Finland 2017 flooding case
The 12 of August 2017 a mesoscale convective storm “Kiira” moved over south of Finland with intensive rainfall (50 mm/h) and strong wind gusts (33 m/s), resulting in flooding and severe damages to the society. The weather situation has been simulated using the Reference HARMONIE-AROME 2.5 km model and different configurations of high-resolution model systems, see [3; Section 1]. The high-resolution runs include both HARMONIE-AROME (nesting and downscaling) and the new OD_DT model (downscaling), with the goal to reach 200-750m horizontal resolution. Several setups of the models have been tested and the forecast output have been distributed to LUMI for further access by end-users, see [3; Section 2] and the storage and tracking tool DCMDB. Model runs have been performed on ECMWF-BOND supercomputers, but also CSC-LUMI has been used recently (e.g. during Q1-2024).
Main conclusions using higher resolution OD-DT model are that the precipitation accumulation amount and placement is better than the Reference HARMONIE-AROME run, for the Finland case. Though, there is a “timing issue”, e.g. approximately 6h delay of maximum precipitation. This is most likely due to the down-scaling of the coupled IFS model, which also has a delay of the event. There is a variability in the skill for the high-resolution runs; e.g. when using different domain size and domain center locations. These settings become even more sensitive when approaching resolution of 200 meters, where one is limited to use smaller domains (due to HPC resources and run-time). A comparison between the Reference model and the OD-DT model (version 0.1.0) shows improved precision for the precipitation when using the OD-DT 750 meter resolution model, see Fig. 3. There are small differences using various OD-DT versions, e.g. versions 0.1.0-0.4.0. More information and results can be found in [3; Section 3].
High-resolution “Consecutive-domains” 500m model runs following the storm-track path have been performed, e.g. the possibility of moving the simulation domain along the expected trajectory of the target systems. More information and animations found in this link.
The hydrological model can use input data from either observations, radars or models in order to estimate the flooding and river run-off. Using models with higher resolution shows a positive impact validating against accumulated precipitation with radar accumulation and station measurements, as shown in [3; Section 4].
Reference and high-resolution configuration setup.
| Reference Harmonie-Arome | Harmonie-Arome | Prototype v0.1.0 -v0.4.0 | |
|---|---|---|---|
| Version | CY43 | CY46h1 | CY46h1 |
| HPC | BOND-ATOS | BOND-ATOS | BOND-ATOS, CSC-LUMI |
| Resolution (horizontal/vertical) | 2.5km / 65 levels | 750m, 500m, 200m / 90 levels | 750m, 500m / 90 levels |
| Time step | 75 sec | 30 / 30 / 15 sec | 20 / 20 sec |
| Domain | 960x1080 (MetCoOp) | 1500x1500, 1000x1000 | 1500x1500, 1000x1000 |
| Coupling | IFS | MetCoOp, IFS | IFS |
| DA | 3DVAR | Nesting/Downscaling | Downscaling |
/scratch/project_465000527/de_33050_common_data/cases/finland_2017/
cy46h1_EDT_ref_fin - The baseline HARMONE-AROME (similar to MEPS) with 2.5 km resolution, 1500 gp’s domain. Using IFS 9 km boundaries and data assimilation.
cy46h1_EDT_s0d_n750_1500gp - Downscaled HARMONIE-AROME run with 750 m resolution, 1500 gp’s domain. Boundaries from 2.5 km HARMONIE-AROME model. No data assimilation (e.g. no observations are used to correct the model initial state). Here using a 0-days spinup period.
cy46h1_EDT_s0d_n500_1500gp - Downscaled HARMONIE-AROME run with 500 m resolution, 1500 gp’s domain. Boundaries from 2.5 km HARMONIE-AROME model. No data assimilation (e.g. no observations are used to correct the model initial state). Here using a 0-days spinup period.
cy46h1_EDT_s0d_n200_1500gp_pgdMPI - Nested HARMONIE-AROME run with 200 m resolution, 1500 gp’s domain. Boundaries from the 500 m model run (here above). No data assimilation (e.g. no observations are used to correct the model initial state). Here using a 0-days spinup period.
Prot_F17_750_1500 - Prototype (v0.1.0) downscaled model run with 750 m, 1500 gp’s domain. No data assimilation. Boundaries from IFS 9 km model.
Prot_F17_500_1500 - Prototype (v0.1.0) downscaled model run with 500 m, 1500 gp’s domain. No data assimilation. Boundaries from IFS 9 km model.
caseFin2107 - Newest prototype (v0.4.0) downscaled model run with 750 m, 1500 gp’s domain. No data assimilation. Boundaries from IFS 9 km model.
Figure 1. Catchment areas (black and red) for hydrology model input; Results from red area is presented in figure 2.
Figure 2. Precipitation accumulation for a 24h period: 12 Aug 2017, 00Z - 13 Aug 2017, 00Z. Reference HARMONIE-AROME 2.5km (left), Downscaled Harmonie-Arome 750m (middle) and radar measurements (right).
Figure 3. Daily precipitation amounts (mm/day) for site measurement at Virolahti site, radars, HARMONIE-AROME 750m model (cy46h1_EDT_s0d_n750) and Reference HARMONIE-AROME 2.5km model.