• Open Access

Identification of NWP humidity biases using high-peaking water vapour channels from IASI


  • F. I. Hilton,

    1. Met Office, FitzRoy Road, Exeter EX1 3PB, UK
    Search for more papers by this author
    • The contribution of these authors was written in the course of their employment at the Met Office, UK, and is published with the permission of the Controller of HMSO and the Queen's Printer for Scotland

  • S. M. Newman,

    Corresponding author
    1. Met Office, FitzRoy Road, Exeter EX1 3PB, UK
    • Met Office, FitzRoy Road, Exeter EX1 3PB, UK.
    Search for more papers by this author
    • The contribution of these authors was written in the course of their employment at the Met Office, UK, and is published with the permission of the Controller of HMSO and the Queen's Printer for Scotland

  • A. D. Collard

    1. European Centre for Medium-Range Weather Forecasts, Shinfield Park, Reading RG2 9AX, UK
    Current affiliation:
    1. IMSG/EMC/NCEP, NOAA, WWB #207, 5200 Auth Road, Camp Springs, MD 20746, USA
    Search for more papers by this author


High-peaking water vapour channels from the Infrared Atmospheric Sounding Interferometer (IASI) have been used to diagnose biases in the Met Office short-range forecasts of upper tropospheric humidity during the Joint Airborne IASI Validation Experiment (JAIVEx) field campaign. These biases are compared with those from the European Centre for Medium-Range Weather Forecasts (ECMWF). The Met Office forecasts are found to be too dry relative to IASI observations and in comparison with ECMWF forecasts globally. IASI is shown to be a useful tool for investigating such biases, and subsequent changes to the Met Office data assimilation scheme have successfully reduced this forecast bias. © 2011 Royal Meteorological Society and British Crown copyright, the Met Office

1. Introduction

The Infrared Atmospheric Sounding Interferometer (IASI) is the first infrared interferometer on an operational meteorological satellite: the European Organisation for the Exploitation of Meteorological Satellites (EUMETSAT)'s MetOp (Klaes et al., 2007). IASI (Siméoni et al., 1997) is a cross-track scanning Michelson interferometer measuring between 645 and 2760 cm−1 at 0.25 cm−1 intervals and 0.5 cm−1 spectral resolution. Hyperspectral instruments such as IASI and the Atmospheric Infrared Sounder (AIRS) on National Aeronautics and Space Administration (NASA)'s Aqua satellite are important for numerical weather prediction (NWP) and have demonstrated strong positive impact on weather forecasts (Hilton et al., 2012). Thanks to high spectral resolution and low radiometric noise, higher vertical resolution information on the temperature and water vapour content of the atmosphere is provided than has been possible with previous operational infrared sounders such as the High-resolution Infrared Radiation Sounder (HIRS).

Post-launch IASI calibration/validation (cal/val) activities included a dedicated field campaign, the Joint Airborne IASI Validation Experiment (JAIVEx). The campaign was based in Houston, Texas, during April and May 2007 (see more details in the study by Larar et al. (2010)). Two research aircraft, the Facility for Airborne Atmospheric Measurements (FAAM) BAe 146 from the UK and NASA's WB-57 high altitude platform, participated in collecting a colocated dataset of interferometer and atmospheric state observations. Associated fields of meteorological variables from operational NWP models were supplied by the Met Office and European Centre for Medium-Range Weather Forecasts (ECMWF).

This paper describes biases in NWP humidity fields that were initially identified from JAIVEx case studies (Section 2). In Section 3 we expand the analysis to explore humidity biases on a global scale. We discuss the impact of subsequent updates to the Met Office humidity variational analysis scheme in Section 4, and summarize this work in Section 5.

2. Case study results

The primary aim of the JAIVEx campaign was to investigate the performance of the IASI instrument. A key part of this analysis involved a quantitative comparison between radiative transfer simulations and IASI radiances. We used the Line-By-Line Radiative Transfer Model (LBLRTM, Clough et al. (2005)) for this purpose, initialized with closely colocated profiles of temperature and water vapour, together with trace gases such as ozone. Parameters such as surface temperature and emissivity were derived from FAAM aircraft measurements at low altitude. FAAM measurements of the atmospheric state from in situ aircraft probes and dropsondes were possible up to the maximum operating ceiling of approximately 10 km. Above this altitude, NWP short-range forecast fields from the Met Office and ECMWF global models were used to complete the atmospheric profile input to LBLRTM.

It was found that in some spectral regions, such as the relatively transparent 800–1200 cm−1 (8–12 µm) window, the radiance simulations are relatively unaffected by the choice of NWP fields. In more opaque spectral regions, however, where the temperature and water vapour weighting functions peak in the upper troposphere/lower stratosphere, the simulated radiances were acutely sensitive to variations in the forecast profiles. The underlying cause of this sensitivity is shown in Figure 1. It is apparent that the Met Office and ECMWF temperature fields for this case study are very similar (to within ± 1 K). However, the Met Office forecast exhibits a significant dry bias relative to ECMWF by as much as a factor of 5.

Figure 1.

Comparison of temperature and water vapour fields for JAIVEx case on 30 April 2007. The left-hand panels compare the fields above 400 hPa (see legend), with the differences shown on the right

The effect of these forecast humidity differences on the LBLRTM simulated spectra is shown in Figure 2, where the observed minus calculated spectral residuals are shown for one of the JAIVEx cases (results averaged over 16 IASI footprints) over the Gulf of Mexico. The 1400–1800 cm−1 (7.1–5.6 µm) region is shown, which is dominated by water vapour absorption and emission. While the LBLRTM simulated spectrum is a close match to the IASI observations when initialized with ECMWF forecast fields (brightness temperature residuals largely limited to within ± 1 K), much larger residuals (approaching − 4 K at some frequencies) are seen when using Met Office upper atmosphere fields.

Figure 2.

Case study from 30 April 2007, Gulf of Mexico, 1529 UTC overpass. Upper panel: IASI clear-sky brightness temperature spectrum overlaid with simulation using ECMWF upper atmosphere fields. Lower panel: residual differences (IASI observed-calculated LBLRTM spectrum) for both Met Office and ECMWF upper atmosphere fields, see legend. Statistics are averaged over 16 footprints. Overlaid as connected data points are Met Office observed-calculated biases from the operational IASI assimilation scheme for clear ocean observations between latitudes 10–40°N

In order to verify that these results were not peculiar to the small region comprising the case study, Figure 2 also compares the residuals with statistics over a wider geographical region covering the Gulf of Mexico from the same day, computed from monitoring statistics output by the Met Office operational IASI assimilation scheme (Hilton et al., 2009). At this time the operational assimilation of IASI data at the Met Office employed the RTTOV-7 fast radiative transfer model (Saunders et al., 2002) with coefficients based on the kCompressed Radiative Transfer Algorithm (kCARTA) (Strow et al., 1998). Averaged observed minus forward-modelled 6-h forecast background (O-B) values for a subset of water vapour channels show very similar behaviour (i.e. a large negative bias for channels with water vapour weighting functions peaking higher in the atmosphere).

Further case studies corresponding to UK winter and Alaska winter conditions were also analysed (not shown). These cases also revealed a significant dry bias for Met Office short-range (6-h) forecast humidity fields relative to ECMWF.

3. Globally widespread model biases

It is instructive to consider whether the results in Section 2 are consistent with analysis on a global scale, i.e. whether the humidity biases identified from IASI water vapour channels can be attributed to fundamental properties of the NWP model and assimilation scheme.

Figure 3 presents mean O-B statistics from the Met Office and ECMWF (Collard and McNally, 2009) operational IASI assimilation schemes for three latitude bands. It is clear firstly that there is a tendency for Met Office biases to be negative for the higher-peaking water vapour channels, in contrast to a small positive bias for ECMWF for this period. Secondly, the same behaviour is seen for all three geographic ranges. A large negative O-B for the Met Office forecast is consistent with the findings of the JAIVEx campaign, and with a dry bias in the upper level humidity fields. The ECMWF positive bias suggests that the upper troposphere was slightly too moist.

