamical processes of the UCLA general circulation model. Methods Comput. Phys. 17: 173265.
  • Ayotte KW. 2008. Computational modelling for wind energy assessment. J. Wind Eng. Ind. Aerodyn. 96: 15711590.
  • Bañuelos-Ruedas F, Angeles-Camacho C, Rios-Marcuello S. 2010. Analysis and validation of the methodology used in the extrapolation of wind speed data at different heights. Renew. Sustain. Energy Rev. 14(8): 23832391.
  • Barantiev D, Novitsky M, Batchvarova E. 2011. Meteorological observations of the coastal boundary layer structure at the Bulgarian Black Sea coast. Adv. Sci. Res. 6: 251259.
  • Beaucage P, Brower MC, Tensen J. 2012. Evaluation of four numerical wind flow models for wind resource mapping. Wind Energy, DOI: 10.1002/we.1568.
  • Bellasio R, Maffeis G, Scire JS, Longoni MG, Bianconi R, Quaranta N. 2005. Algorithms to account for topographic shading effects and surface temperature dependence on terrain elevation in diagnostic meteorological models. Bound.-Lay. Meteorol. 114: 595614.
  • Black T. 1994. The new NMC mesoscale Eta model: description and forecasting examples. Weather Forecast. 9: 265278.
  • Black PG, D'Asaro EA, Sanford TB, Drennan WM, Zhang JA, French JR, Niiler PP, Terrill EJ, Walsh EJ. 2007. Air–sea exchange in hurricanes—Synthesis of observations from the coupled boundary layer air–sea transfer experiment. Bull. Am. Meteorol. Soc. 88(3): 357374, DOI: 10.1175/BAMS-88-3-357.
  • Crawford TL, Dobosy RJ. 1992. A sensitive fast response probe to measure turbulence and heat flux from any airplane. Bound.-Lay. Meteorol. 59: 257278.
  • Doms G, Schattler U. 2008. A Description of the Nonhydrostatic Regional Model LM, Part I: Dynamics and Numerics. DeutschenWetterdienstes (DWD): Offenbach.
  • Draxl C, Hahmann AN, Peña A, Giebel G. 2012. Evaluating winds and vertical wind shear from Weather Research and Forecasting model forecasts using seven planetary boundary layer schemes. Wind Energy, DOI: 10.1002/we.1555.
  • Ek MB, Mitchell KE, Lin Y, Rogers E, Grunmann P, Koren V, Gayno G, Tarpley JD. 2003. Implementation of Noah land surface model advances in the NCEP operational mesoscale Eta model. J. Geophys. Res. 108(D22): 8851, DOI: 10.1029/2002JD003296.
  • Farrugia RN. 2003. The wind shear exponent in a Mediterranean island climate. Renew. Energy 28: 647653.
  • Ferrier BS, Lin Y, Black T, Rogers E, Di Mego G. 2002. Implementation of a new grid–scale cloud and precipitation scheme in the NCEP Eta model. In Preprints, 15th Conference on Numerical Weather Prediction, San Antonio, TX. American Meteorological Society; 280283.
  • Gilliam RC, Childs PP, Huber AH, Raman S. 2005. Metropolitan–scale transport and dispersion from the New York world trade center following September 11, 2001. Part I: an evaluation of the CALMET meteorological model. Pure Appl. Geophys. 162: 19812003.
  • Gioli B, Miglietta F, De Martino B, Hutjes RWA, Dolman HAJ, Lindroth A, Schumacher M, Sanz MJ, Manca G, Peressotti A, Dumas EJ. 2004. Comparison between tower and aircraft–based eddy covariance fuxes in five european regions. Agric. For. Meteorol. 127: 116.
  • Gioli B, Miglietta F, Vaccari FP, Zaldei A, De Martino B. 2006. The Sky Arrow ERA, an innovative airborne platform to monitor mass, momentum and energy exchange of ecosystems. Ann. Geophys. 49: 109116.
