# Land-use biodiversity loss accounts for the prospective MRIO, 2035

> **Status: under development.** The prospective MRIO database and the impact accounts in this folder are preliminary. Structure, values and documentation may change in later versions.

## 1. Prospective MRIO

The prospective MRIO table was constructed with the method of Peng et al. (2026) [[1]](#ref-1). Scenario information was integrated into a historical MRIO table in three steps: preprocessing and harmonization, Leontief-based projection, and optimization-based projection. The historical baseline is EXIOBASE version 3.8.2 for the year 2019 [[2]](#ref-2). The scenario information comes from the outputs for the year 2035 of the Integrated Model to Assess the Global Environment (IMAGE) [[3]](#ref-3) under the Shared Socioeconomic Pathway 2 (SSP2).

The integrated scenario information covers:
- energy supply and demand,
- industry outputs,
- freight transport,
- agricultural production and demand,
- prices and international trade of energy and agricultural products.

The prospective MRIO table for 2035 is industry by industry, with 20 regions × 70 sectors, and is expressed in billions of 2010 USD (10⁹ USD2010). Final demand is given as one column per final demand region.

## 2. Impact calculation

1. **Land occupation per region-sector.** Areas are in Mha for built-up land, cropland, pasture and forest (timber plantation), following the land allocation of Friedli (2026) [[4]](#ref-4). The regional areas equal the IMAGE 2035 land cover (SSP2 - M). Areas are distributed to sectors as follows:
   - cropland and pasture: EXIOBASE 3.8.2 land shares and IMAGE crop and ruminant production (physical productivity);
   - raw milk: monetary productivity (base-year area per unit of output), with the ruminant and raw-milk areas scaled to the IMAGE pasture total;
   - built-up land: monetary output shares;
   - forest: the forestry sector receives the IMAGE timber plantation area.
2. **Characterization.** GLAM terrestrial land-use characterization factors are used (EQ Land use, damage level, global PDF·y/(m²·y)) [[5]](#ref-5). They are aggregated from the 49 EXIOBASE regions to the 20 MRIO regions by area weighting:
   - cropland: arable and fallow CFs, with all `Cropland - fallowed area…` rows of EXIOBASE classified as fallow, as in the GLAM flow bridge;
   - built-up: urban CF;
   - pasture: grassland/pasture/meadow CF;
   - forest: forest, intensive CF.

   The CFs are constant over time. The impact is area × CF of the producing region, in global PDF.
3. **Input–output calculation** with pymrio [[6]](#ref-6):
   - x = Z·ι + Y·ι,
   - A = Z·x̂⁻¹,
   - L = (I − A)⁻¹,
   - S = F·x̂⁻¹, with F the direct impact per region-sector from step 2,
   - M = S·L,
   - D_cba, D_pba, D_imp and D_exp by `pymrio.calc_accounts`,
   - diagonalized matrices D_r = diag(s)·L·diag(y_r) for each final demand region r.

## 3. Files

| File | Content |
|---|---|
| `2035/pmrio20x70_2035_<indicator>_accounts.parquet` | 1 row (indicator) × 8 400 columns (account, region, sector): D_cba, D_pba, D_imp, D_exp, M, S, in this order, with 1 400 columns each |
| `2035/pmrio20x70_2035_<indicator>_meta.yaml` | Metadata: units, method, account definitions, dimensions |
| `2035/<REGION>_2035_<indicator>.parquet` | Diagonalized impacts of the final demand region `<REGION>` (D_r), 1 400 × 1 400. Rows: producing region, producing sector (where the impact occurs) and year. Columns: final supply region, final supply sector. |

Indicators: `BiodiversityLossPDF_from_LandUse` (total), `…_from_BuiltUp`, `…_from_Cropland`, `…_from_Forest`, `…_from_Pasture`.

| Account | Dimension of the columns | Unit |
|---|---|---|
| S | producing region, producing sector | PDF per 10⁹ USD2010 |
| M | final supply region, final supply sector | PDF per 10⁹ USD2010 of final demand |
| D_cba | final demand region, final supply sector (summed over final supply regions) | PDF |
| D_pba | producing region, producing sector | PDF |
| D_imp | final demand region, final supply sector | PDF |
| D_exp | producing region, producing sector | PDF |

