Model Card

This is a template for Hugging Face model cards tailored for geospatial foundation models. Copy the markdown below and paste it into your model's README.md on Hugging Face Hub. Replace the placeholders in {braces} with your own information.

Template

---
# === Basic Information (Required) ===
language:
- {lang_0}
- {lang_1}

license: {license}
license_name: {license_name}
license_link: {license_link}

library_name: {library_name}
provider: {provider}
funder: {funder}

tags:
- {tag_0}
- {tag_1}
- {tag_2}

# === Embedding Properties (Required) ===
embedding_spatial_types:
- {embedding_spatial_type_0}
- {embedding_spatial_type_1}

embedding_temporal_type:
- {embedding_temporal_type_0}
- {embedding_temporal_type_1}

embedding_spatial_context: {embedding_spatial_context}
embedding_temporal_context: {embedding_temporal_context}
embedding_dimension: {embedding_dimension}

# === Model Details (Optional) ===
description: {description}
compression: {compression}
intention: {intention}
cautions: {cautions}
precomputed_embeddings: {precomputed_embeddings}
publication_link: {publication_link}
model_architecture: {model_architecture}

# === Pretraining (Optional) ===
pretraining:
  data_types:
  - {data_type_0}
  - {data_type_1}
  product_names:
  - {product_name_0}
  - {product_name_1}
  training_strategy: {training_strategy}
  training_resource: {training_resource}
  spatial_extent: {spatial_extent}
  temporal_extent: {temporal_extent}
  patch_size: {patch_size}
  temporal_context: {temporal_context}
  batch_size: {batch_size}

# === Inference (Optional) ===
inference:
  data_types:
  - {data_type_0}
  - {data_type_1}
  product_names:
  - {product_name_0}
  - {product_name_1}
  patch_size: {patch_size}
  temporal_context: {temporal_context}

# === Standard HF Fields (Optional) ===
datasets:
- {dataset_0}

metrics:
- {metric_0}

base_model: {base_model}

# === Evaluation Results (Optional) ===
model-index:
- name: {model_id}
  results:
  - task:
      type: {task_type}
      name: {task_name}
    dataset:
      type: {dataset_type}
      name: {dataset_name}
      config: {dataset_config}
      split: {dataset_split}
      revision: {dataset_revision}
      args:
        {arg_0}: {value_0}
    metrics:
      - type: {metric_type}
        value: {metric_value}
        name: {metric_name}
        config: {metric_config}
        args:
          {arg_0}: {value_0}
        verifyToken: {verify_token}
    source:
      name: {source_name}
      url: {source_url}
---

Field Reference

Basic Information

Field	Required	Description	Example
language	No	Language codes	fr, en
license	Yes	License identifier from HF licenses	apache-2.0
license_name	No	Custom license ID (if license = other)	my-license-1.0
license_link	No	Path or URL to license file (if license = other)	LICENSE.md
library_name	No	Library from HF model libraries	keras
provider	Yes	Organization or individual that developed the model	NASA
funder	No	Funding institutions	NSF, ESA
tags	No	Searchable tags	SSL, Geospatial Foundation Model, multispectral

Embedding Properties

Field	Required	Description	Acceptable Values
embedding_spatial_types	Yes	Spatial type of embeddings	pixel, patch, scene
embedding_temporal_type	Yes	Temporal type of embeddings	single-date, multi-date
embedding_spatial_context	Yes	Spatial context scope	spatial context determined by embedding spatial type, spatial context beyond embedding spatial type
embedding_temporal_context	Yes	Temporal context scope	temporal context determined by embedding spatial type, spatial context beyond embedding temporal type
embedding_dimension	Yes	Embedding vector size (integer)	768

Model Details

Field	Required	Description	Example
description	Yes	Free text explanation of the model
compression	No	Description of storage compression used
intention	No	Intended use case and how training data was sampled	land cover, oceans, urban
cautions	No	Constraints or cautions for users	model not trained on snow, loses accuracy with high cloud coverage
precomputed_embeddings	No	Link to precomputed embeddings, or no	yes: https://... or no
publication_link	No	URL to related publication
model_architecture	No	Description of model architecture	ViT-L/14

Pretraining (Optional)

Field	Description	Acceptable Values / Example
data_types	Types of data used for training	RGB, multispectral, hyperspectral, SAR, LiDAR, DEM, climate data, text, semantic data
product_names	Data products used	sentinel-2-l2a
training_strategy	Training approach	Contrastive, MIM, Barlow Twins
training_resource	Training resource requirements (energy, GPU, etc.)
spatial_extent	Bounding box(es) in EPSG 4326
temporal_extent	Date range	01-01-2020 to 31-12-2023
patch_size	Patch size (integer)	224
temporal_context	Temporal context for training	single-date, multi-date
batch_size	Batch size (integer)	32

Inference (Optional)

Field	Description	Acceptable Values / Example
data_types	Types of data supported for inference	RGB, multispectral, hyperspectral, SAR, LiDAR, DEM, climate data, text, semantic data
product_names	Data products supported	sentinel-2-l2a
patch_size	Patch size (integer)	224
temporal_context	Temporal context for inference	single-date, multi-date

Evaluation Results (Optional)

Use model-index to encode evaluation results for downstream tasks.

Field	Required	Description	Example
model-index.name	Yes	Model identifier
task.type	Yes	Task type	crop field segmentation
task.name	No	Task name	Field Segmentation
dataset.type	Yes	Dataset type	Field boundary labels
dataset.name	Yes	Dataset name	PASTIS
dataset.config	No	Dataset subset for load_dataset()
dataset.split	No	Dataset split	test
dataset.revision	No	Dataset revision hash
metrics.type	Yes	Metric ID from HF metrics	wer
metrics.value	Yes	Metric value	20.90
metrics.name	No	Metric display name	Test WER
source.name	No	Source of evaluation results	PANGAEA
source.url	If source provided	Link to source	https://arxiv.org/...