Idea-re/fermi-lat-synthetic-daily-sky-maps

---
license: mit
pretty_name: SPECTRA Fermi-LAT Simulated Sky Maps
task_categories:
- time-series-forecasting
- other
language:
- en
tags:
- astronomy
- gamma-ray
- fermi-lat
- simulation
- convlstm
- anomaly-detection
- time-series
---

# SPECTRA Fermi-LAT Simulated Sky Maps

This repository contains the simulated dataset associated with the manuscript:

**“Self-Supervised ConvLSTM for Fermi Large Area Telescope Transient Detection”**

The associated manuscript is currently **under review at Astronomy and Computing**.

## Overview

This dataset was generated to support the development and validation of self-supervised, spatio-temporal anomaly-detection methods for gamma-ray transient searches in Fermi-LAT-like observations.

Starting from long-duration `gtobssim` simulations within the Fermitools ecosystem, the pipeline produces daily full-sky maps of:

- **counts**
- **exposure**

The primary dataset released here is the original **energy-resolved tensor** with shape:

`(3649, 2, 10, 360, 180)`

where:

- `3649` = valid daily frames over approximately 10 years
- `2` = channels (`counts`, `exposure`)
- `10` = logarithmically spaced energy bins spanning 100 MeV–500 GeV
- `360 x 180` = all-sky spatial grid in celestial coordinates
- spatial resolution = **1 degree/pixel**

One daily frame was discarded during preprocessing and quality control; therefore, the public release contains **3649** valid daily observations.

An energy-integrated representation can be derived by summing over the energy axis, yielding:

`(3649, 2, 360, 180)`

This reduced representation is the one used in the main ConvLSTM experiments discussed in the manuscript, whereas the full tensor released here preserves the complete spectral structure of the simulation.

## Scientific context

The dataset is designed as a controlled testbed for:

- self-supervised next-frame prediction
- residual-based anomaly detection
- transient-search benchmarking
- future extensions to energy-resolved spatio-temporal modeling

It reproduces the structure of daily Fermi-LAT-like observations while keeping the simulation environment fully controlled.

## Data description

### Tensor axes

The main tensor follows the convention:

`(time, channel, energy, x, y)`

with:

- `time = 3649` daily frames
- `channel = 2` → `counts`, `exposure`
- `energy = 10` logarithmic bins
- `x = 360`
- `y = 180`

### Channels

- `channel 0`: counts map
- `channel 1`: exposure map

### Energy range

- 100 MeV to 500 GeV
- 10 logarithmically spaced bins

### Spatial grid

- celestial coordinates
- Cartesian (CAR) projection
- `360 x 180` pixels
- `1 degree/pixel`

### Time coverage

- 3649 valid daily maps
- approximately 10 years of simulated observations

## Repository contents

The repository structure is:

- `README.md`
- `metadata.json`
- `data/`
- `spectra_tensor_part_00.npy`
- `spectra_tensor_part_01.npy`
- `spectra_tensor_part_02.npy`
- `spectra_tensor_part_03.npy`
- `load_example.py`

## Intended use

This dataset is intended for research purposes, including:

- anomaly detection in gamma-ray astronomy
- self-supervised spatio-temporal modeling
- benchmarking of ConvLSTM and related architectures
- methodological studies on synthetic Fermi-LAT-like sky maps

## Citation

If you use this dataset, please cite the associated manuscript.

Suggested citation text:

**Garinei et al., “Self-Supervised ConvLSTM for Fermi Large Area Telescope Transient Detection”, under review at Astronomy and Computing.**

## Important notes

- This dataset is **simulated**, not a redistribution of official Fermi-LAT survey data products.
- The exposure channel is included because exposure variations are an essential part of the predictive and anomaly-detection problem addressed in the manuscript.
- The repository is intended to improve reproducibility and public accessibility of the simulation products described in the paper.

## Contact

For questions regarding the dataset, simulation pipeline, or manuscript status, please contact the corresponding author listed in the associated paper.