The New Zealand Limited Area Model is used to investigate the impact of assimilating NOAA-15 and -16 Advanced Television and Infrared Observation Satellite (TIROS) Operational Vertical Sounder (ATOVS) radiances on surface air temperature over Canterbury, New Zealand, for two föehn cases in January 2004. For both cases, the simulated westerly-northwesterly wind crossing the Southern Alps and descending in the lee (i.e., a föehn) was stronger with ATOVS data (pass 2) than without ATOVS data (pass 1). Also, for one case, the timing of the passage of a cold front over Canterbury was more accurately forecast in pass 2. The associated differences in the potential height ΔH and winds ΔV over South Island between pass 1 and pass 2 for both cases developed from small differences in the initial conditions. It is suggested the dynamical forcing of the Southern Alps contributes to the amplification of ΔH and ΔV. The enhanced ΔV led to stronger adiabatic descent in the lee (or a stronger föehn) with stronger adiabatic warming and surface diabatic heating in pass 2. Additionally, the later passage of the cold front in pass 2 during one case allowed a longer period of heating of the surface air ahead of the cold front. As a result, large well-organized differences in surface air temperature between pass 1 and pass 2 (ΔT of 4–10 K) occurred over Canterbury. Thus, the Southern Alps acted to amplify the impact of assimilating ATOVS radiances on simulated surface air temperature over Canterbury under föehn conditions. Verification with surface observations at five climate stations over Canterbury showed a positive impact of ATOVS radiance assimilation for the two cases.
New Zealand lies in the midlatitudes of the southwest Pacific, surrounded by vast oceans. The two main islands are North Island and South Island (Fig. 1a). North Island has a “spine” of mountain ranges running through the middle, with gentle rolling farmland, interspersed with very steep gullies. The central North Island is dominated by the Volcanic Plateau (with the highest peaks being ~3000 m), an active volcanic area. The Southern Alps form the backbone of South Island with the highest mountaintop of 3755 m. The rolling hills and plains to the east of the Southern Alps in the central South Island is known as the Canterbury, New Zealand, region. With a prevailing westerly wind, Canterbury is generally on the lee side of the Southern Alps (Fig. 2a).
The Advanced Television and Infrared Observation Satellite (TIROS) Operational Vertical Sounder (ATOVS) radiances provide information about air temperature and humidity profiles. Assimilation of these data may improve the initial conditions of a numerical weather model, especially over oceans where other observation types, such as radiosondes, which could give temperature and humidity profile information, are rare. Direct assimilation of satellite infrared and microwave radiances in global–regional models has been conducted in many major meteorological centers for some years. This has shown positive impacts on the analysis and forecast of temperature and moisture profiles, geopotential heights, and even on rainfall forecast scores, especially over the Southern Hemisphere (e.g., Cameron et al. 2005; McNally et al. 2006; McNally 2009; Le Marshall et al. 2006; Derber and Wu 1998; Qi and Sun 2006).
The vast oceans surrounding New Zealand have very few routine rawinsonde observations. The ATOVS radiances from satellites undoubtedly fill this gap to some extent. To quantify the impact of assimilating National Oceanic and Atmospheric Administration (NOAA) satellites NOAA-15 and -16 ATOVS radiances on forecasts using the New Zealand Limited Area Model (NZLAM; Webster et al. 2008), a pair of simulations was conducted from 1 January to 28 February 2004. One simulation (pass 1) did not assimilate the ATOVS radiances; the other simulation (pass 2) did.
