Adding cloud to the Met Office variational analysis system Sharpe, M. C. Met Office, FitzRoy Road, Exeter EX1 3PB, United Kingdom; martin.sharpe@metoffice.gov.uk The strategy for assimilating cloud information in the Met Office 3D and 4D variational data assimilation systems is to use a total-water control variable in the analysis. Assimilating any moisture information then requires an operator to partition total-water increments into vapour and cloud components, for use by observation operators. Conceptually, this partitioning is part of the incrementing operator, which maps increments from linear model space to non-linear model space and, as such, is separate from the observation operators but may be non-linear. An operator that considers liquid cloud only and allows calculation of increments to bulk cloud fraction has been developed, based on the Met Office Unified Model diagnostic large-scale cloud scheme. Its formulation, linearisation and application within the Met Office system are described and test results presented. 1. Introduction 2. Incrementing operator formulation The current Met Office variational analysis system produces water vapour analyses using a single moisture control variable (relative humidity). Motivated by the possibility of assimilating information from cloud cover analyses and cloud affected satellite observations, a revised incrementing operator for cloud and moisture is being developed. The new operator is designed to be used with a total-moisture control variable and provide increments to water vapour and cloud water, for use by the observation operators. While the calculation of these increments may be nonlinear, a linearisation and adjoint are required; the adjoint is a necessary part of the assimilation algorithm and the linear forward code may be used as a computational expedient. The Met Office variational analysis system is designed to use non-linear operators and their linear versions, in the mapping from linear model space to observation space. The diagnostic liquidcloud scheme used by the forecast model provided a suitable basis from which to develop an operator that would derive cloud and moisture quantities from total-moisture increments produced by the linear model, but a number of adaptations were required: The revised incrementing operator, based on the Met Office Unified Model diagnostic large-scale cloud scheme described by Smith 1990, currently considers total-moisture increments as representing changes to only water vapour (q) and liquid cloud (qcl), allowing changes in liquid cloud fraction (Cl) to be diagnosed. The sections that follow outline the formulation of this operator and its linearisation. Some early results and potential methods for introducing ice cloud into the operator are also presented along with an outline of the requirements for further testing and development of the scheme. 1. remove threshold dependant switches: to enable straightforward linearisation and allow cloud observations to influence the minimisation, even in regions where the original cloud scheme would report a zero gradient with respect to its input variables 2. use temperature as an input variable: to allow the temperature increments output from the linear model to be applied in a scheme whose original formulation takes an input expressed as liquid water temperature; TL = T – qcl.Lc/Cp 3. output saturated states: to allow adaptation 1 to be used while still reproducing the features seen in forecast model output and regularizing the scheme to give sensible gradients in saturated regions Adaptations 1 & 2 have been achieved by changing the function used to describe the gridbox joint-pdf of TL and total specific humidity (qT), from the triangular distribution used in the forecast model scheme to a sech2 distribution. The third adaptation has been achieved by changing the values of ∂qsat/∂T used; from those at TL, as in the original formulation, to those at T. With these adaptations applied, Figure 1 shows the relationships between cloud fraction and relative humidity for the simplified scheme alongside those used in the forecast model, in which slightly different schemes are used for global and local area domains. later calculations to take T increments their as input. The resulting cloud increments are then consistent with the changes in total-moisture and energy, relative to the background states used by the initial calculation. Equation 1 shows how increments (denoted by primes) are calculated for qcl. This method of adding increments to ‘pseudo-background’ values derived using TL as an input variable, also applies to other moisture variables. Equation 1: q’cl is calculated using ‘pseudobackground’ values derived from background values of TL, p & qT. 3. Linearisation Although the need for a linear operator was part of the reason for adapting the forecast model cloud schemes, the non-linear equations are solved using a fixed-point iteration. This results in a technical difficulty with linearisation; the need to use non-linear iterates in a line-by-line approach would require either a large amount of extra storage or repetition of the iterative calculations. However, taking an alternative approach uses approximate gradients, moving the analysis minimisation away from the intended solution. Figure 1: The cloud fraction-vs-relative humidity diagnostic resulting from the global forecast model cloud scheme (black), the local area model scheme (blue), and the simplified version (red) Early on in the testing of the simplified cloud scheme, on which the incrementing operator is based, using TL as its input variable was found to provide a better energy constraint than using T; giving cloud liquid water values much more like those produced in the forecast models. Unlike the forecast model schemes, the simplified version is invertible; the original value of T L can be recovered by applying the scheme to the T values obtained when using TL as its input. This means that an initial calculation, using T L as input, allows Following an approach suggested by Polavarapu & Tanguay 1998 the analytic equations are linearised with the intention of using the resulting equations as an approximation to the line-by-line linearisation of the non-linear code. This is equivalent to assuming that the non-linear iteration has converged to the analytic solution. Linearisation test results using line-by-line and analytic approaches showed that, even in regions of large