Configuring GARD
After building GARD, it is run on the command line following this syntax:
./gard downscale_options.txt
where downscale_options.txt is a Fortran style namelist (see example in GARD/run).
The namelist options are described in the tables below and a sample file is available online. :
Parameters
| Name | Type | Required? | Default | Description |
|---|---|---|---|---|
| training_file | string | No | Options File Name | Used to specify a separate namelist file for the training section shown below |
| observation_file | string | No | Options File Name | Used to specify a separate namelist file for the observation section shown below |
| prediction_file | string | No | Options File Name | Used to specify a separate namelist file for the prediction section shown below |
| output_file | string | No | downscaled_output.nc | Downscaled output filename. |
| pass_through_var | integer | No | 1 | Select the variable to be passed through as the output if pass_through=True |
| n_analogs | integer | No (1) | -1 | The number of analogs to include |
| n_log_analogs | integer | No (1) | -1 | The number of analogs to include when computing threshold exceedence from analogs |
| analog_threshold | real | No (1) | -1 | If specified, GARD will compute the probability of exceeding this threshold |
| start_date | string | Yes | n/a | Start date for prediction period |
| end_date | string | Yes | n/a | End date for prediction period |
| start_train | string | Yes | n/a | Start date for training period |
| end_train | string | Yes | n/a | End date for training period |
| start_transform | string | Yes | n/a | Start date for transform period |
| end_transform | string | Yes | n/a | End date for transform period |
| pure_regression | logical | No | FALSE | Use pure regression downscaling approach |
| pure_analog | logical | No | FALSE | Use pure analog downscaling approach |
| analog_regression | logical | No | TRUE | Use analog regression downscaling approach |
| pass_through | logical | No | False | Pass a given predictor variable through as the output without downscaling |
| sample_analog | logical | No | FALSE | |
| logistic_from_analog_exceedance | logical | No | FALSE | |
| logistic_threshold | logical | No | -9999 | |
| weight_analogs | logical | No | TRUE | |
| debug | logical | No | TRUE | prints more output at runtime and outputs files including the coefficients used in each analog regression (or analog values) as well as the predictor data |
| interactive | logical | No | TRUE | Print downscaling status as a percentage on the command line |
Notes:
- Must specify one of n_analogs, n_log_analogs, or analog_threshold
Training Parameters
See description of training data suggestions.
| Name | Type | Required? | Default | Description |
|---|---|---|---|---|
| name | string | Yes | n/a | Name of the training parameters dataset |
| preloaded | string | No | n/a | filepath of the preloaded training parameters dataset |
| interpolation_method | integer | No | 1 | nearest neighbor= 1, bilenear =2 |
| normalization_method | integer | No | 0 | no normalization = 0, mean/stddev from: training data = 1 |
| time_indices | integer | Yes | -1 | list of timesteps in file to aggregate over (GEFS only) |
| time_weights | real | No | 1 | list of averaging weights for individual time indices |
| agg_method | integer | No | 0 | per variable aggregation method when aggregating over time_indices: mean = 0, minimum = 1, maximum = 2, sum = 3 (GEFS only) |
| nvars | integer | Yes | -1 | number of variables to be used in training |
| data_type | string | Yes | n/a | dataset type: GEFS or GCM |
| lat_name | string | Yes | n/a | netCDF variable name for latitude |
| lon_name | string | Yes | n/a | netCDF variable name for longitude |
| time_name | string | Yes | n/a | netCDF variable name for time |
| nfiles | integer | Yes | -1 | number of files in each file list |
| input_transformations | integer | No | 0 | no transform = 0, quantile mapping = 1, log transform = 2, cube root = 3, fifth root = 4 |
| var_names | string | Yes | n/a | variables names to use in training (one for each variable) |
| file_list | string | Yes | n/a | path to file containing a list of training files (one for each variable) |
| selected_time | integer | No | -1 | if set, only this time step will be read from each input file (GEFS only) |
| calendar | string | Yes | standard | |
| calendar_start_year | integer | No | 1900 | |
| timezone_offset | real | No | 0 | hours to offset time variable (e.g. time_name) to account for timezone. |
Prediction Parameters
See description of predictor data suggestions.
| Name | Type | Required? | Default | Description |
|---|---|---|---|---|
| name | string | Yes | n/a | Name of the prediction parameters dataset |
| preloaded | string | No | n/a | filepath of the preloaded prediction parameters dataset |
| interpolation_method | integer | No | 1 | nearest neighbor= 1, bilenear =2 |
| normalization_method | integer | No | 0 | no normalization = 0, mean/stddev from: prediction data = 1, training data = 2 |
| time_indices | integer | Yes | -1 | list of timesteps in file to aggregate over (GEFS only) |
| time_weights | real | No | 1 | list of averaging weights for individual time indices |
| agg_method | integer | No | 0 | per variable aggregation method when aggregating over time_indices: mean = 0, minimum = 1, maximum = 2, sum = 3 (GEFS only) |
| nvars | integer | Yes | -1 | number of prediction parameters to use in downscaling |
| data_type | string | Yes | n/a | dataset type: GEFS or GCM |
| lat_name | string | Yes | n/a | netCDF variable name for latitude |
| lon_name | string | Yes | n/a | netCDF variable name for longitude |
| time_name | string | Yes | n/a | netCDF variable name for time |
| nfiles | integer | Yes | -1 | number of files in each file list |
| transformations | integer | No | 0 | no transform = 0, quantile mapping = 1, log transform = 2, cube root = 3, fifth root = 4 |
| input_transformations | integer | No | 0 | no transform = 0, quantile mapping = 1, log transform = 2, cube root = 3, fifth root = 4 |
| var_names | string | Yes | n/a | variables names to use in prediction (one for each variable) |
| file_list | string | Yes | n/a | path to file containing a list of prediction filepaths (one for each variable) |
| selected_time | integer | No | -1 | if set, only this time step will be read from each input file (GEFS only) |
| calendar | string | Yes | standard | |
| calendar_start_year | integer | No | 1900 | |
| timezone_offset | real | No | 0 | hours to offset time variable (e.g. time_name) to account for timezone. |
Observation Parameters
See description of observational data suggestions.
| Name | Type | Required? | Default | Description |
|---|---|---|---|---|
| name | string | Yes | n/a | Name of the observation parameters dataset |
| preloaded | string | No | n/a | filepath of the preloaded observation parameters dataset |
| nvars | integer | Yes | -1 | number of observation variables to downscale (currently must be 1) |
| nfiles | integer | Yes | -1 | number of files in each file list |
| data_type | string | Yes | n/a | dataset type, typically "obs" |
| lat_name | string | Yes | n/a | netCDF variable name for latitude |
| lon_name | string | Yes | n/a | netCDF variable name for longitude |
| time_name | string | Yes | n/a | netCDF variable name for time |
| input_transformations | integer | No | 0 | no transform = 0, quantile mapping = 1, log transform = 2, cube root = 3, fifth root = 4 |
| var_names | string | Yes | n/a | variables names to use in obs dataset |
| file_list | string | Yes | n/a | path to file containing a list of obs filepaths |
| calendar | string | Yes | standard | |
| calendar_start_year | integer | No | 1900 |