Most of the previous studies about the positive impact of radiance assimilation are mainly through statistical analysis of temperature, humidity, and geopotential height profiles or at middle levels (e.g., Cameron et al. 2005; McNally et al. 2006; McNally 2009; Le Marshall et al. 2006; Derber and Wu 1998; and others). However, analysis of the impact of satellite radiance assimilation on simulated land surface air temperature is rare in the literature. This is perhaps due to the fact that surface air temperature is affected not only by atmospheric processes, but also by the land surface processes and by the complicated land–air interaction as well. An improvement in the atmospheric fields such as air temperature may not necessarily lead to improvement of the land surface air temperature. Figure 2a shows the mean absolute differences (MAD) in the surface air temperature between pass 1 and pass 2 forecasts for January and February 2004. For most of the land areas, the MAD of surface air T was 0.5–0.7 K. However, over Canterbury, the MAD of T is ~0.2 K higher than elsewhere. Checks of the maximum absolute surface air temperature difference (called ΔTmx in this paper) between pass 1 and pass 2 on each day revealed two days with large and well organized absolute ΔTmx (6–10 K) over Canterbury. These days were 19 January 2004 (hereafter case 1; Fig. 2b) and 30 January 2004 (hereafter case 2; Fig. 2c). For both cases the absolute ΔTmx over Canterbury was 2–3 times larger than that over most other land areas of New Zealand on the same two days. For other days, no well-organized ΔTmx patterns with comparable magnitudes were found. The main objective of this study is to answer the following question: “Why and how does the assimilation of ATOVS radiances result in the large forecast ΔTmx over Canterbury on these two days?” In contrast to the previous studies mentioned above we have focused on specific cases and analyzed the variations in the geopotential heights and winds due to ATOVS radiance assimilation that caused the large ΔTmx. The analysis presented here suggests that during föehn conditions, surface air temperature forecasts in the lee of the Southern Alps can be impacted significantly by assimilating ATOVS radiances (i.e., the Southern Alps amplify the impact of ATOVS assimilation under föehn conditions).
The rest of the paper is organized as follows: a brief description of the model and data assimilation scheme in section 2, a description of the background weather in section 3, differences in simulated temperature and winds between pass 1 and pass 2 in section 4, comparison of the simulated surface air temperature with observations over land in section 5, the impact of assimilating ATOVS radiances on simulated geopotential heights and winds in section 6, and finally the summary.
2. Model and data assimilation description
The Met Office Unified Model (UM) has a nonhydrostatic, fully compressible, deep-atmosphere formulation of the Navier–Stokes equations using a terrain-following, height-based vertical coordinate. It employs a horizontally staggered Arakawa C-grid and a vertically staggered Charney–Phillips grid; semi-Lagrangian advection for all prognostic variables, except density, with conservative and monotone treatment of tracers; predictor–corrector implementation of a two-time-level, semi-implicit time integration scheme; and three-dimensional iterative solution of a variable-coefficient elliptic equation for the pressure increment at each time step [see Davies et al. (2005) and Webster et al. (2003) for detailed descriptions]. The simulations were made using the NZLAM, a regional configuration of the UM, shown in Fig. 1a. It has 324 by 324 horizontal grid points with a spacing of 0.11° (about 12 km) and 38 levels in the vertical, with the model top at about 39 km. The highest vertical resolution is concentrated near the ground such that 10 levels span the lowest 2 km of the atmosphere. A global run provided the lateral boundary conditions for the pass-1 and pass-2 simulations. The large-scale cloud and precipitation scheme is described in Wilson and Ballard (1999) and includes a prognostic treatment of ice microphysics. Although the convection scheme is based on Gregory and Rowntree (1990), significant modifications have been made to the basic scheme as described in Webster et al. (2003).
