The Austrian RADiation monitoring network (ARAD) has been established to
advance the national climate monitoring and to support satellite retrieval,
atmospheric modeling and the development of solar energy techniques.
Measurements cover the downward solar and thermal infrared radiation using
instruments according to Baseline Surface Radiation Network (BSRN)
standards. A unique feature of ARAD is its vertical dimension of five
stations, covering an altitude range between about 200 m a.s.l (Vienna) and
3100 m a.s.l. (BSRN site Sonnblick). The paper outlines the aims and scopes
of ARAD, its measurement and calibration standards, methods, strategies and
station locations. ARAD network operation uses innovative data processing
for quality assurance and quality control, utilizing manual and automated
control algorithms. A combined uncertainty estimate for the broadband
shortwave radiation fluxes at all five ARAD stations, using the methodology
specified by the Guide to the Expression of Uncertainty in Measurement
indicates that relative accuracies range from 1.5 to 2.9 % for large
signals (global, direct: 1000 W m
Radiative processes are key for both the natural and the anthropogenic dimension of the climate system as well as its changes. Firstly, the Earth's radiation budget plays an essential role in determining the thermal state of the land surface, the atmosphere and the oceans, thereby also strongly influencing the circulation of the latter two (Ohmura et al., 1998; Ramanathan, 1987). Secondly, the main anthropogenic influence on climate arises through modification of the Earth's radiation budget components (Intergovernmental Panel on Climate Change, 2014). Worldwide in situ ground-based and space-based measurements of radiative fluxes are thus performed to (a) improve the process understanding of the present climate system (e.g., Wild et al., 2015), (b) provide data for the validation and calibration of weather and climate models in order to understand processes in the past and future and/or to improve weather forecasts (Chevallier and Morcrette, 2000; Haiden and Trentmann, 2015; Wild et al., 2001) and (c) support solar energy applications (e.g., Gueymard, 2014).
Surface observations have the lowest degree of uncertainty and are therefore used to validate satellite measurements (Gupta et al., 2004). The most prominent, worldwide observational ground-based network for surface radiation fluxes is the Baseline Surface Radiation Network (BSRN), established in the early 1990s (Ohmura et al., 1998) by the World Climate Research Programme (WCRP). BSRN provides measurements with high accuracy and high temporal resolution, and comprises currently 54 sites in different climate regimes (König-Langlo et al., 2013). Another worldwide network providing surface radiative fluxes is the Global Energy Balance Archive (GEBA) (Gilgen and Ohmura, 1999).
Besides these global observational networks, several national surface radiation monitoring networks exist such as SURFRAD in the US (Augustine et al., 2000) or SACRaM in Switzerland (Wacker et al., 2011). These networks provide long-term observations for climate monitoring and aid process-level understanding on regional scales such as e.g., the European Alps.
The European Alpine region is known to react especially sensitively to both observed (Auer et al., 2007) and projected (Gobiet et al., 2014) global climate change. Therefore there is great need for long-term measurements of the radiative fluxes in Europe, particularly in the Alps, to better understand the drivers of observed and projected changes (e.g., Kotlarski et al., 2015; Marty et al., 2002; Philipona, 2013; Wild, 2009). Moreover, state-of-the-art regional climate models still show significant biases, considerably adding to the uncertainty of future climate change projections (e.g., Frei, 2003; Haslinger et al., 2013). To this aim, ultimately a dense network of high-accuracy surface observations of the radiative fluxes, especially in complex, densely populated alpine topography and adjacent lowlands, is needed where possible impacts of climate change lead to great vulnerability and accurate estimates of the solar energy resources are needed due to the growing demand for renewable energies.
