1. Introduction
There has been remarkable improvement in the quality of numerical weather prediction (NWP) model output on the global, regional, and local scales in the past three decades. For example, the skill of upper-air forecasts of the global European Centre for Medium-Range Weather Forecasts (ECMWF) model has increased by more than 3 days within this period (Richardson et al. 2009). Limited-area models with grid spacings between 2 and 20 km provide additional skill, especially with regard to precipitation and wind forecasts. A problem for NWP at all scales, however, is the comparatively low skill within the nowcasting range (0–6 h), although recent developments based on variational analysis and latent heat nudging appear promising (Dixon et al. 2009; Zhao et al. 2009).
Nowcasting methods are capable of incorporating observations almost in real time, make use of the Eulerian or Lagrangian persistence of atmospheric processes, and are computationally efficient. They are superior to NWP models, typically up to a lead time of 2–3 h (Bowler et al. 2006, Pinto et al. 2010). Nowcasting algorithms are not derived from first principles like the dynamical equations of an NWP model, which makes it more difficult to extend them beyond the classical advection nowcast. Nevertheless, because of their near-real-time availability, nowcasts have become important in time-critical applications such as severe weather warnings, flood forecasting (Werner and Cranston 2009), or aviation forecasts (Pinto et al. 2010).
Most nowcasting systems focus on precipitation and on phenomena related to deep convection (Dixon and Wiener 1993; Hand 1996; Golding 1998; Pierce et al. 2000; Bowler et al. 2006; Feng et al. 2007; Schmeits et al. 2008; Dance et al. 2010). Analysis and nowcasting of near-surface temperature (Sun and Crook 2001) or wind (Crook and Sun 2004) have mainly been regarded as a means for predicting convective initiation and the development of deep convection (Wilson and Schreiber 1986; Wilson et al. 2004). There is, however, an increasing requirement for “nonclassical” nowcasts of quantities like temperature, humidity, or global radiation. There is also a need to extend the applicability of nowcasting methods to mountainous terrain. The Integrated Nowcasting through Comprehensive Analysis (INCA) system described in this paper is an attempt to reach these goals. The basic concept is to complement and improve upon NWP direct model output for a range of variables, using real-time observations and high-resolution topographic data. Outside the immediate nowcasting range the nowcasts are blended into downscaled NWP forecasts.
Since a large number of nowcasting systems have been developed in recent years [see Dance et al. (2010) for a brief summary], one may question the need for a new system. However, the need to provide nowcasts in a highly mountainous area like the Alps required the development of new methods. In the case of precipitation, for example, the INCA methodology is useful in areas where radar coverage is poor and/or inhomogeneous but station density is good. Some aspects of the INCA system have been previously documented. Steinheimer and Haiden (2007) investigated the benefits of using high-resolution analysis fields such as convective available potential energy (CAPE) and convective inhibition (CIN) for predicting intensity changes of convective precipitation. Kann et al. (2009) used the INCA method to calibrate NWP ensemble forecasts. The objective of this paper is to provide a more comprehensive description of the system and to present new methods for the treatment of orographic effects in nowcasting. Section 2 gives a brief overview of the main characteristics of the system and the input data used. The analysis part of INCA, which employs new methods of dealing with orographic effects at high spatial resolution, is described in section 3. A radar–rain gauge combination method that accounts for the inhomogeneous radar coverage in the eastern Alps is presented. Nowcasting methods are presented in section 4, and some verification results with regard to both analyses and forecasts are given in section 5.
2. The INCA system
a. General characteristics
INCA is a multivariable analysis and nowcasting system. It provides near-real-time analyses and forecasts of the fields given in Table 1. Its objective is to improve numerical forecast products in the nowcasting (0–4 h) and very short (up to about 12 h) ranges. It also adds value to NWP forecasts for up to 72 h through downscaling and bias correction. Spatial interpolation is based on distance weighting both in physical and variable space (potential temperature, precipitation). In the case of temperature, humidity, and wind, nowcasts start with a three-dimensional analysis based on a first guess obtained from NWP output, with observation corrections superimposed. For other fields the nowcast starts with an analysis that combines remote sensing and surface station data.
INCA analysis and forecast fields, their input, and their update frequency: NWP = output of an NWP model, SFC = surface station observations, RAD = radar data, and SAT = satellite data. Two-letter acronyms indicate INCA fields.
There is limited interdependency between the fields. In the nowcasting of temperature the cloudiness analysis and nowcast are taken into account. The surface cooling caused by convective cells due to the evaporation of precipitation enters the analysis and nowcasting of temperature. Additional, derived fields include convective parameters such as the lifted condensation level (LCL), or CAPE, as described by Steinheimer and Haiden (2007). Snowfall line and ground temperature are computed for nowcasts of precipitation type (snow, rain, snow–rain mix, freezing rain).
