1. Introduction
The development of operational mesoscale NWP models that represent convection explicitly (Lean et al. 2008), require the development of new, high-resolution data assimilation systems (Dance 2004; Park and Županski 2003; Sun 2005) to provide initial conditions giving detailed information on appropriate scales. Data assimilation can aid accurate representation of small-scale features only if high-resolution observations are available (Baxter 2009). Doppler radars can provide very high temporal and spatial resolution observations, but typically these are not available until precipitation is present. It may be possible to improve storm forecasts by assimilating Doppler velocities from insects during convective development.
Doppler radars provide measurements of the radial velocity component (parallel to the radar beam) of a backscatter target. Raw radial winds are too dense for assimilation in mesoscale models, particularly close to the radar, but from them various velocity products can be created. Examples include volume velocity processing (VVP; Waldteufel and Corbin 1979), velocity–azimuth display (VAD; Browning and Wexler 1968), and superobbed or thinned radial wind vectors (e.g., Salonen et al. 2008, 2009). Thinning is also required to reduce observation error correlations (Stewart et al. 2008).
Several studies have examined the impact of assimilating precipitation-derived radial winds using three-dimensional variational data assimilation (3D-Var). Results have shown neutral to positive impacts on analyses and forecasts (Montmerle and Faccani 2009; Xiao et al. 2008). Using insects instead of precipitation for radar targets has the advantage of providing observations during fine weather. However, the use of insects has associated problems. Insects are capable of flight at speeds of up to several meters per second, depending on species (e.g., Achtemeier 1991; Drake and Farrow 1988; Dudley 2002). They can also assume a common orientation at an angle to the wind (Chapman et al. 2010; Rennie et al. 2010b). To assimilate radial velocities from insects these issues must be considered.
In the United Kingdom, it is advisable to restrict the use of insect-derived observations to daytime. Small insects (with mass of a few milligrams) dominate the airborne biota (Wood et al. 2009) and achieve air speeds (by their own effort) typically less than 1 m s−1 (Dudley 2002, p. 78). Daytime studies of insect echoes from the U.K. weather radar network (Rennie et al. 2010b) revealed that during summer velocity measurements of useable quality are possible up to 2 km MSL and up to 30-km range from the radar. The data coverage may be less than that from insect observations in other countries, for various possible reasons, but the data are essentially free—retrieved without modification to the U.K. weather radars if they already have Doppler capability. However, insects return a much weaker signal than precipitation; the signal is close to the radar detection limit and difficult to separate from ground clutter (Rennie et al. 2010a). Comparison of VADs with model data indicated that ground clutter contamination often affected the most cluttered radars (Rennie et al. 2010b), although future upgrades to radars and processing software could yield much improvement in this area. Nevertheless daytime U.K. insects offer a small range of observations with reasonably small errors.
This study presents a preliminary exploration of the potential of assimilating radial wind observations from insects using the Met Office operational C-band Doppler radars. Case studies were chosen with insects present early in the day, to provide wind information during convective development that resulted in heavy precipitation. The assimilation and forecast experiments were conducted on three suitable days, using the Met Office Unified Model (UM), 3D-Var assimilation, and a domain with 1.5-km horizontal grid spacing. There were very few suitable cases due to unusually wet summers during the data collection period (2008–09; Met Office Climate Summary available online at http://www.metoffice.gov.uk/climate/uk/), with mostly frontal rather than convective rain. The scarcity of cases might imply that the benefit of assimilating insect winds to improve convective precipitation forecasts would be small. However, a previous study (Hand et al. 2004) indicated that most U.K. summer rain is typically convective—in contrast to our experience. Furthermore, convection increases insect observations because insects use the updrafts to gain altitude. Therefore, the usefulness of insect-derived winds should be greater in years when convective showers are prevalent (and more likely to be the cause of flooding).
This paper is organized as follows. Section 2 describes the model, observations. and data assimilation. Section 3 describes the case study days and the setup for two experiments. The first experiment involved assimilating insect-derived winds during one assimilation cycle. The second experiment involved full forecasts for the three case studies. Section 4 presents the analysis of results using various methods, and section 5 discusses the effect of assimilating insect-derived winds. Conclusions are available in section 6.
2. Observations and assimilation
a. Model
The Met Office’s Unified Model (nonhydrostatic) version 7.3 was used with no convective parameterization. The domain covered the southern part of England and Wales (Fig. 1). The model had a fixed horizontal resolution of 288 × 360 cell grid with 0.036° spacing (approximately 1.5 km). The vertical coordinates consisted of 70 terrain-following levels that were closely spaced near the ground. The upper boundary was 40 km MSL.
Map of England and Wales showing locations of the four Doppler radars. The inset indicates the location of the map. The area shown is approximately the model domain.
Citation: Monthly Weather Review 139, 4; 10.1175/2010MWR3482.1
Map of England and Wales showing locations of the four Doppler radars. The inset indicates the location of the map. The area shown is approximately the model domain.
Citation: Monthly Weather Review 139, 4; 10.1175/2010MWR3482.1
Map of England and Wales showing locations of the four Doppler radars. The inset indicates the location of the map. The area shown is approximately the model domain.
Citation: Monthly Weather Review 139, 4; 10.1175/2010MWR3482.1
The initial conditions (ICs) used to initialize the runs were derived from archived output of runs from a 4-km resolution, whole U.K. version of the Unified Model. Thereafter the 1.5-km model was run with hourly assimilation cycling to produce updated analyses. The boundary conditions (BCs) used during the forecasts were supplied from the same U.K. 4-km runs. The forecast length was constrained to 11 h by the duration that BCs were available.
b. Radar observations
The four Doppler radars in the operational network (Chenies, Clee Hill, Cobbacombe, and Dean Hill) at the time of these experiments are located as shown in Fig. 1. They yielded 5 conical scans every 5 min, at nominal elevations of 1°, 2°, 4°, 6°, and 9° for all apart from Chenies, which were 1°, 2°, 4°, 5°, and 5.5°. Each scan had a resolution of an azimuthal 360° by 167 range gates of 600 m. The beamwidth was 1°. Only observations closest to the hour (the assimilation time) were used. The number of raw observations varied according to beam elevation and the amount of ground clutter. The airborne insect population also varied daily (Rennie et al. 2010b), and followed a diurnal cycle (Wood et al. 2009), but decreased near the onset of precipitation. Total insect-derived radial wind observations per radar for all elevations ranged between 1000 and 15 000; the total for all radars ranged between 15 000 and 40 000.