increments, the two linearisations agreed to 3 s.f. and globally averaged statistics were also similar. Figure 2 shows an example of the linearisation test output, for globally averaged correlations between qcl increments calculated using the non-linear and linear simplified schemes. The conclusion from the linearisation tests is that although the analytic linearisation may not be perfect, with respect to the non-linear iteration code, it is acceptable for use in the analysis system given that the simplified cloud scheme is already an approximation to that used in the forecast model. largely similar. More impact on verification scores may be obtained when the scheme is used alongside a newly implemented latent heating scheme for the linear model and applied in a system with finer grid spacing. Figure 2: Area-weighted correlations between nonlinear and linearised qcl increments for lineby-line linearisation (left hand figure) and for the analytic linearisation (right hand figure) 4. Testing Initial testing of the simplified cloud scheme confirmed that the adaptations do result in an adequate reproduction of the cloud and water vapour values diagnosed in the forecast model and the linearisation tests confirmed that using the analytic method is a reasonable approach. Initial results of using the new incrementing operator within a global 4D-Var system (but not using cloud observations) showed small differences in the evolution of the penalty function during minimisation; apparently fitting the observations less closely. The operator was then trialed in a reduced resolution (N48) 4D-Var version of the Met Office global NWP system, over a month-long period during the summer of 2004. In this trial, a change to the initialization of the forecast model was also introduced; to derive increments in water vapour and liquid cloud, given total-water analysis increments, by applying the same method as is used in the incrementing operator. Although the moisture observations used in the trial related only to water vapour, the results are of value in assessing the effectiveness of initialising cloud fields from the total-water analysis. (When the trial was carried out, the linear model used in the Met Office 4D-Var system did not include any latent heating terms.) The trial results showed a neutral overall impact on rms scores in forecast verification against both analyses and observations. However, the background fit to observations was improved for nearly all observations that are sensitive to water vapour, as shown in Figure 3. Additionally, as would be expected from the improved background fit to observations, the size of the analysis increments was reduced for all variables, even though the overall synoptic evolution of both the test and control runs was Figure 3: The improvement in initial fit to observations when applying the incrementing operator for vapour and liquid cloud over a 31 day global NWP system (points lower down on the plot show those observations that have the largest effect on the analysis) 5. Further developments to include frozen cloud in the incrementing operator In order to take full advantage of current and future observations, especially those from satellites and cloud analysis products, the incrementing operator must be extended to diagnose increments to cloud ice content and grid-box ice-cloud fraction. It is hoped that a diagnostic approach will simulate the prognostic ice scheme used in the forecast model sufficiently well that the liquid cloud scheme already developed will not need to be changed radically in order to incorporate a scheme for ice. Given cloud ice increments, calculating the gridbox ice-cloud fraction for use with cloud analysis data is straightforward; following the method applied in the forecast model. Work in this area has focused, so far, on assuming an equilibrium between sublimation & deposition, as used (but not run to convergence) in the existing Met Office forecast system. An alternative approach, intended to make better use of background information, is also being investigated. It is based on diagnosing the cloud ice content by using the scheme for liquid cloud in combination with a parameterization of increments to the partitioning of cloud water between liquid and ice. Although this work is still in its early stages, it is clear that the form of the equilibrium assumption will need to be modified in order to obtain a smooth transition between cold and warm regions (where it does not apply). Additionally, to obtain ice cloud increments that are not unreasonably large, it may be necessary to constrain the calculated increments to the equilibrium state by requiring them to reproduce approximate conservation properties of cloud ice evolution in the forecast model. If using the parameterized partitioning approach, care is required in constraining the incremented partitioning to remain within physical limits. Developing the incrementing operator to include ice cloud, in terms of cloud water and grid-box cloud fraction, is under way and must include work to derive increments to bulk cloud fraction; combining increments in liquid and frozen cloud fractions Testing the completed operator alongside the planned Met Office prognostic cloud scheme will be necessary, as reformulation of the cloud incrementing operator may be required; when the new cloud scheme is used in the forecast model, the incrementing operator must adequately approximate increments that would be obtained by applying the forecast model scheme. 6. Future work Acknowledgements: Testing the incrementing operator for vapour and liquid cloud alongside a newly implemented latent heating scheme, in the linear model used with the Met Office 4DVar system, and in a system with finer grid spacing is under way Testing the performance of the incrementing operator for vapour and liquid cloud in assimilating cloud information is planned to consider 23 & 31 GHz AMSU-A channels and the 19GHz SSMI channel, when the required work on satellite data assimilation code is completed Sue Ballard & Damian Wilson: Guidance on cloud scheme formulation Rick Rawlins: Analysis of results from the global 4D-Var trial References: Polavarapu, S., Tanguay, M., Linearizing iterative processes for four-dimensional data-assimilation schemes. QJRMS, 124, 1715-1742, 1998 Smith, R. N. B., A scheme for predicting layer clouds and their water content in a general circulation model. QJRMS, 116, 435-460, 1990