The topography is constructed from a bilinear interpolation to the 1-km INCA grid of the 30″ digital elevation dataset provided by the U.S. Geological Survey. No smoothing other than the one implicit in the bilinear interpolation is applied. The high resolution of 1 km is an essential characteristic of INCA. It enables the system to directly assimilate observations at most of the stations, since at this resolution the real elevation and exposition of a location come close to that given by the topography on the numerical grid. It allows us to resolve major Alpine valleys such that the modeled valley floor is close to the actual valley floor height. It was subjectively concluded that the resolution is sufficient to approximately reproduce slope inclinations on major valley sidewalls. In the Alps, the typical base-to-crest length scale is of the order of 5 km, which means that sidewall slopes extend over several grid intervals.
The main conceptual difference between the analysis part of INCA and the Austrian Vienna Enhanced Resolution Analysis system (VERA; Steinacker et al. 2006; Schneider et al. 2008) is that INCA relies on NWP model output and high-resolution remote sensing data to interpolate between observations, while VERA is model independent and based on the variational principle applied to higher-order spatial derivatives. It uses a fingerprint technique to incorporate conceptual or climatological information, or upscaled radar data.
b. Coordinate system
Several different INCA domains exist in central Europe. In this paper we refer to the Austrian operational domain, which has a mesh size of 1 km and covers an area of 600 km × 350 km, centered over the eastern Alps (Fig. 1a). In the vertical, a z system is used where z is the height above the “valley-floor surface” shown in Fig. 2. It is a spatially slowly varying reference surface (Fig. 1b), which is smooth compared to the actual topography and connects major valley floors (Haiden 1998). It separates the topography into a base topography and a relative topography and is computed by assigning to every grid point the minimum elevation found within a radius of 10 km. The resulting field is smoothed with a running-average window of 20 km × 20 km. Over flat terrain, the topography and valley-floor surface coincide. Other nowcasting systems that include the vertical dimension are usually based more directly on an NWP model and therefore employ a terrain-following coordinate (Dixon et al. 2009). The coordinate based on the valley-floor surface is needed for the downward extrapolation of three-dimensional NWP forecast fields into valleys, and as a reference height for the parameterization of vertical profiles of temperature and precipitation.
(a) Topography and (b) valley-floor surface of the Austrian INCA domain.
Citation: Weather and Forecasting 26, 2; 10.1175/2010WAF2222451.1
North–south topographic cross sections in a band of 20-km width in the middle of the domain (near 13.5°E). Also shown are the valley-floor surface (boldface) and the z-coordinate surfaces (dashed).
Citation: Weather and Forecasting 26, 2; 10.1175/2010WAF2222451.1
The vertical grid spacing is uniform with Δz = 200 m, and 21 levels including the surface, covering the lowest 4000 m above the local valley-floor surface. The coordinate system used in the wind analysis is a z system with horizontal coordinate surfaces intersecting the terrain. There are 32 levels at a constant spacing of Δz = 125 m. The irregular shape of the grid volumes intersecting the terrain (Fig. 3) is taken into account in the computation of divergence, which is part of the relaxation procedure (Steppeler et al. 2002).
Schematic representation of the coordinate system used in the INCA wind analysis, showing “shaved elements” (subterranean parts dotted) generated by the intersection of z surfaces with the terrain (boldface line).
Citation: Weather and Forecasting 26, 2; 10.1175/2010WAF2222451.1
c. Data sources
1) NWP model output
NWP fields are provided by the Austrian operational version of the Aire Limitée Adaptation Dynamique Développement International (ALADIN) limited-area model described by Wang et al. (2006). It has a horizontal resolution of 9.6 km, 60 levels in the vertical, and is run 4 times per day with a forecast range of 72 h. Postprocessed fields are available roughly 4 h after analysis time. Forecast fields used in INCA are geopotential, temperature, relative humidity, and wind components (3D fields), as well as 2-m temperature and relative humidity, 10-m wind components, precipitation, total cloudiness, low cloudiness, and ground temperature (2D fields). The analysis and nowcasting methods of INCA are not specific to ALADIN. The Swiss version, INCA-CH, for example, uses Consortium for Small-Scale Modeling (COSMO; Steppeler et al. 2003) fields as a first guess. However, some of the empirical parameters and coefficients used in the analysis procedures are potentially sensitive to NWP model resolution and may have to be recalibrated before use in another NWP system.
2) Surface station observations
The Central Institute for Meteorology and Geodynamics (Zentralanstalt für Meteorologie und Geodynamik; ZAMG) operates a network of ∼250 Teilautomatische Wetterstationen (TAWES) semiautomated weather stations in Austria. The network covers most of the topographic elevation range (100–3800 m), with its highest stations at Brunnenkogel (3440 m) and Sonnblick (3105 m). The average horizontal distance between stations is 18 km. The vertical distribution of stations is somewhat biased toward elevations less than 2000 m but there is a sufficient number of mountain stations to allow the construction of three-dimensional correction fields to the NWP model output. Meteorological observations used are 2-m temperature, relative humidity, dewpoint, 10-m wind speed and direction, precipitation amount, and sunshine duration. In addition to TAWES stations, the Austrian network of hydrometeorological stations provides real-time precipitation and temperature data at ∼100 locations (Fig. 4).