Large parts of the European Alps are within the national territory of Austria. The Austrian Alps are also part of the Greater Alpine Region (GAR) with a unique and outstanding long-term instrumental climatological data set (Auer et al., 2007). Significant Austrian contributions to the field of solar radiation and albedo measurements and respective developments of instrumentation date back to the 1950s and 1970s (e.g., Dirmhirn, 1951, 1957; Dirmhirn and Eaton, 1975; Dirmhirn and Trojer, 1955; Sauberer and Dirmhirn, 1952). In the meantime the standard meteorological station network of the Austrian Zentralanstalt für Meteorologie und Geodynamik (ZAMG), which includes the measurement of global horizontal irradiance using a Schenk star pyranometer, has evolved, from 20 stations in 1980 to 235 stations in 2015, the latter corresponding to a mean horizontal station distance of about 20 km. In the 1970s and 1980s, surface radiation studies in Austria were focused on establishing radiation climatologies and increasing the process understanding based on the available station data (e.g., Neuwirth, 1980, 1983). From the mid-1990s until today, work related to ultraviolet (UV) radiation and health risks (e.g., Blumthaler et al., 1996; Rieder et al., 2010; Weihs et al., 2008, 2013), to the role of aerosols in alpine valleys (e.g., Blumthaler et al., 1997; Wuttke et al., 2012) and studies dealing with anthropogenic influences on radiative fluxes and solar energy (e.g., Weihs et al., 2015) predominated in Austria.
With this background the Austrian RADiation monitoring network (ARAD) was founded in the year 2010, following an initiative of ZAMG in cooperation with the universities of Graz and Innsbruck, and the University of Natural Resources and Life Sciences (BOKU) in Vienna. In its original concept, ARAD has to fulfill three major requirements: (i) to provide a state-of-the-art national radiation monitoring network; (ii) to provide data for atmospheric model/satellite data calibration and validation and (iii) to provide high-accuracy data for further technical developments in the field of solar energy utilization.
This paper presents the ARAD network with its aims, scopes and innovative approaches. First, Sect. 2 details the concept of ARAD, the stations network and the measurement setup. Section 3 illustrates the data processing and quality control. An uncertainty analysis, presented in Sect. 4, explores the range for the combined expanded uncertainty of all ARAD stations. In Sect. 5 the data policy, allocation and some examples of data usage are presented. Finally, Sect. 6 discusses some key points before the summary concludes the paper in Sect. 7.
Topographic map (color-coded elevations) with ARAD stations (red points; the station at Klagenfurt is planned to start operating in 2016), TAWES stations measuring GLO (black points) and the coarse resolution subregions (CRSs) defined as regions with common climatic variability (see text).
As given in the Guide to the Expression of Uncertainty in Measurement (GUM)
(JCGM, 2008) and in Gupta (2012) the following metrological terms are used
throughout the manuscript: (1)
The general aim of ARAD is to establish a high-accuracy long-term monitoring network for solar and terrestrial surface radiation in Austria in order to assess the status as well as the temporal and spatial changes of radiative fluxes at/to the surface. To this aim, three general principles served as the basis for establishing the ARAD network: (i) to explore the potential of existing radiation measurements, or measurements with linkage to radiation (e.g., aerosols), (ii) to continue long-term observational series of radiation with known high quality and (iii) to capture the spatial patterns of the radiation climate in Austria. Consequently, observations at five sites are routinely performed, following, as far as practicable, the guidelines for surface radiation measurements established by the BSRN (McArthur, 2005). BSRN guidelines detail the standardization of observation methods (instrument specifications, site location, measurement and calibration procedures, maintenance procedures and intervals) and data acquisition, storage and quality control. Although ARAD stays close to BSRN guidelines, operational logistics at some sites require slight deviations from these.
The accuracy targets for BSRN irradiance measurements are
(McArthur, 2005) 0.5 % or 1.5 W m
The ARAD network comprises currently five monitoring sites: Wien Hohe Warte (WHW; 198 m a.s.l.), Graz/University (GRZ; 398 m a.s.l.), Innsbruck/University (IBK; 578 m a.s.l.), Kanzelhöhe (KSO; 1540 m a.s.l., since 2013) and Sonnblick (SON; 3109 m a.s.l.) which have been operating since 2011.
List of measurement instruments used at ARAD stations.