Radial wind observations were extracted from raw scans (Fig. 2a) using specially designed processing. The processing, described in more detail in Rennie et al. (2010a,b), extracted insect echoes by removing noise and ground clutter, then cleaning the data using spatial filters. An example of processed observations is shown in Fig. 2b. Separating ground clutter and insects was most critical, and it was possible that not all clutter was removed, particularly for Clee Hill, Cobbacombe, and Dean Hill, which are very cluttered. If necessary, insect and precipitation echoes were separated using a reflectivity threshold. Precipitation-derived winds were not assimilated.
(a) Raw radial velocity from Cobbacombe at 1000 UTC 28 Jul 2008 at 2° elevation. The two dark blue rays indicate missing data. (b) Processed radial velocity of echoes identified as insects. (c) Superobbed radial winds from (b).
Citation: Monthly Weather Review 139, 4; 10.1175/2010MWR3482.1
(a) Raw radial velocity from Cobbacombe at 1000 UTC 28 Jul 2008 at 2° elevation. The two dark blue rays indicate missing data. (b) Processed radial velocity of echoes identified as insects. (c) Superobbed radial winds from (b).
Citation: Monthly Weather Review 139, 4; 10.1175/2010MWR3482.1
(a) Raw radial velocity from Cobbacombe at 1000 UTC 28 Jul 2008 at 2° elevation. The two dark blue rays indicate missing data. (b) Processed radial velocity of echoes identified as insects. (c) Superobbed radial winds from (b).
Citation: Monthly Weather Review 139, 4; 10.1175/2010MWR3482.1
c. Observation processing and assimilation
There were a range of observations assimilated in the “control” runs, as per operational NWP. Observation types in 3D-Var included surface (synoptic stations), ground GPS, measurements from aircraft [i.e., Aircraft Meteorological Data Relay (AMDAR)], vertical profiles from radiosondes and wind profilers, scatterometer winds, atmospheric motion vectors by satellite tracking of clouds, etc., satellite radiances plus use of radar-derived surface precipitation rates, and a 3D cloud cover analysis in latent heat and moisture nudging (Dixon et al. 2009). The experimental runs were identical to the control runs, but also included assimilation of radial winds from insects.
Radial winds for each analysis time (AT) were read in from an external file. All other observation types were extracted from a database and taken from between AT − 30 min and AT + 29 min. In preparation for assimilation, all observations underwent checking and filtering, which could include buddy checking and comparison with the model background (i.e., the 1-h forecast from the previous AT). Only observations sufficiently similar to the background were assimilated; a large disparity indicates possible gross errors and would cause slow convergence during assimilation (Stewart et al. 2009). For example, radial winds more than 10 m s−1 different from the background were excluded. Observations also underwent temporal and spatial thinning, which for radial winds involved superobbing.
Superobbing was conducted as follows. The innovations (observation difference from the model background or O − B) were averaged over a default area of 3° by 5 range gates (3 km), which provided the increment for the superob. The increment was added to the model-equivalent radial wind located closest to the center of the cell, to produce the superob (Fig. 2c). The resulting superob resolution approached the grid resolution of the model but was azimuthally dense close to the radar. Azimuthally adjacent superobservation cells were less than 1 km apart. The number of superobs depends on superobbing and thinning; in the present study the number assimilated per cycle ranged between 2000 and 6000. The observation error estimate for each superobbing cell was computed as the variance of the departure of individual innovations from the cell’s mean innovation. To this was added the representativeness error of the model, which was the long-term variance of O − B at each range from the radar. The total error ranged between 1.7 and 3.5 m s−1. No cross correlations of observation errors between adjacent superobbing cells were considered. No additional error was added as special treatment for using insect-derived winds.
Observations were assimilated using a 3D-Var First Guess at Appropriate Time (FGAT) scheme (Lorenc et al. 2000) with cloud nudging and latent heat nudging for moisture (cloud and precipitation, respectively) observations (Dixon et al. 2009). The assimilation of radial winds used the existing covariance statistics with no additional effort to tune the background errors. The correlation length scales for the model control variables range between 90 and 180 km.
3. Case studies and experiments
Three cases were selected for assimilation experiments: 27 and 28 July 2008 and 29 June 2009. These were the only days in 2008–09 with suitable insect echoes, and convection that led to subsequent heavy showers. On 29 June 2009, there were residual overnight showers present during the morning. The precipitation echoes were excluded in order to isolate the impact from insect-derived observations.
The Met Office operational (4-km model) forecasts for these days indicated a moderate degree of skill when verified against radar-derived precipitation. Convective showers were predicted to develop during the afternoon, but the timing and location of showers was slightly wrong. This indicated that the large-scale forcing supplied by the ICs and BCs did not contain gross errors that would make these poor case study choices. Ideally, assimilating additional observations would improve the location and timing of convective showers. However, results will be sensitive to the observation time and location relative to convective features.
To review the synoptic situation for each day, information was drawn from various sources. This included synoptic analysis charts produced by the Met Office at 0000 and 1200 UTC (Met Office, Crown copyright), skew T–logp diagrams from radiosonde ascents (sourced from University of Wyoming Department of Atmospheric Science), visible satellite images [e.g., Moderate Resolution Imaging Spectroradiometer (MODIS)] courtesy of the Natural Environment Research Council (NERC) Satellite Receiving Station, Dundee University, Scotland (more information available online at http://www.sat.dundee.ac.uk/), and precipitation from radar composite images (Met Office, Crown copyright). The synoptic analyses gave an overview of conditions, the skew T–logp indicated CAPE, and satellite and precipitation images pinpointed where convection and rainfall began.
a. Case studies
1) 27 July 2008
The synoptic analyses on 27 July (Fig. 3a) showed a high over Norway and a low to the southwest of Iceland with several trailing fronts to the west of Ireland. Over the United Kingdom there was a weak pressure gradient, with a broad trough. During the day a small low shifted northward toward the United Kingdom, but remained to the southwest. Cumulus clouds appeared between 0948 and 1129 UTC. Radiosonde ascents at midday (south coast and central England) showed no CAPE, although small changes to the air masses could produce CAPE. Small localized heavy convective showers broke out from about 1400 UTC in eastern England, at the location and times shown in Fig. 4a. The showers and areas of convection initiation advected northwest with the wind (Fig. 5a). Further showers broke out toward the west of England after 1800 UTC (not shown).
Analysis charts at 1200 UTC for (a) 27 Jul 2008, (b) 28 Jul 2008, and (c) 29 Jun 2009. (Met Office Crown Copyright.)
Citation: Monthly Weather Review 139, 4; 10.1175/2010MWR3482.1
Analysis charts at 1200 UTC for (a) 27 Jul 2008, (b) 28 Jul 2008, and (c) 29 Jun 2009. (Met Office Crown Copyright.)