Stations used operationally in the hourly temperature and humidity analysis [filled circles represent TAWES and surface synoptic observation (SYNOP) stations; open circles represent hydrological stations].
Citation: Weather and Forecasting 26, 2; 10.1175/2010WAF2222451.1
The following measurement methods are used, with the accuracy obtained in the laboratory given in parentheses. Temperature is measured with linearized negative temperature coefficient (NTC) thermistors (±0.1°C), and relative humidity and dewpoint with capacitive sensors (±5% at humidities <90%) and dewpoint mirrors (±0.2°C), respectively. For wind, mechanical anemometers and sonic instruments are used (±0.5 m s−1 up to 5 m s−1, ±5% above). For precipitation measurement, TAWES stations are equipped with tipping-bucket rain gauges and weighting rain gauges (±10%). Sunshine duration is obtained from heliometer measurements (±2%). In the case of precipitation, there are additional uncertainties and systematic errors due to wind effects, wetting, evaporation, and splashing, which typically amount to 5%–10% in summer, and 10%–50% in snow conditions [see Pappenberger et al. (2009) for a discussion of precipitation observation uncertainty].
3) Radar and satellite data
The radar composite used in INCA is generated from four C-band (5.33 cm) radars operated by Austria’s civil aviation authority. A fifth radar has recently been put into operation but it is not yet part of the composite. The dataset is supplemented with radar data from neighboring countries. Due to the mountainous topography, radar data quality is poor in several areas in western Austria, especially during the wintertime when much of the precipitation growth takes place at low levels. Ground clutter has already been statistically filtered when the data arrives at ZAMG. Brightband effects and range-dependent attenuation are addressed in the precipitation analysis. The maximum constant-altitude plan position indicator (max-CAPPI) product is used as the main radar input for INCA.
The Meteosat Second Generation (MSG) satellite products used in INCA are “cloud type” (Derrien and Le Gléau 2005), which consists of 17 categories, and the visible satellite image (VIS). Cloud type differentiates between three cloud levels (low, medium, and high) as well as different degrees of opaqueness. It also diagnoses whether clouds are more likely of a convective or stratiform nature. The accuracy of the cloud type product is high (93%) for low clouds, for high clouds (79%), and semitransparent clouds (92%), but rather low (37%) for midlevels clouds (Météo-France 2009). The VIS image is used to downscale the infrared-based, and thus coarser-resolution, cloud types during the day. Both satellite products are used at a 15-min time step.
d. Surface-layer index
INCA surface-layer index as computed by (1). Values greater than 0.9 characterize valley and basin floors, and generally flat or hilly terrain. Values less than 0.1 characterize elevated mountain slopes, ridges, and peaks.
Citation: Weather and Forecasting 26, 2; 10.1175/2010WAF2222451.1
3. Analysis methods
a. Temperature and humidity
In parts of a valley atmosphere not represented in the NWP model, the first-guess temperature profile is constructed by performing a downward shift of the PBL temperature profile to the level of the valley-floor surface.
Citation: Weather and Forecasting 26, 2; 10.1175/2010WAF2222451.1
The parameter DTSCALE is a surface-layer temperature surplus (or deficit) scale that is by default set to a minimum value of 1 K but may have larger values depending on the insolation and wind speed. The surface-layer index ISFC in (5a) and (5b) reduces the amount of correction attributed to the surface layer at slope, ridge, and peak locations (Fig. 5).
Example of an INCA temperature analysis during a typical mixed foehn–inversion situation at 0800 UTC 21 Nov 2007. Cold-air pools are present in valleys and basins, while foehn-induced subsidence and mixing create high temperatures at elevations between 1500 and 2000 m.
Citation: Weather and Forecasting 26, 2; 10.1175/2010WAF2222451.1
An additional effect that is considered is the evaporative cooling due to precipitation, which becomes most significant in the case of isolated convective cells. The amount of cooling is parameterized thermodynamically as a function of precipitation amount and boundary layer humidity.
b. Wind
c. Precipitation
The INCA precipitation analysis incorporates station data, radar data, and elevation effects. It is an attempt to combine the quantitative accuracy (compared to radar) of rain gauge measurements with the spatial accuracy provided by the radar field. Any combination method has to deal with the weaknesses of both types of observations as well, namely the limited representativeness and insufficient density of rain gauge stations and the quantitative uncertainty of precipitation estimation by radar. These problems are further enhanced by the Alpine topography. The individual steps of the analysis procedure are given below.
1) Interpolation of station data
Example of a 15-min INCA precipitation analysis based on the combination of rain gauge and radar data (1930 UTC 19 Jun 2009): (top left) pure rain gauge interpolation, (top right) uncorrected max-CAPPI radar composite, (bottom left) corrected radar field, and (bottom right) final INCA precipitation analysis. Rain gauge locations are shown by black dots, and locations of radar stations are indicated by red triangles.