Reading example:

```python
import pandas as pd, yaml
acc = pd.read_parquet("2035/pmrio20x70_2035_BiodiversityLossPDF_from_LandUse_accounts.parquet")
d_cba = acc["D_cba"]                      # 1 x 1400
diag = pd.read_parquet("2035/CHN_2035_BiodiversityLossPDF_from_LandUse.parquet")   # 1400 x 1400
meta = yaml.safe_load(open("2035/pmrio20x70_2035_BiodiversityLossPDF_from_LandUse_meta.yaml"))
```

Global total: 0.1796 PDF.

## 4. Key limitations

- **Development status.** The construction method of the prospective MRIO is peer reviewed [[1]](#ref-1). This version of the prospective MRIO and the land allocation are under development and have not been peer reviewed.
- **Scenario and time.**
  - Only one scenario is included (SSP2).
  - The CFs are constant over time.
- **Characterization factors.** GLAM provides identical land-use CFs for the five EXIOBASE rest-of-world regions. RAF, RAM, RAS, ME and UKR therefore share the built-up, forest and pasture CFs, and their cropland CFs differ by less than 1 %; together they account for 46 % of the global production-based impact. Differences between these regions reflect land area and land-use composition, not ecological vulnerability.
- **Sector definition.** "Ruminant livestock farming" also represents ruminant products (meat and dairy), following the IMAGE commodity "Ruminant". Consumption-based results for this final supply sector therefore include final demand for ruminant products, and the final supply sector "Processing of ruminant livestock" is small (3.0 % of the global consumption-based impact, compared with 31.5 % for "Ruminant livestock farming").
- **Output–land consistency.** In a few region-sectors, the MRIO output is small relative to the allocated land, for example CHN forestry, RUS paddy rice and OCE forestry. Their land per unit of output is 12–20 times the median of the same sector across regions. Totals are unaffected; the attribution of these impacts to final demand is uncertain.
- **Land allocation.**
  - Built-up land is allocated by monetary output shares.
  - Forest use outside the forestry sector is not included.
  - Services and downstream sectors have no direct land use.

## 5. References

<a id="ref-1"></a>[1] Peng, S., Daioglou, V., Stadler, K., Roshandel, R., Bruckner, M., & Pfister, S. (2026). Constructing prospective multiregional input-output tables via scenario-constrained optimization for future environmental footprint assessment. *Environmental Science & Technology*, accepted for publication.

<a id="ref-2"></a>[2] Stadler, K., Wood, R., Bulavskaya, T., Södersten, C.-J., Simas, M., Schmidt, S., et al. (2021). EXIOBASE 3 (3.8.2). Zenodo. https://doi.org/10.5281/zenodo.5589597

<a id="ref-3"></a>[3] Stehfest, E., van Vuuren, D., Kram, T., Bouwman, L., et al. (2014). *Integrated Assessment of Global Environmental Change with IMAGE 3.0: Model description and policy applications*. PBL Netherlands Environmental Assessment Agency, The Hague. https://www.pbl.nl/sites/default/files/downloads/pbl-2014-integrated_assessment_of_global_environmental_change_with_image30_735.pdf

<a id="ref-4"></a>[4] Friedli, L. (2026). *Assessing biodiversity hotspots of future consumption*. Bachelor thesis, ETH Zurich (unpublished student thesis).

<a id="ref-5"></a>[5] GLAM (Global Guidance for Life Cycle Impact Assessment Indicators and Methods, UNEP Life Cycle Initiative), land-use characterization factors mapped to EXIOBASE regions: `glam_exio_char_202511.tsv`, with flow bridge `flow_bridge_202603.tsv`, provided by K. Stadler (NTNU).

<a id="ref-6"></a>[6] Stadler, K. (2021). Pymrio – a Python based multi-regional input-output analysis toolbox. *Journal of Open Research Software*, 9(1), 8. https://doi.org/10.5334/jors.251
