A parameterization of runway visual range (RVR) has been developed using relevant meteorological parameters such as visibility (Vk), relative humidity (RH), temperature (T), precipitation intensity (PI), and precipitation type (PT) measured in years between 2009 and 2011 at Toronto Pearson International Airport during the Canadian Airport Nowcasting Project. The FD12P probe measured PI, Vk, and PT. The observed Vk and PT were tested against data reported by hourly surface observations (SAs). The measured Vk has correlated well with the SA with a correlation coefficient (r) of 0.76 for Vk < 5 km, but the FD12P underestimated visibility by about 20% with a mean difference (MD) of about 196 m. For Vk < 2 km, the FD12P overestimated visibility by about 7% with an MD of 60 m. The SA reported slightly more snow events—22% as compared to 17%—but the FD12P reported many more snow grain cases than the SA. Both the SA and the FD12P reported rain at similar frequency—4% and 5%, respectively. Using a theoretical approach, a parameterization that can be used to determine RVR as a function of Vk has been developed. Using the observed T, RH, and dewpoint temperature (Td), a new parameterization for predicting Vk/RVR in fog has been also developed. These parameterizations agreed with observations (r ≈ 0.8). The parameterizations have been tested using the Canadian Environmental Multiscale Regional model. The results show that when PI, RH, and T are reasonably predicted and the fog events are correctly diagnosed, the model can be used to forecast RVR.
Reduced visibility due to snow, rain, fog, and haze is an important consideration in the landing and takeoff of aircraft. Closing a major airport and diverting incoming traffic because of reduced visibility is a costly operation. Even if the airport may not be closed, under marginal visibility conditions the safety of airport operations is diminished. Takeoff and landing of aircrafts at runway locations in some major airports are normally managed using a site-specific visibility measurement called runway visual range (RVR). In Canada, low-visibility operations are in effect when the measured RVR is between 600 ft (182.9 m) and 1200 ft (365.8 m). Arrivals and departures are not authorized at all for RVR below 600 ft. RVR is an instrumentally derived value, based on standard calibrations, that represents the horizontal distance a pilot will see down the runway from the approach end. It is based on the sighting of either high-intensity runway lights (Va) or on the visual contrast of other targets (e.g., a black object) (Vk)—whichever yields a greater visual range. RVR is reported in increments of 100 ft (30.5 m) up to 1000 ft (304.8 m), increments of 200 ft (61.0 m) from 1000 to 3000 ft (914.4 m), and increments of 500 ft (152.4 m) above 3000–6000 ft. RVR is not normally reported for ranges greater than 6000 ft.
Traditionally, visibility measured using instruments such as the Vaisala FD12P is represented by Vk, as mentioned earlier. Currently, some numerical weather prediction (NWP) models also forecast Vk, but no systematic way of forecasting RVR has been fully developed. This is partly due to the fact that RVR prediction not only depends on atmospheric extinction (β), but also on runway light intensity (I) and background light (BL), which are very difficult to forecast. This will be discussed in more detail later. In this paper, new parameterizations that can be used to forecast RVR under snow, rain, and fog conditions will be developed and tested using direct observations of RVR and visibility collected at Toronto Pearson International Airport (CYYZ) during the Canadian Airport Nowcasting (CAN-Now) research project (Isaac et al. 2011).
Runway visual range, visibility, precipitation intensity, and other standard meteorological observation data were collected at CYYZ, Ontario, Canada, starting in 2007 as part of the CAN-Now project. The main CAN-Now instrument site is shown in Fig. 1. RVR data are measured at all runway locations indicated in Fig. 1 and reported every 5 min in real time on the NAV CANADA aviation weather website (http://atm.navcanada.ca/atm/iwv/CYYZ). Hourly and subhourly weather observations were reported by an observer. The data presented in this paper were collected in the periods from November to December 2009, January to April 2010, and January to May 2011.