The experiment consisted of two NZLAM simulations (i.e., pass 1 and pass 2 mentioned earlier) run for two months over January and February 2004. The pass 1 and pass 2 performed a 3-hourly data assimilation cycle using an incremental three-dimensional variational data assimilation (3DVAR) first guess at appropriate time (FGAT) analysis scheme (Lorenc et al. 2000). The background error covariance used was supplied by the Met Office and is homogeneous, isotropic, and stationary in the model domain. The horizontal correlations are represented by a second order autoregressive function with proscribed length scales for each of the 3DVAR control variables (130 km for streamfunction and unbalanced pressure, 180 km for velocity potential, and 90 km for relative humidity). Vertical error correlations, error variances, and the regression coefficients used to improve the estimate of the balanced pressure derived from the wind field were estimated using a forecast differences method [based on the National Meteorological Center (NMC) method; Parrish and Derber 1992] from the Met Office’s limited-area North Atlantic and Europe configuration of the Unified Model. [See Lorenc et al. (2000) and the Met Office’s VAR Scientific Documentation Papers 11, 12, 13, and 20 for more detailed information.] The 48-h forecasts were made following the 0000 and 1200 NZST (New Zealand standard time; UTC + 12 h) analyses and fields were saved hourly to allow validation of the NZLAM forecasts against observations. Both pass 1 and pass 2 assimilated the following observations: surface (including buoys, land synops, ship synops, and scatterometer winds), aircraft, and radiosondes. Pass 2 additionally assimilated Advanced Microwave Sounding Unit-A (AMSU-A) and –B (AMSU-B) observations from the NOAA-15 and NOAA-16 satellites. These were thinned to one field of view every 80 km with NOAA-16 preferred over NOAA-15 and clear-sky fields of view preferred over partially cloudy ones. The exact channels used in each accepted field of view depend upon the details of the scene: both cloud detection results and surface characteristics (including surface type, i.e., land or sea, and, over land, the orography). For NOAA-15 the selection was made from channels 5, 6, 7, 8, 9, 10, and 18. For NOAA-16 the selection was made from channels 5, 6, 7, 8, 9, 10, 11, 18, 19, and 20. The bias correction method consists of a scan position dependent correction; an airmass-dependent correction based on two model predictors, the 200–50- and the 850–300-hPa thicknesses; and a constant term (Eyre 1992). Bias corrections were determined independently for each ATOVS channel from data collected during the pass-1 run.
Unless stated otherwise, the NZLAM hourly output data used in this study were produced by forecast runs started following analyses at 1200 NZST 18 (case 1) and 29 (case 2) January 2004.
3. The weather situation
In this section, a description of the synoptic weather situations for the two cases is presented based on the surface analyses from the MetService of New Zealand, the observed surface winds and air temperature at six climate stations (Fig. 1b), and the pass-2 simulation.
a. Surface analysis and observations
For case 1 on 19 January, two cold fronts swept up South Island from southwest to northeast bringing a change from warm northwesterly to cold southwesterly winds (Figs. 3a,b). The first cold front (called front 1) reached the middle of South Island about 0000 NZST and affected Canterbury mainly during the nighttime and early morning, while the second one (called front 2) reached the southern edge of South Island around 0000 NZST and affected Canterbury during the daytime on 19 January (Figs. 3a,b). Our following analyses mainly cover the daytime on 19 January and thus only front 2 is considered.
Over Canterbury, a cold front is also called a southerly change, because its passage usually turns airflow from westerly or northwesterly to southerly or southwesterly with a pronounced drop in air temperature (Smith et al. 1991; Sturman et al. 1990). Over central Canterbury, only six climate stations had hourly surface winds on 19 January (Fig. 1b). The approximate positions of front 2 over Canterbury at 1000 and 1400 NZST are shown in Figs. 1c,d, which indicate a southwest to northeast orientation. Similar orientations of cold fronts over Canterbury can also be found in previous publications (Smith et al. 1991; Sturman et al. 1990). The cold front 2 over land gradually moved upslope (i.e., northwestward; Figs. 1c,d) while the part of front 2 that was over the sea moved northeastward (Figs. 3a,b). At all of the five climate stations, except at Snowdon, hourly observations indicated that the air temperature dropped 4–8 K in 2–3 h after the passage of front 2 (Fig. 4, solid lines).
For case 2 on 30 January (Figs. 3c,d), a warm front moved from the windward side to the lee side of South Island in the morning. Behind the front was a warm northwesterly flow that crossed the Southern Alps and descended in the lee over Canterbury during the daytime. At the five stations over Canterbury, the observed surface air temperature ranged from 22° to 27°C in the early afternoon (Fig. 5, solid lines).
At 0100 NZST 19 January, the southern portion of the simulated front 2 (Fig. 6b, thick solid line) in pass 2 moved to the southeast coastal region of the lower South Island, consistent with the analysis–observations (Fig. 3a). The southwesterly winds behind front 2 were ~2 m s−1 stronger than the southerly winds in front of it (Fig. 6f).
