Accounting for correlated error in the assimilation of high-resolution sounder data

Size: px
Start display at page:

Download "Accounting for correlated error in the assimilation of high-resolution sounder data"

Transcription

1 QuarterlyJournalof the RoyalMeteorologicalSociety Q. J. R. Meteorol. Soc. : , October 4 B DOI:.2/qj.236 Accounting for correlated error in the assimilation of high-resolution sounder data P. P. Weston,* W. Bell and J. R. Eyre Met Office, Exeter, UK *Correspondence to: P. P. Weston, Met Office, FitzRoy Road, Exeter EX 3PB, UK. peter.weston@metoffice.gov.uk This article is published with the permission of the Controller of HMSO and the Queen s Printer for Scotland. Until January 3, data from the high-resolution sounder IASI were assimilated with a diagonal observation-error covariance matrix within the Met Office 4D-Var assimilation scheme, assuming no correlation between channels. The errors were inflated to account indirectly for known inter-channel error correlations. This is sub-optimal as it artificially down-weights observations from these instruments. The true nature of these correlations for IASI are estimated here using data from the Met Office 4D-Var assimilation scheme and a posteriori diagnostics based on analysis and background departures. The diagnosed matrices are symmetrised and reconditioned, to make them suitable for use in the operational assimilation scheme. These matrices have been tested in full assimilation experiments. The results of these experiments show that using the new matrices improves forecast accuracy due to more weight in the assimilation being given to the IASI observations, particularly those from water-vapour-sensitive channels. Key Words: data assimilation; observation errors; error correlations; IASI Received 7 June 3; Revised 22 November 3; Accepted 25 November 3; Published online in Wiley Online Library 2 April 4. Introduction Numerical weather prediction (NWP) models require accurate initial conditions in order to produce accurate forecasts. The true state of the atmosphere at any given time is unknown, so data assimilation is used to blend observations and atmospheric state estimates from previous short-range forecasts (model background) to produce the best estimate of the initial state, which is called the analysis. This analysis is then used as an initial condition for the forecast runs. The Met Office currently employ an incremental four-dimensional variational (4D-Var) assimilation scheme (Rawlins et al., 7). Satellite data are among the most important sources of observations used in data assimilation for NWP due to their uniform global coverage and relatively high accuracy. In contrast, conventional observations are clustered over densely populated areas but are sparse elsewhere. A good understanding and accurate specification of the observation errors are vital so that the observations are suitably weighted against the model background. Errors in satellite data arise from a variety of sources including instrument noise, representativeness errors, forward model errors and other preprocessing errors. The statistics of errors from instrument noise are well known through in-orbit measurements of calibration targets. However, understanding of the characteristics of the other sources of error is limited. The total error can be estimated using statistical approaches such as that proposed by Desroziers et al. (5). Data from high spectral resolution sounders, such as the Infrared Atmospheric Sounding Instrument (IASI, on the EUMETSAT MetOp-A satellite), are assimilated into NWP models (e.g. Hilton et al., 9), and it has been shown that these instruments are amongst those which provide the largest benefit to NWP forecast accuracy (Joo et al., 3). The error covariance matrices used have usually been diagonal, assuming no correlations between different channels. However, previous work at Reading University and the Met Office (Stewart, 9; Stewart et al., 3) and at ECMWF (Bormann and Bauer, ; Bormann et al., ) demonstrated that correlations exist in IASI data, particularly for channels sensitive to water vapour. These correlations were suspected to exist when the data were initially assimilated, so the assumed error values were artificially inflated to indirectly account for this. Accounting for the correlations directly should allow more weight to be applied to IASI observations, particularly those from water-vapour-sensitive channels, thus improving the use of these instruments and the accuracy of the analysis. Section 2 outlines the theory and methods used to diagnose the observation-error covariances. Section 3 shows the results of performing a diagnostic technique described by Desroziers et al. (5) on IASI data to estimate the true structure of the observation-error covariance matrices. Section 4 shows the effects of accounting for correlations in a D-Var assimilation and how the diagnostic matrices are then modified to make them suitable for use in the Met Office 4D-Var assimilation scheme. Section 5 describes results from full assimilation experiments which account

2 Accounting for Correlated Error in Sounder Data Assimilation 242 for inter-channel error correlations in the assimilation of IASI observations. Finally section 6 draws conclusions from this study. 2. Theory and methods 2.. Data assimilation Data assimilation is the process of combining information from the model background and observations to produce an analysis. This is done by minimising a cost function of the form J(x) = ( x x b ) T B ( x x b) 2 + { } T y H(x) R { y H(x) } (), 2 where x is the model state vector, x b is the background state vector produced from a previous short-range forecast, y is the vector of observations, B is the background-error covariance matrix, R is the observation-error covariance matrix and H is the observation operator. The cost function displayed here is the 3D-Var cost function, but can easily be extended to four dimensions (Ide et al., 997). Also, additional terms can be added, e.g. to enable variational bias correction (Dee, 4). The observation- and background-error covariance matrices provide information on the error characteristics of the observations and background and therefore how much weight is given to each source of information. The second derivative of the cost function (Eq. ()) is the Hessian and is given by J = B + H T R H, (2) where H is the Jacobian matrix of the observation operator (Ide et al., 997). The conditioning of the Hessian matrix is linked to the speed at which the minimisation of the cost function converges (Haben et al., ). This is quantified by the condition number, which is defined as the ratio of the largest eigenvalue to the smallest eigenvalue of the matrix and effectively describes the inverse of the distance between the matrix and the set of singular or non-invertible matrices. Therefore, the larger the condition number of (2) is, the worse the conditioning and hence the slower the convergence of the cost function. However, at most NWP centres, including the Met Office, an incremental cost function is used, in which J is a function of δx = x x g,whereδx is the increment and x g is a known reference state (Bannister, 8). In addition, a control variable transform is applied before the cost function is minimised, and this takes the form δx = Uχ = U h U v U p χ, where U p is the parameter transform, U v is the vertical spatial transform and U h is the horizontal spatial transform (Lorenc et al., ). The control variable transform simplifies the cost function and the corresponding Hessian then becomes J = I + B T/2 H T R HB /2, (3) so that the smallest eigenvalue the Hessian can have is. The conditioning of the Hessian is then given by the conditioning of the second term B T/2 H T R HB /2 which is directly dependent on the conditioning of the R matrix. Therefore using an illconditioned R matrix in 4D-Var can result in convergence taking longer Observation errors Each observation assimilated into the model has a corresponding error. The two main types of observation error are systematic and random errors. Systematic errors are biases in the data. Observations can be successfully assimilated only if their bias is reduced to acceptable levels. All satellite data which are assimilated into the model are put through a bias correction procedure, which uses atmospheric predictors to remove any biases (Harris and Kelly, ). Deficiencies in the bias correction can lead to some systematic errors appearing in the assimilation, which can lead to some correlations in the residual errors, which are treated as random. Random errors are those which are represented in the R matrix. An observation error is defined as ɛ o = y o H(x t ), (4) where ɛ o is the observation error, y o is an observation, H is the observation operator and x t is the truth at model resolution. This can be split up into two parts using the fact that y t, the errorfree observation, is equal to H t (x t ), the error-free observation operator acting on the truth in model space, as follows: ɛ o = y o H(x t ) = y o H(x t ) y t + H t (x t ) = (y o y t ) + { H t (x t ) H(x t ) } = ɛ e + ɛ f, (5) where ɛ e is the error due to instrument noise and ɛ f is the error in the observation operator. Assuming no correlation between instrument noise and errors in the observation operator, the corresponding error covariance matrices are R = E [ ɛ o ɛ ot] = E [ ɛ e ɛ et] + E [ɛ f ɛ f T] [ + E ɛ e ɛ f T] + E [ ɛ f ɛ et] R = E + F. (6) For the majority of observations, the instrument noise is not correlated between channels, meaning that its corresponding covariance matrix, E, is diagonal. However, IASI measurements are apodised (Chamberlain, 979) which reduces the noise but introduces correlations between neighbouring channels, so in this case E is a band diagonal matrix with bands of non-zero covariances surrounding the diagonal. The channel selection used at the Met Office was chosen so that modelling the effect of the apodisation could be avoided by not choosing adjacent channels (Collard, 7). Errors in the observation operator can be caused by: forward model errors: errors in the radiative transfer model; representativeness errors: caused when there is a scale mismatch between the observations and the model; pre-processing errors: caused by an inaccurate or incomplete state vector, e.g. uncertainties in retrieved skin temperature, and all of these sources of error can contribute to correlations between different channels errors. This means that F can and probably will contain correlations represented by non-zero offdiagonal elements Estimation of observation errors To estimate the structure of the full R matrix, a diagnostic procedure introduced by Desroziers et al. (5) has been used. This uses observation minus background (O B) and observation minus analysis (O A) statistics to produce observation error variances and covariances, using the expression in Eq. (7), R = E [ {y H (x a ) }{ y H ( x b)} T ], (7) where the notation is as in sections 2. and 2.2.