The instruments at the CAN-Now site include a Campbell Scientific CR3000 (CR3k) data acquisition system equipped with a relative humidity (RH) and air temperature (T) probe (HMP45C212). The FD12P measures precipitation, precipitation type, and visibility, and is also equipped with an LM21 that measures BL (see Boudala and Isaac 2009 for further description of the instrument). The accuracy of the RH probe at 20°C is ±2% (for RH = 0%–90%) and ±3% (for RH = 90%–100%). The accuracy of temperature measurement is ±0.1% (for T = −50° to 50°C).
b. Visibility, relative humidity, and dewpoint temperature
The instruments measuring RH, T, and visibility report 1-min-averaged values, but the dewpoint temperature (Td) is not directly measured except for ones reported hourly or in special observations by an observer. Thus, 1-min Td is calculated using the following well-known approximation, given as
where T is in degrees Celsius and RH is in percent.
To compare 1-min-averaged data (CR3k) with those reported by hourly surface observation (SA), the entire dataset was interpolated to common time intervals; these are shown in Fig. 2. Figure 2a shows RH CR3k plotted against RH SA. Similar plots of the ambient T and Td measurements are shown in Figs. 2c and 2d, respectively, and the visibilities are shown in Fig. 2b. There is a significant bias between the two RH measurements near saturation (100%). The RH CR3k never exceeded 97.5% even during heavy fog while the observer reported 100%, and hence this is a concern for visibility parameterization in fog. While this is within the uncertainty of the measurement (3%), visibility compared more favorably to changes in RH SA over that of RH CR3k when the relative humidity is near saturation during fog. This will be discussed more later. Therefore, the 1-min RH CR3k data have been adjusted accordingly. The comparisons of T and Td are much better than RH (Figs. 2c,d), but the 1-min Td is determined using Eq. (1). Generally, the measured visibility using the FD12P corresponds quite well to that reported by the observer, with a correlation coefficient near 0.76 for visibility less than 15 km (Fig. 2b). However, on average, the observer sees higher visibility by about 20% with a mean difference (MD) of about 700 m, which is quite significant. When compared for visibilities less than 5 km, although the correlation coefficient remained unchanged, the MD decreased to about 196 m, which is more reasonable. For visibility less than 1 km, the MD becomes smaller near 60 m, and this time the FD12P overestimates the visibility by about 7%. Hence, at very low visibilities, on average, the FD12P appears to overestimate the visibility. However, most of the discrepancy, particularly the quantization of the data, is probably related to the poor resolution of the visibility markers used by the observer and timing and spatial differences between the observer and the instrument reports. It should be noted that visibility can vary considerably over short time and distance scales.
c. Precipitation and precipitation type
The reduction of visibility/RVR is due to absorption and scattering (extinction) of light by particles of differing sizes that are normally associated with fog, haze, snow, and rain. Since scattering and absorption of light strongly depend on particle type (i.e., shape, size, and index of refraction), accurate prediction of extinction/visibility, and hence RVR, requires accurate prediction of the shape and size distribution of these particles. Currently, some of the more sophisticated NWP models are capable of forecasting snowfall rate and some low clouds and fog, but the model prediction of these quantities is still a big challenge. Therefore, before we attempt to develop some strategy for forecasting RVR, it is necessary to test the ability of the automated instruments, such as the FD12P, to identify precipitation phase. Figure 3 shows the observed frequency distribution of precipitation type based on the FD12P and the observer (Fig. 3a) and precipitation intensity (Fig. 3b) based on the FD12P measurements during the period described in section 2. In Fig. 3a, the FD12P data are interpolated to match the observer’s observation time. Based on the FD12P, no precipitation was detected for approximately 68% of the time and the observer reported a slightly lower no-precipitation frequency of 60%. The frequency of fog occurrence reported by the FD12P, which was close to 2% (Fig. 3a), could have been underestimated since the observer reported near 6%. The FD12P uses a very simple algorithm for diagnosing fog; that is, when there is no precipitation and the 10-min-averaged visibility is less than 1 km, the probe reports fog, but this may not be strictly correct. Fog may exist with light precipitation consisting of many different types of clouds and precipitating particles as indicated in Fig. 3a, and thus these are not reported by the FD12P. The observer also reported more snow events, near 22% compared to 17% that the FD12P reported. However, the FD12P reported many more snow grain (SG) cases than the observer (Fig. 3a), but the observer reported a few cases of snow grain and fog (SGF) that the FD12P did not see. However, both the observer and the FD12P reported rain events at similar frequency (near 4% and 5%, respectively), but the FD12P reported 12 times more drizzle (L) cases. It is possible that some of the drizzle cases reported by the FD12P were mixed with fog. The FD12P also reported some ice pellets (IP) and freezing drizzle (ZL) cases that the observer did not report. Based on the 1-min FD12P data, most precipitation intensities were quite low, with precipitation rates mainly less than 1 mm h−1 (Fig. 3b). These results indicate that automated instruments may have difficulty identifying precipitation type; hence, they should be used with some caution. It was assumed here that the observation provided by the human observer is the ground truth, but it should be recognized that the observer reports have some uncertainties in identifying precipitation phase and in the timely reporting of meteorological conditions as well.