  • Grell GA, Dudhia J, Stauffer DR. 1994. A description of the fifth generation Penn State/NCAR mesoscale model (MM5). NCAR Technical Note NCAR/TN–398+STR, National Center for Atmospheric Research (NCAR): Boulder, CO; 138.
  • Guilherme OC, Guedes RA, Manso MDO. 2009. Estimating wind resource using mesoscale modeling. European Wind Energy Conference (EWEC), Marseille, France.
  • Hiroyuki S, Takeshi I, Atsushi Y, Yukinari F. 2006. An assessment of offshore wind energy potential using mesoscale model. European Wind Energy Conference (EWEC), Athens, Greece.
  • Holtslag AAM, van Ulden AP. 1983. A simple scheme for daytime estimates of the surface fluxes from routine weather data. J. Clim. Appl. Meteorol. 22: 517529.
  • Hu J, Ying Q, Chen J, Mahmudc A, Zhao Z, Chen S–H, Kleeman MJ. 2010. Particulate air quality model predictions using prognostic vs. diagnostic meteorology in central California. Atmos. Environ. 44: 215226.
  • Istituto Superiore per la Protezione e la Ricerca Ambientale (ISPRA). 2010. La realizzazione in Italia del progetto Corine Land Cover 2006. Report No. 131/2010, ISBN: 978-88-448-0477-0. ISPRA: Rome (in Italian).
  • Janjic ZI. 2002. Nonsingular implementation of the Mellor–Yamada level 2.5 scheme in the NCEP meso model. NCEP Office Note No. 437, 61 pp.
  • Janjic ZI. 1990. The step–mountain coordinates: physical package. Mon. Wea. Rev. 118: 14291443.
  • Janjic ZI. 1996. The Mellor-Yamada level 2.5 scheme in the NCEP Eta Model. 11th Conference on Numerical Weather Prediction, Norfolk, VA, 19–23 August 1996. American Meteorological Society: Boston, MA; 333–334.
  • Janjic ZI, Mesinger F. 1984. Finite-difference methods for the shallow water equations on various horizontal grids. Numerical Methods for Weather Prediction, Vol. 1, Seminar, ECMWF, 1983, Reading, UK; 29101.
  • Janjic ZI, Mesinger F. 1989. Response to small-scale forcing on two staggered grids used in finite–difference models of the atmosphere. Q. J. Roy. Meteorol. Soc. 115: 11671176.
  • Kain JS. 2004. The Kain–Fritsch convective parameterization: an update. J. Appl. Meteorol. 43(1): 170181.
  • Liu MK, Yocke MA. 1980. Siting of wind turbine generators in complex terrain. J. Energy. 4: 1016.
  • Mahrt L. 1982. Momentum balance of gravity flows. J. Atmos. Sci. 39: 27012711.
  • Majewski D. 2009. HRM—User's Guide. DeutschenWetterdienstes (DWD): Offenbach.
  • Mari R, Bottai L, Busillo C, Calastrini F, Gozzini B, Gualtieri G. 2011. A GIS-based interactive web decision support system for planning wind farms in Tuscany (Italy). Renew. Energ. 36: 754763.
  • Martin S, Bange J, Beyrich F. 2011. Meteorological profiling of the lower troposphere using the research UAV “M(2)AV Carolo”. Atmos. Meas. Tech. 4: 705716.
  • Maselli F, Gioli B, Chiesi M, Vaccari F, Zaldei A, Fibbi L, Bindi M, Miglietta F. 2010. Validating an integrated strategy to model net land carbon exchange against aircraft flux measurements. Remote Sens. Environ. 114: 11081116.
  • Mason MS, Wood GS, Fletcher DF. 2010. Numerical investigation of the influence of topography on simulated downburst wind fields. J. Wind Eng. Ind. Aerodyn. 98: 2133.
  • Morales L, Lang F, Mattar C. 2012. Mesoscale wind speed simulation using CALMET model and reanalysis information: an application to wind potential. Renew. Energy 48: 5771.
  • Petersen EL, Mortensen NG, Landberg L, Højstrup J, Frank HP. 1998. Wind power meteorology. Part II: siting and models. Wind Energy 1(2): 5572.