On the windward side, low-level winds were mainly westerly in the early morning and changed to southwesterly in late morning (Figs. 7a,b). The southwesterly winds gradually decreased in wind speed and changed to westerly over the windward coastal region and slopes and to northwesterly over the mountain ridges and the leeside slopes (Figs. 7a,b), most likely due to island blocking (Yang et al. 2011; Smith 1982) and the greater frictional effects over the land surface. The westerly and northwesterly flow passed over the mountain ridges and descended in the lee with adiabatic warming, leading to the occurrence of föehn conditions over Canterbury (McGowan et al. 2002) for pass 2. Pass 1 is similar to pass 2 except for weaker simulated winds across the Southern Alps that will be described later in sections 4 and 6.
The föehn is clearly shown in the cross-section diagrams across the central South Island (Figs. 8b and 9b ). After passing the mountain ridge, the descending northwesterly airflow in the lee led to a higher potential temperature tongue extending from 850 hPa to low levels (adiabatic warming). In addition, the airflow descent depressed cloud formation, allowing strong surface heating during daytime (Yang et al. 2011). Figure 10a shows the surface sensible heat flux at 1100 NZST 29 (2300 UTC 28) January 2004. Over most of the Canterbury Plain, the sensible heat flux was 150–250 W m−2, which was quite stronger than over the central windward slope (about 50–100 W m−2). As a result, leeside potential temperatures were 4–6 K higher than on the windward side over the slopes (Figs. 8b and 9b).
The simulated cold fronts over Canterbury during daytime (marked by the wind shear between the northerly and the southerly) showed a northeast–southwest orientation (Figs. 7b), consistent with observations (Figs. 1c,d). Front 2 for pass 2 at 1100 NZST was at the Rangiora coast on the lee side in the middle South Island (Figs. 1c and 7b). At 1300 NZST, front 2 moved to middle slopes with a cold southerly behind (Fig. 9b). The gradual upslope movement of the simulated front 2 over land is consistent with the observations. The cold fronts were embedded within a trough that extended from a low to the south-southeast of South Island (Figs. 11a,b) toward the northwest. Differences in geopotential heights were found around the trough between pass 1 and pass 2 and will be investigated further in section 6.
For pass 2 of case 2, at 1300 NZST the simulated warm front (the shear line between the northerly–northwesterly and the southeasterly) was over the middle to lower slopes of the Southern Alps on the lee side (Fig. 12b), consistent with observations (Fig. 3d). Similar to case 1, the northwesterly flow over the mountain ridges descended in the lee over Canterbury with adiabatic warming (Fig. 13b), leading to a föehn on that day. A trough was simulated over South Island extending from the southeast to the northwest (Figs. 11c,d) for both pass 1 and pass 2. On the windward side of South Island, the geopotential height at low levels for pass 1 was lower than pass 2, while it was higher on the lee side for pass 1. This will be investigated further in section 6.
4. Differences in the simulated surface air temperature
a. Case 1
Figure 14 shows the differences in the simulated surface air temperature ΔT and winds ΔV between pass 1 and pass 2 (pass 2 − pass 1). For case 1 (Figs. 14a,b), two bands of ΔT were shown on the east side of South Island with relatively large positive values. One was over the sea (1–4 K) with northwest–southeast orientation, corresponding to the southern portion of cold front 2 (Figs. 3b and 7b) over sea, and then moved northeastward. Immediately behind this band the ΔV showed northerly or easterly, implying weaker southwesterly winds behind front 2 over the coast and the nearby sea for pass 2 than pass 1.
The other band was over Canterbury (4–10 K) with a southwest–northeast orientation, almost parallel to front 2 over land (Figs. 1c,d) and corresponding well to the absolute ΔTmx shown in Fig. 2b. The maximum ΔT in the band over land were more than twice that over the sea or other land areas. As shown in the cross section (Figs. 8c and 9c), the maximum ΔT in the band extended from the surface to a few hundred meters in the vertical and gradually moved upslope. The magnitude of ΔT gradually increased from morning to the early afternoon as the surface heating became stronger.
