In this study, Observing System Simulation Experiments (OSSEs) are conducted to analyze the impact of assimilating surface sensitive infrared radiance observations over land and sea ice. This type of assimilation has not yet been successfully implemented at operational weather centers. Infrared radiance from AIRS (Atmospheric Infrared Sounder) and IASI (Infrared Atmospheric Sounding Interferometer) is simulated from the Nature Run (NR) provided by European Centre for Medium-Range Weather Forecasts and assimilated in a 3D-Var. analysis system. A control simulation was generated excluding the new data source, but including all data assimilated operationally at the Canadian Meteorological Center. Experiments were conducted allowing surface sensitive channels to be assimilated over all surfaces or excluding Polar Regions. Resulting forecasts were intercompared and validated against NR fields. Results indicate significant positive impacts in the tropics and Southern Hemisphere extratropics and more modest impacts in the Northern Hemisphere extratropics. Some limitations of the OSSE approach are identified, linked to the different forecast systems used for the NR and the assimilation and higher cloud contamination in Polar Regions. This analysis provides useful insight in preparation for the assimilation of real radiance observations.
Observing System Simulation Experiments (OSSEs) are an effective means for evaluating the potential impact of proposed observing systems on numerical weather prediction (NWP) (Arnold Jr. and Dey [
OSSEs involving satellite observations date back to 1970s, when many experiments were conducted (Gordon et al. [
Environment Canada (Garand et al. [
The purpose of this work is to study the potential impact of assimilating surface sensitive infrared radiance observations over land and sea ice. Currently, this is not achieved yet at operational NWP centers, which represents an underuse of that rich information on low level temperature and humidity. Over land, only radiance which is not sensitive to the surface or to low clouds is assimilated. The challenge is to demonstrate the added value on top of that provided by other data types such as surface, aircraft, ground profiler observations, and satellite observations. The strategy is to work first with simulated data to get a sense of the limitations before actually attempting this type of assimilation in a real system. The additional data come from hyperspectral infrared sounders AIRS (Atmospheric Infrared Sounder) and IASI (Infrared Atmospheric Sounding Interferometer). Section
The present study is based on the OSSE configuration used in previous studies [
Assigned observation error as a function of wavenumber (cm−1) for the 142 IASI channels assimilated. Black dots identify surface sensitive channels.
A set of three simulations are performed: the control simulation “CNTLEXP0” and two experimental simulations “OSSEXP1” and “OSSEXP2” (labels appearing in graphs), hereafter referred to as CNTL, EXP1, and EXP2, respectively. As detailed in our previous OSSE work [
In addition to these observations, infrared radiance from surface sensitive channels of AIRS and IASI has been assimilated over land and sea ice under restrictive conditions. High resolution topography data is used as one of the limiting criteria. A local standard deviation of 50 m in the height of topography (at 50 km scale) is imposed in order to limit the assimilation to relatively flat, uniform terrain. That criterion eliminates about 35% of land masses. Furthermore, the additional AIRS and IASI radiance observations are assimilated in regions characterized by high surface emissivity, that is, >0.97, for surface sensitive channels in order to limit errors due to that important parameter. Over sea ice, a complete ice cover is required (ice fraction > 0.99). Over land, a local estimate of surface skin temperature is obtained using a window channel in regions assumed clear. That estimate must be within 4 K of that from the model first guess for assimilation to be allowed. Simulated radiance is assumed to be unbiased since it was computed from the same radiative transfer model (RTM) as the one used in the assimilation. This choice may contribute to an overestimation of positive impact, but we recall that a different forecast model and a different SST are used in the NR and assimilation to avoid the situation of “identical twin” experiments (Cardinali et al. [
The first two-month experiment, EXP1, included the assimilation of surface sensitive infrared radiance observations over all surfaces where no clouds are detected. This is the only difference with respect to the control experiment in which these observations are assimilated over ice-free oceans only. Standard deviation (std.) differences between CNTL and EXP1 forecasts, computed against the ECMWF Nature Run, were computed. Figure
Global distribution of std. difference of 72 h 500 hPa temperature between CNTL and (a) EXP1 and (b) EXP2 evaluated against the Nature Run. Blue regions indicate deterioration while red regions are indicative of improvement with respect to the control.