Citation: Weather and Forecasting 26, 2; 10.1175/2010WAF2222451.1
2) Climatological scaling of the radar data
3) Rescaling of radar data using the latest observations
The climatologically scaled radar field is rescaled based on the comparison between station observations and the radar field at the station location. A maximum spatial shift of 4 km in either direction is permitted to take into account effects due to the finite settling time of hydrometeors, wind drift, and some uncertainty in the radar data localization. If a different radar pixel than the nearest one fits significantly better to the station observation, that value is used in subsequent calculations. A spatial shift vector is computed for every station location and interpolated to the grid using a distance-weighting algorithm. By applying this field of shift vectors to the radar precipitation field, it becomes slightly distorted to better match the rain gauge observations.
4) Final combination
5) Elevation dependence
The inclusion of elevation effects turned out to be crucial for a realistic estimation of the spatial distribution of precipitation in Alpine areas, especially for hydrological applications. Based on feedback from hydrological simulations that used INCA precipitation analyses as an input, it was possible to constrain and optimize this parameterization.
In a first step, a “station topography” zSTA(i, j) is created. It is the topography represented by the reporting stations and is computed by IDS interpolation of INCA elevations at the station locations. Similarly, a “valley precipitation” field PVAL(i, j) is computed by an IDS interpolation identical to (16), but in which the summation extends only over those stations that are located not more than 300 m above the valley-floor surface. This field represents the reference precipitation at the valley-floor level. In the following we drop the gridpoint indices (i, j) for better legibility, where it is understood that each dependent variable is a gridded field.
d. Precipitation type
Due to limited horizontal resolution, direct NWP model output of precipitation type may be of limited use in steep terrain and deep, narrow valleys where the model topography differs too much from the actual topography. This is an issue in flood forecasting, where the amount of precipitation stored as snow on the ground and not immediately contributing to runoff needs to be estimated, and in road weather forecasts as well. For such applications a distinction between rain and snow may be insufficient. In cases where the atmosphere is well mixed and the lapse rate close to moist adiabatic, the boundary between snowfall and rainfall will be relatively narrow. However, in more stable cases, or when the snowfall line has worked its way downward due to latent heat effects, there will be a broader height range with temperatures close to 0°C and associated conditions of snow–rain mix or wet snow.
In INCA the distinction between rain and snow is based on the vertical profile of the wet-bulb temperature Tw at each grid point, derived from the 3D temperature and humidity fields. Following Steinacker (1983) a snow–rain mix is assumed in the height range where 0°C ≤ Tw ≤ +2°C, marking the transition from snowfall into rain. Freezing rain is assumed to occur if rain falls into a surface layer with subzero air temperature, or if the ground temperature is below zero. In the latter case, however, the air temperature must not exceed a critical value set to +2°C. The analysis of ground surface temperature in INCA is based on observations of the +5-cm air temperature, −10-cm soil temperature, and 2-m air temperature. Outside the nowcasting range, the NWP forecast of ground surface temperature is used (corrected for the actual terrain height based on 2-m temperature).
e. Cloudiness
The INCA cloudiness analysis is actually an analysis of the insolation fraction SP as measured by surface stations, where MSG cloud-type data are used for spatial interpolation. The approach is similar to the INCA precipitation analysis in the sense that no NWP model output is used in the analysis, only in the forecast, and remote sensing data are calibrated using station observations. Also, a certain spatial shift (5 km) between a station location and a satellite pixel is allowed in order to take into account uncertainties in the timing and satellite imagery navigation as well as the slanted path of the sunbeam. Specific to cloudiness is the method of performing spatial interpolations of station observations separately for each set of stations that are located beneath the same cloud type. Thus, the relationship between cloud type and sunshine fraction on a given day does not get “smeared out” in areas where different cloud types are bordering each other. A weak smoothing is applied to the resulting field, consistent with the spatial shift between the station and satellite pixel described above. During nighttime, when no station observations of sunshine fraction are available, the SP field is constructed by combining cloud-type information with a scaling based on a monthly varying climatology instead of real-time observations. The relative mean absolute error of the cloudiness analysis, as obtained by cross validation, is about 18%. The cloudiness analysis provides input for the INCA global radiation analysis (Haiden et al. 2009).
4. Nowcasting methods
a. Lagrangian persistence
NWP filtering of motion vectors according to (30). The boldface vector is the NWP model wind, and black arrows show examples of permitted motion vectors. Vectors outside the elliptic area are rejected. The dotted line shows the wind speed deviation parameter, Δ.
Citation: Weather and Forecasting 26, 2; 10.1175/2010WAF2222451.1
To obtain a continuous sequence of forecast fields, a transition from the extrapolation forecast to the NWP forecast is constructed through a prescribed weighting function that gives full weight to the extrapolation forecast during the first 2 h and decreases linearly to zero at 6 h. Attempts to improve upon the fixed weighting by making the time scale of the transition dependent on the magnitudes of NWP and nowcasting errors has as yet not shown any benefit.
