WRF-VPRM v0.3 (sample output)

This dataset contains input and sample output data from WRF-VPRM (Weather Research and Forecasting model coupled with the Vegetation Photosynthesis and Respiration Model) simulations conducted over complex alpine topography. It accompanies the manuscript "Sensitivity of CO2 exchange in WRF-VPRM to model resolution and parameter settings over Alpine topography" (Geoscientific Model Development, in review).

Contents

Input data:

- CAMS subset — CO2 initial and boundary conditions from the Copernicus Atmosphere Monitoring Service
- VPRM input — CORINE land cover, MODIS-derived vegetation indices (EVI, LSWI), and pre-processed VPRM parameter fields (generated via pyVPRM and vprm_shapeshifter)
- FLUXNET2015 — reference site data used for VPRM parameter optimisation and model validation

Sample model output:

- One day (2012-07-27) of WRF-VPRM hourly output across the nested domains (54 km, 27 km, 9 km, 3 km, 1 km). The full multi-month simulation output is ~5 TB and is not redistributed.

Not included here (archived in the companion records):

- Modified WRF source code → DOI 10.5281/zenodo.22253904 (https://doi.org/10.5281/zenodo.22253904)
- Modified WPS source code → DOI 10.5281/zenodo.22257104 (https://doi.org/10.5281/zenodo.22257104)
- Workflow, namelists, post-processing scripts, and extracted site-level / domain-averaged time series CSVs used to generate the paper figures → DOI 10.5281/zenodo.22253621 (https://doi.org/10.5281/zenodo.22253621)

Model setup

The simulations use a coupled WRF-VPRM framework to simulate CO2 fluxes over the Alps. Key components:

- Meteorological boundary conditions from ERA5 reanalysis
- CO2 initial and boundary conditions from CAMS (Copernicus Atmosphere Monitoring Service)
- VPRM parameters optimised using differential evolution against FLUXNET2015 site observations
- MODIS satellite data for vegetation indices (EVI, LSWI)
- CORINE Land Cover 2018 for land-surface classification

Note on prior versions

Versions v0.1 and v2 of this record bundled WRF and WPS source code, but those bundles contained upstream code rather than the modified WRF-P and WRF-VPRM-CLC branches actually used in the paper. The modified source is now archived in the three separate records linked above. This v0.3 release retains only the data files.

What's Changed in v0.3

Supersedes v0.2 (DOI 10.5281/zenodo.20395680). **Only `WRFOUT.zip` changes**;
`Fluxnet2015.zip`, `CAMS.zip` and `VPRM_input.zip` are unaffected and carry over
unchanged.

Software: https://github.com/Matthias-Reif-PhD/WRF at tag **v0.3**
(branch `WRF-P`; pairs with WPS v0.3). 

1. Corrected: the online VPRM `Topt` bug

In `chem/module_ghg_fluxes.F` the per-channel optimum temperature `Topt` was assigned with
Fortran `DATA` statements placed *inside* a runtime `SELECT CASE` / `if` block, in both the
`VPRM` and `dflux_dT` subroutines.

`DATA` is a **compile-time initializer** and ignores runtime control flow. With several
`DATA Topt` statements for one array, the compiler keeps only the **first in source order** —
the `VPRM_table_US` values `Topt = 20,20,20,20,20,22,18,0`. ifort issued no warning.

**Every parallel VPRM channel therefore ran with the same wrong `Topt`**, regardless of which
channel was being computed, even though the other parameters (PAR0, lambda, alpha, beta) did
vary correctly per channel.

### Fields affected in v0.2

| field | v0.2 status |
|---|---|
| `EBIO_GEE`, `EBIO_GEE_2` … `EBIO_GEE_5` | **wrong** — ran at Topt=20/18 instead of their intended per-site values |
| `EBIO_GEE_DPDT` | **wrong** (same cause) |
| `EBIO_RES*` | unaffected — respiration has no `Topt` dependence |
| `EBIO_GEE_REF`, `EBIO_GEE_DPDT_REF` | **unaffected** — their intended `Topt` happens to equal the value the bug forced |

Magnitude of the correction at midday 2012-07-27: 7–34 % relative change in the affected GEE
channels, varying by resolution and channel.

2. Added: online Topt-percentile ensemble

Five new parallel channels carry the p10/p25/p50/p75/p90 Topt-percentile ensemble, computed
**online** in the same run rather than by offline recomputation:

```
EBIO_GEE_P10  EBIO_GEE_P25  EBIO_GEE_P50  EBIO_GEE_P75  EBIO_GEE_P90
EBIO_RES_P10  EBIO_RES_P25  EBIO_RES_P50  EBIO_RES_P75  EBIO_RES_P90
```

Parameters come from `WRF_VPRM_post/vprm_params_topt_p{10,25,50,75,90}.csv`, built by
`build_member_param_csvs.py` from the per-PFT Topt and PAR0/lambda percentile tables. Each
member carries a self-consistent (Topt, PAR0, lambda) triple — they are coupled by
equifinality and must not be varied independently.

Note the five `EBIO_RES_P*` fields are **identical to one another** by construction: alpha and
beta are not varied across members, because respiration carries no `Topt` dependence.

Existing channel names and indices are unchanged; the new fields are appended.

3. Validation

Every resolution was cross-validated against an independent offline recomputation of the VPRM
equations (`WRF_VPRM_post/recompute_vprm_fluxes.py`) driven by each run's own archived T2 and
SWDOWN:

| resolution | max abs. GPP difference | RECO difference |
|---|---|---|
| 54 km | 0.004 % | 0.000 % |
| 9 km  | 0.015 % | 0.000 % |
| 3 km  | 0.025 % | 0.000 % |
| 1 km  | 0.028 % | 0.000 % |

RECO agrees exactly; GPP agrees at float32 level. At 54 km and 27 km — where the run is serial
or near-serial and the decomposition matches the archive — the unaffected control channels
(`EBIO_GEE_REF`, `EBIO_RES_REF`, `*_DPDT_REF`) reproduce v0.2 **bit-for-bit**, while the
corrected channels change. That is the sharpest available confirmation that the fix altered
exactly what it should and nothing else.