To the northwest of the band of ΔT over land were northerly or northwesterly ΔV, implying stronger northerly–northwesterly winds for pass 2. The stronger northwesterly extended from the surface up to 700 hPa (Figs. 8c and 9c). This led to stronger adiabatic warming and surface diabatic heating from late morning to early after afternoon for pass 2 than for pass 1. In addition, the pass-2-simulated front 2 over Canterbury arrived later than that in pass 1 (Figs. 7b,d, 8, and 9). The pass-1 simulation of front 2 reached the middle slope and upper slope at 1100 and 1300 NZST, while the pass-2 simulation of front 2 only reached the coast at 1100 NZST and the middle slope at 1300 NZST. The later arrival of front 2 for pass 2 allowed more time (from morning to early afternoon) for adiabatic warming and surface diabatic heating ahead of it over the Canterbury Plains, leading to higher surface air temperature for pass 2 compared to pass 1. Thus, the large ΔT over Canterbury was due to pass 2 having stronger adiabatic descent warming and stronger surface diabatic heating than pass 1 and also due to the fact the cold front arrival over land was later for pass 2 than pass 1.
b. Case 2
A large band of ΔT (4–8 K) was found over Canterbury (Figs. 14c,d), almost parallel to the orientation of the warm front (shear line) over land (Fig. 3d) and corresponding closely to the absolute ΔTmx shown in Fig. 2c. For pass 2, significant northwesterly flow was simulated over the mountain ridges and descended in the lee with strong adiabatic warming (Figs. 12a and 13b) and strong surface heating due to less cloud, leading to stronger (about 50–100 W m−2) surface sensible heat flux over the Canterbury Plain than over the windward slope (Fig. 10b). However, in contrast to pass 2, no adiabatic descent of airflow was simulated in the lee for pass 1. In fact, the simulated pass 1 winds showed weak upward motion over the leeside slope (Fig. 13a). This led to lower air temperatures in the boundary layer for pass 1 than for pass 2 over Canterbury (Figs. 13c and 14c,d). Similarly to case 1, the gradual increase in the surface diabatic heating led to larger and deeper positive ΔT in the boundary layer in the early afternoon than in the morning. These analyses showed that the large positive ΔT over Canterbury for case 2 were again due to the descent of air from aloft causing greater adiabatic warming and land surface heating over Canterbury for pass 2 compared to pass 1.
The above analyses have shown higher surface air temperatures over Canterbury for pass-2 simulations compared to pass 1. We now consider which simulation is better or more reliable. Here we have used the observed diurnal winds and air temperature at five climate stations (Fig. 1b) to validate the simulations (Figs. 4 and 5).
a. Case 1
The approximate times for front 2 to cross the five climate stations are indicated by the vertical solid lines, dotted lines, and dashed lines for the observations, pass 1, and pass 2, respectively, in Fig. 4. The time at which front 2 crossed a station for the two experiments was determined by when the wind direction changed from northwesterly to southerly or southwesterly at Lake Tekapo, Hanmer, and Rangora, New Zealand, and by the abrupt wind speed increase (2–4 m s−1) in the southwesterly flow at Winchmore and Christchurch, New Zealand.
For pass 1, the arrival time of front 2 at the five stations was 1–5 h earlier than indicated by the observations. The simulated front arrival times for pass 2 were generally closer to the observations than for pass 1. The earlier arrival time of front 2 for pass 1 led to an earlier drop in simulated surface air temperature than for pass 2 and the observations. This led to a higher simulated surface air temperature for pass 2 than for pass 1.
At Lake Tekapo, the error in the simulated southerly change onset for pass 2 was about 3 h early, which was larger than at the other four stations. This is probably due to the relatively low resolution of NZLAM (12 km), which does not completely resolve the small-scale mountains around Lake Tekapo. The early passage of the simulated cold front for pass 2 erroneously curtailed the warming conditions and brought in cold air well before the surface air could be heated up as indicated by the observations. This led to larger errors in the simulated surface air temperature for pass 2 at Lake Tekapo than at the other stations.
In terms of the southerly change onset and the corresponding surface air temperature change associated with the passage of front 2, the above analyses showed that overall pass 2’s performance was better than pass 1.
b. Case 2
Figure 5 shows the diurnal temperature on 30 January at the five climate stations for the observations, pass 1, and pass 2. Pass 1 had quite large errors (5°–10°C) in the simulated surface air temperature, especially in the early afternoon. Pass 2 greatly corrected the errors with the simulated air temperature much closer to observations. This indicated that the adiabatic warming and surface diabatic heating were more accurately simulated for pass 2 than for pass 1 over the Canterbury Plains.