Since the INCA precipitation analysis includes an elevation dependence, the extrapolation method described above cannot be applied directly to the final analysis but must be applied to intermediate fields that do not contain the elevation dependence. Thus, the “valley” precipitation is advected, and the elevation dependence is newly computed for each nowcast step. This ensures that the topographic patterns in the precipitation field remain stationary. As a result, precipitation patterns in the nowcast intensify as they move over higher terrain and weaken when they move over a valley. Note that this only applies to the station interpolation component of the combined field, not the radar component. Thus, convective cells are largely unaffected by this orographic effect in the extrapolation.
Nowcasts of cloudiness, or rather, sunshine fraction, are based on cloud motion vectors derived from consecutive visible (during daytime) and infrared (during nighttime) satellite images. During sunrise and sunset, a time-weighted combination of both motion vector fields is used. As in the case of precipitation, the extrapolation is performed for a forecast range of up to +6 h using a 15-min time step. The nowcasting of cloudiness includes a consistency check with the precipitation nowcast.
b. NWP model trend
As in the case of precipitation and cloudiness, the temperature nowcast is blended into the NWP forecast. The time scale of the weighting function depends on static stability, varying from 3 h under well-mixed conditions to values of up to 12 h for pronounced inversion conditions. This dependency accounts for observed variations in the persistence of temperature forecast errors under different synoptic conditions. The humidity nowcast is analogous to temperature, without the cloudiness correction factor. Wind is treated similarly to temperature and humidity. The NWP trend is put on top of the analysis, and a fixed weighting function provides a transition into the NWP forecast.
5. Verification
There has been a renewed level of interest in the critical analysis of methods and measures used in meteorological forecast verification. Classical verification measures and scores are scrutinized with regard to statistical properties such as propriety and equitability (Mason 2008), and new spatial verification methods are being developed and compared (Gilleland et al. 2009). Here, we use one of the new object-oriented methods, Structure–Amplitude–Location (SAL; Wernli et al. 2008) to evaluate the INCA forecast skill and compare it with that of the ALADIN limited-area model, in addition to classical point verification. The analysis skill of INCA is quantified by cross validation based on single-station denial, looking at mean error (bias), mean absolute error (MAE), and root-mean-square error (RMSE).
a. Analysis verification
Evaluations have been performed for the analysis fields of temperature, precipitation, and wind. The verification is based on two months (July 2009 and January 2010), representing a summer and a winter month. Analyses of wind and temperature are at hourly time intervals, while the precipitation analysis has a temporal resolution of 15 min.
1) Temperature
For the cross validation of INCA temperature analyses, a total of 174 stations were used. The values of bias, MAE, and RMSE for different fields are given in Table 2 for July 2009 and January 2010. The bias in the temperature is small for both months over the averaging period. This is true not just for the average over all stations but also for individual stations, where the bias is typically less than 0.2°–0.3°C. MAE and RMSE are ∼20% higher in the winter month, due to the more frequent occurrence of stable stratifications and inversions. The distribution of individual analysis errors is shown by the scatterplot in Fig. 10. The majority of points lie within an interval of ±3°C from the diagonal, but differences of up to 5°–6°C do occur sometimes. Such large errors are associated with stable nighttime conditions and are found mainly in deep Alpine valleys, at locations where the distance to the nearest TAWES station is relatively large. Data points below 10°C in the diagram are mostly from mountaintop stations, where a stable boundary layer does not form. Consequently, maximum errors are smaller there.
Distribution of individual analysis–observation pairs in the INCA temperature cross validation for July 2009. For better legibility only data from every third analysis (0000, 0300, … , 2100 UTC) are shown.
Citation: Weather and Forecasting 26, 2; 10.1175/2010WAF2222451.1
Results of the INCA analysis cross validation for a summer and a winter month. Shown are the bias (mean error), MAE, and RMSE. For the wind components and for precipitation relative values are given as well.
2) Wind
The cross validation of the INCA wind has been performed for a limited region only. Due to the nonlocal effects of the divergence reduction algorithm, a full analysis had to be computed for each single-station denial of each analysis. The moderately hilly northern part of Austria (47.7°–49°N, 13°–17°E) was chosen as a verification area. Analyses are based on the 0000 UTC runs from ALADIN and all available station data. For the cross validation a sample of 36 stations was used. Values of bias, MAE, and RMSE have been calculated separately for the two wind components (Table 2). Since the magnitude of the analysis error strongly depends of the wind magnitude, relative measures with respect to the mean wind at each station are also given. They are computed as differences between the analyzed and observed winds at each station, divided by the observed period-mean wind speed at that station, and then averaged over all stations. Relative MAEs are on the order of 50%, and somewhat higher in July than in January. Like in the case of precipitation, as discussed below, there is a rather large uncertainty in the analysis of point values. The absolute MAE is about 1 m s−1.