  • Pielke RA, Cotton WR, Walko RL, Trembaek CJ, Lyons WA, Grasso LD, Nieholls ME, Moran MD, Wesley DA, Lee TJ, Copeland JH. 1992. A comprehensive meteorological modeling system—RAMS. Meteor. Atmos. Phys. 49: 6991.
  • Prieto M, Navarro J, Copeland J. 2007. A comparison of wind field simulations through a coupling of MM5/CALMET in a complex terrain. European Wind Energy Conference (EWEC), Milan, Italy.
  • Sarrat C, Noilhan J, Lacarrère P, Ceschia E, Ciais P, Dolman AJ, Elbers JA, Gerbig C, Gioli B, Lauvaux T, Miglietta F, Neininger B, Ramonet M, Vellinga O, Bonnefond JM. 2009. Mesoscale modelling of the CO2 interactions between the surface and the atmosphere applied to the April 2007 CERES field experiment. Biogeosciences 6: 633646, DOI: 10.5194/bg-6-633-2009.
  • Scire JS, Robe FR, Fermau ME, Yamartino RJ. 1999. A User's Guide for the CALMET Meteorological Model (Version 5.0). Earth Tech Inc.: Concord, MA.
  • Shimada S, Ohsawa T, Yatsu K. 2009. A study on the ability of mesoscale model MM5 for offshore wind resource assessment in Japanese coastal waters. European Wind Energy Conference (EWEC), Marseille, France.
  • Skamarock WC, Klemp JB, Dudhia J, Gill DO, Barker DM, Wang W, Powers JG. 2005. A Description of the Advanced Research WRF Version 2 NCAR Technical Note. National Center for Atmospheric Research (NCAR): Boulder, CO.
  • Soares CC, Chagas GO, Guedes RA. 2011. Estimating wind resource using mesoscale modeling. European Wind Energy Conference (EWEC), Brussels, Belgium.
  • Surapipith V, Riddhiraksa N, Uttamung P. 2011. Meteorology for CALPUFF at a seaside industrial complex. 14th Conference on Harmonisation within Atmospheric Dispersion Modelling for Regulatory Purposes, 2–6 October 2011, Kos, Greece.
  • Sykes RI, Parker SF, Henn DS, Cerasoli CP. 1998. PC–SCIPUFF Version 1.0. Titan ARAP Report 717, DTRA Technical Report, Titan Corporation, Alexandria, VA.
  • Tammelin B, Vihma T, Atlaskin E, Badger J, Fortelius C, Gregow H, Horttanainen M, Hyvönen R, Kilpinen J, Latikka J, Ljungberg K, Mortensen NG, Niemelä S, Ruosteenoja K, Salonen K, Suomi I, Venäläinen A. 2013. Production of the Finnish Wind Atlas. Wind Energy 16: 1935, DOI: 10.1002/we.517.
  • US EPA. 2004. User's Guide for the AERMOD Meteorological Preprocessor (AERMET) EPA–454/B–03e002. U.S. Environmental Protection Agency: Research Triangle Park, NC.
  • Yim S, Fung J, Lau A, Kot S. 2007. Developing a high–resolution wind map for a complex terrain with a coupled MM5/CALMET system. J. Geophys. Res. 112: D05106.
  • Vellinga OS, Dobosy RJ, Dumas EJ, Gioli B, Elbers JA, Hutjes RWA. 2013. Calibration and quality assurance of flux observations from a small research aircraft. J. Atmos. Oceanic Technol 30: 161181.
  • Wang W, Shaw WJ, Seiple TE, Rishel JP, Xie Y. 2008. An evaluation of a diagnostic wind model (CALMET). J. Appl. Meteorol. Climatol. 47: 17391757.
  • Wieringa J. 1992. Updating the Davenport roughness classification. J. Wind Eng. Ind. Aerodyn. 41–44: 357368.
  • Zhang L, Lemeur R. 1992. Effect of aerodynamic resistance on energy balance and Penman-Monteith estimates of evapotranspiration in greenhouse conditions. Agricultural and Forest Meteorology 58(3–4): 209228.