3 2422 P. P. Weston et al..5. Operational Standard Deviation.5 Diagnosed 6 8 Noise Channel Index Figure. Diagnostic IASI error standard deviations from 4D-Var output, IASI instrument noise and previously used operational values. N.B. The previously used operational error standard deviations for the water-vapour-sensitive channels indexed from 7 to 38 are 4 K Correlation Figure 2. Diagnostic IASI error correlation matrix from 4D-Var output. Two assumptions are made in the derivation of Eq. (7). The first is that observation errors and background errors are independent. The second is that the R and B matrices used to produce the analysis are exactly correct. In this application of the diagnostic, this assumption will be violated due to the artificial inflation of the previously used assumed observation errors. This violated assumption can lead to some unrealistic results which will be commented on further in section 4.2. Additionally it has been shown by Desroziers et al. (9) that the observations and background errors must have sufficiently different scales for the results of this diagnostic to be reliable. 3. Results of diagnostics The Desroziers diagnostic was performed on data from the 4D-Var data assimilation system at the Met Office. The model background and analysis fields in 4D-Var are specified on the model grid and the observation operator interpolates these values to observation times and locations before running the radiative transfer model to transform model variables to radiances. It is these background and analysis values in observation space, together with the bias-corrected observations, which are used to produce the diagnostics. The observations used have been run through operational quality control (QC) and so there are some cloudy profiles where only a subset of the channels out of the complete selection are used. Diagnostics using a set of such observations and those produced from only completely clear observations yielded very similar results. All of the diagnostics have been produced using days of data, which corresponds to approximately 75 profiles or equivalently approximately 53 million observations treating each channel separately. It appears that data obtained over one day provides a sufficiently large sample to produce robust diagnostics which are almost identical to those produced from more data. Also, there is very little daily or seasonal variability in the diagnostics. Figure shows that the diagnosed error standard deviations are much smaller than the values used operationally from the initial implementation of IASI data in November 7 until January 3 in 4D-Var. This is a result of the artificial inflation of the previously used operational standard deviations to account for the inter-channel error correlations that were not modelled directly. The inflation was larger for the water-vapour-sensitive channels since this was where the correlations were thought to be largest. There is also significant inflation in the temperaturesounding channels which was implemented to be consistent with the assumed observation errors of similar channels on other sounders such as the Advanced Microwave Sounding Unit and the Atmospheric Infrared Sounder (AMSU-A and AIRS; Hilton et al., 9), and to compensate for inaccuracies in the assumed B. Prior to the development of objective methods for the estimation of R,such ad hoc error inflation was commonly used to optimise the influence of new observations. The instrument noise in Figure has been derived from principal component residuals by EUMETSAT (). For the temperature-sounding channels (selected channel indexes 86) in the spectral range cm,where the diagnosed standard deviations are close to the instrument noise, other sources of error are small compared to instrument noise. However, for the surface-sensitive channels (indexed 87 2) in the spectral range cm, there is a slightly larger difference showing that the other contributions to the errors are slightly larger for these channels. Finally for the water-vapour-sensitive channels (indexed 3 38) in the spectral range cm, the diagnosed standard deviations are much larger than the instrument noise, showing that the other sources of error are large for these channels. Figure 2 shows that the strongest positive error correlations are between water vapour and surface-sensitive channels. These are shown by the red blocks surrounding the diagonal towards the bottom right corner of the matrix. Also apparent are the weaker error correlations between temperature-sounding channels shown by the paler colours in the top left corner of the matrix. The correlations are caused by a combination of forward model error, representativeness error, apodisation and other pre-processing errors. However, it is difficult to split these up to see which source of error is the dominant cause. One way to attempt to isolate the horizontal representativeness errors is to compare the results of the diagnostic from the D-Var and 4D-Var, as done by Stewart et al. (3). However, there are some other differences in addition to the representativeness errors between the D-Var and 4D-Var, such as different state vector variables which could contribute to the differences in the diagnostics.

4 Accounting for Correlated Error in Sounder Data Assimilation Percentage Channel Index Figure 3. Percentage difference in diagnostic IASI error standard deviations from 4D-Var output run at N26 and N48 resolutions Correlation Figure 4. The correlation matrix of the difference in diagnostic IASI error covariance matrices from 4D-Var output run at N26 and N48 resolutions. Another way to partially isolate the contribution from horizontal representativeness errors is to evaluate the Desroziers diagnostic on first-guess and analysis departures from 4D-Var run at different resolutions. The Met Office 4D-Var system was run at N48 ( 27 km) resolution using an N48 background and N26 ( 6 km) resolution using an N26 background with the same QC and thinning, and hence the same set of observations. The only differences between the two runs are the forecast and data assimilation model resolutions. Performing the Desroziers diagnostic on output from these runs and inspecting differences will illustrate to what extent different IASI channels are affected by errors of representativeness. It will also show which of the correlations are caused by representativeness errors. The diagnosed errors from the N48 analysis would be expected to be larger than those from the N26 analysis. The larger the differences, the larger the representativeness error in that channel. Figure 3 shows that there is a negligible increase in representativeness error between the two resolutions in the highest-peaking temperature-sounding channels (indexed 5). The lower-peaking temperature-sounding channels (indexed 6 86) show some evidence of increased representativeness error at the lower resolution, with the increases generally larger for the lowest peaking of these channels (higher indices). The surface-sensitive channels (indexed 87 2) have relatively large changes in representativeness errors. The mid-tropospheric peaking water-vapour-sensitive channels (indexed 9 2 and 28 38) have the largest changes in representativeness errors, whereas the lowest- and highest-peaking water-vapour-sensitive channels (indexed 3 8 and respectively) have smaller changes in representativeness errors. These results are what were expected with small representativeness errors in temperature-sounding channels and larger representativeness errors in surface-sensitive and water-vapoursensitive channels, although there are a couple of anomalies to this general pattern with the small representativeness errors for the highest- and lowest-peaking water-vapour-sensitive channels. Figure 4 shows that the representativeness errors are responsible for correlations between the lowest-peaking temperaturesounding channels, surface-sensitive channels and the watervapour-sensitive channels. The unrealistic results such as the negative values on the diagonal (which cause the blue lines running the length and breadth of the matrix) for some high-peaking channels and some correlation values being larger than are caused by the violated assumptions in the Desroziers method and are not significant results. What should be stressed is that the differences between representativeness errors at N48 and N26 resolutions do not reveal the full extent of the representativeness errors. They merely suggest which channels are affected by these errors the most and which are affected by them the least. Also, the effective spatial resolution given by the footprint size of IASI varies from 2 2 km at nadir to 39 km at the edge of the scan. The diagnostics calculated to produce Figures 3 and 4 were averaged over all scan positions, meaning that quantifying representativeness error using the above method will not be entirely accurate. An alternative way to isolate horizontal representativeness error would be to keep the model resolution fixed and compare diagnostics using observations from different scan positions; this will be the subject of future investigations. 4. Matrix modifications 4.. Effect of accounting for the correlations Modelling the correlations will have an effect on the weights given to the observations in the assimilation. However, will this effect be the same for all observations or will it vary depending on the relationship between the O B departures for different channels? A pair of D-Var experiments was run using the same observations and background, but one run used a diagonal R matrix and one used a correlated R matrix (with the same diagonal values as diagnosed by Stewart et al., 3). This means that comparisons between the two experiments will isolate the effect that modelling the correlations has on the retrieval. Figure 5(a) shows that the O Bs fluctuate about zero for all channels including the surface- and water-vapour-sensitive channels (indexed 87 38) which have the most correlated observation errors. Figure 5(b) shows the normalised departures which are the original departures pre-multiplied by the inverse of the R matrix for both the diagonal and correlated matrices. This forms part of the increment which is added to the background to form the analysis. In this case, these normalised departures are larger in magnitude when using the correlated matrix (particularly for the most correlated channels). This results in larger retrieval increments for both temperature and

5 2424 P. P. Weston et al. (a) O-B (K) (a) O-B (K) (c) Pressure (hpa) (b) Normalised value (K - ) Correlated matrix Diagonal matrix Retrieval - Background (K) (d) Pressure (hpa) Correlated matrix Diagonal matrix Retrieval - Background (g/kg) Figure 5. (a) Raw and (b) normalised first-guess (O B) departures for all assimilated channels, and (c) temperature and (d) specific humidity retrieval increments, for an IASI profile located in the Atlantic Ocean. Solid blue and dotted red lines represent results when using the correlated matrix and diagonal matrix respectively. specific humidity (illustrated by Figure 5(c) and (d)) when accounting for the correlations, which implies that the retrieval is pulling closer to the observations. Thus accounting for the correlations increases the weight given to the observations, in this case. Figure 6 shows the opposite case, where accounting for the correlations results in smaller retrieval increments thus decreasing the weight given to the observations. The O Bs are all negative for the most correlated channels and the normalised departures are smaller in magnitude when using the correlated matrix. This results in the retrievals staying closer to the background for both temperature and specific humidity. These two examples show that accounting for correlations can both down- and up-weight the observations in individual cases thus supporting the findings of Bormann and Collard (2). However, it is also interesting to look at the effect which accounting for the correlations has on average over a large number of observations. Figure 7 shows that the standard deviations of the retrieval departures (O As) for most channels are increased when accounting for correlations. The effect is largest in the water vapour channels where the correlations are strongest. This means that, in general, accounting for the correlations results in a small down-weighting of the observations. Hence, directly accounting for correlations and inflating the errors both result in a downweighting of the observations overall, highlighting why error inflation is commonly found to be a pragmatic solution to counteract error correlations. However, the examples above show that error inflation will never fully compensate for neglecting error correlations. (b) (c) Pressure (hpa) Normalised value (K - ) (d) Correlated matrix Diagonal matrix Retrieval - Background (K) Pressure (hpa) Correlated matrix Diagonal matrix Retrieval - Background (g/kg) Figure 6. As Figure 5, but for a different profile located in the Atlantic Ocean Using the matrices in the assimilation The matrices diagnosed from 4D-Var output were then tested in the Met Office s assimilation scheme. All of the diagnosed matrices were asymmetric and some were not positive definite. These are examples of unrealistic features in the diagnostic covariance matrices as a result of violated assumptions in the Desroziers diagnostic mentioned in section 2.3. The matrices have to be valid covariance matrices when used in the 4D-Var assimilation scheme and hence must be both symmetric and positive definite. So, before the matrices were used in the assimilation, they were symmetrised (by taking the mean of the original matrix and its transpose) and any negative eigenvalues of the matrix were modified so that they were positive. A potential problem with accounting for correlated observation errors in the assimilation is the computational cost of inverting the full R matrix. This process can be done by the very efficient Cholesky decomposition method (Golub and Van Loan, 996), but has to be done separately for each observation and at each minimisation iteration. It has to be done for each observation because the channel selection can vary for each observation depending on the height and amount of retrieved cloud. In locations where there is detected cloud in the field of view, those channels whose temperature Jacobian matrices have % of their integrated value below the cloud top are rejected, as explained by Pavelin et al. (8). Put simply, only those channels which have most of their sensitivity above the cloud are used. In these cases, when using the full R matrix, it is only the sub-matrix formed by the relevant channel selection which will need to be inverted. The matrix has to be inverted for each minimisation iteration because the only alternative is to store the inverses of all possible