3. Theoretical derivation of RVR
a. The Koschmieder method
Visibility based on the Koschmieder (1924) method, which is based on scattering of light by a black object that is being observed, is given as
where ɛ is the threshold visual contrast, which is usually taken to be 0.05. Since this threshold is a recommended value by the World Meteorological Organization (WMO), many of the visibility-measuring instruments, such as the FD12P probe, use this value. Since the expression given in Eq. (2) is derived based on natural daylight, it is sometimes referred to as “daytime visibility.”
b. The Allard method
In the absence of natural daylight, such as during nighttime, another method is used that is based on undirected artificial light, which is normally referred to as the Allard (1876) method; it is given as (see Boudala and Isaac 2009)
where ET is the threshold illuminance of the light source when it is just visible, and I is the luminous intensity of the light source. The luminous intensity I normally depends on the intensity of the light source. For typical airport operations, there are six different light intensity settings used: 0, 15, 120, 500, 2500, and 10 000 candela (cd). The threshold illuminance ET depends on the background light BL and is normally approximated as
The background light BL is directly measured for determination of RVR and this will be discussed in the next section.
c. Background light observation
Figure 4 shows time series of the observed frequency distribution of BL during daytime at the Pearson International Airport during years between 2009 and 2011 (see section 2 for time periods). The most frequent (32%) BL value is 1000 cd m−2, which is a value that is commonly associated with a normal day. Very bright days (BL > 12 000 cd m−2) occurred less than 2% of the time during this period.
Figure 5 shows times series of precipitation type and visibility (Fig. 5a), and background light and cloud amount (Fig. 5b) for four consecutive days (14–17 December 2009). During the night, the background light goes down to a value near 4 cd m−2, as indicated in Fig. 5b. When fog occurs, it is always associated with low visibility, as indicated in Fig. 5a; this has important implications on the intensity light settings that will be discussed in the next section. As indicated in Fig. 5, the effect of cloud cover on BL is relatively weak, at least for this winter case.
d. The parameterization of RVR as a function of Vk
Based on the earlier discussions, RVR can be given as
Following Boudala and Isaac (2009), Vk can be related to Va using Eq. (3). Equation (3) is nonlinear, but it can be numerically solved if proper values for BL and I are chosen. Runway lights can generally be set at six different intensity levels. In Canada, RVR reported by the Air Traffic Service (ATS) is based on the runway lights being set at intensity level 3 (500 cd) unless the lights are at level 4 (2500 cd) or level 5 (10 000 cd). This means that RVR is calculated assuming that the lights are at level 3 even if the lights are off (level zero) or at levels 1 or 2. The level 5 light setting can be requested by pilots during low-visibility conditions, but it is not typically used. The 10-min Meteorological Aerodrome Report (METAR) RVR, however, is based on the lights at level 5 for all conditions; therefore, only levels 3, 4, and 5 will be considered here.
