# Irrecoverable Earth: Global Carbon Reserves (v2, 2025)

## Overview

This dataset maps global "irrecoverable carbon" — carbon stocks that, if lost, cannot be recovered by 2050, making their conservation critical for meeting climate goals. It also includes "manageable carbon" (stocks that can be influenced by human management) and "vulnerable carbon" (total stocks vulnerable to release upon land conversion). The three are nested pools: **manageable ⊃ vulnerable ⊃ irrecoverable**.

This collection serves the **Conservation International v2 (2025) release** (Zenodo 17645053), the revised-methodology update of Noon et al. (2022). The earlier v1 (2021) release has been retired as superseded by v2.

- **Irrecoverable carbon**: 2018, 2022, 2023, 2024
- **Vulnerable & manageable carbon**: 2010, 2018, 2024

(Irrecoverable 2010 is not included — its Zenodo source TIFF has a corrupt IFD and cannot be read.)

## Source & Attribution

**Source**: Conservation International
**Project Page**: https://www.conservation.org/irrecoverable-carbon
**Data (v2, 2025 update)**: https://doi.org/10.5281/zenodo.17645053
**Citation**: Noon, M.L., Goldstein, A., Ledezma, J.C. et al. Mapping the irrecoverable carbon in Earth's ecosystems. *Nat Sustain* **5**, 37–46 (2022). https://doi.org/10.1038/s41893-021-00803-6
**License**: Creative Commons Attribution Non Commercial 4.0 International (CC BY-NC 4.0) — non-commercial use only.

**Key findings from the 2025 update**: 5.4% (7.4 Gt) of Earth's irrecoverable carbon was lost between 2018–2024; CI's published global total is now 128.0 Gt (CI corrected the original 2018 baseline from 139.1 to 135.3 Gt).

---

## Data Formats & Access

**Base URL**: `https://s3-west.nrp-nautilus.io/public-carbon/`

Two products per layer: the source Cloud-Optimized GeoTIFF (carbon **density**, Mg C ha⁻¹, ~300 m, unchanged from source) and H3 hexagonal parquet (carbon **total**, Mg C per cell).

### Cloud-Optimized GeoTIFFs

**Path Pattern**: `v2/cogs/{category}_c_total_{year}.tif` — pixel values are carbon DENSITY in Mg C ha⁻¹.

| File | Year |
|------|------|
| `v2/cogs/irrecoverable_c_total_{2024,2023,2022,2018}.tif` | 2024, 2023, 2022, 2018 |
| `v2/cogs/vulnerable_c_total_{2024,2018,2010}.tif` | 2024, 2018, 2010 |
| `v2/cogs/manageable_c_total_{2024,2018,2010}.tif` | 2024, 2018, 2010 |

All COGs are compressed with ZSTD. (Irrecoverable 2010 is absent — corrupt source TIFF.)

### H3 Hexagonal Parquet

Carbon **totals** at H3 **resolution 9** (native), with parent rollups at resolutions 8, 7, 6, 5 as columns, hive-partitioned by H3 resolution 0 (`h0`).

**Path Pattern**: `{dataset-name}/hex/h0={cell}/data_0.parquet`

| Dataset | Path prefix |
|---------|-------------|
| Irrecoverable carbon 2024 | `irrecoverable-carbon-2024/hex/` |
| Irrecoverable carbon 2023 | `irrecoverable-carbon-2023/hex/` |
| Irrecoverable carbon 2022 | `irrecoverable-carbon-2022/hex/` |
| Irrecoverable carbon 2018 | `irrecoverable-carbon-2018/hex/` |
| Vulnerable carbon 2024    | `vulnerable-carbon-2024/hex/` |
| Vulnerable carbon 2018    | `vulnerable-carbon-2018/hex/` |
| Vulnerable carbon 2010    | `vulnerable-carbon-2010/hex/` |
| Manageable carbon 2024    | `manageable-carbon-2024/hex/` |
| Manageable carbon 2018    | `manageable-carbon-2018/hex/` |
| Manageable carbon 2010    | `manageable-carbon-2010/hex/` |