6. The causes
In this section we investigate the reasons why stronger winds were simulated to pass the Southern Alps and descend in the lee over Canterbury for pass 2 than pass 1 for both cases. For case 1, we investigate why front 2 of case 1 was slower moving (and thus arrived later over Canterbury as occurred) for pass 2 than pass 1.
a. Case 1
The assimilation of ATOVS radiance impacts temperature and humidity profiles directly and also the pressure–geopotential heights and wind fields through model adjustment after the simulation starts. The winds and geopotential heights at 850 hPa were quite similar between pass 1 and pass 2 (Figs. 11a,b). However, obvious differences in heights ΔH and winds ΔV were found later in the forecast (Figs. 15a,c,e). At 1300 NZST 18 January (which coincides with the end of the 3-h analysis window of NZLAM and beginning of the simulation used in this study), a trough with a northwest–southeast orientation on 850 hPa (Fig. 15a, black thick dashed line) was southwest of South Island. A few ΔH maxima were found along the trough. The two ΔH maxima affecting South Island 10 h later were denoted by H1 and H2. Associated with the eastward moving of the trough, H1 and H2 gradually amplified in magnitude and moved eastward. At 0100 NZST 19 January, H2 moved over the lower South Island (i.e., in this area the pass 2 trough was deeper than for pass 1) as seen in Figs. 15c and 6e with a corresponding pattern in ΔV seen. This actually led to weaker southwesterly winds for pass 2 than pass 1 at 850 hPa in this area (Figs. 6a,c,e). At the surface, a negative pressure maximum developed in this area (Fig. 6f) associated with the ΔH above. It corresponded to a weaker pressure gradient behind the cold front 2 for pass 2 than pass 1 (Figs. 6b,d), thus a weaker southwesterly (Fig. 6f), and slower-moving front 2 for pass 2 than pass 1 (Figs. 6b,d). With the eastward movement of this trough, the mesoscale negative ΔH (H2) gradually moved eastward and amplified, leading to a later arrival of front 2 for pass 2 over Canterbury (and the nearby sea) during the daytime (Figs. 7b,d).
At 1100 NZST 19 January, the trough was over the middle of South Island (Fig. 15e). The H1 moved to the north of South Island and H2 moved over the Canterbury Plain and its nearby sea (Fig. 15e). The distribution of the ΔH reflected a stronger northwest–southeast pressure gradient (tighter geopotential height contours) over the windward side for pass 2 than pass 1 on 850 hPa (Figs. 7a,c), leading to a stronger southwesterly for pass 2 (Fig. 7e). On the lee side, the negative ΔH maxima (Fig. 15e) indicated a weaker west–east geopotential height gradient at the 850-hPa level over the coastal region behind the cold front (Figs. 7a,c) for pass 2 and thus weaker southwesterly winds (Fig. 7e). At the surface, associated with the ΔH maxima above, a weaker west–east pressure gradient for pass 2 was also found over Canterbury and the nearby sea behind the cold front (Figs. 7b,d) and thus weaker southwesterly (Fig. 7f), leading to a slower movement of front 2 for pass 2 than pass 1 as described earlier.
For both pass 1 and pass 2, the southwesterly gradually weakened and changed to westerly over the windward coastal region and then to northwesterly over the Southern Alps and the lee (Figs. 7a–d). In fact, the island blocking increased pressure on the windward slope due to forced ascent of airflow, and decreased pressure on the leeside slope due to the airflow descent. This in fact increased the pressure gradient (or tightened the isobars) along the Southern Alps, a positive feedback of the wind variations to the potential height due to mountain dynamical forcing (Fig. 7). The intensified pressure gradient over the Southern Alps in turn increased the airflow speed passing across it. For pass 2, the windward incoming wind speed was stronger, thus the island blocking was also more significant. This was shown in Fig. 7f, where at the surface a westerly ΔV was found toward a ΔP maximum on the windward coastal region, indicating stronger island blocking for pass 2. This led to a larger northwest–southeast pressure gradient along the Southern Alps for pass 2. Thus, the stronger wind speed of airflow past the Southern Alps for pass 2 than for pass 1 was due to a stronger incoming airflow on the windward side and a stronger pressure gradient over the mountains.