Citation: Monthly Weather Review 139, 4; 10.1175/2010MWR3482.1
Analysis charts at 1200 UTC for (a) 27 Jul 2008, (b) 28 Jul 2008, and (c) 29 Jun 2009. (Met Office Crown Copyright.)
Citation: Monthly Weather Review 139, 4; 10.1175/2010MWR3482.1
Convective shower breakout time for (a) 27 Jul 2008, (b) 28 Jul 2008, and (c) 29 Jun 2009. Showers first appeared at the times indicated by the locations. Times are in UTC.
Citation: Monthly Weather Review 139, 4; 10.1175/2010MWR3482.1
Convective shower breakout time for (a) 27 Jul 2008, (b) 28 Jul 2008, and (c) 29 Jun 2009. Showers first appeared at the times indicated by the locations. Times are in UTC.
Citation: Monthly Weather Review 139, 4; 10.1175/2010MWR3482.1
Convective shower breakout time for (a) 27 Jul 2008, (b) 28 Jul 2008, and (c) 29 Jun 2009. Showers first appeared at the times indicated by the locations. Times are in UTC.
Citation: Monthly Weather Review 139, 4; 10.1175/2010MWR3482.1
Rain rate shortly after convective shower breakout for 27 Jul 2008 at (a) 1500 and (b) 1700 UTC, and for 28 Jul 2008 at (c) 1400 and (d) 1600 UTC. The rain rate for 29 Jun 2009 is shown in Figs. 10a,b. Rain rate is mm h−1.
Citation: Monthly Weather Review 139, 4; 10.1175/2010MWR3482.1
Rain rate shortly after convective shower breakout for 27 Jul 2008 at (a) 1500 and (b) 1700 UTC, and for 28 Jul 2008 at (c) 1400 and (d) 1600 UTC. The rain rate for 29 Jun 2009 is shown in Figs. 10a,b. Rain rate is mm h−1.
Citation: Monthly Weather Review 139, 4; 10.1175/2010MWR3482.1
Rain rate shortly after convective shower breakout for 27 Jul 2008 at (a) 1500 and (b) 1700 UTC, and for 28 Jul 2008 at (c) 1400 and (d) 1600 UTC. The rain rate for 29 Jun 2009 is shown in Figs. 10a,b. Rain rate is mm h−1.
Citation: Monthly Weather Review 139, 4; 10.1175/2010MWR3482.1
2) 28 July 2008
The synoptic analysis (0000 UTC 28 July) showed a cold front to the southwest of the United Kingdom that crossed the south coast after 1200 UTC (Fig. 3b), with a trough lying across England at 0000 UTC 29 July. A central England radiosonde ascent measured a 17 J kg−1 of CAPE at 0000 UTC 28 July, but none at midday. However, in the south amounts of CAPE in the order of 100–200 J kg−1 were observed from several ascents at midday. Cumulus clouds had appeared by midday. Precipitation initiated with isolated showers in southern England at 1040 UTC and in eastern England at midday (Fig. 4b). After 1300 UTC more showers had appeared around these locations (Fig. 5b). These showers continued to develop as they advected northwest into Wales. Some showers also developed in the north of Wales, at about 1500 UTC. Frontal showers followed by more stratiform rain crossed into the southwest of England in the late afternoon and crossed the south coast around 1900 UTC.
3) 29 June 2009
The midnight synoptic analysis showed an upper-level cold front with a lower-level front trailing from a low to the west of the United Kingdom, with a similar situation at midday (Fig. 3c). A rainband advected northward from the southwest of England overnight. This rain continued north along the Welsh border and Wales, dissipating by 1400 UTC. CAPE was measured in central England at 56 J kg−1 at midday, and in lesser amounts between 4 and 25 J kg−1 at various locations during the morning. Cumulus clouds formed in eastern England and in a line from the southeast to north Wales around midday. The first showers appeared at 1400 UTC in the southeast (Fig. 4c), then in Wales near the central border region, and by 1430 UTC several showers appeared in a line between these two locations (see Figs. 10a,b). This line continued to develop and advected northward, clearing away by 2100 or 2200 UTC. The heaviest showers developed in the east central region and formed a larger precipitation cell. There were no showers in any other region (e.g., eastern England or the southwest) until showers advected in from France around 2000 UTC.
b. Experiments
There were two experiments. The first isolated the influence of the radial winds in one assimilation cycle. The second experiment involved full forecasts with multiple assimilation cycles for all three cases.
The single-cycle experiment demonstrated how the observation information was spread during the assimilation and forecast. For this first experiment, 28 July 2008 was selected because there were no showers within the United Kingdom until around midday and most convective development was localized, making it easier to analyze. The wind was generally southeasterly throughout the domain at heights for which there were insect-derived wind observations. The model was initialized at 0500 UTC and spun up with hourly assimilation cycles from 0600 to 0800 UTC, making a short forecast each cycle to create the background for the subsequent analysis. Radial winds were only assimilated for the 0800 UTC cycle. An 11-h forecast was run from AT = 0800 UTC. The output is compared to the same day’s control run that was produced in the full forecast experiment, as the two runs only diverged at the 0800 UTC cycle.
For the second, full forecast experiment, there was a control run (denoted C) without radial winds, and an experimental run (denoted RW) with radial winds for each case study day. All were initialized at 0500 UTC, with assimilation cycles hourly from 0600 UTC. The earliest radial wind observations were at 0600 UTC, so the control and experimental runs began to diverge from 0600 UTC. Assimilation cycles continued until 1100 UTC (2009 case) or 1200 UTC (2008 cases), and the forecast was run for 11 h from each AT starting at 0800 UTC, as detailed in Table 1. The 29 June 2009 run was curtailed because of missing radar-derived precipitation products required for assimilation after 1100 UTC.
Overview of the three case studies’ assimilation and forecast cycle times, including when BCs were updated.
The forecasts used BCs produced from archived runs of the Met Office 4-km resolution model. Although the forecasts were run hourly, updates to the BCs were available less frequently. For the 2009 case, BCs were produced every 6 h from ATs of 0300 UTC and 0900 UTC from T + 1 onward. Hence, the 0600 to 0900 UTC forecasts used different BCs for the 1000 and 1100 UTC forecasts. For the 2008 cases, BCs were available every 3 h from T + 0, so the BCs were also updated at 1200 UTC. Consequently, the 11-h forecasts starting from 0800 and 1200 UTC could differ substantially from the 0900 to 1100 UTC forecasts as the boundary forcing changed.
4. Results
a. Single-cycle experiment
The single-cycle experiment included assimilation of radial winds only at 0800 UTC in order to isolate the impact of the additional observations. For the data assimilation cycle including radial winds, the cost function minimization required twice the number of iterations to reach convergence, and the penalty values were in the order of twice the size. This difference was observed for all assimilations cycles for all experiments, and is accountable through the large number of additional observations.