As can be seen in Fig. 1, ARAD covers currently four out of five so-called coarse-resolution subregions (CRSs) defined in the HISTALP project (HIstorical Instrumental climatological Surface Time series, (Auer et al., 2007). CRSs define subregions with common long-term trends for multiple climate parameters (air pressure, air temperature, precipitation, sunshine duration and cloudiness) within the GAR. There are four CRSs defined in the horizontal domain (regions: northwest, northeast, southeast, southwest) and one in the vertical domain (high-elevation summit sites). Currently all CRSs, besides the southwest region, are represented by at least one ARAD site. For comparison reasons, Fig. 1 also shows the location of the automated stations of the national meteorological observing network of Austria (semi-automatic weather stations, TAWES, operated by ZAMG) measuring global radiation by Schenk star pyranometers.
The designated ARAD site Klagenfurt (which will be operational by 2016) will provide, along with the ARAD sites KSO and SON, a unique vertical transect of radiation measurements south of the Alpine main ridge, within relatively small horizontal distance (approximately 100 km) and with very similar mesoscale synoptic conditions. This transect will allow vertical gradients of radiative fluxes with and without cloud effects to be investigated, as well as effects of boundary layer dynamics and thickness, related to synoptic conditions and anthropogenic emissions, and turbidity effects on radiative fluxes.
It is important to mention that one of the ARAD sites, SON, is also included
in the BSRN network (since January 2013;
Typical ARAD station: suntracker with different radiation sensors. The picture in the lower right corner shows a pyranometer without radiation shield exposing the heating and ventilation system (PMOD-VHS).
List of metadata of the ARAD stations. Mean annual air temperatures (Temp.) and precipitation sums (Precip.) are given for the climatological period 1981–2010.
ARAD sites are equipped with four broadband radiation sensors which are suitable to BSRN requirements, mounted on a suntracker (see Fig. 2), for measurements of global (GLO), direct (DIR) and diffuse (DIF) solar radiation and downward longwave radiation (DLW). The suntracker allows correct tracking of the solar path, guarantees the continuous alignment of the pyrheliometer to record DIR and ensures continuous shading of the pyranometer for measurements of DIF and the pyrgeometer for measurements of DLW. All radiation sensors used within ARAD are state-of-the-art thermopile instruments with specifications well within the limits recommended and accepted by BSRN. Following the ISO 9060 classification, all pyranometers used within ARAD are secondary standard instruments and all pyrheliometers are first-class instruments. Further details on instrument specifications can be found in Table 1.
Within ARAD all horizontally mounted sensors are placed in ventilation units to reduce solar loading and thermal offsets; in addition they are heated and ventilated (see Fig. 2) in order to reduce interference by rain drops, dew, rime, ice and snow deposition on the instrument dome and to reduce temperature gradients that are known to occur between the glass dome and the instrument body due to infrared (IR) losses (Philipona, 2002). For ARAD pyranometers and pyrgeometers we mostly use the heating and ventilation system PMOD-VHS developed by the Physikalisch-Meteorologisches Observatorium Davos (PMOD) in Switzerland (described in detail in Philipona (2002) and the ventilation unit SBL 480 developed by the German manufacturer Eigenbrodt (see Fig. 3); however slight deviations in instrumentation occur depending on the station (see Table 2 for site-specific information). At KSO all radiometers are equipped with ventilation units CV 2 developed by manufacturer Kipp & Zonen. For one of the pyranometers and the pyrgeometer at the ARAD GRZ station we use a slightly different, self-designed system. ARAD sensors use a general heating power of 10 W, with the exception of the SON station, where the power is increased to 25 W due to frequent snow and frost conditions.
The modified Eigenbrodt heating and ventilation system with external sensor leveling as used for the ARAD stations (left) and some special modification to reduce snow and ice effects at ARAD/BSRN SON station: mounting plastic tubes at the inlet of the heating and ventilation system beneath the instruments (lower right) delay the buildup of rime, leading to reduced ventilation and the buildup of snow/ice on the pyranometer domes, deteriorating the radiation signal (upper right).
A heating and ventilation system that works correctly should prevent so-called nighttime zero offsets that occur due to IR losses at the glass dome. Thus, the nighttime signal of the pyranometers is a direct quality indicator for the thermal control system. We note that the heating effect on the dome due to the fan motors is nullified when wind speeds are moderate to high.