Figure 6 shows numerical solutions of Eq. (3) for nighttime (BL = 6.8 cd m−2) (Fig. 6a), normal day (BL = 1000 cd m−2) (Fig. 6b), bright day (BL = 12 000 cd m−2) (Fig. 6c) using three runway light settings, and for runway light level 3 combining night and day times (Fig. 6d). Also shown in Fig. 6d are the results of Eq. (6) that will be discussed in more detail later. During the night (Fig. 6a), the nighttime visibility Va is much larger than the daytime visibility Vk, particularly at lower visibilities. However, the effect of the light intensity becomes weaker as the visibility decreases. The implication of this is that the choice of light intensity setting does not significantly affect the relationship between Va and Vk at low visibilities that may be due to heavy fog or snow. During the day (Figs. 6b,c), Vk is generally larger than Va except when the intensity is at level 5 and the visibilities are less than about 1.5 km. If the light setting is at level 3 (Fig. 6d), the only relevant curve is the nighttime curve because RVR is taken to be the maximum [Eqs. (4) and (6)]. Table 1 shows power-law relationships fitted to the solutions given in Fig. 6. If the light setting is known, it is possible to use these relationships to predict the RVR values at the assumed background light levels. However, in many cases the light intensity setting may not be known. Based on the earlier discussions and using the definition of RVR in Eq. (5), for most ATS operations in Canada, assuming normal day and night conditions at level 3 light intensity setting in Table 1, RVR may be approximated as
As indicated in Fig. 6d, during the daytime, the parameterization will follow the green line, and during nighttime, it will follow the red line. Note here that for low-visibility cases (visibility < 500 m), the parameterization closely follows the normal day line (blue dash), and otherwise assumes the RVR is the same as the daytime visibility Vk. During nighttime, the RVR is approximated by the blue curve—the nighttime visibility. This assumption will be tested in the following section using observation data.
4. Verification of the parameterization
a. Two case studies
Figure 7 shows the time series of visibility Vk measured using the FD12P; measured RVR at the runway locations 06R, 06L, 33R, 24R, and 24L (locations indicated in Fig. 1); and the parameterized RVR (RVRP) determined using Eq. (6) (Fig. 7a). The precipitation type is based on the FD12P and human observer (Fig. 7b), and precipitation intensity is based on the FD12P (Fig. 7c) for 25 November 2009 at CYYZ. To switch from nighttime to daytime conditions, the background light BL measurement was used. The visibility and RVR observation data had temporal resolutions of 1 and 5 min, respectively. The human observer typically reported every hour, but several special reports are also included. The visibility during the night near 0300 and 0900 UTC decreased to near 200 m because of a mixture of fog and drizzle. Note that the FD12P reported snow during this time period, which seems to be a false alarm based on the human observer and the observed T. The visibility/RVR improved during the day after the fog lifted, although there was occasional drizzle and precipitation, particularly during the day near 1600 UTC. Note that the observed visibility Vk is significantly lower than the observed RVR during the night, but similar to the observed RVR during the day. Considering that the parameterized RVR is based on an approximate solution, the agreement with the observed RVR at all runway locations is quite reasonable. Recall also that RVR is reported only when it is less than 6000 ft (1.83 km), since it is limited by the length of the runway and runway lights and hence this is not a typical situation. As a result, it is not possible to test the RVR parameterization at values higher than this observational limit.
Another similar example is given in Fig. 8. These observations were made at CYYZ on 25 October 2010. Similar to the previous example, the visibility/RVR during the night decreased to near 100/500 m because of fog, which lasted for almost 7 h from 0500 to 1200 UTC. The earlier reduction of visibility/RVR appeared to be due to both rain and fog, although the observer missed the rain event. Both the observed and parameterized RVR values during the reduced visibility period were significantly higher than the observed visibility, reaching a factor of 5 when the visibility was approaching 100 m. The parameterized RVR in this case also shows an agreement as compared to the observed RVR reported at all locations.
b. Using the entire dataset
To test the parameterization using the entire dataset, the measured visibility data were interpolated to the nearest RVR observation time. In Fig. 9 there are some examples of scatterplots of RVR measured at the runways showing the close proximity to each other and pointing in the same direction (Figs. 9a–c). Also shown is the measured visibility at the CAN-Now site compared against RVRs measured at three locations for nighttime and daytime conditions (Figs. 9d–f). There is a significant variability in RVR even for runways in close proximity and pointing in the same direction. On average, during the daytime the observed RVR is close to the observed visibility, but during the nighttime RVR is in greater agreement with the previous discussions. There is considerable scatter in the plots that may be attributed to variations in location, runway intensity light setting, and the RVR reporting scheme used, as discussed earlier.
Figure 10 shows scatterplots of the measured and parameterized RVRs at six different runway locations: 06L (Fig. 10a), 06R (Fig. 10b), 33L (Fig. 10c), 33R (Fig. 10d), 24L (Fig. 10e), and 24R (Fig. 10f). The parameterization agreed with observations with a correlation coefficient (r) near 0.8, which is very encouraging considering the variability of RVR shown in Fig. 9. The runways that are farther away from the visibility measurement location (not shown in this figure) showed relatively lower correlations, as would be expected. Quantization of the data in the plot may also be related to the RVR reporting scheme mentioned earlier.
