Finland SOC status from a multimodel ensemble recalibration

Gridded soil organic carbon (SOC) maps of Finland — HIKET / NextGenC Kriged maps of soil organic carbon stock for Finnish forest land, 1985–2024, from six soil decomposition models (SP1, TP2, TP3, Yasso07, Yasso15, Yasso20) and their equal-weight ensemble. Produced by the HIKET Bayesian model-intercomparison project for the NextGenC reporting effort. This is a demonstration / illustrative product — see Interpretation & limitations below before using the maps quantitatively. Contents- 14 multiband GeoTIFFs: for each model M and the ENSEMBLE, M_SOC_mean.tif (mean SOC, tC ha⁻¹) and M_SOC_sd.tif (total uncertainty, 1 SD). 40 bands each = years 1985–2024 (band n = year 1984 + n; year also stored as the layer name, e.g. y1985).- ENSEMBLE_SOC_delta10yr.tif — single-band annual SOC change (tC ha⁻¹ yr⁻¹) over 2015–2024, the per-cell OLS slope of the ensemble mean (~76 % of cells gaining, mean +0.13).- Per-plot value tables (the numbers behind the maps): <MODEL>_SOC_mean.csv / _sd.csv (447 plots × 40 years) and Finland_SOC_matrices.ods (all 12 matrices).- thumbnails/ — one PNG preview per raster (1985–2024 time-average; the delta is the 2015–2024 change). Peat = black, water/sea = white.- README.md (full file documentation) and LICENSE. A methods note is provided alongside the archive as SOC_maps_README.pdf. Raster specification- CRS: ETRS89-LAEA Europe (EPSG:3035), equal-area. The 2 km grid is snapped to the EEA reference grid (cell edges on multiples of the resolution from the LAEA origin).- Resolution 2000 m; grid 330 × 589, masked to Finnish forest land; Float32; DEFLATE-compressed; sea / outside-Finland = NoData; units tonnes C per hectare. Method. Posterior-predictive SOC (mean + SD per plot per year) from each model's Bayesian calibration is interpolated by ordinary kriging in log space (one pooled variogram per model), back-transformed to tC ha⁻¹. Per-model SD fuses the kriging (interpolation) variance with the interpolated model posterior uncertainty; the ensemble SD applies the law of total variance (within-model + between-model structural spread). Kriging is performed natively in EPSG:3035. Mineral-soil forest SOC is targeted: peat/water plots are dropped from the input and cells >50 % peatland (Luke MS-NFI paatyyppi) or >50 % water (SYKE Ranta10 lakes) are masked. Interpretation & limitations. Spatial detail is limited by the plot network: plots are ~27 km apart on average and SOC is spatially weakly autocorrelated at that spacing (variogram ~85–90 % nugget). The maps therefore reproduce the national mean field with plot-anchored local deviations; smooth sub-regional gradients are not resolvable from these points, and detail between plots is interpolation, not measurement. Read the _sd layer as the honest statement of confidence — it inflates between plots and in data-sparse regions. Point SOC predictability is intrinsically low in this system (calibration R² ≈ 0.05–0.11). Note: the per-plot CSV/ODS _sd is the parameter spread on the predictive mean (observation error not added). The TP3 layers use the forward-Euler posterior, superseded by an exact matrix-exponential re-calibration; SOC levels change negligibly. Credits & attribution. Project: HIKET — Bayesian calibration and structural intercomparison of SOC models for the Finnish greenhouse-gas inventory, produced for NextGenC. Litter inputs: Tupek et al. (Zenodo DOI 10.5281/zenodo.19736499). Mask source data (not redistributed; please credit if reused): peat — Luke Multi-source NFI (paatyyppi), © Natural Resources Institute Finland, CC-BY-4.0; water — SYKE Ranta10 / Shoreline 1:10 000 lakes, © Finnish Environment Institute (SYKE), CC-BY-4.0. License: Creative Commons Attribution 4.0 International (CC-BY-4.0). Contact: [e-mail]