Figure 3.

IASI observed-calculated biases for ECMWF and the Met Office operational assimilation schemes for all clear, night-time, sea observations on 30 April 2007, showing channels in the strongly absorbing water vapour band between 1300 and 1600 cm−1. The biases are divided into latitude bands as follows: 70–20°N Northern Hemisphere; 20°N–20°S Tropics; 20–70°S Southern Hemisphere. The plotted data are from the 314-channel subset of Collard (2007)

Further evidence for the global nature of the Met Office dry bias is presented in Figure 4, where the monthly mean analysis humidity fields for the Met Office and ECMWF from April 2007 are compared. It is striking that the Met Office analysis is extremely dry near the tropopause relative to ECMWF. Other monthly periods show similar behaviour.

Figure 4.

Zonal mean analysis relative humidity fields for April 2007. The upper plot shows the Met Office field as a function of pressure (hPa), while the lower plot shows the difference (Met Office-ECMWF). The black contours at the tropopause level indicate a relative humidity difference more negative than − 8%

4. Discussion

As evidenced in Figure 4, the Met Office analyses are drier than analyses for ECMWF and NCEP (not shown) around the tropopause across all latitudes during 2007, and have much weaker gradients in humidity. Investigations into the reasons for the dry bias found evidence that the main cause is the data assimilation scheme rather than a systematic dry bias in the Met Office Unified Model (UM) itself: long climate runs of the UM without any data assimilation relax towards a climatology more similar to the ECMWF and NCEP analyses, with stronger humidity gradients and generally moister conditions (Sean Milton, 2008, pers. comm.). This is suggestive of the introduction and reinforcement of the dry bias at analysis time.

Several factors in the 4D-Var assimilation scheme (Rawlins et al., 2007) could account for drying of the upper troposphere. After implementation, 4D-Var was found to be producing large negative (drying) increments at around 8 km. In order to counteract this unwanted behaviour, the stratospheric water vapour was constrained to be between 1 and 3 mg kg−1, which removed approximately 80–90% of the drying increments. However, the tropopause has a simple definition in the scheme, identified at the time by a globally constant value of 2.5 potential vorticity units (PVU), provided that this was between 400 and 100 hPa. This definition led to increments in the upper troposphere instead of the stratosphere, the main effect of which was to dry the upper troposphere, particularly in the extra-tropics (David Jackson, 2008, pers. comm.).

In November 2009, a package of changes to the Met Office humidity assimilation scheme, designed to address the dry bias, was made operational. In particular, the application of stratospheric water vapour increments was removed altogether above a raised definition of the tropopause at 5 PVU. In addition, a less conservative approach was taken to the assimilation of relative humidity data from certain types of radiosonde, allowing the data to be used at temperatures as cold as − 80 °C (Vaisala RS-90/92) or − 60 °C (Vaisala RS-80) instead of the previous − 40 °C threshold. The former change was expected to improve the bias to high-peaking IASI water vapour channels by about 0.3–0.4 K and the latter by a further 0.1 K.

Figure 5 demonstrates the change in bias between the Met Office 6-h forecast and IASI following the introduction of these measures, showing that the expected 0.5 K improvement was seen in the high-peaking water vapour channels. Figure 5 also shows, however, that a cold (dry) bias remains in the Met Office short-range forecasts around the tropopause level.

Figure 5.