To assess the effect of radial winds, the 0800 UTC analysis and forecast were compared for runs with (RW) and without (C) radial wind observations. The difference in the near-surface wind field (RW minus C) at AT = 0800 UTC (Fig. 6a) indicated a decrease in wind speed around the locations of Clee Hill, Dean Hill, and Cobbacombe, and only a slight change around Chenies. The effect of the radial wind observations from all four radars was clearly visible in the u and υ components of the assimilation increments (not shown).
(a) Difference (RW minus C) in near-surface wind at analysis time for the single-cycle experiment: 28 Jul 2008 with radial winds only at 0800 UTC. Color represents difference in wind speed. (b) Difference in wind for same experiment at VT = 1200 UTC. Every 10th wind vector is shown. Note that the vector scale differs for (a) and (b).
Citation: Monthly Weather Review 139, 4; 10.1175/2010MWR3482.1
(a) Difference (RW minus C) in near-surface wind at analysis time for the single-cycle experiment: 28 Jul 2008 with radial winds only at 0800 UTC. Color represents difference in wind speed. (b) Difference in wind for same experiment at VT = 1200 UTC. Every 10th wind vector is shown. Note that the vector scale differs for (a) and (b).
Citation: Monthly Weather Review 139, 4; 10.1175/2010MWR3482.1
(a) Difference (RW minus C) in near-surface wind at analysis time for the single-cycle experiment: 28 Jul 2008 with radial winds only at 0800 UTC. Color represents difference in wind speed. (b) Difference in wind for same experiment at VT = 1200 UTC. Every 10th wind vector is shown. Note that the vector scale differs for (a) and (b).
Citation: Monthly Weather Review 139, 4; 10.1175/2010MWR3482.1
The observations from each radar affected a 100-km-wide area (Fig. 6a), although the observations occupied less than half this width. This indicates how the background error covariances in the assimilation scheme spread the information. For this amount of spread, it is to be expected that any small-scale features in the radial wind field would be smeared out. The vertical extent of influence of the radial winds rapidly diminished above ≈2 km—the maximum height of the observations (not shown). The observation information was advected with the wind, so that the differences (RW minus C) initially stayed in western part of the domain (Fig. 6b). As rain initiated, the slight changes to small-scale features eventually yielded localized wind difference magnitudes up to 10 m s−1 locally, such as off the coast south of Wales, because of phase differences.
There are several points to note for this experiment. The lower speed of the observed radial winds may indicate clutter contamination in this example. Such a result is consistent with the findings of Rennie et al. (2010b), which showed that VADs from the three most cluttered radars had on average a lower speed than the model background. Note that superobbing will also reduce the maximum speed of radial wind observations. Also, in such cases, where wind is from the southeast and convective development is located in the south or east, the observation information is advected away from the region of interest. This reduces the potential observation impact.
To confirm the small impact on convective development, comparison of precipitation during the forecast to T + 10 (not shown) indicated only small perturbations of size, intensity, or location of precipitation cells. The convergence lines from which they formed were essentially unchanged, so that cell clusters appeared in about the same place at the same time. A larger impact may result from successive assimilation of radial winds over several cycles, or from inclusion of upstream data (e.g., radars in eastern or southeastern England or France).
b. Full forecast experiments
This section is separated according to three different methods used to assess the three full forecast experiments.
1) Analysis interpretation
The observation impact is here estimated from the change to the analyses. Several comparisons were undertaken. Radial wind superobs were compared with the RW model background (analogous to the O − B term) and other wind observations. The effect of radial winds on the assimilation increments and analysis was examined (analogous to that depicted in Fig. 6a). To distinguish the impact of the radial winds, RW analyses and increments were compared with the control C. Each day is discussed individually.
(i) 27 July 2008
On this day the wind field was variable across the domain near the surface, and a mesoscale anticyclonic circulation was visible above 500 m. The wind speed over land was around 5 m s−1 at 500–1500 m AGL. The insect speeds from radial wind superobs were typically within 2 m s−1 of the RW background winds, except at Chenies where speeds were generally 2–3 m s−1 faster. The difference in direction was less than 90°. The analysis increments at early ATs [0800 UTC (see Fig. 7a) and 0900 UTC] showed the changes to the wind field were collocated with the radar observations. After assimilation of radial winds the analysis RW wind direction differed from that of C by less than 45°. Wind profiler measurements near Cobbacombe were consistent with the insect-derived winds; they also showed a wind speed comparable to the model, but a difference in direction.
Wind increments at AT = 0800 UTC for the three cases. 27 Jul 2008: (a) x-wind component and (b) y-wind component. 28 Jul 2008: (c) x-wind component and (d) y-wind component. 29 Jun 2009: (e) x-wind component and (f) y-wind component. Speeds are in m s−1 and the color scales differ. The zero velocity contour is delineated.
Citation: Monthly Weather Review 139, 4; 10.1175/2010MWR3482.1
Wind increments at AT = 0800 UTC for the three cases. 27 Jul 2008: (a) x-wind component and (b) y-wind component. 28 Jul 2008: (c) x-wind component and (d) y-wind component. 29 Jun 2009: (e) x-wind component and (f) y-wind component. Speeds are in m s−1 and the color scales differ. The zero velocity contour is delineated.
Citation: Monthly Weather Review 139, 4; 10.1175/2010MWR3482.1
Wind increments at AT = 0800 UTC for the three cases. 27 Jul 2008: (a) x-wind component and (b) y-wind component. 28 Jul 2008: (c) x-wind component and (d) y-wind component. 29 Jun 2009: (e) x-wind component and (f) y-wind component. Speeds are in m s−1 and the color scales differ. The zero velocity contour is delineated.
Citation: Monthly Weather Review 139, 4; 10.1175/2010MWR3482.1
(ii) 28 July 2008
The wind was southeasterly across the domain, turning more easterly over land. The model wind and insect speeds were both ≈8 m s−1 at AT = 0800 UTC. The radial wind assimilation resulted in a change of direction; near the radars the RW wind was approximately easterly where the C wind was southeasterly (see wind increments for RW at AT = 0800 UTC in Fig. 7b). The RW wind showed a small decrease in speed that was greatest near the cluttered radars. Overall, this trend was consistent for all ATs, and the changes propagated downstream. Measurements from radiosondes and wind profilers in the southwest of England yielded a comparable difference in direction to the model, and the speed also differed by ±1–2 m s−1.This means that although insect-derived winds showed a lower speed than the model background, ground clutter contamination might not have caused this difference.