b. Case 2
Figures 15b,d,f show the differences in the ΔH and ΔV on 850 hPa. At 1300 NZST 29 January (which coincided with the end of the 3-h analysis window of NZLAM and the beginning of the simulation used in this study), the two small ΔH maxima that affected South Island later are denoted by H3 and H4 and were close to the trough axis. The magnitudes of both ΔH and ΔV of H3 amplified while moving southeastward. The H4 was also amplified while moving eastward. At 1300 NZST 30 January, H3 was on the windward side and H4 was on the lee side of South Island (Fig. 15f). Associated with the quite large ΔHs, obvious differences in the pressure patterns and winds at 850 hPa were found between pass 1 and pass 2 (Figs. 11c,d). The 850-hPa geopotential height for pass 1 over South Island was larger than over both the windward and leeward sides. On the windward side, the pressure pattern–distribution for pass 1 led to northeasterly winds at low levels offshore (Figs. 12c,d). The low-level incoming northeasterly showed decreased wind speed over the windward coastal region due to island blocking with a small pressure ridge at the surface of the central west coast (Fig. 12d) and a height ridge over the windward upper slope at 850 hPa (Fig. 12c), which is a positive feedback of wind speed variation to pressure and geopotential height. However, the deceleration of the low-level winds meant that the flow was no longer geostrophic, making the winds more easterly and directed away from the island, especially at the surface. Thus, conditions were not favorable for the low-level winds to ascend and pass the mountains (Fig. 13a). On the lee side, the near-surface winds were weak southerlies, which met the weak ascending oncoming airflow over the mountains (Fig. 13a).
For pass 2 the pressure over South Island was lower than for pass 1 (Figs. 11c,d). This led to northwesterly winds toward the west coast at 850 hPa for pass 2 (Fig. 12a). Near the surface on the windward side, the oncoming winds were weaker with more northerly direction for pass 2 than for pass 1 (Figs. 12b,d). However, the winds were still overall northeasterly for pass 2, so conditions were not favorable for the surface winds to ascend and pass the mountains as discussed earlier for pass 1. Indeed, significant northwesterly winds flowing over the Southern Alps occurred only above 900 hPa for pass 2 (Fig. 13b). Similar to case 1, the island blocking to the northwesterly winds led to tight height–pressure contours along the Southern Alps (Figs. 12a,b), which is a positive feedback of the wind speed variation to the height field due to mountain dynamical forcing. The orientation of the pressure–height contours over the Southern Alps for pass 2 was quite different from pass 1 because of differences in wind directions (Fig. 12).
The analyses for both cases showed that the differences in winds between pass 1 and pass 2 over South Island were due to the small ΔHs and ΔVs that developed from the initial conditions. The ΔH and ΔV maxima and their amplification may be attributed to three reasons. First, they were coincident with a baroclinic trough. The amplification of ΔH and ΔV might be a result of quasi-exponential growth of the baroclinically active scales (Tribbia and Baumhefner 2004). Second, some differences in rainfall between pass 1 and pass 2 were found associated with troughs around the major ΔH maxima areas (not shown). Thus, moist processes (Zhang et al. 2002, 2003) may be another reason for the amplification. Third, when the trough moved over South Island, the island forced adiabatic ascent on the windward side and descent of airflow in the lee led to variations in the pressure and thus in the wind speed, further enhancing the ΔH. Thus, the Southern Alps played a dynamical role in amplifying the ΔH and ΔV over South Island under these föehn conditions. The amplification of ΔV led to stronger adiabatic descent in the lee with stronger adiabatic warming and surface diabatic heating in pass 2 during the daytime. This led to large differences in the surface air temperature over Canterbury between pass 1 and pass 2 (ΔT of 4–10 K). Thus, the impact of assimilating ATOVS radiance on simulated surface air temperature in leeside areas was amplified by mountains under a föehn condition.