Met Office IASI monitoring statistics (mean observed-calculated brightness temperatures) for a subset of IASI channels, presented for the period 23 October to 20 November 2009. The ‘MetDB channel number’ encompasses the following spectral ranges: 1–132 mainly CO2 temperature sounding, 133–146 window channels, 147–161 ozone, 162–280 mainly H2O channels. The high-peaking water vapour channels lie in the range illustrated by the red box. A set of changes to the Met Office humidity assimilation scheme (amongst others) became operational on 10 November 2009

It is expected that changes to the Met Office analysis scheme will bring further improvements. One option is to introduce a new humidity control variable into the 4D-Var scheme. Figure 1 of McNally et al. (2006) presents mean O-B departures for observations from the AIRS against ECMWF forecasts from August 2003. The figure shows negative bias for more than 50% of the AIRS channels in the 1250–1650 cm−1 water vapour band, unlike the IASI statistics from ECMWF in Figure 3. This is unlikely to be an instrument effect, as AIRS and IASI show biases relative to each other of less than 0.2 K across the spectrum (Hilton et al., 2012). In October 2003, ECMWF changed the humidity formulation of their 4D-Var, from specific humidity to a relative humidity term normalized by the standard deviation of forecast error appropriate for the degree of humidity (Hólm et al., 2002). The move towards a too-moist analysis resulted from a change to the ECMWF model physics (Tompkins et al., 2005) coupled with the staged introduction of an adaptive variational bias correction scheme (Auligné et al., 2007) in January and September 2006. In the latter, radiance observations were bias corrected against the model background, effectively removing any anchoring constraint that the data may have upon the analysis. (Tony McNally, 2010, pers. comm.).

Trials at the Met Office (Ingleby et al., 2011, pers. comm.) of a new formulation of humidity control variable in 4D-Var similar to that of Hólm et al., (2002) have shown positive results. The fit of background and analysis to humidity-sensitive satellite channels is improved—for upper tropospheric channels the root mean square fit to background is improved by about 3%. This new formulation was applied operationally at the Met Office beginning in July 2011. A further possibility would be to use well-calibrated observations, such as those from IASI or AIRS, as anchoring observations for the upper tropospheric humidity. Initial crude tests assimilating all satellite humidity channels without bias correction demonstrated an improvement in fit to high-peaking IASI water vapour channels of around 0.4 K, albeit with a detrimental effect on forecast impact. In practice only a very few channels would be carefully selected for assimilation without bias correction. However, this method would require significant further testing as a similar experiment at ECMWF with just two AIRS water vapour channels also showed negative forecast impact.

5. Summary

Since launch, IASI has demonstrated high absolute calibration accuracy (Larar et al., 2010) and excellent stability suitable for long-term climate monitoring (Wang et al., 2010). This work shows that IASI operational observed-calculated brightness temperature statistics are a valuable tool for identifying and monitoring NWP model biases for both localized case studies and on a global scale. Daily monitoring reports from NWP centres are available in near real time. (Monitoring reports for radiance assimilation are available through the Satellite Application Facility for Numerical Weather Prediction (NWP SAF). This is a EUMETSAT-funded activity that exists to coordinate research and development between NWP centres. The plots can be accessed via http://research.metoffice.gov.uk/research/interproj/nwpsaf/monitoring.html.). In particular, statistics for IASI high-peaking water vapour channels are shown to be very sensitive to model humidity in the upper troposphere/lower stratosphere region. This work provided important evidence to prompt changes to the Met Office humidity assimilation scheme, which have helped to reduce a forecast dry bias around the tropopause level. The remaining bias will be tackled through future enhancements to the Met Office scheme.


This work benefited greatly from discussions and investigations involving Sean Milton, Rick Rawlins, David Jackson, Bruce Ingleby and Sid Clough at the Met Office and Tony McNally at ECMWF. We thank Nigel Atkinson and Peter Schlüssel for timely provision of IASI data during the JAIVEx campaign, and the contributions of many people involved in JAIVEx are gratefully acknowledged. This work was partially funded under EUMETSAT contract Eum/CO/06/1596/ PS. Airborne data were obtained using the BAe-146-301 Atmospheric Research Aircraft (ARA) flown by Directflight Ltd and managed by the Facility for Airborne Atmospheric Measurements (FAAM), which is a joint entity of the Natural Environment Research Council (NERC) and the Met Office.