Since RVR is parameterized as a function of Vk, accurate estimation of RVR highly depends on the accuracy of Vk, which may be measured or estimated based on meteorological parameters. For most forecasting/nowcasting applications, Vk is parameterized using model-predicted moisture variables, such as RH, during a foggy day or precipitation rate during snow or rain. Hence, identifying the precipitation type and understanding its impact on visibility is critical for visibility prediction. These issues will be discussed in following sections.
5. Parameterization of RVR in snow and fog
a. In snow
As discussed earlier, the prediction of RVR using Eqs. (3) and (4) is not trivial, but the parameterization of RVR as a function of Vk facilitates using the predicted Vk to estimate RVR. It has been shown that visibility in snow or extinction βs can be related to snowfall rate and T, following Boudala and Isaac (2009), as
where βs is in kilometers, T in degrees Celsius, and S is the snowfall intensity in millimeters per hour. Equation (7) can be converted to Vk in snow using Eq. (2) assuming ɛ = 0.05, and this will be referred to as visBI09 from here on. The application of Eq. (7) in NWP models will be discussed in section 6.
b. In fog
Prediction of RVR in fog is difficult since the prediction of Vk in fog is also difficult. Visibility in warm fog can be predicted if the microphysical parameters such as liquid water content (Eldrige1969; Tomasi and Tampieri 1976) and cloud particle number concentration (Gultepe et al. 2006, 2009) are known. For ice fog, visibility/RVR can be predicted if the ice water content and mean particle diameter of fog-forming particles are known (e.g., Boudala and Isaac 2009). However, most of the current operational NWP models are not capable of resolving the small-scale fog-forming dynamical and microphysical phenomena. The cloud parameterization schemes used in these models usually do not work well near the surface (see section 8 for further discussion). As an alternative, visibility may be related to RH, Td, and T. To explore this possibility, the data collected during the time periods mentioned in section 2 at CYYZ have been analyzed.
Figure 11 shows 1-min-averaged β measured using the FD12P, measured RH (Fig. 11a), and the dewpoint-to-ambient temperature ratio (Td/T) (Fig. 11b). The fog events were determined using the FD12P and the data from the human observer. There is a positive correlation between β and both RH and Td/T, particularly at higher values of Td/T and RH. Using the data shown in Figs. 11a,b, β due to fog can be parameterized as
where φ is defined as
and Td and T are the observed dewpoint and ambient temperature in kelvins, respectively. The correlation coefficient (r) is 0.75 only when RH was used. However, when both RH and φ are used, the correlation coefficient increased to 0.83 [see Eq. (8a)], the F statistical value, which is a measure of goodness of fit, increased from 11 457 to 15 769, and the estimated error variance decreased from 1.3 to 1 km—all indicating improvements. The statistical significance testing, or P value, remained at zero, indicating the statistical significance of both variables. The “min” expression included in Eq. (8b) is used in order to avoid infinity at the saturation point. As indicated in Fig. 11, the extinction β becomes very sensitive to changes in RH when the RH exceeds 95%; hence, accurate prediction or measurement of RH is very crucial for such a parameterization. Unfortunately, as discussed earlier, the accuracy of RH measurements becomes even smaller for RH > 90%. The application of this parameterization and the others discussed in the previous sections will be discussed in the following section.
6. NWP model application
The NWP model data used in this study were generated using the Canadian Global Environmental Multiscale Regional (GEM-Reg) model at 15-km resolution in its regional configuration (Côté et al. 1998a,b). This version of the model is described in more detail by Mailhot et al. (2006). The regional forecast model is run at the Canadian Meteorological Centre (CMC) 4 times a day (0000, 0600, 1200, and 1800 UTC), out to 48 h. The grid-scale condensation scheme is based on Sundqvist (1978). In this scheme, cloud formation actually begins before the grid-resolved humidity reaches saturation. The threshold relative humidity is 80% at the lowest model level, but varies in the vertical. Precipitation occurs instantaneously at the ground if the mixing ratio exceeds a certain threshold. In principle, fog may be related to the liquid water mixing ratios predicted at lower model levels. However, because of the coarse vertical and horizontal resolution of the model, the lower model level moisture field is poorly predicted and hence fog prediction is not always possible.