In this study, through two case studies we have investigated the impact of ATOVS radiance assimilation in NZLAM on simulated surface air temperatures over the Canterbury region and the amplification of this effect by the Southern Alps, New Zealand. Under föehn conditions, adiabatic descent of air from aloft can occur over the Canterbury Plains with adiabatic warming and less cloud, leading to strong land surface diabatic heating during the daytime. These two heating conditions occurred for the two cases during the daytime on 19 (case 1) and 30 (case 2) January 2004. In addition, for case 1 a cold front (southerly change) swept across South Island with a northeast–southwest orientation over Canterbury during the daytime.
For case 1, the westerly-northwesterly winds over the Southern Alps simulated by pass 2 (which included ATOVS data) were stronger, thus enabling more significant heating (adiabatic warming and surface diabatic heating) in the lee over Canterbury, than for pass 1, and the passage of the simulated cold front over Canterbury was 1–3 h later. This later passage of the cold front in the pass-2 simulation allowed more time for the surface air ahead of the cold front to be heated by these two heating conditions from late morning to early afternoon. This led to large surface air temperature differences between pass 2 and pass 1, ΔT of 4–10 K, over Canterbury at matching times during the day.
Similarly for case 2, significant heating conditions due to the descent of northwesterly winds from aloft were simulated for pass 2 over the Canterbury region, but not for pass 1. This again led to higher surface air temperature (ΔT, 4–8 K) in this area during the daytime for pass 2, especially in the early afternoon when the surface diabatic heating was the strongest.
The differences in the winds ΔV and potential height ΔH over South Island that led to the large ΔT over Canterbury between pass 1 and pass 2 for the two cases developed from small values in ΔV and ΔH at the initial conditions associated with a trough–low affecting South Island. The amplification of the magnitudes of ΔH and ΔV may be attributed to baroclinic and moist processes as described in previous publications (Tribbia and Baumhefner 2004; Zhang et al. 2002, 2003). In this study, the dynamical forcing of the Southern Alps was suggested as another process for the amplification of the differences in these fields through adiabatic ascent of airflow on the windward side and descent in the lee. The enhanced ΔV led to stronger adiabatic descent in the lee with stronger adiabatic warming and surface diabatic heating in pass 2 during the daytime. Additionally, the later passage of the cold front in pass 2 of case 1 allowed more time for the surface air ahead of the cold front to be heated by these two heating conditions. As a result, large well-organized differences in the surface air temperature between pass 1 and pass 2 (ΔT of 4–10 K) occurred over Canterbury. In other words, the Southern Alps played a role in amplifying the impact of assimilating ATOVS radiance on simulated surface air temperature over Canterbury under these föehn conditions. Without the Southern Alps, the magnitude of ΔT over Canterbury would have been smaller for the two cases. Under a föehn condition, the simulation of the surface air temperature in the lee is more sensitive to the initial conditions than in other areas.
Verification of the simulated surface winds and air temperatures was conducted by comparing them with observations at five climate stations in the mid-Canterbury region. In terms of the timing of the arrival of the southerly change and the simulation of the surface air temperature within a period 2–3 h after the southerly change for case 1, pass 2 was closer to observations than pass 1. For case 2, the daytime surface air temperatures simulated by pass 2 over the Canterbury region were much closer to observations than those simulated by pass 1.
However, the improvement in the two cases studied does not mean there is always a positive impact by assimilating ATOVS radiance on the simulated surface air temperature over Canterbury. Table 1 shows the mean absolute errors (MAE) at the five climate stations over Canterbury during January and February 2004 for pass 1 and pass 2. Regarding the MAEs, pass 2 had only slight improvement at four stations, but was slightly worse at Hanmer, suggestive of no improvements, and it was even worse than pass 2 for most times during the two months. In fact, assimilating ATOVS radiance of NOAA-15 and -16 for NZLAM domain resulted in only small differences at the initial conditions, which needed suitable amplification to have significant impact through baroclinic and moist processes, and the mountain dynamical forcing suggested in this study. Most times during the two months might lack suitable amplifications. Additionally, the simulated surface fluxes by the land surface model coupled with NZLAM might have large errors from time to time, which might overshadow the small improvement at the initial conditions without suitable amplifications.
This research was carried out under research collaboration SC0128 with the Met Office and funded by the New Zealand Foundation for Research, Science and Technology under Contract C01X081. We thank the anonymous reviewers for their useful comments and suggestions for improving the manuscript.