Upload README.md with huggingface_hub

README.md

@@ -328,6 +328,64 @@ ea = ds["tp"].sel(step=24, latitude=slice(25, -14),
 | 2026 | Jan--Feb | 51 | ~7,344 | ~0.8 GB |
 | **Total** | | **720** | **144,228** | **~17.8 GB** |
 
+## Catalog / Index
+
+A lightweight **catalog.parquet** (~1.8 MB) at the repo root indexes all 150,246 parquet files
+across the dataset. Use it to discover available dates, runs, and members without listing the
+full repo tree.
+
+**Download**: [`catalog.parquet`](catalog.parquet)
+
+| Column | Example | Description |
+|--------|---------|-------------|
+| `year` | `2024` | Forecast year |
+| `month` | `03` | Forecast month |
+| `date` | `20240301` | Forecast date (YYYYMMDD) |
+| `run` | `00z` | Run hour |
+| `member` | `control` | Ensemble member name |
+| `filename` | `2024030100z-control.parquet` | Parquet filename |
+| `hf_path` | `run_par_ecmwf/2024/03/20240301/00z/...` | Full path in this repo |
+| `size_bytes` | `107520` | File size in bytes |
+
+### Quick Start with the Catalog
+
+```python
+import pandas as pd
+from huggingface_hub import hf_hub_download
+
+# 1. Load the catalog (~1.8 MB, indexes all 150k+ files)
+catalog_path = hf_hub_download(
+    repo_id="E4DRR/gik-ecmwf-par", repo_type="dataset",
+    filename="catalog.parquet"
+)
+catalog = pd.read_parquet(catalog_path)
+
+# 2. Explore what's available
+print(catalog.groupby(["year", "month"]).size())       # files per month
+print(catalog["run"].unique())                          # ['00z','06z','12z','18z']
+print(catalog["member"].nunique())                      # 51
+
+# 3. Filter for a specific date + run
+subset = catalog[(catalog["date"] == "20250101") & (catalog["run"] == "00z")]
+print(subset[["member", "filename", "size_bytes"]])
+
+# 4. Download a specific parquet using its hf_path
+row = subset.iloc[0]
+parquet_path = hf_hub_download(
+    repo_id="E4DRR/gik-ecmwf-par", repo_type="dataset",
+    filename=row["hf_path"]
+)
+```
+
+### Coverage Summary (from catalog)
+
+| Year | Months | Dates | Files | Total Size |
+|------|--------|-------|-------|------------|
+| 2024 | Mar--Dec | 306 | ~62,424 | ~6.5 GB |
+| 2025 | Jan--Dec | 365 | ~75,504 | ~7.8 GB |
+| 2026 | Jan--Mar 7 | 66 | ~12,318 | ~1.3 GB |
+| **Total** | | **737** | **150,246** | **~18.5 GB** |
+
 ## How It Works
 
 The **Grib-Index-Kerchunk (GIK)** method applies the same principle as video streaming to weather data: