Atmospheric forcing noise

Atmospheric forcing noise#

This section describes atmospheric forcing noise for eCLM-PDAF.

The implementation of the atmospheric forcing perturbation routines was first introduced by Yorck Ewerdwalbesloh.

General Description#

eCLM implements perturbations for atmospheric forcing variables (temperature, precipitation, longwave radiation, and shortwave radiation) to support ensemble-based data assimilation in eCLM-PDAF. The perturbation scheme maintains physical consistency through cross-variable correlations and spatiotemporal coherence.

Physical Consistency#

Perturbed variables are cross-correlated to preserve physical relationships between forcing fields. For example, a positive perturbation of incoming shortwave radiation corresponds to a negative perturbation of longwave radiation and a positive perturbation of temperature. The cross-correlation structure follows Reichle et al. (2007) and Han et al. (2014).

Spatial Correlation#

Perturbations are spatially correlated using an isotropic correlation function based on grid cell separation distance. Precipitation uses a shorter correlation length scale than other variables to reflect its more localized spatial structure. Correlation parameters are configurable.

Temporal Correlation#

Temporal correlation is generated using a first-order autoregressive (AR-1) model following Evensen (2009). The default temporal persistence corresponds to a decorrelation timescale of one day, which is adjustable based on application requirements.

Implementation#

Perturbation fields are read from preprocessed noise files and applied to atmospheric forcing variables within the eCLM model during runtime.

1. Noise-Based Forcing Perturbation#

When having an ensemble run, the memory consumption is too large if the forcing data is perturbed one by one so that I get a file for each month for each member idea: perturb the forcings in the CLM sourcecode with a noise file. for each perturbed variable (temperature, precipitation, longwave and shortwave radiation), a stream is introduced.

Adds spatiotemporal noise to atmospheric forcing data for ensemble data assimilation.

Modified files:

  • shr_stream_mod.F90

  • shr_strdata_mod.F90

  • shr_dmodel_mod.F90

Key Features:

  • Stores ensemble metadata in stream structure:

    • caseId - ensemble member ID

    • dt - forcing time resolution

    • numEns - ensemble size that the perturbation file was created for. Example: Running with 50 ensemble members with a noise file created for an ensemble of up to 200 members. Then, numEns should be set to 200 in the stream file for right usage of perturbation dimension.

  • Time-shifting mechanism: Different ensemble members read different temporal frames from the same noise file

    • Formula: frame = nt + caseId * 24/dt

    • Example: For 3-hourly data (dt=3), member 0 starts at frame nt, member 1 at nt+8, member 2 at nt+16, etc.

  • Detects noise fields by checking if model field name contains "noise"

  • Adjusts time coordinate reading to account for ensemble-extended noise files

Preparing Atmospheric Forcing Noise#

This section describes the workflow for generating and configuring spatiotemporally correlated noise fields for atmospheric forcing perturbations.

Noise Generation Tool#

The correlatedNoise.py (HPSCTerrSys/eCLM_atm-forcing-generator#8) script generates monthly NetCDF files containing spatiotemporally correlated noise fields for all ensemble members.

Key Configuration Parameters:

  • Ensemble size: Number of ensemble members to generate noise for (must be specified)

  • Temporal resolution (dt): Time step in hours (e.g., 3 for 3-hourly data, 1 for hourly data)

  • Spatial resolution (ds): Grid spacing in km (e.g., 12.5 for 12.5 km grid, 3 for 3 km grid)

  • Spatial sampling (nn): Number of points used for noise generation before upscaling (limited by computational resources)

  • Temporal persistence (rho): AR-1 correlation coefficient (adjust based on forcing time resolution)

  • Covariance matrix: Cross-variable covariance structure and correlation coefficients (configurable)

  • File paths: Input/output directories (must be adjusted for local system)

Output: One NetCDF file per month containing noise fields for all ensemble members and all perturbed variables.

Stream File Configuration#

Stream files configure how eCLM reads and applies noise fields to atmospheric forcing data. Three separate stream files are required (precipitation, solar radiation, and other variables). Each stream file must specify the following metadata fields:

Required Metadata:

  • timeInformation: Temporal resolution of forcing data in hours (e.g., 3 for 3-hourly data)

  • caseId: Ensemble member index (e.g., 0 for first member, 1 for second member)

  • numEns: Total ensemble size used when generating the noise files

  • Perturbation variables: tbot_noise (temperature), lwdn_noise (longwave radiation), precn_noise (precipitation), swdn_noise (shortwave radiation)

Stream File Examples#

The following examples demonstrate proper stream file configuration for precipitation, solar radiation, and other atmospheric variables. Each example must be adapted with correct file paths, temporal parameters, and ensemble member IDs for your specific application.