(iii) 29 June 2009
There was a band of strong south-southeasterly wind (10–15 m s−1) across the domain, predominantly in the western part. In the east the wind was slower—a few meters per second. Chenies was outside the range of the wind band, and Cobbacombe was on its western edge. At 0800 UTC the radial wind observations clearly reduced the wind near Cobbacombe and Clee Hill. The comparison of C and RW indicated a small decrease in wind speed at Dean Hill and a small increase in wind speed at Chenies (RW increments at AT = 0800 UTC are shown in Fig. 7c). The direction was only slightly changed. Cobbacombe observations particularly limited the western extent of the strong wind band. Later RW ATs showed similarly lower speeds around Cobbacombe and Clee Hill. Radiosondes and wind profilers observed a wind direction similar to the background’s and a slightly lower speed. These observations were comparable to the insect-derived wind observations.
2) Forecast verification against surface observations
Forecasts from each AT were verified against synoptic surface observations from a 1-h time window centered on the VT. Verification was conducted for forecast ranges of 2–6 or 7 h. Comparison with observations was made by first interpolating the model fields to the observation location. The verification statistics defined below were averaged over all observations within the domain. This analysis only focused on temperature, relative humidity (RH), and wind velocity.
(a) RW and C temperature MSE (in K2) for 27 Jul 2008, plotted for each AT at different forecast ranges. MSE increases with forecast range and decreases with analysis time. (b) VRMSE for 28 Jul 2008, for each AT at different forecast ranges. For (a) and (b) the legend denotes the analysis times. The C values are connected with solid lines and RW values are connected with dashed lines, except for AT = 1200 in (a) where C is marked with a solid diamond and RW with an open diamond.
Citation: Monthly Weather Review 139, 4; 10.1175/2010MWR3482.1
(a) RW and C temperature MSE (in K2) for 27 Jul 2008, plotted for each AT at different forecast ranges. MSE increases with forecast range and decreases with analysis time. (b) VRMSE for 28 Jul 2008, for each AT at different forecast ranges. For (a) and (b) the legend denotes the analysis times. The C values are connected with solid lines and RW values are connected with dashed lines, except for AT = 1200 in (a) where C is marked with a solid diamond and RW with an open diamond.
Citation: Monthly Weather Review 139, 4; 10.1175/2010MWR3482.1
(a) RW and C temperature MSE (in K2) for 27 Jul 2008, plotted for each AT at different forecast ranges. MSE increases with forecast range and decreases with analysis time. (b) VRMSE for 28 Jul 2008, for each AT at different forecast ranges. For (a) and (b) the legend denotes the analysis times. The C values are connected with solid lines and RW values are connected with dashed lines, except for AT = 1200 in (a) where C is marked with a solid diamond and RW with an open diamond.
Citation: Monthly Weather Review 139, 4; 10.1175/2010MWR3482.1
The RW and C temperature and RH verification statistics for all three case studies showed general trends that are not unexpected. The MSE increased with forecast range, as if the forecast diverged further from truth over time (Fig. 8a). The MSE generally decreased for later analysis times, suggesting that for later assimilation cycles the model converged toward the truth. These trends were true for all cases and for both control and experiment runs. In all cases, the difference between the C and RW statistics was small compared to the magnitude of the error. Overall the impact of radial winds was neutral, with the RW MSE sometimes larger and sometimes smaller than the C MSE. In one case (i.e., 27 July 2008) the RW temperature MSE was frequently less (Fig. 8a) than the C MSE.
The VRMSE showed similar trends to the MSE, though not as distinct (Fig. 8b). There was no obvious increase in VRMSE with forecast range. An artifact visible for 28 July 2008 (Fig. 8b) is that for the latest VTs the error increased substantially. A possible cause of this is a sudden change in the wind at these VTs that was slowly replicated in the forecast. This feature also appeared for 27 July 2008 (not shown). For that day the VRMSE suddenly increased for VT = 1500 UTC, and decreased as the VT approached 1800 UTC. In the case of the 27 July 2008, a highly variable wind field combined with precipitation in the wrong place at 1500 UTC may account for the large discrepancy in forecast and observed wind velocities. Comparison of RW and C VRMSE for all days indicated that radial winds had a neutral impact; the small positive or negative difference was much smaller than the magnitude of the VRMSE.
Finally, the mean and standard deviation of the MSE differences (RW minus C) were calculated (Table 2). On average, the VRMSE was positive, the temperature MSE was slightly positive, and the RH MSE was slightly negative. The standard deviation was much larger than the mean’s difference from zero in all cases. It is probable that no impact could be discerned from the small sample size. To conclude, the average difference between surface observations and forecast was similar for C and RW runs for each case. The impact of assimilating radial wind observations was small and neutral by this measure.
Mean and std dev of the RW minus C difference between MSE and VRMSE statistics. The data from all three cases were combined. The N indicates the number of VRMSE values contributing to the mean.
3) Divergence and precipitation verification
Precipitation forecasts were verified by comparing model precipitation with radar reflectivity-derived precipitation composite from the weather radar network. The analysis wind divergence (negative convergence) field was also examined to assess changes to convective development. Qualitative examination of precipitation rates near the time when convective showers appeared, and during their subsequent development, indicated whether the showers were forecast more accurately by assimilating radial winds. Quantitative verification was conducted by comparing rainfall accumulation over time, and calculating fractional skill scores (FSS; see Roberts and Lean 2008). The FSS examined the fractional precipitation coverage at different accumulation thresholds, over increasing sampling areas. Quantitative analysis was not possible for 29 June 2009 because of missing rainfall accumulation data products.
(i) 27 July 2008
The weather radars first registered showers between eastern England and Dean Hill by 1400 UTC (Fig. 4a). In forecasts from all C and RW ATs (not shown) there were many spurious showers. The first showers appeared in Wales from around 1100 UTC, and probably resulted from orographic-related convergence. A small line of showers appeared at around 1300 UTC in the northeast, possibly related to a sea breeze convergence line. At 1400 UTC, showers had also appeared in the correct locations, although they started too early. As the showers developed, the model continued to produce too many showers in the wrong areas (north, southeast, and west). Forecasts from later ATs produced far fewer showers, particularly in the wrong locations, implying improvement in the forecast. However, the overall precipitation patterns remained. By eye it was not apparent if either C or RW produced a better forecast. The FSS for this day showed that generally RW had slightly lower skill than C, for different accumulation thresholds and sampling radii.
(ii) 28 July 2008
In the radars, showers first appeared in the east and the midsouth just after midday (Fig. 4b). These developed into a larger region of showers with larger cells, which joined and advected slowly northwestward. The model (RW and C) showers first appeared between 1130 and 1200 UTC in the southeast corner, as well as the west coast of Wales, both erroneous locations. Figure 9a displays the surface RH, with convergence lines overlaid that indicate where showers appeared in the forecast. These convergence lines were inferred from the divergence and precipitation at 1200 UTC (not shown). The forecast precipitation at 1200 UTC for successive ATs was reduced in erroneous areas and appeared in the midsouth (near Dean Hill) where it had originally been missing; hence, the 1200 UTC analysis was fairly close to reality. In all RW and C forecasts there were too few showers in eastern England and erroneous showers in Wales. Otherwise the forecast showers in the central region were fairly accurate. The strong, erroneous convergence line in the southeast corner of England was in the wrong location to be influenced by the radial wind information, which was advected away from the area. Figure 9b shows the difference in divergence with and without radial wind observations. Note that negative values represent more convergent areas. The localized large changes in divergence (Fig. 9b) analogous to a small phase shift, and small changes to the moisture fields (not shown), indicated that the convergence lines (Fig. 9a) were perturbed slightly from the modifications to the wind field.
Convective initiation on 28 Jul 2008, analysis at 1100 UTC. (a) Surface relative humidity and wind field (vectors) of RW experiment. Dotted lines indicate locations of convergence, and are approximately where showers initiated during the next hour. (b) Difference in the divergence field (RW minus C). The convergence lines [refer to (a) for location] are slightly modified, but have not moved substantially, hence, the largest divergence changes are localized in the same area.
Citation: Monthly Weather Review 139, 4; 10.1175/2010MWR3482.1
Convective initiation on 28 Jul 2008, analysis at 1100 UTC. (a) Surface relative humidity and wind field (vectors) of RW experiment. Dotted lines indicate locations of convergence, and are approximately where showers initiated during the next hour. (b) Difference in the divergence field (RW minus C). The convergence lines [refer to (a) for location] are slightly modified, but have not moved substantially, hence, the largest divergence changes are localized in the same area.
Citation: Monthly Weather Review 139, 4; 10.1175/2010MWR3482.1
Convective initiation on 28 Jul 2008, analysis at 1100 UTC. (a) Surface relative humidity and wind field (vectors) of RW experiment. Dotted lines indicate locations of convergence, and are approximately where showers initiated during the next hour. (b) Difference in the divergence field (RW minus C). The convergence lines [refer to (a) for location] are slightly modified, but have not moved substantially, hence, the largest divergence changes are localized in the same area.
Citation: Monthly Weather Review 139, 4; 10.1175/2010MWR3482.1
Most apparent from examination of the forecast rainfall is that the radial winds made minor differences to the analyses that altered the forecast in terms of where and when convective precipitation appeared. FSS indicated only marginal differences that were not consistently positive or negative. It is probable that this case is not particularly sensitive to small changes in the wind field.
(iii) 29 June 2009
The radars revealed a line of showers appeared between northwest Wales and the southeast of England from 1400 UTC (Fig. 4c). This line of showers developed into a cluster of heavy rain cells that merged and advected slowly northward. Figure 10 compares the radar precipitation rate (Figs. 10a,b) with the forecast precipitation rate from C and RW runs at VTs of 1500 and 1700 UTC for AT = 1000. In both C and RW the showers appeared too early and in the wrong places; convection initiated along two lines oriented northwestward: one from Dean Hill to west Wales and the other from southeast England to north Wales. At VT = 1500 UTC (Figs. 10c,e) the model showers were overdeveloped, particularly in forecasts from the later ATs (1000 and 1100 UTC). The showers joined together as the forecasts evolved, and became more accurately located. At VT = 1700 UTC (Figs. 10d,f) the forecasts from all ATs were close to accurate: the two lines of showers had converged into one, of approximately the right intensity and location. It was difficult to distinguish whether the C or RW produced more accurate forecasts; in general they were comparable.
Rainfall rate on 29 Jun 2009 at (left) VT = 1500 UTC and (right) VT = 1700 UTC for (top) radar, (middle) C, and (bottom) RW. Model forecasts are from AT = 1000 UTC.
Citation: Monthly Weather Review 139, 4; 10.1175/2010MWR3482.1
Rainfall rate on 29 Jun 2009 at (left) VT = 1500 UTC and (right) VT = 1700 UTC for (top) radar, (middle) C, and (bottom) RW. Model forecasts are from AT = 1000 UTC.
Citation: Monthly Weather Review 139, 4; 10.1175/2010MWR3482.1
Rainfall rate on 29 Jun 2009 at (left) VT = 1500 UTC and (right) VT = 1700 UTC for (top) radar, (middle) C, and (bottom) RW. Model forecasts are from AT = 1000 UTC.
Citation: Monthly Weather Review 139, 4; 10.1175/2010MWR3482.1
Because of missing radar rainfall accumulation products there were no FSS calculations for this case.
5. Discussion
The various analyses of assimilating insect-derived radial winds show a small and neutral impact on the forecast of convective showers. This result contrasts with the finding that the radial wind observations caused changes to the wind velocity in the order of 45° and 2 m s−1 in some instances. The radial winds also dominated the velocity assimilation increments, perhaps as a result of their abundance compared with other wind observation types. So then why was there little effect on the subsequent forecast?
In all cases, the showers were slightly modified. For example, cells appeared slightly earlier or later, intensity was changed, clustering was affected, the location altered, or cells did not appear at all. However, the factors driving convection were not strongly altered and convergence lines remained in about the same location (Fig. 9). Note that no convergence lines were directly observed by the radars. The additional wind observations had little impact in circumstances where their information was advected away from convergence lines (e.g., 28 July 2008). Furthermore, the moisture field affected where showers initiated (not shown), and this is only indirectly affected by wind observations. It is also possible that the BCs dominated the forecast by forcing the large-scale features that assimilation could not influence (Baxter 2009; Baxter et al. 2011).
Quantitative impact evaluation of the observations indicated a neutral impact on average. No strong biases prevailed in any case or statistic. The error of the forecast (e.g., from verification against observations) was much greater than the magnitude of the difference introduced by the radial winds. The mean forecast errors may be of the same order as the surface observation errors. The neutral impact may be considered a positive outcome, since there has been no effort yet to tune the model to this observation type. Furthermore, if any one observation type were able to strongly change the whole field, it could cause instability or severely damage the forecast, rather than improve it.