ARAD pyranometers are calibrated through comparison with the working
standard pyranometer (International Organization for
Standardization, 1992) at ZAMG headquarters in Vienna or sent for calibration
to the instrument manufacturer Kipp & Zonen. The calibration procedure at
ZAMG is based on the direct comparison of 1 min average values (with a
sampling rate of 1 Hz) during periods with GLO exceeding 600 W m
The working standard of ZAMG itself is calibrated using the traditional so-called sun-and-shade calibration method (International Organization for Standardization, 1993), thereby the working standard is calibrated against a cavity radiometer at ZAMG headquarters in Vienna (TMI, serial number 68025) that participates in the international pyrheliometer comparison (IPC) exercise in Davos, Switzerland, every 5 years. This procedure ensures that the reference TMI cavity radiometer is traceable to the World Radiometric Reference (WRR) in Davos, Switzerland (Fröhlich, 1991).
It is important to note that this procedure differs from BSRN suggestions for pyranometer calibration, which recommend the “alternate method” of Forgan (1996). The Forgan method is based on swapping both pyranometers that measure GLO and DIF and calculate the calibration coefficients, based on simultaneous equations of GLO given by the component sum of DIF and DIR, before and after the swap, for specific solar zenith angles. BSRN recommends this procedure as it allows for on-site instrument calibration and avoids thermal shocks of the instrument that may occur when using the traditional sun-and-shade method (Forgan, 1996).
Despite this recommendation the ARAD consortium utilizes the sun-and-shade method for instrument calibration because (i) this approach is also used for calibration of Schenk star pyranometers at TAWES sites of ZAMG and thus provides traceability of all ZAMG radiation sensors to one common standard; (ii) it avoids the use of a traveling-standard pyrheliometer at all sites, which would be necessary when using the Forgan method; (iii) extended comparison intervals under stable clear-sky conditions are used for calibrating the reference pyranometer.
For pyrgeometers, optimal methods of calibration and traceability to an absolute irradiance scale are still under debate (Gröbner et al., 2014, 2015; Philipona, 2015). Thus, BSRN recommends sending pyrgeometers to PMOD Davos for calibration with a blackbody cavity (McArthur, 2005). Every 2 years, ARAD pyrgeometers are sent to Kipp & Zonen for calibration. The calibration procedure is traceable to the World Infrared Standard Group (WISG) in Davos consisting of four reference pyrgeometers, building a long-term stable reference (Gröbner et al., 2014). This calibration procedure avoids the necessity of a site reference instrument because each pyrgeometer is directly traceable to WISG.
Finally, all ARAD pyrheliometers are calibrated by direct comparison (International Organization for Standardization, 1990) against a TMI cavity radiometer (participating regularly at the IPC in Davos) and thus directly traceable to the World Radiometric Reference (WRR).
At BSRN/ARAD site SON, the calibration of all instruments is performed annually, as recommended by BSRN. At other ARAD sites instrument calibration is performed every 2 years. Long-term stability analysis of ARAD radiometer sensitivities (based on different calibration certificates of the same instrument at different times) indicates that the relative change of the calibration factor over 2 years is well below 1 % for all pyranometers and < 0.2 % for all pyrheliometers which operate within the ARAD network.
The radiation components GLO, DIR, DIF and DLW are the key variables measured at each ARAD site. Besides these radiation components, data on body temperature of all sensors and the heating power and ventilation of pyranometers are recorded at all ARAD sites except KSO.
Sample of ARAD data measured at the SON station on 3 November 2014 between 06:00 and 15:00 UTC. GLO denotes the global radiation on a horizontal plane, DIR the direct solar radiation on a plane normal to the direct beam, DIF the diffuse radiation on a horizontal plane and DLW the downward longwave radiation.
Data sampling takes place at 1 Hz (GRZ, IBK, SON, WHW) and 10 Hz frequency (KSO) with the following 1 min statistics calculated and stored: average, standard deviation, minimum and maximum. An example plot of the measured irradiances at the SON station during 1 day is shown in Fig. 4.