Profile of std. difference between CNTL and (a) EXP1 and (b) EXP2 temperature, computed against the Nature Run over the global domain, as a function of lead time into the forecast. Blue regions indicate deterioration while red regions are indicative of improvement with respect to the control.
The question arises: why the assimilation appears to be more problematic in Polar Regions? A recognized difficulty is that of correct identification of cloud conditions. Frequent surface inversions with often nearly isothermal profiles above the boundary layer make both clear sky identification and subsequent assimilation a challenge. Here, as discussed elsewhere [
Difference between all-sky and clear sky brightness temperature (K) assimilated in experiments EXP1 (a) and EXP2 (b) for AIRS channel 787 for a 6 h period centered on 15 February 2006, 18 UTC.
Standard deviation difference between EXP2 temperature (°C) forecasts and Nature Run (a, b) and own cycle (c, d) analyses for the Northern Hemisphere (a, c) and Southern Hemisphere (b, d) extratropics regions. Blue regions indicate deterioration while red regions are indicative of improvement with respect to the control.
In EXP2, the volume of assimilated radiance observations for surface sensitive channels increases by about 15% with respect to CNTL. Both AIRS and IASI sensors provide additional observations over land at about 800 locations per day. Here we examine various common measures of impact pertaining to EXP2. Figure
Profile of standard deviation differences (solid lines) for EXP2 (red) and CNTL (blue) against the NR for the tropics (a) and extratropics south (b) regions at 120 hr; broken lines represent the bias. The level of significance above 90% is indicated. The three variables UU (zonal wind, m/s), GZ (geopotential, dam), and TT (temperature, °C) are identified on the
A common measure used in NWP to evaluate forecasts is anomaly correlation. Figure
Differences in anomaly correlation (%) at 850 hPa (a) and 500 hPa (b) for temperature between EXP2 and CNTL, validated against the NR, as a function of forecast lead time for the four regions indicated. Gray shading above the dotted line indicates that the impact is positive at 95% significance level.
Results derived from time series provide additional insight, depicting the level of consistency in scores from individual forecasts. Figure
Standard deviation time series against the NR of EXP2 (red) and CNTL (blue) 850 hPa temperature (K) for 24 h, 72 h, and 120 h forecasts for the tropics (a) and extratropics south (b) regions. Indicated on the right is the percentage of cases where EXP2 improves over CNTL as well as the global std. difference.
In this study, Observing System Simulation Experiments were conducted to evaluate the potential impact of assimilating infrared radiance observations over land and sea ice. OSSEs offer the advantage of benefiting from a predefined “true” atmospheric state allowing sensitivity studies in a controlled environment. As shown here, OSSEs provide much insight on strengths and limitations to expect with such an assimilation using real observations. It was found, in particular, that there are inherent difficulties in Polar Regions, with negative consequences spreading to lower latitudes. These are likely linked to the assimilation of cloud contaminated radiance. However, it was also shown that, for this specific OSSE, cloud contamination may be enhanced due to the difference between the atmospheric models used in the NR and in the assimilation. A lack of consistency was noted between results evaluated against own and Nature Run analyses for Northern Hemisphere extratropics region. This inconsistency is likely linked to differences in boundary layer physics of the ECMWF and EC models. In contrast, consistent and largely positive impact results were obtained for the tropics and Southern Hemisphere regions. Since Environment Canada is planning to upgrade the resolution of its global model to 15 km, future OSSE work will require higher resolution Nature Run. This is now possible using the recently released two-year Nature Run at 7 km resolution and 30 min intervals by the Global Modeling and Assimilation Office (
Assimilation with real data is a natural continuation to this study. The context is now favorable, with Environment Canada now benefiting from a higher resolution analysis and the implementation of a 4D ensemble variational assimilation system (Buehner et al. [
The authors declare that there is no conflict of interests regarding the publication of this paper.