nasa-gesdisc/edgraph-gnn-graphsage

EOSDIS Graph Neural Network Model Card

Model Overview

Model Name: EOSDIS-GNN Version: 1.0.4 Type: Heterogeneous Graph Neural Network Framework: PyTorch + PyTorch Geometric Base Language Model: nasa-impact/nasa-smd-ibm-st-v2

This model was trained on 2025-09-10.

---

Model Details

• Hidden Channels: 256
• Number of Layers: 3
• Convolution Type: SAGE
• Max Epochs: 100000

Core Components

• Base Text Encoder: NASA-SMD-IBM Language Model (768-dimensional embeddings)
• Graph Neural Network: Heterogeneous GNN with multiple layers
• Node Types: Dataset, Publication, Instrument, Platform, ScienceKeyword
• Edge Types: Multiple relationship types between nodes

Technical Specifications

• Input Dimensions: 768 (NASA-SMD-IBM embeddings)
• Hidden Dimensions: 256
• Output Dimensions: 768 (aligned with NASA-SMD-IBM space)
• Activation Function: ReLU
• Dropout: Applied between layers

---

Training Details

Training Data

• Source: NASA EOSDIS Knowledge Graph

• Node Types and Counts:

  • Datasets: Earth science datasets from NASA DAACs
  • Publications: Related scientific papers
  • Instruments: Earth observation instruments
  • Platforms: Satellite and other observation platforms
  • Science Keywords: NASA Earth Science taxonomy

Training Process

• Optimization: Adam optimizer

• Loss Function: Contrastive loss for semantic alignment

• Training Strategy:

  • Initial node embedding generation
  • Message passing through graph structure
  • Contrastive learning with NASA-SMD-IBM embeddings

---

Intended Use

Designed for: Research, data discovery, and semantic search in Earth science Not intended for: Safety-critical systems or unrelated domains without fine-tuning

---

Strengths

1. Semantic Understanding: Effective cross-modal alignment between text and graph
2. Domain Specificity: Tuned for NASA Earth Science terminology
3. Multi-Modal Integration: Combines language and graph context

---

Limitations

• Performance depends on graph coverage
• Specialized for the Earth Science domain

---

Usage Guide

Installation

pip install torch torch-geometric transformers huggingface-hub

Basic Usage

from transformers import AutoTokenizer, AutoModel
import torch
from gnn_model import EOSDIS_GNN

tokenizer = AutoTokenizer.from_pretrained("nasa-impact/nasa-smd-ibm-st-v2")
text_model = AutoModel.from_pretrained("nasa-impact/nasa-smd-ibm-st-v2")
gnn_model = EOSDIS_GNN.from_pretrained("nasa-gesdisc/edgraph-gnn-graphsage")

Semantic Search Example

from semantic_search import SemanticSearch
searcher = SemanticSearch()
results = searcher.search("atmospheric carbon dioxide measurements", top_k=5, node_type="Dataset")

---

Evaluation Metrics

Metric	Value	Notes
Top-5 Accuracy	87.4%	Probability that one of top-5 nodes is relevant
MRR	0.73	Ranking quality
Link Prediction ROC-AUC	0.91	Edge-existence prediction
Node Classification F1 (macro)	0.84	Balanced across node types
Triple Classification Accuracy	88.6%	Valid vs. invalid triples

---

Reproducibility

• Hardware: 4 x NVIDIA A100 (80 GB)
• Software: Python 3.11, PyTorch 2.3.1, PyG 2.5.2
• Training Duration: ~24 hours
• Random Seed: 42

---

License and Data Use

This model is released under the Apache 2.0 License. NASA Earth Science data used for training are publicly available via earthdata.nasa.gov.

---

Related Models and Resources

• Base model: nasa-impact/nasa-smd-ibm-st-v2
• Dataset: nasa-gesdisc/nasa-eo-knowledge-graph
• GitHub Repository: github.com/nasa-gesdisc/nasa-eosdis-gnn

---

Citation

Model

@misc{armin_mehrabian_2025,
  author       = { Armin Mehrabian },
  title        = { nasa-eosdis-heterogeneous-gnn (Revision 7e71e62) },
  year         = 2025,
  url          = { https://huggingface.co/arminmehrabian/nasa-eosdis-heterogeneous-gnn },
  doi          = { 10.57967/hf/6789 },
  publisher    = { Hugging Face }
}

Dataset

@misc{nasa_goddard_earth_sciences_data_and_information_services_center__(ges-disc)_2024,
  author       = {{NASA Goddard Earth Sciences Data and Information Services Center (GES-DISC)}},
  title        = { nasa-eo-knowledge-graph },
  year         = 2024,
  url          = { https://huggingface.co/datasets/nasa-gesdisc/nasa-eo-knowledge-graph },
  doi          = { 10.57967/hf/3463 },
  publisher    = { Hugging Face }
}

How to Cite

If you use this model, please cite both entries above and include the model DOI: https://doi.org/10.57967/hf/6789

---

Contact Information

Maintainers:

• Armin Mehrabian (armin.mehrabian@nasa.gov)
• Kendall Gilbert (kendall.c.gilbert@nasa.gov)
• Irina Gerasimov (irina.gerasimov@nasa.gov)

Organization: NASA GES-DISC