# $Id: $ #****c* Configuration Files/ecmwf_refrac_1dvar.cf * # # NAME # ecmwf_refrac_1dvar.cf - Default configuration file for 1DVar # implementations in ROPP. ECMWF background data, # refractivity observations. # # SYNOPSIS # <1DVar_program> ... -c ecmwf_refrac_1dvar.cf ... # # DESCRIPTION # This file reflects the default configuration for the 1DVar # implementations within ROPP using ECMWF background data and refractivity # observations. # # NOTES # Running a ROPP 1DVar with this configuration should provide the # same results as running the same tool without any configuration # file. It's main purpose is to document the default settings, and # which configuration options are available to the user. # # EXAMPLE # # # SEE ALSO # # # REFERENCES # # # AUTHOR # Met Office, Exeter, UK. # Any comments on this software should be given via the ROM SAF # Helpdesk at http://www.romsaf.org # # COPYRIGHT # (c) EUMETSAT. All rights reserved. # For further details please refer to the file COPYRIGHT # which you should have received as part of this distribution. # #**** #------------------------------------------------------------------------------- # 1. Input and output files #------------------------------------------------------------------------------- # The names of input and output files can also be specified through command line # arguments; names mspecified through the command line will always overwrite # configuration file settings. The command line arguments corresponding to # configuration options are given in round brackets after the detailed comments # below. # 1.1 Background # -------------- bg_input_file = ropp_bg.nc # Background profile (-b, -bg, --bg) bg_corr_file = ropp_bg_corr.nc # Background error correlations / covariances (--bg-corr) # 1.2 Observations # ---------------- obs_input_file = ropp_obs.nc # Observation profile (-y, -obs, --obs) obs_corr_file = ropp_obs_corr.nc # Observation error correlations / covariances (--obs-corr) # 1.3 Output # ---------- output_file = ropp_out.nc # Retrieved profile (-o) #------------------------------------------------------------------------------- # 2. Error covariance models #------------------------------------------------------------------------------- # 2.1 Background error covariances # -------------------------------- # # Background error covariances can be constructed using the following methods: # # FSFC Fixed Sigmas, Fixed correlations: Both error correlations and # and error standard deviations are read from a background # error correlation file. The error correlation file must # contain both the error correlation matrix as well as the # standard deviations (errors) for all background state vector # elements. # # VSFC Variable Sigmas, Fixed Correlations: Error correlations are read # from an error correlation file, while per profile error # estimates as contained in the background data file are used. # In this case, the error correlation / covariance data files # only require to contain the error correlations. # # RSFC Relative Sigmas, Fixed Correlations: Relative specific humidity (q) # and relative surface pressure (p*) errors, and (absolute) # temperature (T) errors, are read from a background error # correlation file, as are the error correlations. # (All fields must come from the same file.) The relative q (p*) # errors are multiplied by the profile q (p*) values to give the # profile errors. RSFC is therefore a hybrid of FSFC and VSFC. # # Note that error correlation files may contain latitudinally binned error # correlations and standard deviations, allowing to have latitudinally varying # error correlation structures and standard deviations even in the FCFS scenario. # # Specifying (and providing) a properly formatted error correlation / covariance # file for the background data is mandatory for the ROPP 1DVars. bg_covar_method = VSFC # Variable Sigmas, Fixed Correlations # 2.2 Observation error covariances # --------------------------------- # # Bending angle and refractivity error covariances can be constructed using the # following methods: # # FSFC Fixed Sigmas, Fixed correlations: Both error correlations and # and error standard deviations are read from an observation # error correlation file. The error correlation file must # contain both the error correlation matrix as well as the # standard deviations (errors) for all observation vector # elements. # # VSDC Variable Sigmas, Diagonal Correlations: A diagonal error correlation # structure (i.e., no error correlations) is assumed, while per # profile error estimates as contained in the observation data file # are used. In this case, no error correlation / covariance data # file is required. # # VSFC Variable Sigmas, Fixed Correlations: Error correlations are read # from an error correlation file, while per profile error # estimates as contained in the observation data file are used. # In this case, the error correlation / covariance data files # only require to contain the error correlations. # # Note that error correlation files may contain latitudinally binned error # correlations and standard deviations, allowing to have latitudinally varying # error correlation structures and standard deviations in the FSFC and VSFC # scenarios. # # In contrast to the background data, observations do not require an error # correlation / covariance file in scenario VSDC, if the input data already # contains error estimates. obs_covar_method = VSFC # Variable Sigmas, Fixed Correlations #------------------------------------------------------------------------------- # 3. Quality control #------------------------------------------------------------------------------- # 3.1 Valid observation height range # ---------------------------------- # # A valid height range can be specified, outside of which observations are # not used in the 1dVar analysis. min_1dvar_height = 0.0 # minimum observation height (km) max_1dvar_height = 50.0 # maximum observation height (km) # 3.2 Generic quality control # --------------------------- # # Generic quality control checks for the obviouos things - whether