A specific data acquisition system (DAQ), developed by Logotronic
(
A traceable reference voltage generator (RVG) is used to recalibrate the DAQ at all ARAD stations periodically (every
2 years). Results from RVG measurements suggest that a potential zero offset is < 10
Complementary to ARAD radiation data, basic meteorological records of air temperature, air pressure, relative humidity, precipitation, sunshine duration, wind speed and direction and global radiation (measured with a black and white pyranometer, a Schenk star pyranometer) from the collocated TAWES station are also stored as 1 min averages or sums in the same database. For details on meteorological observation methods and related uncertainties the interested reader is referred to Haiden et al. (2011).
ARAD instruments are regularly inspected and maintained by human observers. Maintenance intervals are at a maximum 1 week within the ARAD network. At three sites, the BSRN site SON, KSO and GRZ instruments are maintained on a daily basis, as recommended by BSRN.
Maintenance tasks are defined corresponding to BSRN standards, are regulated in a maintenance manual and follow a defined number of checks. These include (i) cleaning and verifying leveling of all instruments, (ii) visually inspecting cables, ventilation and heating devices and (iii) controlling the sun tracking system and shading of the instruments recording DIF and DLW. Immediately after maintenance, the results of the check are entered in a web interface and stored in a central database for further processing and use. On an annual or biennial basis the heating and ventilation systems are cleaned and a control and service of the sun tracking system is performed following recommendations by the manufacturer. Apart from these routine and regular maintenance tasks, occasional service is required whenever failed checks recorded by a human observer or the automated or manual data quality analysis (see Sect. 3.2) make this necessary.
The harsh weather conditions at the ARAD and BSRN station SON required the
following technical adaptations: (1) for data quality assurance the heating
and ventilation power of a pyranometer or pyrgeometer (see Sect. 2.3) is connected to one single electrical circuit
per instrument. In case of a single failure of the ventilation (due to e.g., ice formation) the heater is also immediately turned off, avoiding
differential heating leading to additional thermal offsets. (2) Before
operational use of the Hukseflux DR02-T pyrheliometer with a heated front
window, different shading and heating experiments were performed to
determine the ideal heating power between maximum frost removal efficiency
and minimum thermal offsets, which was found to be 2.1 W. A comparison of
DIR using the heated DR02-T (corrected with temperature dependence) and an
unheated CHP1 pyrheliometer over 11 clear-sky days during a 2-year period
showed a mean relative difference of
After being measured and recorded, all parameters (see Sect. 2.4.2), are immediately stored in a Sybase database
at ZAMG in Vienna. The ARAD convention is thereby to store both the raw
voltage output signal of the sensors and the irradiance in W m
One exception in this respect is the front-window heated Hukseflux DR02-T
pyrheliometer operated at SON. For this instrument, temperature effects are
not negligible (estimated as 5 % at
Summary of the quality flag statistics (average of 2012–2014) for the measured parameters GLO, DIF, DIR, DLW at four ARAD sites which have been operating since 2012 using the combined daily automated and manual QC including e.g., maintenance times. The last row (“Perfect”) indicates the percentage of timestamps when all four parameters are simultaneously flagged as “good”.
To maximize ARAD data quality we use a combination of a manual and two automated data quality control (QC) methods with daily and hourly resolution (weekend data are controlled at the beginning of the following week). The algorithms used for automated control are an extended version of those provided by Long and Shi (2008); see details below.
The daily manual data QC consists of visual inspections of (i) the graph of
the daily automated QC (see Sect. 3.2.2) from the
previous day, (ii) a near-real time, interactive data plot of all stations
(updated every 5 min) of the 1 min average radiation values and DIR
calculated on the horizontal surface to enable the comparison between GLO
and its component sum (DIR
The daily automated QC consists of a script containing the QC criteria for all recorded radiation fluxes as described in Long and Shi (2008) as well as some additional criteria detailed below. The automated script is run daily for the data recorded over the last 24 h at all ARAD stations. For each station, the script reads a metadata file containing all relevant coefficients for criteria that have been established based on site specific data (e.g., upper climatological limits of fluxes as in Long and Shi, 2008).
In an initial step the automated QC creates quality flags for each recorded element and the records are modified from their initial value 0 (unchecked data) to either one of the following four states: 1 (data checked and passed all tests successfully), 2 (data checked but wrong as at least one test failed), 3 (data checked but of questionable quality), 255 (missing data) and stored in separate tables of the database.
