[Dataset Title/Name]: GEM-Forest v1.0: A Global satellite EMbedding–based map of forests and tree crops for 2020
BYOD
### Contact Details
palubad@natur.cuni.cz
### Dataset description
## GEM-Forest: Global satellite EMbedding–based map of forests and tree crops for 2020
**GEM-Forest v1.0** is a 10-meter resolution global dataset utilizes satellite embeddings from the **Google DeepMind Alpha Earth Foundation (AEF)** model and a Linear SVM classifier to provide a highly accurate map of forests and agricultural tree crops for the year 2020.
Unlike standard land cover products, GEM-Forest specifically identifies **agricultural tree crops** (such as oil palm, rubber, and orchards) to minimize the confusion between forests and agricultural tree crops.
The approach demonstrates strong potential for temporal transferability across the 2017–2025 period covered by AEF embeddings. This capability allows multi-year applications and change detection based on models trained for a single year. GEM-Forest can support national and international policy and regulatory decisions, including the **EU Deforestation Regulation (EUDR)**.
---
### Dataset specifications
| Property | Details |
| :--- | :--- |
| **Spatial resolution** | 10 meters |
| **Coordinate system** | WGS 84 (EPSG:4326) |
| **Temporal scope** | 2020 (Baseline) |
| **Data source** | AEF Satellite Embeddings |
| **Global accuracy** | 91% (Fn/F); >85% (Tree Crops) |
---
### Datasets/classes
The dataset contains three core values, from which two primary products are derived:
#### 1. GEM-FnF2020 (Forest / Non-Forest)
A binary classification following the FAO forest definition.
| Value | Class | Description |
| :--- | :--- | :--- |
| **0** | Non-Forest | Shrubs, grasslands, water, and other non-forested land, no data |
| **1** | Forest | Land >0.5 ha with trees >5m and canopy cover >10%, excluding tree under agricultural and urban land use. |
#### 2. GEM-TC2020 (Forest / Non-Forest / Tree Crops)
A multi-class map, where the tree crops represents a sub-class of non-forest.
| Value | Class | Description |
| :--- | :--- | :--- |
| **0** | Non-Forest (Other) | Built-up, water, bare land, non-woody vegetation, no data. |
| **1** | Forest | Land >0.5 ha with trees >5m and canopy cover >10%, excluding tree under agricultural and urban land use. |
| **2** | Tree Crops | Agricultural plantations (oil palm, rubber, coconut, other palms, European tree crops, such as olives and fruit trees). |
---
### [GEE GEM-Forest Explorer with provided probabilities](https://danielp-cuni.projects.earthengine.app/view/gem-forest)
---
### Citation
> Paluba, D., Marsocci, V., Onačillová, K., Puerta Quintana, Y. T., and Hastie, A. (2026): GEM-Forest: A Global satellite EMbedding–based map of forests and tree crops for 2020, EGUsphere [preprint], https://doi.org/10.5194/egusphere-2026-1401.
### Cloud-optimized GeoTIFFs and other supporting data
> Paluba, D., Marsocci, V., Onačillová, K., Puerta Quintana, Y. T., & Hastie, A. (2026). GEM-Forest: A Global satellite EMbedding–based map of forests and tree crops for 2020 (GEM-Forest products, training & validation data, and model weights) (1.0) [Data set]. Zenodo. https://doi.org/10.5281/zenodo.18921586
### Earth Engine Snippet if dataset already in GEE
### Earth Engine Code editor snippet ([Open in Code Editor](https://code.earthengine.google.com/92180879cd72ba1ffbdac6502046a360))
```javascript
// Access the GEM-Forest Image Collection
var GEM_Forest = ee.ImageCollection("projects/danielp-cuni/assets/GEM-Forest/GEM-Forest_2020");
// Create a mosaic for global display
var GEM_Forest = GEM_Forest.mosaic();
// Masking out non-forest from both products
var GEM_Forest_FnF = GEM_Forest.updateMask(GEM_Forest.eq(1));
var GEM_Forest_TC = GEM_Forest.updateMask(GEM_Forest.gte(1));
// Visualise the maps
Map.addLayer(GEM_Forest_FnF,
{min: 1, max: 1, palette: ['green']},
'GEM-Forest (F/nF 2020)'
);
Map.addLayer(GEM_Forest_TC,
{min: 1, max: 2, palette: ['green','yellow']},
'GEM-Forest (F/nF/Tree crops 2020)'
);
```
### GEE Python API snippet
```python
import ee
import geemap
# Access the GEM-Forest Image Collection
GEM_Forest = ee.ImageCollection("projects/danielp-cuni/assets/GEM-Forest/GEM-Forest_2020")
# Create a mosaic for global display
GEM_Forest = GEM_Forest.mosaic()
# Masking out non-forest from both products
GEM_Forest_FnF = GEM_Forest.updateMask(GEM_Forest.eq(1))
GEM_Forest_TC = GEM_Forest.updateMask(GEM_Forest.gte(1))
# Visualise the maps
Map = geemap.Map()
Map.addLayer(GEM_Forest_FnF,
{'min': 1, 'max': 1, 'palette': ['green']},
'GEM-Forest (F/nF 2020)'
)
Map.addLayer(GEM_Forest_TC,
{'min': 1, 'max': 2, 'palette': ['green','yellow']},
'GEM-Forest (F/nF/Tree crops 2020)'
)
```
### Enter license information
[Creative Commons Attribution 4.0 International](https://creativecommons.org/licenses/by/4.0/)
### Keywords
forest, non-forest, tree crop, embeddings, EUDR, oil palm, rubber, coconut, olives, fruit trees
### Code of Conduct
- [x] I agree to follow this project's Code of Conduct
0 条评论