6 Accounting for Correlated Error in Sounder Data Assimilation 2425 Percentage difference Figure 7. Percentage difference in standard deviations of retrieval departures for each channel averaged over all observations in a 6 h window when using the correlated matrix compared to using the diagonal matrix. Table. The number of iterations, timings and cost function values resulting from using diagonaland full R matrices in 4D-Var. Number of Overall Final value R matrix used iterations time (s) of J Old operational diagonal version Diagnosed version with correlations channel selections ( 5); this requires too much memory to be a viable option. In addition to timings, another indicator of the health of the data assimilation is the number of iterations the minimisation takes to converge. The more iterations, the slower is the convergence of the minimisation. This can be caused by the observations and background being too different or the assumed errors being too small. It can also be caused by ill-conditioning in the Hessian (Eq. (3)). One final indicator is the value of the cost function (Eq. ()) which is calculated from the inverses of both the backgrounderror covariance matrix (B) and R. This means that the smaller the errors used, the larger the cost function will be at the start of the minimisation. The value of the cost function at the end of the minimisation is determined by how close the analysis fits to both the observations and the background. If the observation errors used in the assimilation are correctly specified, then the final cost function value should be equal to the total number of observations divided by 2, as derived by Desroziers and Ivanov () and Talagrand (999). In a typical operational data assimilation run, there are.9 million separate observations assimilated, of which.5 million are from IASI. Therefore, if the observation errors are correctly specified, then the total cost function value should be 95 with a contribution of 25 from IASI alone. Table shows that introducing the full R matrix results in a large increase (+ 28%) in the number of iterations required for the minimisation to converge. This in turn results in a large increase (+ 27%) in the cost of the analysis. What is interesting is that the percentage increase in time is almost exactly matched by the increase in iterations, suggesting that the extra time to invert the full matrices is negligible. Further tests using a fixed number of iterations confirmed that the increase in processing time associated with inverting the full matrices is less than %. The final column of the table shows that the final value of the total cost function increases significantly. Looking at the cost function values in more detail shows that the contribution from IASI increases from to 244. This value is much closer to the predicted optimum value of 25. This suggests that the cause of the increase in iterations is not that the error values used are too small. The other potential reason for a large increase in iterations is the ill-conditioning of the Hessian (Eq. (3)). Given that the Hessian s conditioning is directly proportional to the conditioning of the R matrix, this could be the cause of the observed increase in iterations. To investigate this further, the condition numbers of the full matrices used are examined. The condition number of the full matrix for IASI is and is significantly larger than that of the diagonal matrix, which is 64.. Generally a condition number of would not be considered to be large for a covariance matrix. However, the minimisation is very sensitive to the conditioning of the Hessian and hence the R matrix. This suggests that reducing the condition number of the R matrix could be a way of reducing the number of iterations required for convergence and hence improving the stability of the minimisation. This can be done by modifying the eigenvalues of the matrix Methods of reconditioning Reducing the condition number of a matrix by modifying the eigenvalues can be done in various ways. The main idea is to move the smallest and largest eigenvalues so that they are relatively closer together. The larger eigenvalues are dominant and modifying these will result in significant changes to the structure of the matrix. Therefore the modifications should be concentrated on the smaller eigenvalues. The first method tested is to set a minimum eigenvalue threshold as λ thresh = λ max κ req, (8) where λ thresh is the minimum eigenvalue threshold value, λ max is the existing maximum eigenvalue and κ req is the required condition number. Then all eigenvalues smaller than this threshold are set to the threshold and the matrix is reconstructed using a reverse eigendecomposition using the original eigenvectors and new eigenvalues (Golub and Van Loan, 996). The advantage of this method is that it keeps the largest eigenvalues constant, and so should change the overall structure of the matrix minimally. The disadvantage is that the effect of setting the smallest eigenvalues to a constant threshold results in many of the small diagonal values of the matrix getting set to an almost constant value. This effect seems to be most pronounced for the errors of channels which have the weakest inter-channel error correlations (the temperature-sounding channels). This is intuitive given that the eigenvalues of a diagonal matrix are just the diagonal values themselves and thus increasing the eigenvalues will increase the diagonal elements by the same amount. This unwanted effect means that many channels will have the same error and be given the same weight when they may really have quite different errors and should be given different weights. The second method tested is to increment the diagonal of the matrix (which has the effect of incrementing every eigenvalue by the same amount) by a quantity calculated to give the required condition number: λ inc = λ max λ min κ req, (9) κ req where λ inc is the increment and λ min is the existing minimum eigenvalue. The advantage of this method is that the relationship between the errors of different channels is unchanged, i.e. they are all incremented rather than being set to an almost constant value. The disadvantage is that the largest eigenvalues are changed albeit by a relatively small amount. This has the effect of weakening the correlations between channels slightly more than in the first method.

7 2426 P. P. Weston et al. Initial tests showed that matrices reconditioned using the first method resulted in the minimisation taking either the same or more iterations to converge than using matrices reconditioned using the second method. Also, reducing the condition number significantly using the first method resulted in most of the temperature-sounding channels using the same or very similar error values to each other. However, Figure shows that the diagnosed errors for the temperature-sounding channels vary quite significantly between.4 K for the highest-peaking channels to.25 K for the lowest-peaking channels. Therefore using the first method results in more unrealistic errors than using the second method. For these reasons the second method, using the increment, was chosen to produce the matrices which would be tested more thoroughly. The choice of matrix to use was a compromise between preserving the stability of the minimisation and using the most accurate errors. The lower the condition number chosen, the more stable the minimisation but the less accurate the errors. By testing matrices with different condition numbers and analysing the convergence of the cost function value in each case, it was found empirically that a good compromise was to use a matrix with a condition number of 67 for subsequent testing. This is very similar to the condition number (64) of the previously used diagonal matrix for IASI. Using this matrix in the original minimisation configuration resulted in 69 iterations, which took 548 s and with a final penalty value of Comparing these values with Table shows that, although these values are smaller than when using the ill-conditioned diagnostic matrices, they are still significantly larger than the control using the diagonal matrices. However, using this matrix in the operational minimisation, which has a fixed number of 6 iterations, 3 at lower resolution ( km) and 3 at higher resolution( 6 km), and uses pre-conditioning of the full Hessian, results in good convergence. It is unclear whether the need for the reconditioning of the R matrix is specific to the Met Office data assimilation system at these resolutions or whether this type of adjustment is compensating for other sub-optimalities (e.g. neglected spatial error correlations), and will also be necessary in other assimilation systems. Another potential solution to the problem of slow convergence could be to further pre-condition the Hessian, and this may become a more viable option if correlated observation errors are to be used for more observation types Effect of reconditioning on the matrices Figure 8 shows the eigenvalues of the originally diagnosed matrix and the newly reconditioned matrix. Despite the method of reconditioning incrementing all of the eigenvalues, this has a much larger relative effect on the smallest original eigenvalues than on the largest original eigenvalues. Reconditioning the matrix in this way increases the error variances of all channels by a constant amount (. K 2 ). Figure 9 shows that the corresponding error standard deviations change by differing amounts depending on the original value for each channel (.33 and.5 K when the initial error standard deviations are close to and K respectively). However, this method does preserve the structure of the standard deviations between channels. The equivalent plot (not shown) for the minimum eigenvalue method was examined and the standard deviations of all of the channels which originally had the smallest errors were all very similar after reconditioning this way. Also, the standard deviations of the water-vapour-sensitive channels, which had larger standard deviations originally, are changed by a much smaller amount. Although reconditioning using the chosen method does not affect the off-diagonal elements of the R matrix, comparing Figures 2 and shows that it does affect the correlations markedly. In general the correlations are weakened due to the larger diagonal values in the error covariance matrix. In the Eigenvalue Eigenvector number Figure 8. Eigenvalues of the diagnosed (red) and reconditioned (black) IASI error covariance matrix. Standard Deviation (K) Operational Reconditioned Diagnosed Figure 9. Operational, diagnosed and reconditioned IASI error standard deviations. channels with the strongest correlations, they are reduced by as much as Assimilation trial results The use of correlated observation errors for IASI was tested in two assimilation trials. Both controls were set up using a configuration which was operational in July with a forecast model at

8 Accounting for Correlated Error in Sounder Data Assimilation 2427 (a) 6 8 Percentage change - -2 (b) Correlation Figure. Reconditioned IASI error correlation matrix. N3 ( km) resolution and a two-stage data assimilation run at model resolutions of N8 ( km) followed by N26 ( 6 km). Control and experiment ran for 3 days between December and 3 December, and control and experiment 2 ran for days between 4 June and 4 July. Experiment will be referred to as the winter trial and experiment 2 as the summer trial. Figure shows this change generally improves the forecast accuracy, with reductions in the forecast error of most of the forecast verification metrics against both observations and analyses. The most significant improvements are in the pressure at mean sea level (PMSL) and 5 hpa geopotential height in the Southern Hemisphere with reductions in forecast RMSE of between.3 and 2.5% against observations and between.5 and.5% against analyses for these metrics. There are also smaller reductions in forecast RMSE in most of these metrics in the Northern Hemisphere. There is not a significant change in forecast RMSE of tropical winds against observations but against analyses there is a slight increase in forecast RMSE. This may be because of the correlated errors changing the fine-scale humidity structure in the analysis and this not being carried forward into the forecasts. For this reason, these apparent degradations in the Tropics are not true indicators of a reduction in forecast accuracy. Significant changes to the humidity field in the analysis have been previously found to suffer from similar problems when verifying against own analyses (Geer and Bauer, ). Figure 2 shows that the analysis fit to IASI changes significantly when using the correlated errors. Slightly more weight is being given to the high-peaking temperature-sounding channels, as shown by the reductions in standard deviation of analysis departures of 2%. This is to be expected because the error standard deviations for these channels are smaller in the full matrix than in the diagonal matrix and there are negligible correlations between these channels. The new errors for these channels are smaller by approximately a factor of 2 than the corresponding channels on the AIRS and AMSU-A instruments. This may have an effect on the minimisation convergence, but the difference in the assumed errors is consistent with the difference in instrument noise between these instruments. Much more weight is being given to the water-vapour-sensitive channels, as shown by the reductions in standard deviation of analysis departures of 5 %. This is due to the much smaller error standard deviations and despite the significant correlations in the full matrix for these channels. Conversely, less weight is Percentage change Winter trial Summer trial NH T+24 PMSL NH T+48 PMSL NH T+72 PMSL NH T+96 PMSL NH T+ PMSL NH T+24 H5 NH T+48 H5 NH T+72 H5 NH T+24 W25 TR T+24 W85 TR T+48 W85 TR T+72 W85 TR T+24 W25 SH T+24 PMSL SH T+48 PMSL SH T+72 PMSL SH T+96 PMSL SH T+ PMSL SH T+24 H5 SH T+48 H5 SH T+72 H5 SH T+24 W25 Verification metric Figure. Change in mean percentage forecast root mean squared error (RMSE) and weighted skill against (a) observations and (b) analysis for winter and summer trials of accounting for correlated errors for IASI. The verification metrics used are pressure at mean sea level (PMSL), geopotential height at 5 hpa (H5) and winds at 25 and 85 hpa (W25 and W85) for the Tropics (TR) and Northern and Southern Hemispheres (NH, SH). being given to the lower-peaking temperature-sounding channels and some of the surface-sensitive channels, as shown by the increases in standard deviation of analysis departures of 2%. This is an interesting result because the error standard deviations for these channels are similar in both the full matrix and the diagonal matrix. However, there are significant correlations between these channels in the full matrix, so this supports the discussion in section 4., which suggests that accounting for correlations down-weights observations overall. We now look at the change in background fit to observation types whose treatment is unchanged in the control and the trial; this will give an idea of how accounting for correlated errors affects the short-range forecast of those variables to which those observation types are sensitive. Figure 3 shows that the background fit to AIRS is improved for almost all channels. The improvement is largest ( 2%) for the water-vapoursensitive channels (wavenumbers greater than 25 cm )but also significant (.5.5%) for the surface-sensitive channels (wavenumbers cm ). There is a very small change (.%) in the fit to the temperature-sounding channels (wavenumbers less than 75 cm ). These results show that accounting for correlated errors in IASI data improves the shortrange forecasts of temperature near the surface and of humidity in the troposphere. There is a reduction in the standard deviation of Microwave Humidity Sounder (MHS) first-guess departures of.% for channel 3 (83 ± GHz),.6% for channel 4 (83 ± 3GHz)

