EiffL/GowerStreetDESY3

---
license: mit
task_categories:
- other
tags:
- weak-lensing
- cosmology
- simulation
- convergence-maps
- des-y3
- lsst
- shape-noise
size_categories:
- 100K<n<1M
---

# Gower Street DES Y3 Lensing Tiles

Weak lensing convergence map tiles extracted from the [Gower Street](http://star.ucl.ac.uk/GowerStreetSims/) N-body simulation suite, processed through a Born-approximation raytracing pipeline with DES Y3 MagLim source n(z) distributions.

## Dataset Description

Each sample contains a (4, H, W) convergence map tile covering ~3400 deg², corresponding to 4 DES Y3 MagLim tomographic bins. Tiles are extracted from equatorial HEALPix base faces after harmonic-space filtering and rotation for data augmentation.

The dataset includes **15 configurations** (5 angular scale cuts x 3 noise levels), with ~9400 tiles per configuration from 787 simulations (12 tiles per sim: 3 rotations x 4 equatorial tiles).

### Configurations

Each configuration is identified by `lmax_{lmax}_{noise_level}` and stored in a separate data directory:

| lmax | Tile size | Angular scales | nside | Noise levels |
|------|-----------|---------------|-------|--------------|
| 200 | 128x128 | > 0.9 deg | 128 | noiseless, des_y3, lsst_y10 |
| 400 | 256x256 | > 0.45 deg | 256 | noiseless, des_y3, lsst_y10 |
| 600 | 256x256 | > 0.3 deg | 256 | noiseless, des_y3, lsst_y10 |
| 800 | 512x512 | > 0.23 deg | 512 | noiseless, des_y3, lsst_y10 |
| 1000 | 512x512 | > 0.18 deg | 512 | noiseless, des_y3, lsst_y10 |

### Noise levels

Shape noise arises from the intrinsic ellipticity dispersion of source galaxies. For a HEALPix pixel at resolution nside, the noise standard deviation per pixel per tomographic bin is:

```
sigma_pix = sigma_e / sqrt(2 * n_eff * A_pix)
```

where `sigma_e` is the per-component intrinsic ellipticity dispersion, `n_eff` is the effective galaxy number density (in sr⁻¹), and `A_pix = 4pi / N_pix` is the pixel solid angle. The factor of 2 accounts for two ellipticity components. Noise is Gaussian and independent per pixel.

Shape noise is added to the full-sky nside=1024 convergence map **before** harmonic filtering, so the noise is band-limited consistently with the signal. For a given (sim_id, noise_level), the same noise realization is shared across all lmax cuts and orientations. RNG seed: `sim_id * 1000 + noise_level_index`.

#### `noiseless`

No shape noise added. Pure signal from the Born-approximation raytracing.

#### `des_y3` — DES Year 3 ([Amon et al. 2022](https://arxiv.org/abs/2105.13543), Table 1)

Per-bin effective number density and intrinsic ellipticity dispersion from the DES Y3 MagLim sample:

| Bin | n_eff (arcmin⁻²) | sigma_e |
|-----|-------------------|---------|
| 0 | 1.476 | 0.243 |
| 1 | 1.479 | 0.262 |
| 2 | 1.484 | 0.259 |
| 3 | 1.461 | 0.301 |

#### `lsst_y10` — LSST Year 10 ([DESC SRD](https://arxiv.org/abs/1809.01669))

| Bin | n_eff (arcmin⁻²) | sigma_e |
|-----|-------------------|---------|
| 0-3 | 6.75 | 0.26 |

Total n_eff = 27 arcmin⁻² split uniformly across 4 bins to match the DES tomographic structure.

### Loading

```python
from datasets import load_dataset

# Load a specific (lmax, noise_level) configuration
ds = load_dataset("EiffL/GowerStreetDESY3", data_dir="data/lmax_600_des_y3")

sample = ds["train"][0]
kappa = sample["kappa"] # (4, 256, 256) convergence map
omega_m = sample["Omega_m"] # Matter density parameter
noise = sample["noise_level"] # "des_y3"

# Load noiseless version at same angular scale
ds_clean = load_dataset("EiffL/GowerStreetDESY3", data_dir="data/lmax_600_noiseless")

# Load LSST-depth version
ds_lsst = load_dataset("EiffL/GowerStreetDESY3", data_dir="data/lmax_600_lsst_y10")
```

### Fields

| Field | Type | Description |
|-------|------|-------------|
| `kappa` | array (4, H, W) float32 | Convergence map tiles, 4 tomographic bins |
| `sim_id` | int | Gower Street simulation ID (1-791) |
| `orientation_id` | int | Rotation orientation (0-2) |
| `tile_id` | int | Equatorial tile index (0-3) |
| `noise_level` | string | Noise level: "noiseless", "des_y3", or "lsst_y10" |
| `Omega_m` | float | Matter density parameter |
| `sigma_8` | float | RMS density fluctuation amplitude |
| `S8` | float | S8 = sigma_8 * sqrt(Omega_m / 0.3) |
| `w` | float | Dark energy equation of state |
| `h` | float | Hubble parameter H0/100 |
| `n_s` | float | Scalar spectral index |
| `Omega_b` | float | Baryon density parameter |
| `m_nu` | float | Sum of neutrino masses (eV) |

## Pipeline

1. **N-body simulations**: Gower Street suite (791 simulations with varying cosmological parameters)
2. **Raytracing**: Born-approximation lensing through particle lightcone shells (nside=2048 input, nside=1024 output), weighted by DES Y3 MagLim n(z) distributions (4 tomographic bins)
3. **Shape noise injection**: Gaussian noise added per pixel at nside=1024, calibrated to DES Y3 or LSST Y10 survey depth
4. **Harmonic filtering**: `map2alm(lmax)` -> `rotate_alm(euler)` -> `alm2map(nside_down)` ensures all tiles see identical harmonic-space processing
5. **Tile extraction**: 3 fixed rotations x 4 equatorial HEALPix base tiles = 12 tiles per simulation per configuration

### Rotations

Three orientations of the sphere provide data augmentation while keeping tiles in equatorial positions (minimal projection distortion):
- Orientation 0: identity (Euler angles 0, 0, 0)
- Orientation 1: 90 deg about y-axis (0, 90, 0)
- Orientation 2: 90 deg about z-axis (90, 0, 0)

## Source

- Simulations: [Gower Street](http://star.ucl.ac.uk/GowerStreetSims/) (Jeffrey et al.)
- n(z): [DES Y3 MagLim](https://desdr-server.ncsa.illinois.edu/despublic/y3a2_files/datavectors/)
- Pipeline: [LensingFoM](https://github.com/EiffL/LensingFoM)

## Citation

If you use this dataset, please cite the Gower Street simulations paper and DES Y3 data release.