Precipitation type in the model is diagnosed based on the Bourgouin (2000) scheme using the vertical temperature profile. The solar and infrared radiation are based on Fouquart and Bonnel (1980) and Garand (1983). In this paper, visibility in the model is determined using the parameterizations for extinction as follows: for fog βf see Eq. (8) in this paper, for rain (R) is based on Marshall and Palmer’s (1948) exponential size distribution, and for snow (S) βs [Eq. (7)] is based on Boudala and Isaac (2009).
In the model, fog is diagnosed only if the total precipitation (P = S + R) is less than 0.5 mm h−1 and T > 0°C or P is less than 0.25 mm h−1 and T < 0°C, provided that RH > 97%. The relative humidity threshold used in the model for diagnosing fog has been arbitrarily chosen for this test. Some models may have a more sophisticated method for diagnosing or predicting fog events and hence any suitable method can be used. The visibility Vk is then calculated as
assuming that the threshold visual contrast is 0.05 in Eq. (2). The extinction due to both rain and snow βrs is given as βrs = βr + βs, and hence β in the parameterization is represented as
For visibility prediction in fog using Eq. (8), we have used the average values of the three bottom model levels of T and Td instead of the surface values.
Figure 12 shows time series of parameterized visibility visBI09: the visibilities measured (VkFD12P) and based on human observation (SA) and GEM-Reg (GEM-RegBI09) (Fig. 12a); precipitation type based on the FD12P and human observation (Fig. 12b); RVR derived based on the FD12P (RVRFD12P) and GEM-Reg using Eq. (6); observed RVRs in three locations (24R, 24L, and 33R) (Fig. 12c); and precipitation intensity based on the FD12P and GEM-Reg (Fig. 12d). The data displayed in the figure were collected on 5 and 6 February 2011 at CYYZ. In snow, visibility/RVR is closely linked to the precipitation intensity, as can be seen in Figs. 12a,c,d. Precipitation started at 2000 UTC on 5 February and ended at 0300 UTC on 6 February. The model captured the starting and ending times of the precipitation remarkably well (Fig. 12b). The observed precipitation type is mainly snow based on both human observer and FD12P (Fig. 12b). The observed snowfall intensity reached near 4 mm h−1 and the associated minimum visibility and RVR reached near 400 and 600 m, respectively (Figs. 12c and a), where the T is close to −4°C. Note that the parameterized visibility, visBI09, correlated well with observations (Fig. 12a), which implies that the reduction of the visibility was mainly caused by snow. Although the model captured the visibility trend, the predicted minimum visibility only reached about 700 m. This is because the predicted precipitation only reached a maximum of 2 mm h−1, which is about half the observed value. If the precipitation intensity was predicted well (perfect forecast), the visibility forecast would closely follow the parameterization visBI09. The maximum of the observed precipitation intensity is also shifted to the left and this is also reflected in the predicted visibility (see Figs. 12a,d). The magnitudes of the simulated RVRs are very close to the observed values (2000–700 m), although the minimum is shifted to the left in the same manner as the predicted visibility and precipitation intensity. Considering the predicted RVR is based on visBI09 derived using predicted T and S, the agreement with observations is quite good and encouraging.