### Usage Examples

**DuckDB — compare irrecoverable carbon across years (totals in Gt C):**
```sql
SELECT year, SUM(carbon) / 1e9 AS total_Gt_C
FROM (
    SELECT 2024 AS year, carbon FROM read_parquet('https://s3-west.nrp-nautilus.io/public-carbon/irrecoverable-carbon-2024/hex/h0=*/data_0.parquet')
    UNION ALL SELECT 2023, carbon FROM read_parquet('https://s3-west.nrp-nautilus.io/public-carbon/irrecoverable-carbon-2023/hex/h0=*/data_0.parquet')
    UNION ALL SELECT 2022, carbon FROM read_parquet('https://s3-west.nrp-nautilus.io/public-carbon/irrecoverable-carbon-2022/hex/h0=*/data_0.parquet')
    UNION ALL SELECT 2018, carbon FROM read_parquet('https://s3-west.nrp-nautilus.io/public-carbon/irrecoverable-carbon-2018/hex/h0=*/data_0.parquet')
)
GROUP BY year ORDER BY year;
```

---

## Data Dictionary

### Hexagonal Parquet Schema

| Column | Type | Description |
|--------|------|-------------|
| `carbon` | double | Total carbon stock in the H3 cell, in Mg C (1 Mg = 1 tonne). |
| `h9` | uint64 | H3 cell ID at resolution 9 (native; exactly one row per cell). |
| `h8`–`h5` | uint64 | H3 parent cell IDs at resolutions 8, 7, 6, 5 (rollups). |
| `h0` | int64 | H3 resolution-0 cell ID — hive partition key. |

`SUM(carbon)` over hex rows is directly valid as a carbon TOTAL (no GROUP BY or dedup) — exactly one row per `h9` cell. Roll up to coarser resolutions with `GROUP BY h8` (or h7/h6/h5) `+ SUM(carbon)`.

### How `carbon` is computed (and how it validates)

The source COGs store carbon **density** (Mg C ha⁻¹). Each ~300 m pixel's density is multiplied by the pixel's ground area (latitude-dependent on the WGS84 grid) and apportioned to the H3 cells it overlaps — an area-integral that yields total Mg C per cell.

**Validation (data-workflows #330):** hex `SUM(carbon)` matches an independent full-resolution area-integral of the source COG to a uniform **+0.22%** on every layer (a benign H3-cell-area vs exact-pixel discretization offset). These straight area-integrals run **~2–3% above** CI's published committed-loss headline figures — e.g. irrecoverable 2024 = 131.9 Gt C here vs CI's 128.0 Gt. Consequently the per-year series is **not strictly monotonic** in a raw integral (e.g. irrecoverable rises slightly 2023→2024); this reflects the shipped density rasters, not our processing. **For committed-loss change figures, cite CI's summary, not a difference of these totals.**

### Terminology

- **Irrecoverable Carbon**: Ecosystem carbon that, if lost, could not be recovered by 2050 through natural regeneration or restoration.
- **Vulnerable Carbon**: Total carbon stocks (biomass + soil) vulnerable to release upon land conversion.
- **Manageable Carbon**: Carbon stocks in ecosystems that can be managed by human activities (forestry, agriculture, conservation).

### v2 (2025) methodology vs the original 2021 release

1. **Bug fix**: corrected overestimation of irrecoverable carbon in boreal grasslands/shrublands in Asia (2018 total 139.1 → 135.3 Gt).
2. **Updated coastal extents**: Global Mangrove Watch 2022 and UNEP-WCMC 2022 (seagrasses, salt marshes).
3. **Annual tracking**: added 2022, 2023, 2024 (irrecoverable); 2024 (vulnerable, manageable).
4. **Protection status**: WDPA data through January 2023–October 2025.

---

## Background & Methodology

"Irrecoverable carbon" identifies stocks where preventing loss is more effective than restoration — the overlap of high carbon density (climate value), low recoverability (cannot regenerate by 2050), and high threat (risk of conversion). The dataset integrates biomass (Spawn et al., Walker et al.), soil carbon (SoilGrids250m), ecosystem vulnerability, and ESA CCI Land Cover for annual conversion tracking.