9 2428 P. P. Weston et al. Percentage Change Wavenumber cm - Figure 2. Percentage change in analysis fit to IASI observations for each IASI channel averaged over the entire winter trial period. Percentage Change Wavenumber cm - Figure 3. Percentage change in background fit to AIRS observations averaged over the entire winter trial period. and.7% for channel 5 (83 ± 7 GHz). These channels peak in the lower to mid-troposphere, with the most improved fit in channel 3 which peaks in the mid-troposphere. This shows that the extra weight being given to water-vapour-sensitive IASI observations consistently improves the short-range forecasts of humidity in the mid-troposphere. 6. Conclusions Inter-channel error correlations in IASI data have been diagnosed using the Desroziers diagnostic. The results show that there are very small correlations between high-peaking temperaturesounding channels, weak correlations between lower-peaking temperature-sounding channels, stronger correlations between surface-sensitive channels and very strong correlations between water-vapour-sensitive channels. These results are consistent with previous estimates of error correlations for IASI. It has been shown that correlations between most of the channels are caused by horizontal representativeness errors, although forward model errors, vertical representativeness errors, apodisation and other pre-processing errors could also contribute. Comparisons between estimates of the diagonal of the full error covariance matrix and the previously used operational values highlight the artificial inflation of the latter to account indirectly for the neglected correlations. Further investigations into the direct effects of accounting for correlations show that this inflation of errors was justified since both accounting for correlations and inflating the diagonal errors down-weight the observations on average. It has also been shown that the instrument noise is the main source of error in the temperature-sounding channels, while it is a minor source of error in both the surface-sensitive and water-vapour-sensitive channels. Using the symmetrised versions of the diagnostic matrices in the Met Office 4D-Var assimilation scheme results in slow convergence of the minimisation. Improving the conditioning of the matrices by adding an artificial error term to the diagonals of the diagnosed matrices and using these modified matrices helps to alleviate this problem. It remains to be seen whether there are alternative solutions to this problem, for example by performing further pre-conditioning to the Hessian. The use of these reconditioned matrices has been tested in two assimilation trials. The treatment of the correlated errors for IASI in 4D-Var leads to a significant improvement in forecast accuracy, as shown by the reductions in forecast RMSE in different variables and at different forecast lead times. In addition, background fits to most AIRS and MHS channels are improved, indicating general improvements to the model temperature and humidity fields. The reason behind these improvements is that more weight is given to IASI observations, particularly those from water-vapour-sensitive channels, in the 4D-Var assimilation scheme. For this reason the use of correlated errors for IASI was implemented into the Met Office operational system in January 3. Future work will concentrate on implementing the operational use of correlated observation errors for the other hyperspectral IR sounders assimilated at the Met Office: AIRS and CrIS (the Crosstrack Infrared Sounder). This method could also be applied to improve the assimilation of data from microwave sounders such as AMSU-A, MHS and the Advanced Technology Microwave Sounder (ATMS). An extension to the work presented at the end of section 3 would be to completely isolate the representativeness errors from other sources of error. To do this, the Desroziers diagnostic could be run on output from a 4D-Var run where the observations resolution is degraded to the model resolution. This would eliminate any contribution from horizontal representativeness error and, by comparing these diagnostics to Figures and 2, the full characteristics of representativeness error could be estimated. Acknowledgements The authors acknowledge the technical and scientific guidance provided by Laura Stewart, James Cameron, Brett Candy, Fiona Smith, Ed Pavelin, Nigel Atkinson and Gordon Inverarity. The authors are also grateful to the two anonymous reviewers whose comments helped to improve the manuscript. References Bannister RN. 8. A review of forecast-error covariance statistics in atmospheric variational data assimilation. II: Modelling the forecast-error covariance statistics.q. J. R. Meteorol. Soc. 34: Bormann N, Bauer P.. Estimates of spatial and interchannel observationerror characteristics for current sounder radiances for numerical weather prediction. I: Methods and application to ATOVS data. Q. J. R. Meteorol. Soc. 36: Bormann N, Collard AD. 2. Experimentation with inter-channel error correlations with AIRS and IASI at ECMWF. In Proceedings of International TOVS Study Conference 8, Toulouse, France. (accessed 8 December 3). Bormann N, Collard AD, Bauer P.. Estimates of spatial and interchannel observation-error characteristics for current sounder radiances for numerical weather prediction. II: Application to AIRS and IASI data. Q. J. R. Meteorol. Soc. 36: Chamberlain JE The Principles of Interferometric Spectroscopy. John Wiley & Sons Ltd: Chichester, UK. Collard AD. 7. Selection of IASI channels for use in numerical weather prediction.q. J. R. Meteorol. Soc. 33: Dee DP. 4. Variational bias correction of radiance data in the ECMWF system. Proceedings of Workshop on Assimilation of High Spectral Resolution Sounders in NWP: ECMWF: Reading, UK.

10 Accounting for Correlated Error in Sounder Data Assimilation 2429 Desroziers G, Ivanov S.. Diagnosis and adaptive tuning of observationerror parameters in a variational assimilation. Q. J. R. Meteorol. Soc. 27: Desroziers G, Berre L, Chapnik B, Poli P. 5. Diagnosis of observation, background and analysis-error statistics in observation space. Q. J. R. Meteorol. Soc. 3: Desroziers G, Berre L, Chapnik B. 9. Objective validation of a data assimilation system: Diagnosing sub-optimality. Proceedings of Workshop on Diagnostics of Data Assimilation System Performance: ECMWF: Reading, UK. EUMETSAT.. EPS Product Validation Report: IASI L PCC PPF. EUMET- SAT: Darmstadt, Germany. idcplg?idcservice=get FILE&dDocName=PDF IASI L PCC PPF& RevisionSelectionMethod=LatestReleased&Rendition=Web (accessed 25 January ). Geer AJ, Bauer P.. Enhanced use of all-sky microwave observations sensitive to water vapour, cloud and precipitation, EUMETSAT/ECMWF Fellowship Programme Research report. ECMWF: Reading, UK. Golub GH, Van Loan CF Matrix Computations (3rd edn). Johns Hopkins University Press: Baltimore, MD. Haben SA, Lawless AS, Nichols NK.. Conditioning of incremental variational data assimilation, with application to the Met Office system. Tellus 63: Harris BA, Kelly G.. A satellite radiance-bias correction scheme for data assimilation.q. J. R. Meteorol. Soc. 27: Hilton F, Atkinson NC, English SJ, Eyre JR. 9. Assimilation of IASI at the Met Office and assessment of its impact through observing system experiments.q. J. R. Meteorol. Soc. 35: Ide K, Courtier P, Ghil M, Lorenc AC Unified notation for data assimilation: Operational, sequential and variational. J. Meteorol. Soc. Japan 75: Joo S, Eyre JR, Marriott R. 3. The impact of Metop and other satellite data within the Met Office global NWP system using an adjoint-based sensitivity method.mon. Weather Rev. 4: Lorenc AC, Ballard SP, Bell RS, Ingleby NB, Andrews PLF, Barker DM, Bray JR, Clayton AM, Dalby T, Li D, Payne TJ, Saunders FW.. The Met Office global three-dimensional variational data assimilation scheme. Q. J. R. Meteorol. Soc. 26: Pavelin EG, English SJ, Eyre JR. 8. The assimilation of cloud-affected infrared satellite radiances for numerical weather prediction. Q. J. R. Meteorol. Soc. 34: Rawlins F, Ballard SP, Bovis KJ, Clayton AM, Li D, Inverarity GW, Lorenc AC, Payne TJ. 7. The Met Office global four-dimensional variational data assimilation scheme.q. J. R. Meteorol. Soc. 33: Stewart LM. 9. Correlated observation errors in data assimilation, PhD thesis. University of Reading: Reading, UK. Stewart LM, Dance S, Nichols NK, Eyre JR, Cameron J. 3. Estimating interchannel observation error correlations for IASI radiance data in the Met Office system. Q. J. R. Meteorol. Soc. in press, doi:.2/qj.22. Talagrand O A posteriori evaluation and verification of analysis and assimilation algorithms. Proceedings of Workshop on Diagnosis of Data Assimilation Systems: ECMWF: Reading, UK.