Precipitation Noise Stream File#

File: user_datm.streams.precip_noise.stream_0000.txt

<dataSource>
   GENERIC
</dataSource>
<domainInfo>
  <variableNames>
     time    time
        xc      lon
        yc      lat
        area    area
        mask    mask
  </variableNames>
  <filePath>
     /path/to/domain/file
  </filePath>
  <fileNames>
     domain.lnd.EUR-11_EUR-11.230216_mask.nc
  </fileNames>
</domainInfo>
<fieldInfo>
   <variableNames>
     PRECTmms precn_noise
   </variableNames>
   <filePath>
     /path/to/noise/files
   </filePath>
   <fileNames>
2003-01.nc
2003-02.nc
2003-03.nc
2003-04.nc
2003-05.nc
2003-06.nc
2003-07.nc
2003-08.nc
2003-09.nc
2003-10.nc
2003-11.nc
2003-12.nc
   </fileNames>
   <offset>
     0 
   </offset>
   <timeInformation>
     3 
   </timeInformation>
   <caseId>
     0 
   </caseId>
   <numEns>
     64 
   </numEns>
</fieldInfo>
Solar Radiation Noise Stream File#

File: user_datm.streams.solar_noise.stream_0000.txt

<dataSource>
   GENERIC
</dataSource>
<domainInfo>
  <variableNames>
     time    time
        xc      lon
        yc      lat
        area    area
        mask    mask
  </variableNames>
  <filePath>
     /path/to/domain/file
  </filePath>
  <fileNames>
     domain.lnd.EUR-11_EUR-11.230216_mask.nc
  </fileNames>
</domainInfo>
<fieldInfo>
   <variableNames>
     FSDS swdn_noise
   </variableNames>
   <filePath>
     /path/to/noise/files
   </filePath>
   <fileNames>
2003-01.nc
2003-02.nc
2003-03.nc
2003-04.nc
2003-05.nc
2003-06.nc
2003-07.nc
2003-08.nc
2003-09.nc
2003-10.nc
2003-11.nc
2003-12.nc
   </fileNames>
   <offset>
     0 
   </offset>
   <timeInformation>
     3 
   </timeInformation>
   <caseId>
     0 
   </caseId>
   <numEns>
     64 
   </numEns>
</fieldInfo>
Temperature and Longwave Radiation Noise Stream File#

File: user_datm.streams.other_noise.stream_0000.txt

<dataSource>
   GENERIC
</dataSource>
<domainInfo>
  <variableNames>
     time    time
        xc      lon
        yc      lat
        area    area
        mask    mask
  </variableNames>
  <filePath>
     /path/to/domain/file
  </filePath>
  <fileNames>
     domain.lnd.EUR-11_EUR-11.230216_mask.nc
  </fileNames>
</domainInfo>
<fieldInfo>
   <variableNames>
        TBOT     tbot_noise
        FLDS     lwdn_noise
   </variableNames>
   <filePath>
     /path/to/noise/files
   </filePath>
   <fileNames>
2003-01.nc
2003-02.nc
2003-03.nc
2003-04.nc
2003-05.nc
2003-06.nc
2003-07.nc
2003-08.nc
2003-09.nc
2003-10.nc
2003-11.nc
2003-12.nc
   </fileNames>
   <offset>
     0 
   </offset>
   <timeInformation>
     3 
   </timeInformation>
   <caseId>
     0 
   </caseId>
   <numEns>
     64 
   </numEns>
</fieldInfo>

2. DATM Integration (datm_comp_mod.F90)#

Passes ensemble information (inst_index, dt_option, ninst) from the data atmosphere (DATM) component to the stream infrastructure, enabling the noise perturbation mechanism to operate correctly within CESM’s multi-instance framework.

Design Rationale#

  1. Backward compatibility: All changes are guarded by USE_PDAF preprocessor directives

  2. Efficiency: Uses a single noise file for all ensemble members by time-shifting, avoiding storage duplication

Modified Source Code#

  • src/csm_share/streams/shr_stream_mod.F90

  • src/csm_share/streams/shr_strdata_mod.F90

  • src/csm_share/streams/shr_dmodel_mod.F90

  • src/datm/datm_comp_mod.F90

Contents