3) Precipitation
In the case of precipitation, Table 2 shows values for the whole of Austria. With 257 stations used in the cross validation, the data density is higher than for the other parameters. While the bias equals zero again both for July and January, the RMSE reacts to the increased spatial variability associated with summertime convective precipitation. The relative measures show that the error with regard to point values for short-duration analyses (15 min) is rather large: around 50% in summer, and more than 100% in winter. This can be compared to results of Skok and Vrhovec (2006) who found point errors of up to 50% in cross validation for 24-h totals, relatively independent of the interpolation method. However, operational use of INCA precipitation analyses for hydrological simulations in small catchments of the order of 100 km2 indicates that areal averages are significantly more reliable than point values. Cross validations were also performed for two subdomains, one in the western part of Austria (Tyrol) and one in the east (Lower Austria). Differences between the mountainous west and the lowlands in the east are small in winter. In summer, the RMSE is significantly higher in the lowlands even though the MAE is virtually the same in both areas. This is because the RMSE is more sensitive to outliers, and there were an unusually high number of localized, heavy convective precipitation events in the eastern parts of Austria during July 2009. Figure 11 illustrates the correspondence between withheld observations and analyses for July 2009. As expected, the correlation is significantly weaker than in the case of temperature (Fig. 10), and errors of a factor 2 occur quite frequently in the analysis of 15-min precipitation amounts.
Distribution of individual analysis–observation pairs in the INCA 15-min precipitation cross validation for July 2009. For better legibility only data from every 12th analysis (0000, 0300, … , 2100 UTC) have been included in the plot. The line pattern is due to the finite resolution (0.1 mm) of the rain gauge data. Only values ≥0.1 mm are shown.
Citation: Weather and Forecasting 26, 2; 10.1175/2010WAF2222451.1
b. Nowcast verification
The SAL verification method (Wernli et al. 2008) has been applied to INCA precipitation forecasts for different subdomains in Austria for the same two months as in the cross validation. Lead times up to 12 h are considered. The location score shows little variability both for summer and winter. The average value over a mean (summer or winter) 12-h forecast period is approximately 0.3, with slightly lower values at the beginning of the period. The amplitude score also shows a rather similar pattern of behavior for summer and winter and for the different subdomains. There is an underestimation (about −0.5) of precipitation amounts in the first 6 h and an overestimation (about 0.5) in the second half of the forecast period due to the increasing influence of the NWP model for longer lead times. The structure score, which takes into account the shape and extension of precipitation fields, shows more distinct changes. For the first 4–5 h (i.e., the nowcasting period) it is close to 0, which means the structure is well represented. For longer lead times it increases to values between 1.0 and 1.5, indicating the structural loss in the INCA precipitation forecasts, when the kinematically extrapolated fields merge into NWP forecasts outside the nowcasting range. For lead times of up to 30 h, a more detailed evaluation is presented in Wittmann et al. (2010).
In terms of classical point verification against station observations, Fig. 12 shows the MAEs of the INCA and ALADIN temperature forecasts for lead times up to 12 h. The INCA MAE is significantly smaller than the ALADIN MAE over the first 3 h in summer and the first 6 h in winter. The residual benefit of INCA outside the nowcasting range is due to the downscaling procedure illustrated in Fig. 6. For precipitation, substantial improvement is usually found for the first 2–3 h.
INCA (dashed lines) and ALADIN (solid lines) MAEs of temperature, averaged over all stations, for July 2009 (gray) and January 2010 (black).
Citation: Weather and Forecasting 26, 2; 10.1175/2010WAF2222451.1
6. Summary and conclusions
The central European multivariable analysis and nowcasting system INCA adds value to NWP forecasts by providing analyses, nowcasts, and downscaled forecasts. Analysis methods for temperature and precipitation include topographic effects, which makes the system applicable to mountainous terrain. For temperature and humidity it takes into account the ability of the local topography to support a pronounced surface layer. For precipitation, the local elevation dependence is included as part of a nonlinear combination of rain gauge and radar data. Nowcasting of precipitation and cloudiness is based on the classical advection method with motion vectors derived from the cross correlation of previous analyses. Nowcasting of temperature, humidity, and wind is Eulerian, and based on modified trends of the NWP model.
Based on cross validation of INCA analyses, verification of INCA nowcasts, and qualitative assessments during day-to-day applications of the system, the following main conclusions can be drawn.
Temperature is analyzed with an average accuracy of 1°–1.5°C, with the smallest errors occurring in lowland areas and at exposed mountain stations. In Alpine valleys the mean absolute error is larger (1.5°–2.5°C), mostly due to the lack of precise information about inversion heights.
A combination of radar and rain gauge information and the inclusion of a parameterization of elevation effects is essential for providing useful precipitation analyses in the eastern Alpine region. However, the relationship between scaled radar values and rain gauge observations shows rather high variability, and the mean relative analysis error for 15-min precipitation amounts is on the order of 50%–100%.
Both wind and precipitation analyses suffer from representativeness problems, due to substantial subgrid-scale variations in these fields, in spite of the 1-km resolution of the analysis system.
Comparison of INCA and NWP forecast skill shows that substantial improvements due to nowcasting are achieved in the first 2–3 h for precipitation, and in the first 6 h for temperature.
