GlacioVision Dataset

This dataset presents a comprehensive, multi-modal collection of glacier-related data for three glaciers of Pakistan, which include Siachen, Baltoro, and Chiantar, over the period 2019-2024. It is designed to support research in glaciology, remote sensing, climate science, and machine learning, particularly for tasks such as glacier segmentation, elevation modeling, and multi-modal data fusion.

The dataset integrates three primary data sources:

Satellite imagery from Google Earth Engine (Sentinel-1 and Sentinel-2)

Climate reanalysis data from Copernicus Atmosphere Data Store (CAMS EAC4)

Elevation measurements from NASA Earthdata (GEDI L2A)

These data sources are processed into spatially aligned formats and organized into three complementary dataset representations: raw data, patch-based data, and inference-ready data. Additionally, the full data processing pipeline is provided to ensure reproducibility.

Dataset Contents

The dataset is organized into the following main components:

1. Patch-Based Dataset (Primary)

A machine learning–ready dataset consisting of spatial patches extracted from multi-modal raster data.

Patch size: 256 × 256 pixels

Overlap: 50%

Inputs:

Multi-band satellite + Elevation imagery (10 bands)

Climate data (year-wise)

Outputs:

Glacier segmentation masks

Elevation patches

This format is optimized for deep learning applications.

2. Raw Dataset

Full-resolution, spatially continuous data before patch extraction.

Multi-band satellite + Elevation raster

Glacier masks

Elevation rasters

Climate CSV files

This representation preserves original spatial structure and is suitable for geospatial analysis.

3. Inference Dataset

Structured dataset for model evaluation and real-world deployment scenarios.

4. Processing Pipeline

All scripts used for:

Climate data extraction and preprocessing

Satellite data acquisition and compositing

Elevation extraction and interpolation

Raster stacking and alignment

Patch generation

Data Characteristics

Temporal Coverage

2019-2024

Spatial Coverage

Three glacier regions (Siachen, Baltoro, Chiantar)

Spatial Resolution

25 meters (satellite and stacked data)

Coordinate Reference System

EPSG:32643

Data Sources and Methodology

Climate Data

Climate variables are derived from CAMS global reanalysis (EAC4). Data is extracted for each glacier region using bounding box coordinates and processed into yearly inputs using a hydrological year window:

October (previous year) to September (current year)

Variables include temperature, pressure, humidity, snow properties, and atmospheric parameters.

Satellite Data

Satellite imagery is obtained via Google Earth Engine:

Sentinel-2: Optical bands (B2, B3, B4, B8, B11, B12) with NDSI

Sentinel-1: SAR bands (VV, VH)

Data is filtered for cloud cover (<30%) and aggregated using median compositing over June–September.

Elevation Data

Elevation data is derived from GEDI L2A measurements:

Extracted from HDF5 files

Filtered using quality and degradation flags

Interpolated into continuous Elevation maps using IDW (Inverse Distance Weighting)

Data Integration

All data sources are:

Reprojected to a common CRS (EPSG:32643)

Resampled to consistent resolution

Spatially aligned and stacked into multi-band rasters

Final stacked raster includes:

Sentinel-2 bands

NDSI

Sentinel-1 bands

Elevation band

Use Cases

This dataset supports:

Glacier segmentation (semantic segmentation)

Elevation prediction and reconstruction

Multi-modal deep learning

Climate-glacier interaction studies

Temporal analysis of glacier dynamics

File Formats

GeoTIFF (.tif): Satellite imagery, Elevation maps, masks

CSV (.csv): Climate data, extracted elevation points

Python scripts (.py): Processing pipeline

Keywords

Glacier Monitoring, Remote Sensing, Sentinel-1, Sentinel-2, Elevation maps, Climate Data, Multi-Modal Dataset, Deep Learning, Geospatial Data, Time Series

Notes

Climate data is shared per year and reused across patches

Elevation data is interpolated and may include smoothing artifacts

Satellite data is limited to summer months (June-September)

Dataset includes overlapping patches for improved model training

Acknowledgments

We acknowledge:

Copernicus Atmosphere Data Store (CAMS)

Google Earth Engine

NASA GEDI mission

Open-source geospatial libraries (Rasterio, Pandas, QGIS)