Previous studies have examined the impact of wind observation assimilation. Assimilation of wind profiles and other wind observations has been shown to reduce errors in several severe-weather case studies using a 20-km resolution model with hourly assimilation/forecast cycling (Benjamin et al. 2004). However, impact could be neutral, particularly when the weather was benign. Xiao et al. (2005) observed a positive impact for a short-term forecast of a heavy rainfall event in Korea, when assimilating radial winds using 3D-Var and a 10-km resolution mesoscale model. Further radial wind assimilation tests by the Korean Meteorological Agency indicated a positive impact on forecasts of convective events, particularly short-range quantitative precipitation forecast (QPF; Xiao et al. 2008). Assimilation of radial winds from one radar into 4- and 12-km resolution versions of the Met Office UM (Rihan et al. 2008) produced an impact in the analysis and forecast that varied with the observation error. The influence on the location of showers decreased with increasing forecast lead time. The impact of the radial winds on the features of interest was also case dependent. Radial wind assimilation into a model with 2.5-km grid spacing, using precipitation echoes from 17 Doppler radars in France (Montmerle and Faccani 2009), indicated a mostly neutral impact. Positive impacts on analyses and precipitation forecasts were seen particularly in cases where the convective development was located close to the radar. However, the correlation length was in the order of 15 km, which is much shorter than the O(100 km) used in the present study.
The present study has several points of comparison to these previous studies. It used a high-resolution (1.5 km) version of the UM that produced the convective precipitation without parameterization. Furthermore, the radial wind observations were assimilated before the convective precipitation appeared. The impact seen here using insect echoes was small and neutral. Montmerle and Faccani (2009) noted forecast improvements particularly when the low-level convergence structures were well observed at the onset of precipitation, whereas degradation could occur in unobserved areas because the background could not be corrected realistically. Insects offer a means to observe low-level structures in areas not covered by precipitation observations, which could yield positive impacts.
The findings presented in this study only indicate the potential magnitude of the impact of assimilating radial winds, due to inherent limitations. First, while the observations were processed carefully using the limited tools available, the radial winds might have suffered from ground clutter contamination. Flawed observations are not anticipated to produce a more accurate forecast. It was apparent that observations from highly cluttered radars frequently induced a lower wind speed compared to the control runs; however, any bias appeared to be low since insect-derived wind observations were comparable to those from wind profilers and radiosondes. A bias correction for radial winds is not feasible because the possible opposing biases from insect flight and ground clutter contamination cannot be reliably quantized or located. Second, there were only three case studies with weak convective forcing, as evinced by the low CAPE. A case with strong local forcing might demonstrate a greater gain from insect-derived winds, but no such case arose during this project. However, if insect-derived winds are routinely assimilated, then cases like those presented here will be common.
The use of these observations is naturally limited to the range and height to which insects are detected. The U.K. single-polarization C-band radars yield insect observations that cover an area with diameter ∼40 km, which limits the likelihood of observing convective development. Vertical coverage is limited to the convective boundary layer (1–2 km) so may offer little information about the vertical wind structure. Here, the assimilation of observations over a series of short assimilation cycles means the information is advected across the domain to cover a larger region. This has the potential for affecting the vertical structure, for example by shifting convergence lines. Coverage will improve when there are more Doppler radars, and with the U.K.’s dense radar network this coverage should be reasonable. Dual polarization will also increase coverage; insects were detected to 40–50 km with a dual-polarized C-band radar in southeast England in 2007 (Rennie et al. 2010b). Furthermore, with the use of four-dimensional (4D)-Var, a more complete wind field will be retrieved (Sun and Crook 1997).
There remains a scope for improving the assimilation by optimizing the background error covariance matrices for radial wind assimilation, including greater consideration of the balances imposed by the assimilation. For example, hydrostatic balance may not be appropriate for grid spacing less than 20 km in regions where convection is occurring (Vetra-Carvalho et al. 2010). A reduction of the background correlation length scales could also be considered. In the current formulation used in the limited-area version of the Met Office variational analysis scheme the background error covariance is closely tied to the correlation length scale, so reducing the length scale increases the background error covariance at the observation location and therefore it is hard to tell whether a closer fit to the observations is due to increased background error or reduced length scales. Of course, a closer fit to the observations does not necessarily mean an improved forecast. A Met Office work program investigating new definitions of background error correlations in the high-resolution models and the fit to synoptic scales as well as convective scales is planned. The observation errors might also be reduced, particularly if the raw observation processing is enhanced. For example, dual polarization would facilitate reduced ground clutter contamination. Future improvements to the radar network and to processing methods would improve the quality and quantity of insect observations, and so increase the observation impact. While a positive impact could not be demonstrated here, results suggest that radial winds from insects could improve a convective forecast by small-scale changes to the precipitation field.
6. Conclusions
In this study radial winds from four Doppler radars were assimilated with 3D-Var into a high-resolution version of the Met Office UM. The observations were derived exclusively from insect echoes, which were present in the convective boundary layer and detected out to a range of approximately 20 km. There were three case studies comprising days where insects were present during the morning, and convective showers appeared during the afternoon.
In one experiment, radial winds were assimilated during one assimilation cycle only. Comparison with a control run indicated that the new observation information was spread around the vicinity of the radars and advected with the wind field. In another experiment, for the 3 days radial winds were assimilated at every hour during the morning and a forecast was run for 11 h from each AT. Control runs (resembling operational assimilation and forecasting) were identical, but had no radial wind assimilation. The effect of the radial winds was evident through perturbations to the wind field at the analysis and forecast times, and in the forecast precipitation field. Overall the impact was neutral, with no clear improvement or degradation of the forecast. The additional observations did not strongly affect the location of convergence lines from which the precipitation evolved. This result is considered positive as it demonstrates that assimilating these observations did not destabilize the model.
Insect-derived observations must be treated with care to account for their potential biases from nonzero airspeed and ground clutter contamination. Future efforts utilizing advances in radar hardware and software, including dual-polarization and clutter rejection techniques, will provide two advantages. First, these will improve the extraction of insect echoes to yield more wind observations, and second the observation errors will be reduced or better quantified. Insects can provide wind observations with high spatial and temporal resolution during fine weather, which could fill data gaps to improve the short-range forecast of high-impact weather events.
Acknowledgments
This research was funded by NERC Grant NE/002137/1 as part of the Flood Risk from Extreme Events Programme. We thank the Met Office and BADC for providing the radar data. We wish to acknowledge Malcolm Kitchen, Humphrey Lean, Nigel Roberts, Mark Dixon, and Alan Grant for their assistance and advice in setting up the experiments and interpreting the verification statistics.
REFERENCES
Achtemeier, G. L., 1991: The use of insects as tracers for “clear-air” boundary layer studies by Doppler radar. J. Atmos. Oceanic Technol., 8, 746–765.
Baxter, G. M., 2009: 4D-Var for high resolution, nested models with a range of scales. Ph.D. thesis, University of Reading, Reading, United Kingdom, 184 pp.