Estimating interchannel observation-error correlations for IASI radiance data in the Met Office system

Estimating interchannel observation-error correlations for IASI radiance data in the Met Office system Estimating interchannel observation-error correlations for IASI radiance data in the Met Office system A BC DEF B E B E A E E E E E E E E C B E E E E E E DE E E E B E C E AB E BE E E D ED E E C E E E B

More information

Spatial and inter-channel observation error characteristics for AMSU-A and IASI and applications in the ECMWF system

Spatial and inter-channel observation error characteristics for AMSU-A and IASI and applications in the ECMWF system Spatial and inter-channel observation error characteristics for AMSU-A and IASI and applications in the system Niels Bormann, Andrew Collard, Peter Bauer Outline 1) Estimation of observation errors AMSU-A

More information

Bias correction of satellite data at the Met Office

Bias correction of satellite data at the Met Office Bias correction of satellite data at the Met Office Nigel Atkinson, James Cameron, Brett Candy and Stephen English Met Office, Fitzroy Road, Exeter, EX1 3PB, United Kingdom 1. Introduction At the Met Office,

More information

Assimilation of IASI data at the Met Office. Fiona Hilton Nigel Atkinson ITSC-XVI, Angra dos Reis, Brazil 07/05/08

Assimilation of IASI data at the Met Office. Fiona Hilton Nigel Atkinson ITSC-XVI, Angra dos Reis, Brazil 07/05/08 Assimilation of IASI data at the Met Office Fiona Hilton Nigel Atkinson ITSC-XVI, Angra dos Reis, Brazil 07/05/08 Thanks to my other colleagues! Andrew Collard (ECMWF) Brett Candy, Steve English, James

More information

DIAGNOSING OBSERVATION ERROR STATISTICS FOR NUMERICAL WEATHER PREDICTION

DIAGNOSING OBSERVATION ERROR STATISTICS FOR NUMERICAL WEATHER PREDICTION DIAGNOSING OBSERVATION ERROR STATISTICS FOR NUMERICAL WEATHER PREDICTION J. Waller, S. Dance, N. Nichols (University of Reading) D. Simonin, S. Ballard, G. Kelly (Met Office) 1 AIMS 2 OBSERVATION ERRORS

More information

ASSIMILATION OF CLOUDY AMSU-A MICROWAVE RADIANCES IN 4D-VAR 1. Stephen English, Una O Keeffe and Martin Sharpe

ASSIMILATION OF CLOUDY AMSU-A MICROWAVE RADIANCES IN 4D-VAR 1. Stephen English, Una O Keeffe and Martin Sharpe ASSIMILATION OF CLOUDY AMSU-A MICROWAVE RADIANCES IN 4D-VAR 1 Stephen English, Una O Keeffe and Martin Sharpe Met Office, FitzRoy Road, Exeter, EX1 3PB Abstract The assimilation of cloud-affected satellite

More information

Estimates of observation errors and their correlations in clear and cloudy regions for microwave imager radiances from NWP

Estimates of observation errors and their correlations in clear and cloudy regions for microwave imager radiances from NWP Estimates of observation errors and their correlations in clear and cloudy regions for microwave imager radiances from NWP Niels Bormann, Alan J. Geer and Peter Bauer ECMWF, Shinfield Park, Reading RG2

More information

New Applications and Challenges In Data Assimilation

New Applications and Challenges In Data Assimilation New Applications and Challenges In Data Assimilation Met Office Nancy Nichols University of Reading 1. Observation Part Errors 1. Applications Coupled Ocean-Atmosphere Ensemble covariances for coupled

More information

ON DIAGNOSING OBSERVATION ERROR STATISTICS WITH LOCAL ENSEMBLE DATA ASSIMILATION

ON DIAGNOSING OBSERVATION ERROR STATISTICS WITH LOCAL ENSEMBLE DATA ASSIMILATION ON DIAGNOSING OBSERVATION ERROR STATISTICS WITH LOCAL ENSEMBLE DATA ASSIMILATION J. A. Waller, S. L. Dance, N. K. Nichols University of Reading 1 INTRODUCTION INTRODUCTION Motivation Only % of observations

More information

DIAGNOSING OBSERVATION ERROR STATISTICS FOR NUMERICAL WEATHER PREDICTION

DIAGNOSING OBSERVATION ERROR STATISTICS FOR NUMERICAL WEATHER PREDICTION DIAGNOSING OBSERVATION ERROR STATISTICS FOR NUMERICAL WEATHER PREDICTION J. Waller, S. Dance, N. Nichols (University of Reading) D. Simonin, S. Ballard, G. Kelly (Met Office) EMS Annual Meeting: European

More information

Background and observation error covariances Data Assimilation & Inverse Problems from Weather Forecasting to Neuroscience

Background and observation error covariances Data Assimilation & Inverse Problems from Weather Forecasting to Neuroscience Background and observation error covariances Data Assimilation & Inverse Problems from Weather Forecasting to Neuroscience Sarah Dance School of Mathematical and Physical Sciences, University of Reading

More information

Plans for the Assimilation of Cloud-Affected Infrared Soundings at the Met Office

Plans for the Assimilation of Cloud-Affected Infrared Soundings at the Met Office Plans for the Assimilation of Cloud-Affected Infrared Soundings at the Met Office Ed Pavelin and Stephen English Met Office, Exeter, UK Abstract A practical approach to the assimilation of cloud-affected

More information

Extending the use of surface-sensitive microwave channels in the ECMWF system

Extending the use of surface-sensitive microwave channels in the ECMWF system Extending the use of surface-sensitive microwave channels in the ECMWF system Enza Di Tomaso and Niels Bormann European Centre for Medium-range Weather Forecasts Shinfield Park, Reading, RG2 9AX, United

More information

Satellite Radiance Data Assimilation at the Met Office

Satellite Radiance Data Assimilation at the Met Office Satellite Radiance Data Assimilation at the Met Office Ed Pavelin, Stephen English, Brett Candy, Fiona Hilton Outline Summary of satellite data used in the Met Office NWP system Processing and quality

More information

A New Infrared Atmospheric Sounding Interferometer Channel Selection and Assessment of Its Impact on Met Office NWP Forecasts

A New Infrared Atmospheric Sounding Interferometer Channel Selection and Assessment of Its Impact on Met Office NWP Forecasts ADVANCES IN ATMOSPHERIC SCIENCES, VOL. 34, NOVEMBER 2017, 1265 1281 Original Paper A New Infrared Atmospheric Sounding Interferometer Channel Selection and Assessment of Its Impact on Met Office NWP Forecasts

More information

Of Chessboards and Ghosts Signatures of micro-vibrations from IASI monitoring in NWP?

Of Chessboards and Ghosts Signatures of micro-vibrations from IASI monitoring in NWP? Of Chessboards and Ghosts Signatures of micro-vibrations from IASI monitoring in NWP? Niels Bormann 1, James R.N. Cameron 2, and Anthony P. McNally 1 1 European Centre for Medium-range Weather Forecasts

More information

ASSIMILATION EXPERIMENTS WITH DATA FROM THREE CONICALLY SCANNING MICROWAVE INSTRUMENTS (SSMIS, AMSR-E, TMI) IN THE ECMWF SYSTEM

ASSIMILATION EXPERIMENTS WITH DATA FROM THREE CONICALLY SCANNING MICROWAVE INSTRUMENTS (SSMIS, AMSR-E, TMI) IN THE ECMWF SYSTEM ASSIMILATION EXPERIMENTS WITH DATA FROM THREE CONICALLY SCANNING MICROWAVE INSTRUMENTS (SSMIS, AMSR-E, TMI) IN THE ECMWF SYSTEM Niels Bormann 1, Graeme Kelly 1, Peter Bauer 1, and Bill Bell 2 1 ECMWF,

More information

PCA assimilation techniques applied to MTG-IRS

PCA assimilation techniques applied to MTG-IRS PCA assimilation techniques applied to MTG-IRS Marco Matricardi ECMWF Shinfield Park, Reading, UK WORKSHOP Assimilation of Hyper-spectral Geostationary Satellite Observation ECMWF Reading UK 22-25 May

More information

Quantifying observation error correlations in remotely sensed data

Quantifying observation error correlations in remotely sensed data Quantifying observation error correlations in remotely sensed data Conference or Workshop Item Published Version Presentation slides Stewart, L., Cameron, J., Dance, S. L., English, S., Eyre, J. and Nichols,

More information

Improved Use of AIRS Data at ECMWF

Improved Use of AIRS Data at ECMWF Improved Use of AIRS Data at ECMWF A.D. Collard, A.P. McNally European Centre for Medium-Range Weather Forecasts, Reading, U.K. W.W. Wolf QSS Group, Inc., NOAA Science Center, 5200 Auth Road, Camp Springs

More information

Quantifying observation error correlations in remotely sensed data

Quantifying observation error correlations in remotely sensed data Quantifying observation error correlations in remotely sensed data Conference or Workshop Item Published Version Presentation slides Stewart, L., Cameron, J., Dance, S. L., English, S., Eyre, J. and Nichols,

More information

Bias correction of satellite data at the Met Office

Bias correction of satellite data at the Met Office Bias correction of satellite data at the Met Office Nigel Atkinson, James Cameron, Brett Candy and Steve English ECMWF/EUMETSAT NWP-SAF Workshop on Bias estimation and correction in data assimilation,

More information

Assimilation of hyperspectral infrared sounder radiances in the French global numerical weather prediction ARPEGE model

Assimilation of hyperspectral infrared sounder radiances in the French global numerical weather prediction ARPEGE model Assimilation of hyperspectral infrared sounder radiances in the French global numerical weather prediction ARPEGE model N. Fourrié, V. Guidard, M. Dahoui, T. Pangaud, P. Poli and F. Rabier CNRM-GAME, Météo-France

More information

OBSERVING SYSTEM EXPERIMENTS ON ATOVS ORBIT CONSTELLATIONS

OBSERVING SYSTEM EXPERIMENTS ON ATOVS ORBIT CONSTELLATIONS OBSERVING SYSTEM EXPERIMENTS ON ATOVS ORBIT CONSTELLATIONS Enza Di Tomaso and Niels Bormann European Centre for Medium-range Weather Forecasts Shinfield Park, Reading, RG2 9AX, United Kingdom Abstract