In a second step a graphical summary of the daily automated QC is produced. This graph shows the time series of the 1 min averaged radiation fluxes along with the cloud observations from the hourly SYNOP report from the closest available station and information about failed QC criteria from the automated QC as detailed below.
A third step in the QC procedure comprises the checking of 61 quality
criteria. These criteria can be grouped into four types: (1) integrity tests
(e.g., missing values), (2) outlier detections, (3) min/max tests (e.g., exceedance of possible limits) and (4) comparison tests (e.g., GLO vs.
component sum of DIR horizontal
Finally measured irradiance signals are converted to units of W m
Table 3 summarizes the 3-year average quality flag statistics for all stations which have been operating since 2012. The fraction of data flagged as good equals, depending on the measured parameter, 74–85 % at SON and 91–99 % at the other stations. Note, that reduced values at the GRZ station compared to WHW and IBK do not reflect a lower quality but more frequent daily maintenance (data are flagged as “wrong” during maintenance). The fraction of observations where data of all four parameters are simultaneously flagged as “good” ranges from 67 % at the SON to 88–94 % at other ARAD stations. Lower data quality at the BSRN SON station reflects the difficult measurement conditions at this exposed high-alpine site.
Contributions to uncertainties affecting shortwave measurements at all five ARAD stations in the period 1 July 2014 to 30 June 2015.
Same as Table 4 but relative contributions of uncertainty components to
total combined uncertainty. Uncertainties are given for a small (ss: 50
Same as Tables 4 and 5 but products of sensitivity factor
The hourly automated QC consists essentially of a strongly reduced set of criteria from the daily automated QC, and no quality flags are recorded in a database; instead, an email containing the error information is sent to the on-site staff in case of failure of any QC criteria. Hourly checked criteria consist of all tests described in Appendix A plus a check for missing data. The outlier detection (test (4) in Appendix A) is thereby not applied to the irradiance signals as during the daily automated QC but to the difference between instrument housing temperature and air temperature to detect eventual thermal offsets.
In the following paragraphs we detail uncertainty estimates for all shortwave radiative fluxes measured at all five ARAD sites over the full annual cycle from 1 July 2014 to 30 June 2015.
The measurement equation for the shortwave fluxes (GLO, DIF and DIR) reads
The first term
The measurement uncertainty of the raw signal
The third term in Eq. (2)
The estimation of the thermal offset uncertainty
To date no routine measurement of the pyrgeometer dome temperatures nor
capping experiments are available for ARAD sites. However, the latter are
intended for a field campaign in 2016. Thus the detection of (daytime)
thermal offsets is limited to the analysis of nighttime pyranometer data, as
deviations from zero are assumed to be triggered by thermal offsets
(Michalsky, 2005; Philipona, 2002). Unfortunately, the
limited resolution and accuracy of the DAQ at most ARAD sites except KSO
(see Sect. 2.4.2), corresponding to approximately 1 W m
The last term in Eq. (2)
Same as Tables 4–6, but combined uncertainties affecting shortwave measurements at all ARAD sites. Relative combined expanded values are marked in bold to allow the comparison with BSRN target accuracies (see Sect. 2.1).
The leveling-induced uncertainty is estimated as follows: from operational
experience (during the daily maintenance phase) in the ARAD network we estimate
an average maximum slope of the sensor tilt of 1
Results of the relative contributions of the individual standard
uncertainties
Relative and absolute numbers of the combined expanded uncertainty for all
ARAD sites (Table 7) can be directly compared with the BSRN target
accuracies (see Sect. 2.1). Values for the combined
expanded uncertainties of the ARAD sites range between 1.46 and 2.87 % (or 15 to 29 W m
ARAD data are available upon request from ZAMG for “bona fide
research purposes”, without commercial application or intentions in mind. A
formal agreement has to be signed by the user to confirm the intentions of
use and additional obligations related to the publication and use of the
data. Quality-controlled data from the ARAD and BSRN SON station are transferred
on a monthly basis to the BSRN archive and are available there. An RSS feed
for the latest data from SON is available here:
Within a research project at ZAMG, a method provided by Forgan and Dyson (2004) to use 1 min radiation data statistics based on 1 Hz sampling as recorded in ARAD (average, minimum, maximum, standard deviation) to improve the calculation of sunshine duration from pyrheliometer measurements was tested and compared to 1 Hz records of a NIP pyrheliometer and measurements from a conventional sunshine duration sensor (Haenni-Solar, used at the TAWES stations of ZAMG) at the WHW station for a period of 4 months. The excellent agreement between the calculations based on the Forgan and Dyson (2004) method and the NIP data (average daily differences of 3.7 min (0.8 %) of sunshine duration) suggests that ARAD stations can serve as regional references for sunshine duration measurements in Austria, aiding the continuous improvement and verification of the routine TAWES measurements.