Baxter, G. M., S. L. Dance, A. S. Lawless, and N. K. Nichols, 2011: Four-dimensional variational data assimilation for high resolution nested models. Comput. Fluids, in press, doi:10.1016/j.compfluid.2011.01.023.
Benjamin, S. G., B. E. Schwartz, E. J. Szoke, and S. E. Koch, 2004: The value of wind profiler data in U.S. weather forecasting. Bull. Amer. Meteor. Soc., 85, 1871–1886.
Browning, K. A., and R. Wexler, 1968: A determination of kinematic properties of a wind field using Doppler radar. J. Appl. Meteor., 7, 105–113.
Caumont, O., and V. Ducrocq, 2008: What should be considered when simulating Doppler velocities measured by ground-based weather radar?. J. Appl. Meteor. Climatol., 47, 2256–2262.
Caumont, O., and Coauthors, 2006: A radar simulator for high-resolution nonhydrostatic models. J. Atmos. Oceanic Technol., 23, 1049–1067.
Chapman, J. W., R. L. Nesbit, L. E. Burgin, D. R. Reynolds, A. D. Smith, D. R. Middleton, and J. K. Hill, 2010: Flight orientation behaviors promote optimal migration trajectories in high-flying insects. Science, 327, 685–685.
Dance, S. L., 2004: Issues in high-resolution limited area data assimilation for quantitative precipitation forecasting. Physica D, 196, 1–27.
Dixon, M., Z. Li, H. Lean, N. Roberts, and S. Ballard, 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.
Doviak, R., and D. Zrnić, 1993: Doppler Radar and Weather Observation. 2nd ed. Academic Press, 562 pp.
Drake, V. A., and R. A. Farrow, 1988: The influence of atmospheric structure and motions on insect migration. Annu. Rev. Entomol., 33, 183–210.
Dudley, R., 2002: The Biomechanics of Insect Flight: Form, Function, Evolution. Princeton University Press, 536 pp.
Hand, W. H., N. I. Fox, and C. G. Collier, 2004: A study of twentieth-century extreme rainfall events in the United Kingdom with implications for forecasting. Meteor. Appl., 11, 15–31.
Lean, H. W., P. A. Clark, M. Dixon, N. M. Roberts, A. Fitch, R. Forbes, and C. Halliwell, 2008: Characteristics of high-resolution versions of the Met Office Unified Model for forecasting convection over the United Kingdom. Mon. Wea. Rev., 136, 3408–3424.
Lorenc, A. C., and Coauthors, 2000: The Met Office global three-dimensional variational data assimilation scheme. Quart. J. Roy. Meteor. Soc., 126, 2991–3012.
Montmerle, T., and C. Faccani, 2009: Mesoscale assimilation of radial velocities from Doppler radars in a preoperational framework. Mon. Wea. Rev., 137, 1939–1953.
Park, S. K., and D. Županski, 2003: Four-dimensional variational data assimilation for mesoscale and storm-scale applications. Meteor. Atmos. Phys., 82, 173–208.
Rennie, S. J., A. J. Illingworth, and S. L. Dance, 2010a: On differentiating ground clutter and insect echoes from Doppler weather radars using archived data. Atmos. Meas. Tech. Discuss., 3, 1843–1860.
Rennie, S. J., A. J. Illingworth, S. L. Dance, and S. P. Ballard, 2010b: The accuracy of Doppler radar wind retrievals using insects as targets. Meteor. Appl., 17, 419–432.
Rihan, F. A., C. G. Collier, S. P. Ballard, and S. J. Swarbrick, 2008: Assimilation of Doppler radial winds into a 3D-Var system: Errors and impact of radial velocities on the variational analysis and model forecasts. Quart. J. Roy. Meteor. Soc., 134, 1701–1716.
Roberts, N. M., and H. W. Lean, 2008: Scale-selective verification of rainfall accumulations from high-resolution forecasts of convective events. Mon. Wea. Rev., 136, 78–97.
Salonen, K., H. Järvinen, S. Järvenoja, S. Niemelä, and R. Eresmaa, 2008: Doppler radar radial wind data in NWP model validation. Meteor. Appl., 15, 97–102.
Salonen, K., H. Järvinen, G. Haase, S. Niemelä, and R. Eresmaa, 2009: Doppler radar radial winds in HIRLAM. Part II: Optimizing the super-observation processing. Tellus, 61A, 288–295.
Stewart, L. M., S. L. Dance, and N. K. Nichols, 2008: Correlated observation errors in data assimilation. Int. J. Numer. Methods Fluids, 58, 1521–1527, doi:10.1002/fld.1636.
Stewart, L. M., J. Cameron, S. L. Dance, E. English, J. Eyre, and N. K. Nichols, 2009: Observation error correlations in IASI radiance data. Mathematics Rep. Series 01/2009, Department of Mathematics, University of Reading, 26 pp. [Available online at http://www.reading.ac.uk/maths/research/maths-report-series.aspx.]
Sun, J., 2005: Convective-scale assimilation of radar data: Progresses and challenges. Quart. J. Roy. Meteor. Soc., 131, 3439–3463.
Sun, J., and N. A. Crook, 1997: Dynamical and microphysical retrieval from Doppler radar observations using a cloud model and its adjoint. Part I: Model development and simulated data experiments. J. Atmos. Sci., 54, 1642–1661.
Vetra-Carvalho, S., M. Dixon, S. Migliorini, N. K. Nichols, and S. Ballard, 2010: Breakdown of hydrostatic balance at convective scales in the forecast errors in the Met Office Unified Model. Department of Mathematics, University of Reading, Preprint MPS_2010-10, 18 pp. [Available online at http://www.reading.ac.uk/web/FILES/maths/preprint_10_10_Sanita.pdf.]
Waldteufel, P., and H. Corbin, 1979: On the analysis of single-Doppler radar data. J. Appl. Meteor., 18, 532–542.
Wood, C. R., D. R. Reynolds, P. M. Wells, J. F. Barlow, I. P. Woiwod, and J. W. Chapman, 2009: Flight periodicity and the vertical distribution of high-altitude moth migration over southern Britain. Bull. Entomol. Res., 99, 525–535.
Xiao, Q., Y.-H. Kuo, J. Sun, W.-C. Lee, E. Lim, Y.-R. Guo, and D. M. Barker, 2005: Assimilation of Doppler radar observations with a regional 3D-Var system: Impact of Doppler velocities on forecasts of a heavy rainfall case. J. Appl. Meteor., 44, 768–788.
Xiao, Q., and Coauthors, 2008: Doppler radar data assimilation in KMA’s operational forecasting. Bull. Amer. Meteor. Soc., 89, 39–43.