More information

Effect of Predictor Choice on the AIRS Bias Correction at the Met Office

Effect of Predictor Choice on the AIRS Bias Correction at the Met Office Effect of Predictor Choice on the AIRS Bias Correction at the Met Office Brett Harris Bureau of Meterorology Research Centre, Melbourne, Australia James Cameron, Andrew Collard and Roger Saunders, Met

More information

An Evaluation of FY-3C MWHS-2 and its potential to improve forecast accuracy at ECMWF

An Evaluation of FY-3C MWHS-2 and its potential to improve forecast accuracy at ECMWF An Evaluation of FY-3C MWHS-2 and its potential to improve forecast accuracy at ECMWF Heather Lawrence, final-year EUMETSAT fellow, ECMWF Supervised by: Niels Bormann & Stephen English Slide 1 China s

More information

Initial results from using ATMS and CrIS data at ECMWF

Initial results from using ATMS and CrIS data at ECMWF Initial results from using ATMS and CrIS data at ECMWF Niels Bormann 1, William Bell 1, Anne Fouilloux 1, Tony McNally 1, Ioannis Mallas 1, Nigel Atkinson 2, Steve Swadley 3 Slide 1 1 ECMWF, 2 Met Office,

More information

UPDATES IN THE ASSIMILATION OF GEOSTATIONARY RADIANCES AT ECMWF

UPDATES IN THE ASSIMILATION OF GEOSTATIONARY RADIANCES AT ECMWF UPDATES IN THE ASSIMILATION OF GEOSTATIONARY RADIANCES AT ECMWF Carole Peubey, Tony McNally, Jean-Noël Thépaut, Sakari Uppala and Dick Dee ECMWF, UK Abstract Currently, ECMWF assimilates clear sky radiances

More information

TOWARDS IMPROVED HEIGHT ASSIGNMENT AND QUALITY CONTROL OF AMVS IN MET OFFICE NWP

TOWARDS IMPROVED HEIGHT ASSIGNMENT AND QUALITY CONTROL OF AMVS IN MET OFFICE NWP Proceedings for the 13 th International Winds Workshop 27 June - 1 July 2016, Monterey, California, USA TOWARDS IMPROVED HEIGHT ASSIGNMENT AND QUALITY CONTROL OF AMVS IN MET OFFICE NWP James Cotton, Mary

More information

Direct assimilation of all-sky microwave radiances at ECMWF

Direct assimilation of all-sky microwave radiances at ECMWF Direct assimilation of all-sky microwave radiances at ECMWF Peter Bauer, Alan Geer, Philippe Lopez, Deborah Salmond European Centre for Medium-Range Weather Forecasts Reading, Berkshire, UK Slide 1 17

More information

ACCOUNTING FOR THE SITUATION-DEPENDENCE OF THE AMV OBSERVATION ERROR IN THE ECMWF SYSTEM

ACCOUNTING FOR THE SITUATION-DEPENDENCE OF THE AMV OBSERVATION ERROR IN THE ECMWF SYSTEM ACCOUNTING FOR THE SITUATION-DEPENDENCE OF THE AMV OBSERVATION ERROR IN THE ECMWF SYSTEM Kirsti Salonen and Niels Bormann ECMWF, Shinfield Park, Reading RG2 9AX, United Kingdom Abstract This article reports

More information

Satellite data assimilation for Numerical Weather Prediction II

Satellite data assimilation for Numerical Weather Prediction II Satellite data assimilation for Numerical Weather Prediction II Niels Bormann European Centre for Medium-range Weather Forecasts (ECMWF) (with contributions from Tony McNally, Jean-Noël Thépaut, Slide

More information

Rosemary Munro*, Graeme Kelly, Michael Rohn* and Roger Saunders

Rosemary Munro*, Graeme Kelly, Michael Rohn* and Roger Saunders ASSIMILATION OF METEOSAT RADIANCE DATA WITHIN THE 4DVAR SYSTEM AT ECMWF Rosemary Munro*, Graeme Kelly, Michael Rohn* and Roger Saunders European Centre for Medium Range Weather Forecasts Shinfield Park,

More information

Derivation of AMVs from single-level retrieved MTG-IRS moisture fields

Derivation of AMVs from single-level retrieved MTG-IRS moisture fields Derivation of AMVs from single-level retrieved MTG-IRS moisture fields Laura Stewart MetOffice Reading, Meteorology Building, University of Reading, Reading, RG6 6BB Abstract The potential to derive AMVs

More information

AMVs in the ECMWF system:

AMVs in the ECMWF system: AMVs in the ECMWF system: Overview of the recent operational and research activities Kirsti Salonen and Niels Bormann Slide 1 AMV sample coverage: monitored GOES-15 GOES-13 MET-10 MET-7 MTSAT-2 NOAA-15

More information

Radiance Data Assimilation

Radiance Data Assimilation Radiance Data Assimilation Rahul Mahajan for Andrew Collard NCEP/NWS/NOAA/EMC IMSG DTC GSI Tutorial 12 August 2015 Outline Introduction Different types of satellite data. Basic Concepts for Assimilating

More information

SSMIS 1D-VAR RETRIEVALS. Godelieve Deblonde

SSMIS 1D-VAR RETRIEVALS. Godelieve Deblonde SSMIS 1D-VAR RETRIEVALS Godelieve Deblonde Meteorological Service of Canada, Dorval, Québec, Canada Summary Retrievals using synthetic background fields and observations for the SSMIS (Special Sensor Microwave

More information

Assimilation of Cloud-Affected Infrared Radiances at Environment-Canada

Assimilation of Cloud-Affected Infrared Radiances at Environment-Canada Assimilation of Cloud-Affected Infrared Radiances at Environment-Canada ECMWF-JCSDA Workshop on Assimilating Satellite Observations of Clouds and Precipitation into NWP models ECMWF, Reading (UK) Sylvain

More information

Monitoring and Assimilation of IASI Radiances at ECMWF

Monitoring and Assimilation of IASI Radiances at ECMWF Monitoring and Assimilation of IASI Radiances at ECMWF Andrew Collard and Tony McNally ECMWF Slide 1 Overview Introduction Assimilation Configuration IASI First Guess Departures IASI Forecast Impacts The

More information

Towards a better use of AMSU over land at ECMWF

Towards a better use of AMSU over land at ECMWF Towards a better use of AMSU over land at ECMWF Blazej Krzeminski 1), Niels Bormann 1), Fatima Karbou 2) and Peter Bauer 1) 1) European Centre for Medium-range Weather Forecasts (ECMWF), Shinfield Park,

More information

Assimilation of Cross-track Infrared Sounder radiances at ECMWF

Assimilation of Cross-track Infrared Sounder radiances at ECMWF ssimilation of ross-track Infrared Sounder radiances at EMWF Reima Eresmaa, nthony P. McNally and Niels Bormann European entre for Medium-range Weather Forecasts Reading, Berkshire, United Kingdom Introduction

More information

IASI PC compression Searching for signal in the residuals. Thomas August, Nigel Atkinson, Fiona Smith

IASI PC compression Searching for signal in the residuals. Thomas August, Nigel Atkinson, Fiona Smith IASI PC compression Searching for signal in the residuals Tim.Hultberg@EUMETSAT.INT Thomas August, Nigel Atkinson, Fiona Smith Raw radiance (minus background) Reconstructed radiance (minus background)

More information

Improving the conditioning of estimated covariance matrices

Improving the conditioning of estimated covariance matrices Improving the conditioning of estimated covariance matrices Jemima M. Tabeart Supervised by Sarah L. Dance, Nancy K. Nichols, Amos S. Lawless, Joanne A. Waller (University of Reading) Sue Ballard, David

More information

Masahiro Kazumori, Takashi Kadowaki Numerical Prediction Division Japan Meteorological Agency

Masahiro Kazumori, Takashi Kadowaki Numerical Prediction Division Japan Meteorological Agency Development of an all-sky assimilation of microwave imager and sounder radiances for the Japan Meteorological Agency global numerical weather prediction system Masahiro Kazumori, Takashi Kadowaki Numerical

More information

Land Data Assimilation for operational weather forecasting

Land Data Assimilation for operational weather forecasting Land Data Assimilation for operational weather forecasting Brett Candy Richard Renshaw, JuHyoung Lee & Imtiaz Dharssi * *Centre Australian Weather and Climate Research Contents An overview of the Current

More information

Impact of hyperspectral IR radiances on wind analyses

Impact of hyperspectral IR radiances on wind analyses Impact of hyperspectral IR radiances on wind analyses Kirsti Salonen and Anthony McNally Kirsti.Salonen@ecmwf.int ECMWF November 30, 2017 Motivation The upcoming hyper-spectral IR instruments on geostationary

More information

Accounting for Correlated Satellite Observation Error in NAVGEM

Accounting for Correlated Satellite Observation Error in NAVGEM Accounting for Correlated Satellite Observation Error in NAVGEM Bill Campbell and Liz Satterfield Naval Research Laboratory, Monterey CA ITSC-20 Oct 27 Nov 3, 2015 Lake Geneva, WI, USA 1 Sources of Observation

More information

Doppler radial wind spatially correlated observation error: operational implementation and initial results

Doppler radial wind spatially correlated observation error: operational implementation and initial results Doppler radial wind spatially correlated observation error: operational implementation and initial results D. Simonin, J. Waller, G. Kelly, S. Ballard,, S. Dance, N. Nichols (Met Office, University of

More information

METEOSAT cloud-cleared radiances for use in three/fourdimensional variational data assimilation

METEOSAT cloud-cleared radiances for use in three/fourdimensional variational data assimilation METEOSAT cloud-cleared radiances for use in three/fourdimensional variational data assimilation G. A. Kelly, M. Tomassini and M. Matricardi European Centre for Medium-Range Weather Forecasts, Reading,

More information

IMPROVEMENTS IN FORECASTS AT THE MET OFFICE THROUGH REDUCED WEIGHTS FOR SATELLITE WINDS. P. Butterworth, S. English, F. Hilton and K.