Seasonal mean daily courses based on 10 min averages of the recorded and quality-controlled 1 min average values for all measured irradiance quantities at all ARAD stations for the full annual cycle 1 December 2013 to 30 November 2014.
Correlation between short-range (0–24 h) forecasts and observations of daily averaged values of downward shortwave and longwave flux at the ARAD WHW station. Monthly means for the verification period 1 Janary–31 December 2012 are shown.
Forecasting cloudiness and surface radiation remains a major challenge for numerical weather prediction models, and high-accuracy ground-based data are needed for model development/improvement and validation. BSRN and ARAD measurements are used at the European Centre for Medium-Range Weather Forecasts (ECMWF) to evaluate forecasts of downward fluxes of shortwave and longwave radiation. Figure 6 shows verification results for Vienna for the period 1 January–31 December 2012. Two features during this intercomparison interval are worth noting. (1) There is a strong drop in correlation between observations and model output, in both the shortwave and longwave range, during autumn. This is a recurring issue in the ECMWF model due to difficulties in representing low stratus clouds. As documented in Haiden and Trentmann (2015) this model deficiency affects forecasts over large parts of Central Europe. (2) A transient drop of the shortwave flux occurred in May, triggered by the prevalent weather situation, i.e., a humid and relatively unstable air mass with weak pressure gradients was present over Austria at that time. Because of the absence of significant synoptic-scale forcing, cloud formation was mainly governed by mesoscale processes, which are less well represented in the model due to limited resolution (16 km in 2012). This created relatively large daytime forecast errors, over the course of several days, affecting the monthly average of the shortwave skill. In addition to the monitoring/evaluation of the operational forecast, surface radiation measurements are also used at ECMWF in the evaluation of new model versions, and for assessing the quality of satellite-derived radiation products.
The cloud parameterization scheme for diffuse solar radiation of the solar radiation model STRAHLGRID, developed at ZAMG (Olefs and Schoener, 2012), to provide gridded data sets of near-surface solar direct and diffuse radiation in near-real time (based on the INCA nowcasting system), was developed using ARAD DIF measurements. The so-called HIM ratio factor is thereby defined by the ratio of measured all-sky to modeled clear-sky diffuse radiation correlated with MSG-2 satellite cloud types. More details on the model and model performance can be found in Sect. 2.3 of the paper by Olefs et al. (2013) that also describes the coupling to the operational energy-balance snow-cover model SNOWGRID, showing the added value of ARAD data for applications in alpine hydrology and cryosphere.
Recently, a method to improve area-wide 72 h DIR forecasts, driven by atmospheric data of the ALARO numerical weather prediction model used at ZAMG through a coupling of the high-resolution STRAHLGRID model, was successfully tested at ZAMG. For stations with significant horizon shading, a reduction of the root-mean-square error (RMSE) of 3–40 % could be achieved due to the better resolved topography. For stations in the lowlands, an improved cloudiness scheme in STRAHLGRID based on ARAD data led on average to a RMSE reduction of 25 %.
