Background
In the current operational system, snow analysis is based on 2D-OI separated from the SEKF (for soil moisture analysis) and the 1D-OI (for soil temperature and snow temperature). As a step towards an unified land data assimilation system (de Rosnay et al., 2022), snow analysis has been implemented in the SEKF. This page shows details of the changes and the first result.
Branches
dako_CY48R1.1_SEKF_snow
- ifs-scripts: https://git.ecmwf.int/users/dako/repos/ifs-scripts/browse?at=refs%2Fheads%2Fdako_CY48R1.1_SEKF_snow
- ifs-source: https://git.ecmwf.int/users/dako/repos/ifs-source/browse?at=refs%2Fheads%2Fdako_CY48R1.1_SEKF_snow
- ifs-suites: https://git.ecmwf.int/users/dako/repos/ifs-suites/browse?at=refs%2Fheads%2Fdako_CY48R1.1_SEKF_snow
Configurations to analyse snow water equivalent in the SEKF
- Input analysed snow depth from the 2D-OI (arpifs/obs_preproc/obsgen.F90)
- Calculate and input additional EDA Jacobians related to snow (gen/fetchmars, arpifs/sekf/sekf_geteda_jacobians.F90)
- Covariances between soil moisture (CV) and snow depth (OBS) are set to 0
- Covariances between snow water equivalent (CV) and 2m temperature, relative humidity, soil moisture (OBS) are also set to 0
- These 2 settings mean that snow and soil moisture analysis are independent each other at the moment
- Jacobians between snow water equivalent (CV) and snow depth (OBS) are set to 1/rsn (based on the relationship in the model)
- Analyse snow water equivalent using the analysed snow depth and Jacobians (arpifs/sekf/sm_ekf_main.F90)
- σo=0.01 [m]
- σb=2.0 [kg/m2] (=0.01 [m] if rsn=200 [kg/m3])
- So far, increments for snow water equivalent are not reflected to the analysis field (offline calculation only)
Result (01/12/2021 00UTC)
1.Analysis increments in the 2D-OI | 2.Analysis increments in the SEKF (observation: SD from the 2D-OI) | 3.Analysis increments in the SEKF (observation: SWE from the 2D-OI) |
|---|---|---|
|
|
|
- Similar pattern is seen between the analysis increments for SWE in (1) the 2D-OI and (2) the SEKF.
- Note that large differences are seen in some regions due to inconsistency of snow density between the 2D-OI analysis and trajectory (lower limit of 100kg/m3 in the 2D-OI but 50kg/m3 in the model).
- When SWE from the 2D-OI is assimilated directly in the SEKF (3), such differences are not seen. (The trial branch: dako_CY48R1.1_SEKF_tmp)
- This is not an issue when snow depth observations are assimilated directly in the SEKF
| sekf_inc_l1.grib(SM1) | sekf_inc_l2.grib(SM2) | sekf_inc_l3.grib(SM3) | |
|---|---|---|---|
| Cntl |
|
|
|
| Test |
|
|
|
- Analysis increments for soil moisture are the same (bit-identical) between before and after the changes.
Future work
- Input snow observations directly in the SEKF (similar work as soil moisture analysis using analysed 2m temperature and relative humidity)
- Need to implement a localization in observation space
- Each QC also should be implemented in the SEKF
- Consider flow-dependent background errors