IMPROVEMENTS IN FORECASTS AT THE MET OFFICE THROUGH REDUCED WEIGHTS FOR SATELLITE WINDS. P. Butterworth, S. English, F. Hilton and K. IMPROVEMENTS IN FORECASTS AT THE MET OFFICE THROUGH REDUCED WEIGHTS FOR SATELLITE WINDS P. Butterworth, S. English, F. Hilton and K. Whyte Met Office London Road, Bracknell, RG12 2SZ, UK ABSTRACT Following

More information

All-sky observations: errors, biases, representativeness and gaussianity

All-sky observations: errors, biases, representativeness and gaussianity All-sky observations: errors, biases, representativeness and gaussianity Alan Geer, Peter Bauer, Philippe Lopez Thanks to: Bill Bell, Niels Bormann, Anne Foullioux, Jan Haseler, Tony McNally Slide 1 ECMWF-JCSDA

More information

Diagnosis of observation, background and analysis-error statistics in observation space

Diagnosis of observation, background and analysis-error statistics in observation space Q. J. R. Meteorol. Soc. (2005), 131, pp. 3385 3396 doi: 10.1256/qj.05.108 Diagnosis of observation, background and analysis-error statistics in observation space By G. DESROZIERS, L. BERRE, B. CHAPNIK

More information

remote sensing Article Joanne A. Waller 1, *, Susan P. Ballard 2, Sarah L. Dance 1,3, Graeme Kelly 2, Nancy K. Nichols 1,3 and David Simonin 2

remote sensing Article Joanne A. Waller 1, *, Susan P. Ballard 2, Sarah L. Dance 1,3, Graeme Kelly 2, Nancy K. Nichols 1,3 and David Simonin 2 remote sensing Article Diagnosing Horizontal and Inter-Channel Observation Error Correlations for SEVIRI Observations Using Observation-Minus-Background and Observation-Minus-Analysis Statistics Joanne

More information

Imager-assisted cloud detection for assimilation of Infrared Atmospheric Sounding Interferometer radiances

Imager-assisted cloud detection for assimilation of Infrared Atmospheric Sounding Interferometer radiances Quarterly Journal of the Royal Meteorological Society Q. J. R. Meteorol. Soc. 14: 2342 2352, October 214 A DOI:1.12/qj.234 Imager-assisted cloud detection for assimilation of Infrared Atmospheric Sounding

More information

Does the ATOVS RARS Network Matter for Global NWP? Brett Candy, Nigel Atkinson & Stephen English

Does the ATOVS RARS Network Matter for Global NWP? Brett Candy, Nigel Atkinson & Stephen English Does the ATOVS RARS Network Matter for Global NWP? Brett Candy, Nigel Atkinson & Stephen English Met Office, Exeter, United Kingdom 1. Introduction Along with other global numerical weather prediction

More information

All-sky assimilation of MHS and HIRS sounder radiances

All-sky assimilation of MHS and HIRS sounder radiances All-sky assimilation of MHS and HIRS sounder radiances Alan Geer 1, Fabrizio Baordo 2, Niels Bormann 1, Stephen English 1 1 ECMWF 2 Now at Bureau of Meteorology, Australia All-sky assimilation at ECMWF

More information

Use of AMSU data in the Met Office UK Mesoscale Model

Use of AMSU data in the Met Office UK Mesoscale Model Use of AMSU data in the Met Office UK Mesoscale Model 1. Introduction Brett Candy, Stephen English, Richard Renshaw & Bruce Macpherson The Met Office, Exeter, United Kingdom In common with other global

More information

Assimilation Experiments of One-dimensional Variational Analyses with GPS/MET Refractivity

Assimilation Experiments of One-dimensional Variational Analyses with GPS/MET Refractivity Assimilation Experiments of One-dimensional Variational Analyses with GPS/MET Refractivity Paul Poli 1,3 and Joanna Joiner 2 1 Joint Center for Earth Systems Technology (JCET), University of Maryland Baltimore

More information

Andrew Collard, ECMWF

Andrew Collard, ECMWF Assimilation of AIRS & IASI at ECMWF Andrew Collard, ECMWF Slide 1 Acknowledgements to Tony McNally, Richard Engelen & Rossana Dragani Overview Introduction Assimilation Configuration - Channel Selection

More information

A physically-based observation error covariance matrix for IASI

A physically-based observation error covariance matrix for IASI ITSC20 29 Oct 2015 A physically-based observation error covariance matrix for IASI Hyoung-Wook Chun 1, Reima Eresmaa 2, Anthony P. McNally 2, Niels Bormann 2, and Marco Matricardi 2 1 Korea Institute of

More information

Towards the assimilation of AIRS cloudy radiances

Towards the assimilation of AIRS cloudy radiances Towards the assimilation of AIRS cloudy radiances N. FOURRIÉ 1, M. DAHOUI 1 * and F. RABIER 1 1 : National Center for Meteorological Research (CNRM, METEO FRANCE and CNRS) Numerical Weather Prediction

More information

QUALITY CONTROL OF WINDS FROM METEOSAT 8 AT METEO FRANCE : SOME RESULTS

QUALITY CONTROL OF WINDS FROM METEOSAT 8 AT METEO FRANCE : SOME RESULTS QUALITY CONTROL OF WINDS FROM METEOSAT 8 AT METEO FRANCE : SOME RESULTS Christophe Payan Météo France, Centre National de Recherches Météorologiques, Toulouse, France Astract The quality of a 30-days sample

More information

Assimilation of precipitation-related observations into global NWP models

Assimilation of precipitation-related observations into global NWP models Assimilation of precipitation-related observations into global NWP models Alan Geer, Katrin Lonitz, Philippe Lopez, Fabrizio Baordo, Niels Bormann, Peter Lean, Stephen English Slide 1 H-SAF workshop 4

More information

The impact of assimilation of microwave radiance in HWRF on the forecast over the western Pacific Ocean

The impact of assimilation of microwave radiance in HWRF on the forecast over the western Pacific Ocean The impact of assimilation of microwave radiance in HWRF on the forecast over the western Pacific Ocean Chun-Chieh Chao, 1 Chien-Ben Chou 2 and Huei-Ping Huang 3 1Meteorological Informatics Business Division,

More information

Assimilation of SEVIRI cloud-top parameters in the Met Office regional forecast model

Assimilation of SEVIRI cloud-top parameters in the Met Office regional forecast model Assimilation of SEVIRI cloud-top parameters in the Met Office regional forecast model Ruth B.E. Taylor, Richard J. Renshaw, Roger W. Saunders & Peter N. Francis Met Office, Exeter, U.K. Abstract A system

More information

Experiences from implementing GPS Radio Occultations in Data Assimilation for ICON

Experiences from implementing GPS Radio Occultations in Data Assimilation for ICON Experiences from implementing GPS Radio Occultations in Data Assimilation for ICON Harald Anlauf Research and Development, Data Assimilation Section Deutscher Wetterdienst, Offenbach, Germany IROWG 4th

More information

Evaluation of FY-3B data and an assessment of passband shifts in AMSU-A and MSU during the period

Evaluation of FY-3B data and an assessment of passband shifts in AMSU-A and MSU during the period Interim report of Visiting Scientist mission NWP_11_05 Document NWPSAF-EC-VS-023 Version 0.1 28 March 2012 Evaluation of FY-3B data and an assessment of passband Qifeng Lu 1 and William Bell 2 1. China

More information

Satellite data assimilation for NWP: II

Satellite data assimilation for NWP: II Satellite data assimilation for NWP: II Jean-Noël Thépaut European Centre for Medium-range Weather Forecasts (ECMWF) with contributions from many ECMWF colleagues Slide 1 Special thanks to: Tony McNally,

More information

Progress towards better representation of observation and background errors in 4DVAR

Progress towards better representation of observation and background errors in 4DVAR Progress towards better representation of observation and background errors in 4DVAR Niels Bormann 1, Massimo Bonavita 1, Peter Weston 2, Cristina Lupu 1, Carla Cardinali 1, Tony McNally 1, Kirsti Salonen

More information

The potential impact of ozone sensitive data from MTG-IRS

The potential impact of ozone sensitive data from MTG-IRS The potential impact of ozone sensitive data from MTG-IRS R. Dragani, C. Lupu, C. Peubey, and T. McNally ECMWF rossana.dragani@ecmwf.int ECMWF May 24, 2017 The MTG IRS Long-Wave InfraRed band O 3 Can the

More information

New screening of cold-air outbreak regions used in 4D-Var all-sky assimilation

New screening of cold-air outbreak regions used in 4D-Var all-sky assimilation EUMETSAT/ECMWF Fellowship Programme Research Report No. 35 New screening of cold-air outbreak regions used in 4D-Var all-sky assimilation Katrin Lonitz and Alan Geer February 2015 Series: EUMETSAT/ECMWF

More information

Use of satellite winds at Deutscher Wetterdienst (DWD)

Use of satellite winds at Deutscher Wetterdienst (DWD) Use of satellite winds at Deutscher Wetterdienst (DWD) Alexander Cress Deutscher Wetterdienst, Frankfurter Strasse 135, 63067 Offenbach am Main, Germany alexander.cress@dwd.de Ø Introduction Ø Atmospheric

More information

Use of ATOVS raw radiances in the operational assimilation system at Météo-France

Use of ATOVS raw radiances in the operational assimilation system at Météo-France Use of ATOVS raw radiances in the operational assimilation system at Météo-France Élisabeth Gérard, Florence Rabier, Delphine Lacroix Météo-France, Toulouse, France Zahra Sahlaoui Maroc-Météo, Casablanca,

More information

Assimilating cloud and precipitation: benefits and uncertainties

Assimilating cloud and precipitation: benefits and uncertainties Assimilating cloud and precipitation: benefits and uncertainties Alan Geer Thanks to: Katrin Lonitz, Peter Lean, Richard Forbes, Cristina Lupu, Massimo Bonavita, Mats Hamrud, Philippe Chambon, Fabrizio

More information

Bias correction of satellite data at Météo-France

Bias correction of satellite data at Météo-France Bias correction of satellite data at Météo-France É. Gérard, F. Rabier, D. Lacroix, P. Moll, T. Montmerle, P. Poli CNRM/GMAP 42 Avenue Coriolis, 31057 Toulouse, France 1. Introduction Bias correction at

More information

Assimilation of MIPAS limb radiances at ECMWF using 1d and 2d radiative transfer models