Outside the nowcasting range, INCA improves the NWP temperature forecast through downscaling by 5%–10%.
The development of new, and the modification of existing, analysis and nowcasting methods by accounting for topographic effects allows nowcasting methodology to be applied to Alpine terrain. However, many processes relevant for nowcasting, such as slope and valley winds, or the vertical temperature profile in inversion cases, are not adequately resolved even by the rather dense observational network in the eastern Alps. This limits our current analysis and nowcasting capability in such areas. Further improvements are unlikely to come from denser station networks but rather from more comprehensive use of satellite data, and from high-resolution NWP. The European Union project INCA-CE—A Central European Initiative in Nowcasting Applications, which began in 2010, will provide a framework for such developments. Emphasis will be on improved wind downscaling based on high-resolution NWP and on enhanced use of satellite data in the nowcasting of deep convection. Situation-dependent blending of the nowcast into the NWP forecast will be another priority in future INCA development.
Acknowledgments
We are grateful to three anonymous reviewers for constructive comments that helped to improve this paper. We want to thank Klaus Stadlbacher, Alexander Beck, and Martin Auer for their work on the operational aspects of INCA. We would also like to thank the hydrological authorities of Lower Austria and Salzburg for funding projects that led to improvements in the INCA precipitation and temperature nowcast.
REFERENCES
Bowler, N., Pierce C. , and Seed A. , 2006: STEPS: A probabilistic precipitation forecasting scheme which merges an extrapolation nowcast with downscaled NWP. Quart. J. Roy. Meteor. Soc., 132, 2127–2155.
Crook, N. A., and Sun J. , 2004: Analysis and forecasting of the low-level wind during the Sydney 2000 Forecast Demonstration Project. Wea. Forecasting, 19, 151–167.
Dance, S., Ebert E. , and Scurrah D. , 2010: Thunderstorm strike probability nowcasting. J. Atmos. Oceanic Technol., 27, 79–93.
Derrien, M., and Le Gléau H. , 2005: MSG/SEVIRI cloud mask and type from SAFNWC. Int. J. Remote Sens., 26, 4707–4732.
Dixon, M., and Wiener G. , 1993: TITAN: Thunderstorm Identification, Tracking, Analysis and Nowcasting—A radar-based methodology. J. Atmos. Oceanic Technol., 10, 785–797.
Dixon, M., Li Z. , Lean H. , Roberts N. , and Ballard S. , 2009: Impact of data assimilation on forecasting convection over the United Kingdom using a high-resolution version of the Met Office Unified Model. Mon. Wea. Rev., 137, 1562–1584.
Feng, Y., Wang Y. , Peng T. , and Yan J. , 2007: An algorithm on convective weather potential in the early rainy season over the Pearl River delta in China. Adv. Atmos. Sci., 24, 101–110.
Gilleland, E., Ahijevych D. , Brown B. G. , Casati B. , and Ebert E. E. , 2009: Intercomparison of spatial forecast verification methods. Wea. Forecasting, 24, 1416–1430.
Golding, B. W., 1998: Nimrod: A system for generating automated very short range forecasts. Meteor. Appl., 5, 1–16.
Haiden, T., 1998: Analytical aspects of mixed-layer growth in complex terrain. Preprints, Eighth Conf. on Mountain Meteorology, Flagstaff, AZ, Amer. Meteor. Soc., 368–372.
Haiden, T., and Pistotnik G. , 2009: Intensity-dependent parameterization of elevation effects in precipitation analysis. Adv. Geosci., 20, 33–38.
Haiden, T., Kann A. , Pistotnik G. , Stadlbacher K. , and Wittmann C. , 2009: Integrated Nowcasting through Comprehensive Analysis (INCA)—System description. ZAMG Rep., Zentralanstalt für Meteorologie und Geodynamik, Vienna, Austria, 60 pp. [Available online at http://www.zamg.ac.at/fix/INCA_system.pdf.]
Hand, W. H., 1996: An object-oriented technique for nowcasting heavy showers and thunderstorms. Meteor. Appl., 3, 31–41.
Kann, A., Wittmann C. , Wang Y. , and Ma X. , 2009: Calibrating 2-m temperature of limited-area ensemble forecasts using high-resolution analysis. Mon. Wea. Rev., 137, 3373–3387.
Kann, A., Seidl H. , Wittmann C. , and Haiden T. , 2010: Advances in predicting continental low stratus with a regional NWP model. Wea. Forecasting, 25, 290–302.
Mason, S. J., 2008: Understanding forecast verification statistics. Meteor. Appl., 15, 31–40.
Météo-France, 2009: Validation report for cloud products (CMa-PGE01 v2.0, CT-PGE02 v1.5 & CTTH-PGE03 v2.0). SAF/NWC/CDOP/MFL/SCI/VR/03, Issue 1, Rev. 0, 30 pp.
Pappenberger, F., Ghelli A. , Buizza R. , and Bodis K. , 2009: The skill of probabilistic precipitation forecasts under observational uncertainties within the generalized likelihood uncertainty estimation framework for hydrological applications. J. Hydrometeor., 10, 807–819.