Figure 13 shows time series of model-predicted and -observed RVRs at three locations (24L, 24R, and 33R) (Fig. 13a), observed and simulated relative humidities (Fig. 13b), precipitation type based on both the FD12P and human observer, simulated and observed temperatures (Fig. 13c), and precipitation observed and simulated (Fig. 13d). The data plotted here are collected on 24 and 25 January 2010 at CYYZ. The observed temperature remained well between 0° and 7°C, although the simulated temperatures were slightly higher. The low RVR near 37 h (~1400 UTC) on 25 January occurred mainly because of fog based on both the observer and FD12P (Fig. 13c) but, as indicated by the human observer in the same panel, the fog was mixed with rain and drizzle. As discussed earlier, the FD12P is not able to report fog when it is mixed with precipitation, and hence reported rain or drizzle during this period. The presence of fog is indicated by relative humidity reaching near to 100% based on observation and also model simulation (Fig. 13b). Note that the atmosphere has warmed during this period, indicating the passage of a warm front and associated precipitation fog. The start and end times of the precipitation are well simulated (Fig. 13d), but there are some discrepancies in the magnitude of the precipitation intensity. Note that the minimum in RVR (200 m) occurred near 38 h where the observed precipitation intensities are quite low (<0.5 mm h−1) (Figs. 13a,d), indicating that the RVR minimum is mainly due to fog, which would mainly be linked to the parameterization given in Eq. (8). Incidentally, the simulated precipitation intensities are also low, agreeing with the observed values; as a result, the predicted RVR reached near 200 m, which also agreed well with observation during this period, although slightly shifted to the left following the maximum in the simulated RH (Fig. 13b). During the earlier hours between 24 and 30, the parameterization overestimated the RVR. This is because the model predicted a significant amount of precipitation, which is not supported by observation. Thus, based on our algorithm using the model-predicted variables, the rain droplets would be the main contributors to the total extinction, which would give a relatively smaller extinction than that found for fog, and hence the predicted RVR is larger. Thus, accurate prediction of relevant meteorological parameters—such as precipitation, relative humidity, and the type of precipitation—has a critical importance for prediction of RVR using NWP models. Therefore, the above examples demonstrate that such parameterizations could be applied to predict the runway visual range, provided that the relevant parameters associated with it are reliably predicted.
7. Summary and conclusions
To develop a parameterization that can be used for predicting runway visual range (RVR), relevant meteorological parameters such as visibility (Vk), precipitation intensity, relative humidity (RH), temperature (T), and precipitation type measured at the Toronto Pearson International Airport (CYYZ) during the Canadian Airport Nowcasting (CAN-Now) project have been analyzed. The precipitation intensity Vk and precipitation type were measured using the Vaisala FD12P probe. The observed Vk and precipitation type were tested against data reported by a human observer. Generally, it was found that measured Vk corresponded quite well to those reported by an observer with a correlation coefficient r of near 0.76 for Vk < 15 km. However, the FD12P underestimated Vk by about 20% with a mean difference (MD) of about 1 km. When the two datasets were compared for Vk < 5 km, however, the correlation coefficient was unchanged, but the MD decreased to more a reasonable value of close to 200 m. For Vk < 2 km, however, the FD12P overestimated visibility by about 7% with an MD of 60 m. The observations also indicated that the human observer reported about 6 times more fog events than the FD12P, and as would be expected, the FD12P missed all fog cases that were mixed with drizzle, rain, or snow. The observer reported slightly more snow events—nearly 22% as compared to 17% by the FD12P; however, the FD12P reported significantly more snow grain cases than the observer. Both the observer and the FD12P reported rain events at a similar frequency—near 4% and 5%, respectively—but the FD12P reported 12 times more drizzle cases. It is possible that the FD12P reported some of the drizzle cases at the expense of fog. The FD12P also reported some IP and ZR cases that the observer did not report.
Using a theoretical approach and aviation air traffic service operational procedures, a parameterization that can be used for predicting runway visual range as a function of visibility has been developed. The parameterization was tested using direct measurements of RVR and Vk during years between 2009 and 2011 at the Toronto Pearson International Airport during the CAN-Now project, and the agreement found was quite good, with a correlation coefficient r near 0.8. Predictions of RVR using a new parameterization that adapts a suitable parameterization of Vk as a function of snowfall intensity and temperature during snow has been discussed and shown to work quite well. Using measurements of temperature (T), relative humidity (RH), and dewpoint temperature (Td), a new parameterization that can be used for forecasting Vk, and hence RVR during fog has also been also developed; the parameterization agreed reasonably well with observations, with a correlation coefficient r of near 0.8. These parameterizations were tested using Canadian Environmental Multiscale Regional (GEM-Reg) model data. To test the applicability of these parameterizations in the model, two case studies using model-simulated data and meteorological observations under very low-visibility conditions during snow, fog, drizzle, and rain at CYYZ were presented. The results show that when the relevant meteorological parameters such as precipitation intensity, T, and RH are reasonably predicted, and fog events are correctly diagnosed, the model could be used to predict RVR.
This work was funded by the National Search and Rescue Secretariat, Transport Canada and NAV CANADA.