metadata

language: en
license: mit
tags:
  - flood-segmentation
  - remote-sensing
  - earth-observation
  - dem
  - computer-vision
  - prithvi
  - foundation-model
datasets:
  - custom-flood-dataset
metrics:
  - iou
  - dice
  - f1
library_name: terratorch
pipeline_tag: image-segmentation

Prithvi-Tiny for Flood Risk Segmentation

Model Description

This model is a Prithvi-Tiny fine-tuned for flood risk segmentation from Digital Elevation Models (DEM) and precipitation data.

Prithvi-Tiny foundation model fine-tuned for flood segmentation

Model Details

Architecture: Prithvi-Tiny
Training Epochs: 18
Model Size: 128MB
Task: Semantic Segmentation (Flood Risk Prediction)
Input: DEM + Precipitation data
Output: Flood depth categories

Usage

import torch
from terratorch.tasks import SemanticSegmentationTask

# Load the model
model = SemanticSegmentationTask.load_from_checkpoint("path/to/checkpoint.ckpt")
model.eval()

# Your inference code here

Training Data

The model was trained on flood simulation data from multiple US counties, including:

DEM data from USGS National Map
Precipitation data from NOAA Atlas 14
Simulated flood depth labels

Performance

The model achieves state-of-the-art performance on flood segmentation tasks across diverse geographical regions.

Limitations

Model is trained primarily on US geographical data
Performance may vary on international datasets
Requires specific input preprocessing

Citation

If you use this model in your research, please cite:

@misc{flood-foundation-model-prithvi-tiny,
  title={Flood Risk Foundation Model: Prithvi-Tiny},
  author={FloodRisk-DL Team},
  year={2024},
  url={https://huggingface.co/chrimerss/flood-foundation-prithvi-tiny}
}