Pierce, C. E., Hardaker P. J. , Collier C. G. , and Haggett C. M. , 2000: GANDOLF: A system for generating automated nowcasts of convective precipitation. Meteor. Appl., 7, 341–360.
Pinto, J., Dupree W. , Weygandt S. , Wolfson M. , Benjamin S. , and Steiner M. , 2010: Advances in the Consolidated Storm Prediction for Aviation (CoSPA). Preprints, 14th Conf. on Aviation, Range, and Aerospace Meteorology, Atlanta, GA, Amer. Meteor. Soc., J11.2. [Available online at http://ams.confex.com/ams/pdfpapers/163811.pdf.]
Richardson, D. S., and Coauthors, 2009: Verification statistics and evaluations of ECMWF forecasts in 2008–2009: ECMWF Tech. Memo. 606, 9 pp.
Schmeits, M. J., Kok K. J. , Vogelezang D. H. P. , and Westrhenen R. M. V. , 2008: Probabilistic forecasts of (severe) thunderstorms for the purpose of issuing a weather alarm in the Netherlands. Wea. Forecasting, 23, 1253–1267.
Schneider, S., Chimani B. , Kaufmann H. , Bica B. , Lotteraner C. , Tschannett S. , and Steinacker R. , 2008: Nowcasting of a supercell storm with VERA. Meteor. Atmos. Phys., 102, 23–36.
Shen, S. S. P., Dzikowski P. , Li G. , and Griffith D. , 2001: Interpolation of 1961–97 daily temperature and precipitation data onto Alberta polygons of ecodistrict and soil landscapes of Canada. J. Appl. Meteor., 40, 2162–2177.
Sherman, C. A., 1978: A mass-consistent model for wind field over complex terrain. J. Appl. Meteor., 17, 312–319.
Skok, G., and Vrhovec T. , 2006: Considerations for interpolating rain gauge precipitation onto a regular grid. Meteor. Z., 15, 545–550.
Steinacker, R., 1983: Diagnose und prognose der schneefallgrenze. Wetter Leben, 35, 81–90.
Steinacker, R., and Coauthors, 2006: Downscaling meteorological information over complex terrain with the fingerprint technique by using a priori knowledge. Mon. Wea. Rev., 134, 2758–2771.
Steinheimer, M., and Haiden T. , 2007: Improved nowcasting of precipitation based on convective analysis fields. Adv. Geosci., 10, 125–131.
Steppeler, J., Bitzer H.-W. , Minotte M. , and Bonaventura L. , 2002: Nonhydrostatic atmospheric modeling using a z-coordinate representation. Mon. Wea. Rev., 130, 2143–2149.
Steppeler, J., Doms G. , Schättler U. , Bitzer H. , Gassmann A. , Damrath U. , and Gregoric G. , 2003: Meso-gamma-scale forecasts using the nonhydrostatic model LM. Meteor. Atmos. Phys., 82, 75–96.
Sun, J., and Crook N. A. , 2001: Real-time low-level wind and temperature analysis using WSR-88D data. Wea. Forecasting, 16, 117–132.
Wang, Y., Williamson C. , Garvey D. , Chang S. , and Cogan J. , 2005: Application of a multigrid method to a mass-consistent diagnostic wind model. J. Appl. Meteor., 44, 1078–1089.
Wang, Y., Haiden T. , and Kann A. , 2006: The operational limited area modelling system at ZAMG: ALADIN-AUSTRIA. Österreichische Beiträge zu Meteorologie und Geophysik, Vol. 37, Zentralanstalt für Meteorologie und Geodynamik, 33 pp.
Werner, M., and Cranston M. , 2009: Understanding the values of radar-rainfall nowcasts in flood forecasting and warning in flashy catchments. Meteor. Appl., 16, 41–55.
Wernli, H., Paulat M. , Hagen M. , and Frei C. , 2008: SAL—A novel quality measure for the verification of quantitative precipitation forecasts. Mon. Wea. Rev., 136, 4470–4487.
Wilson, J. W., and Schreiber W. E. , 1986: Initiation of convective storms by radar-observed boundary layer convergence lines. Mon. Wea. Rev., 114, 2516–2536.
Wilson, J. W., Ebert E. E. , Saxen T. R. , Roberts R. D. , Mueller C. K. , Sleigh M. , Pierce C. E. , and Seed A. , 2004: Sydney 2000 Forecast Demonstration Project: Convective storm nowcasting. Wea. Forecasting, 19, 131–150.
Wittmann, C., Haiden T. , and Kann A. , 2010: Evaluating multi-scale precipitation forecasts using high resolution analysis. Adv. Sci. Res., 4, 89–98.
Zhao, Q., Cook J. , Xu Q. , and Harasti P. R. , 2009: Improving short-term storm predictions by assimilating both radar radial-wind and reflectivity observations. Wea. Forecasting, 23, 373–391.