Following the uncertainty analysis documented in this paper, there may be
three future activities: (1) correction of some of the identified
uncertainty sources in the data will be possible during post-processing
(e.g., cosine response, thermal offset, leveling, temperature dependence,
influence of soiling) which will partly need additional measurements (no
correction for historical data possible, e.g., record leveling, with a digital level for
every sun-tracker) or experiments (historical data correction possible by
capping for daytime thermal offsets, soiling, by analyzing data after as opposed to before
maintenance) or simply using the calibration certificates as was
demonstrated (historical data correction possible, e.g., directional
response error), thereby reducing the combined uncertainties given in Table 7
closer towards BSRN targets. (2) It may become necessary to provide
these uncertainty calculations operationally as meta-information, in addition to
the allocated data using measured irradiances in the full time resolution (1 min) and technical specifications of the individual sensors. As some of
the uncertainty components depend also on the actual short-term fluctuations
of the meteorological conditions (e.g., the directional or leveling error is
only relevant when DIR > 0) this would allow the user to better
judge the actual uncertainty of the delivered data snippet, especially in
relation to the uncertainty inherent to e.g., a model run or measurements by
a different instrument. (3) Comparison of redundantly measured parameters
(e.g., GLO vs. DIF when DIR
For some of the uncertainty components such as thermal offsets for
pyranometer measurements, attempts were undertaken to improve the error
correction with the currently available data. In correlation experiments of
the nighttime offsets with the longwave net flux, a rather low coefficient
of determination of 0.31 was found and even multiple correlations, including
additionally e.g., the air temperature, wind speed, case temperatures or the
clear-sky index (Marty and Philipona, 2000), the latter related to
cloud cover, could only increase
A comparable uncertainty estimate for DLW was beyond the scope of this study and is left for future work. However, comparing the average expanded sensitivity uncertainty of our CGR4 pyrgeometers of 4.1 % (see Table 1) with the BSRN target of 2 % suggests that targets are not met because operational and other uncertainties add to this value of sensitivity uncertainty. We note that to date we did not investigate a possible sensitivity of the calibration coefficients of our CGR4 pyrgeometers with respect to integrated water vapor (IWV) related to dome spectral transmissivity as suggested by Gröbner et al. (2014), especially for climates with low IWV values. However, this issue might be of minor importance for ARAD as all ARAD CGR4 pyrgeometers were built after the year 2003. Recent results suggest that those instruments suffer much less from the IWV dependency issue (Gröbner and Wacker, 2013).
We presented the Austrian RADiation monitoring network (ARAD), comprising currently five stations, which largely follows the quality standards and guidelines of BSRN. One ARAD site, Sonnblick (SON), has been part of the BSRN network since January 2013.
Our main conclusion is that ARAD could serve as a useful example for establishing state-of-the-art radiation monitoring at the national level with a multiple-purpose approach. In particular, we hope that our experiences will give a stimulus for radiation monitoring in other countries/regions worldwide, which is of utmost importance for climate change research. Instrumentation, guidelines and tools (such as the data quality control) developed within ARAD are best practices which could be easily adopted for other regions.
The consistently defined measurement methods and automated and manual data quality control mechanisms present some innovative aspects to increase data availability and quality. Special care is taken at the ARAD/BSRN SON station, where rough weather conditions necessitate a strict control, regular maintenance and adapted measurement methods. Average combined quality flags (automated and manual controls) of the last 3 years of operation indicate a high ratio of 74–85 % and 91–99 % of data judged as good by those strict checks for stations SON and the other stations, respectively.
A detailed combined uncertainty analysis of the shortwave radiation fluxes
at all five ARAD stations indicates that the combined expanded uncertainties
range between 1.46 and 2.87 % (or 15 to 29 W m
Besides meeting the requirements of a national climate monitoring network, ARAD data can serve the international climate research and impact communities as well as practical applications related to solar energy. Thus, ARAD provides comprehensive, high-quality and consistent irradiance reference data for comparison with satellite measurements, model simulations (regional climate models, numerical weather prediction models, energy balance models) and other measurement systems (e.g., sunshine duration or radiation).
Here we summarize the quality criteria used in the automated QC (see Sect. 3.2.2) of the ARAD data that are not described in
the study of Long and Shi (2008).
Based on experience, for every 1 min time step For every 1 min time step For every 1 min time step To detect outliers we compare the difference of the actual 1 min
average value of all irradiances and the respective daily averages
(GLO
The ARAD project is financed in part by the Austrian Federal Ministry of Science, Research and Economy. The authors thank Martin Mair for his essential and valuable technical support since the start of the ARAD Project. Edited by: C. Ammann