Assimilation of MIPAS limb radiances at ECMWF using 1d and 2d radiative transfer models Assimilation of MIPAS limb radiances at ECMWF using 1d and 2d radiative transfer models Niels Bormann, Sean Healy, Mats Hamrud, and Jean-Noël Thépaut European Centre for Medium-range Weather Forecasts

More information

THE ASSIMILATION OF SURFACE-SENSITIVE MICROWAVE SOUNDER RADIANCES AT ECMWF

THE ASSIMILATION OF SURFACE-SENSITIVE MICROWAVE SOUNDER RADIANCES AT ECMWF THE ASSIMILATION OF SURFACE-SENSITIVE MICROWAVE SOUNDER RADIANCES AT ECMWF Enza Di Tomaso and Niels Bormann European Centre for Medium-range Weather Forecasts Shinfield Park, Reading, RG2 9AX, United Kingdom

More information

Impact of Megha-Tropique's SAPHIR humidity profiles in the Unified Model Analysis and Forecast System

Impact of Megha-Tropique's SAPHIR humidity profiles in the Unified Model Analysis and Forecast System Impact of Megha-Tropique's SAPHIR humidity profiles in the Unified Model Analysis and Forecast System S. Indira Rani 1, Amy Doherty 2, Nigel Atkinson 2, William Bell 2, Stuart Newman 2, Richard Renshaw

More information

International TOVS Study Conference-XV Proceedings

International TOVS Study Conference-XV Proceedings Relative Information Content of the Advanced Technology Microwave Sounder, the Advanced Microwave Sounding Unit and the Microwave Humidity Sounder Suite Motivation Thomas J. Kleespies National Oceanic

More information

Application of PCA to IASI: An NWP Perspective. Andrew Collard, ECMWF. Acknowledgements to Tony McNally, Jean-Noel Thépaut

Application of PCA to IASI: An NWP Perspective. Andrew Collard, ECMWF. Acknowledgements to Tony McNally, Jean-Noel Thépaut Application of PCA to IASI: An NWP Perspective Andrew Collard, ECMWF Acknowledgements to Tony McNally, Jean-Noel Thépaut Overview Introduction Why use PCA/RR? Expected performance of PCA/RR for IASI Assimilation

More information

All-sky observations: errors, biases, and Gaussianity

All-sky observations: errors, biases, and Gaussianity All-sky observations: errors, biases, and Gaussianity Alan Geer and Peter Bauer ECMWF, Shinfield Park, Reading RG2 9AX, United Kingdom Alan.Geer@ecmwf.int ABSTRACT Microwave imager observations are sensitive

More information

Observation error specifications

Observation error specifications Observation error specifications Gérald Desroziers, with many contributions Météo-France, CNRS 4 av. G. Coriolis, 357 Toulouse Cédex, France gerald.desroziers@meteo.fr ABSTRACT The aim of this paper is

More information

The use of the GPS radio occultation reflection flag for NWP applications

The use of the GPS radio occultation reflection flag for NWP applications Ref: SAF/ROM/METO/REP/RSR/022 Web: www.romsaf.org Date: 27 August 2015 ROM SAF Report 22 The use of the GPS radio occultation reflection flag for NWP applications Sean Healy ECMWF Healy: Reflection Flag

More information

An evaluation of radiative transfer modelling error in AMSU-A data

An evaluation of radiative transfer modelling error in AMSU-A data An evaluation of radiative transfer modelling error in AMSU-A data Cristina Lupu, Alan Geer, Niels Bormann and Stephen English 20 th International TOVS Study Conference, Lake Geneva, USA 28 October 2015

More information

Toward assimilation of CrIS and ATMS in the NCEP Global Model

Toward assimilation of CrIS and ATMS in the NCEP Global Model Toward assimilation of CrIS and ATMS in the NCEP Global Model Andrew Collard 1, John Derber 2, Russ Treadon 2, Nigel Atkinson 3, Jim Jung 4 and Kevin Garrett 5 1 IMSG at NOAA/NCEP/EMC 2 NOAA/NCEP/EMC 3

More information

The impact of satellite data on NWP

The impact of satellite data on NWP The impact of satellite data on NWP Anthony McNally European Centre for Medium-range Weather Forecasts (ECMWF) Shinfield Park, Reading RG2 9AX, UK Anthony.McNally@ecmwf.int ABSTRACT This study looks at

More information

IMPACT OF IASI DATA ON FORECASTING POLAR LOWS

IMPACT OF IASI DATA ON FORECASTING POLAR LOWS IMPACT OF IASI DATA ON FORECASTING POLAR LOWS Roger Randriamampianina rwegian Meteorological Institute, Pb. 43 Blindern, N-0313 Oslo, rway rogerr@met.no Abstract The rwegian THORPEX-IPY aims to significantly

More information

The assimilation of AMSU and SSM/I brightness temperatures in clear skies at the Meteorological Service of Canada

The assimilation of AMSU and SSM/I brightness temperatures in clear skies at the Meteorological Service of Canada The assimilation of AMSU and SSM/I brightness temperatures in clear skies at the Meteorological Service of Canada Abstract David Anselmo and Godelieve Deblonde Meteorological Service of Canada, Dorval,

More information

Retrieval Algorithm Using Super channels

Retrieval Algorithm Using Super channels Retrieval Algorithm Using Super channels Xu Liu NASA Langley Research Center, Hampton VA 23662 D. K. Zhou, A. M. Larar (NASA LaRC) W. L. Smith (HU and UW) P. Schluessel (EUMETSAT) Hank Revercomb (UW) Jonathan

More information

by Howard Berger University of Wisconsin-CIMSS NWP SAF visiting scientist at the Met Office, UK

by Howard Berger University of Wisconsin-CIMSS NWP SAF visiting scientist at the Met Office, UK by Howard Berger University of Wisconsin-CIMSS NWP SAF visiting scientist at the Met Office, UK This documentation was developed within the context of the EUMETSAT Satellite Application Facility on Numerical

More information

RTMIPAS: A fast radiative transfer model for the assimilation of infrared limb radiances from MIPAS

RTMIPAS: A fast radiative transfer model for the assimilation of infrared limb radiances from MIPAS RTMIPAS: A fast radiative transfer model for the assimilation of infrared limb radiances from MIPAS Niels Bormann, Sean Healy, and Marco Matricardi European Centre for Medium-range Weather Forecasts (ECMWF),

More information

Introduction to Data Assimilation

Introduction to Data Assimilation Introduction to Data Assimilation Alan O Neill Data Assimilation Research Centre University of Reading What is data assimilation? Data assimilation is the technique whereby observational data are combined

More information

The assimilation of GPS Radio Occultation measurements at the Met Office

The assimilation of GPS Radio Occultation measurements at the Met Office The assimilation of GPS Radio Occultation measurements at the Met Office M.P. Rennie Met Office Exeter, UK michael.rennie@metoffice.gov.uk ABSTRACT In the past few years GPS radio occultation (GPSRO) measurements

More information

Combining Polar Hyper-spectral and Geostationary Multi-spectral Sounding Data A Method to Optimize Sounding Spatial and Temporal Resolution

Combining Polar Hyper-spectral and Geostationary Multi-spectral Sounding Data A Method to Optimize Sounding Spatial and Temporal Resolution Combining Polar Hyper-spectral and Geostationary Multi-spectral Sounding Data A Method to Optimize Sounding Spatial and Temporal Resolution W. L. Smith 1,2, E. Weisz 1, and J. McNabb 2 1 University of

More information

USE, QUALITY CONTROL AND MONITORING OF SATELLITE WINDS AT UKMO. Pauline Butterworth. Meteorological Office, London Rd, Bracknell RG12 2SZ, UK ABSTRACT

USE, QUALITY CONTROL AND MONITORING OF SATELLITE WINDS AT UKMO. Pauline Butterworth. Meteorological Office, London Rd, Bracknell RG12 2SZ, UK ABSTRACT USE, QUALITY CONTROL AND MONITORING OF SATELLITE WINDS AT UKMO Pauline Butterworth Meteorological Office, London Rd, Bracknell RG12 2SZ, UK ABSTRACT Satellite wind fields derived from geostationary imagery

More information

The Impact of Background Error on Incomplete Observations for 4D-Var Data Assimilation with the FSU GSM

The Impact of Background Error on Incomplete Observations for 4D-Var Data Assimilation with the FSU GSM The Impact of Background Error on Incomplete Observations for 4D-Var Data Assimilation with the FSU GSM I. Michael Navon 1, Dacian N. Daescu 2, and Zhuo Liu 1 1 School of Computational Science and Information

More information

COMPARISON OF SIMULATED RADIANCE FIELDS USING RTTOV AND CRTM AT MICROWAVE FREQUENCIES IN KOPS FRAMEWORK

COMPARISON OF SIMULATED RADIANCE FIELDS USING RTTOV AND CRTM AT MICROWAVE FREQUENCIES IN KOPS FRAMEWORK COMPARISON OF SIMULATED RADIANCE FIELDS USING RTTOV AND CRTM AT MICROWAVE FREQUENCIES IN KOPS FRAMEWORK Ju-Hye Kim 1, Jeon-Ho Kang 1, Hyoung-Wook Chun 1, and Sihye Lee 1 (1) Korea Institute of Atmospheric

More information

Representativity error for temperature and humidity using the Met Office high resolution model

Representativity error for temperature and humidity using the Met Office high resolution model School of Mathematical and Physical Sciences Department of Mathematics and Statistics Preprint MPS-2012-19 12 September 2012 Representativity error for temperature and humidity using the Met Office high

More information

THORPEX Data Assimilation and Observing Strategies Working Group: scientific objectives

THORPEX Data Assimilation and Observing Strategies Working Group: scientific objectives THORPEX Data Assimilation and Observing Strategies Working Group: scientific objectives Pierre Gauthier Co-chair of the DAOS-WG Atmospheric Science and Technology Branch Environment Canada Data assimilation

More information

The role of GPS-RO at ECMWF" ! COSMIC Data Users Workshop!! 30 September 2014! !!! ECMWF

The role of GPS-RO at ECMWF ! COSMIC Data Users Workshop!! 30 September 2014! !!! ECMWF The role of GPS-RO at ECMWF"!!!! COSMIC Data Users Workshop!! 30 September 2014! ECMWF WE ARE Intergovernmental organisation! 34 Member and Cooperating European states! 270 staff at ECMWF, in Reading,

More information