data values # are within reasonable bounds (specified below as minimum and maximum values), # and if the input data is roughly consistent. # # In addition, co-location of background and observations is checked via the # great-circle and temporal distance betwen the nominal locations of both. For # testing purposes, and in case the 1DVar is used with climatological data as # background which is not properly co-located or time stamped, this particular # check can be disabled. genqc_colocation_apply = .true. # Apply colocation checks? genqc_max_distance = 300.0 # Maximum obs vs. bg great circle distance (km) genqc_max_time_sep = 3600.0 # Maximum obs vs. bg temporal seperation (sec) genqc_min_temperature = 150.0 # K genqc_max_temperature = 350.0 # K genqc_min_spec_humidity = 0.0 # g/kg genqc_max_spec_humidity = 50.0 # g/kg genqc_min_impact = 6.2e6 # m genqc_max_impact = 6.6e6 # m genqc_min_bangle = -1.0e-4 # rad genqc_max_bangle = 0.1 # rad genqc_min_geop_refrac = -1.0e3 # m genqc_max_geop_refrac = 1.e5 # m genqc_min_refractivity = 0.0 # N-units genqc_max_refractivity = 500.0 # N-units # 3.3 Background quality control # ------------------------------ # # Background quality control, if applied, rejects observation data points which # deviate by more than * , where the # is calculated from both the assumed observation and forward # modelled background errors. If the number of rejected points exceeds # , the entire profile is considered to be of poor # quality, and is not further processed. bgqc_apply = .true. # Apply background quality control? bgqc_reject_factor = 10.0 # Data rejected if O-B > factor * sigma bgqc_reject_max_percent = 50.0 # Maximum percentage of data rejected # 3.4 Probability of Gross Error # ------------------------------ # # Probability of Gross Error, as a diagnostic quantity, is always calculated # as part of the standard processing (from the O - B differences). It can also # be used for quality control, by calculating weights based on the value of # the gross error probability which are applied to the observations. pge_apply = .false. # Apply PGE for quality control? pge_fg = 0.001 # First guess PGE pge_d = 10.0 # Width of gross error plateau #------------------------------------------------------------------------------- # 4. Preconditioning, convergence checks and minimiser #------------------------------------------------------------------------------- # 4.1 Preconditioning # ------------------- # # Preconditioning accelerates the minimisation significantly and his highly # recommended; it should only be disabled for testing purposes. use_precond = .true. # 4.2 Convergence checks # ---------------------- # # Apart from the minimiser's own convergence check (which is based on the change # of the gradient size), two more convergence checks are implemented: The # minimisation is assumed to have converged if any of the following conditions # # - change of state vector < * # - change of cost function < # # is achieved for at least consecutive iterations / calls of # the cost function evaluation. # # Note that the convergence criteria for the state vector changes are specified # as a fraction of the background error. conv_check_apply = .true. # Apply additional convergence checks? conv_check_n_previous = 2 # Minimum number of iterations required conv_check_max_delta_state = 0.1 # State vector must change less than this * bg error conv_check_max_delta_J = 0.1 # Cost function must change less than this # 4.3 minROPP settings # ------------------ minropp_method = MINROPP # minimisation method # MINROPP (default) or LEVMARQ minropp_log_file = screen # 'screen' for output on the screen, file name otherwise minropp_impres = 0 minropp_n_iter = 1500 minropp_mode = 0 minropp_n_updates = 50 minropp_eps_grad = 1.0e-8 minropp_dx_min = 1.0e-16 #------------------------------------------------------------------------------- # 5. Additional output #------------------------------------------------------------------------------- # # If is .true., the 1DVar will add additional diagnostic # output to the retrieval data file. At present, the following diagnostics are # available: # # - O-B differences # - O-A differences # - PGE extended_1dvar_diag = .false. #------------------------------------------------------------------------------- # 6. Log(pressure) and Log(humidity) options #------------------------------------------------------------------------------- # # If is .true., the 1DVar will perform the minimisation using log(p) # rather than absolute pressure in the state vector # If is .true., the 1DVar will perform the minimisation using log(q) # rather than absolute pressure in the state vector use_logp = .false. use_logq = .false. #------------------------------------------------------------------------------- # 7. Seasonal observation scaling options #------------------------------------------------------------------------------- # # These options allow the observation errors (the standard deviation values) to # be scaled according to the season. This is a sinusoidal correction of the form: # new_error=old_error*(1++*COS(2pi*(season+))). # 'season' is calculated in the code and takes a value between 0 and 1 (the start # and end of the calendar year respectively) # is the amplitude of the sinusoidal scaling factor. # is a constant offset applied to the observation errors, on # which the sinusoidal factor is added. # gives control over the 'phase' of the sinusoid, i.e. a value # of 0.1 will shift the maximum of the sinusoid back one tenth of a year. # This value should be between -1 and 1. # # Take care using this functionality as it is possible to produce negative # standard deviation values. In this case a warning is produced and the program # reverts to the unscaled sigma values. # A recommended approach is to input observation errors for the season which has # the smallest errors and provide a positive value for which is # equal to . This will ensure that the seasonal adjustment will # always result in positive variances. season_amp = 0.0 season_offset = 0.0 season_phase = 0.0