---
license: apache-2.0
task_categories:
  - image-segmentation
  - object-detection
language:
  - en
tags:
  - geospatial
  - remote-sensing
  - olmoearth
  - karst
  - groundwater
  - sentinel-1
  - sentinel-2
  - pennsylvania
  - geology
size_categories:
  - 10K<n<100K
pretty_name: Pennsylvania Karst Features for OlmoEarth Fine-Tuning
configs:
  - config_name: default
    data_files:
      - split: train
        path: data.parquet
---

# OlmoEarth-v1-FT-Karst-Groundwater-Base

Fine-tuning dataset for [OlmoEarth-v1-Base](https://huggingface.co/allenai/OlmoEarth-v1-Base) to detect karst terrain features in Pennsylvania from Sentinel-1 SAR backscatter and Sentinel-2 optical imagery.

**47,020 labeled points · 4 classes · ~50,000 km² · 3 AOI regions**

---

## Quick Start

Two formats are provided for the two upload paths:

### Option A — OlmoEarth Studio Web UI (`studio/import.geojson`)
Upload via Studio's "Import your own dataset" feature. Map the properties to your project's tags during import.

```python
import json
with open("studio/import.geojson") as f:
    data = json.load(f)
```

### Option B — Pipeline / rslearn (`rslearn/annotation_features.geojson`)
For direct use with `olmoearth_run prepare_labeled_windows` or any rslearn-compatible workflow.

```python
import json
with open("rslearn/annotation_features.geojson") as f:
    annotations = json.load(f)
with open("rslearn/annotation_task_features.geojson") as f:
    tasks = json.load(f)
```

---

## Files

| Path | Records | Purpose |
|---|---|---|
| `rslearn/annotation_features.geojson` | 47,020 points | OlmoEarth `oe_*` schema (rslearn / olmoearth_projects pipeline) |
| `rslearn/annotation_task_features.geojson` | 3 polygons | AOI region boundaries with `oe_annotations_task_id` |
| `studio/import.geojson` | 47,020 points | Studio Web UI import format (`task_name`, `karst_status`, etc.) |
| `data.parquet` | 47,020 rows | Flat tabular preview for HF Data Studio |

---

## Schemas

### rslearn / pipeline schema (`rslearn/annotation_features.geojson`)

```json
{
  "type": "Feature",
  "geometry": {"type": "Point", "coordinates": [lon, lat]},
  "properties": {
    "oe_labels": {"category": 0},
    "oe_start_time": "2020-01-01 00:00:00+00:00",
    "oe_end_time": "2025-12-31 00:00:00+00:00",
    "oe_annotations_task_id": "uuid",
    "confidence": 0.83,
    "quality_flags": ["on_carbonate", "dense_cluster", "source_pageode"]
  }
}
```

### Studio import schema (`studio/import.geojson`)

```json
{
  "type": "Feature",
  "geometry": {"type": "Point", "coordinates": [lon, lat]},
  "properties": {
    "task_name": "pa_karst_central",
    "observation_time": "2020-01-01T00:00:00Z",
    "sample_category": "sinkhole",
    "sample_number": 0.83,
    "sample_true_false": false
  }
}
```

Field mapping:
- `sample_category` → karst class (training field, enum: `surface_depression`, `sinkhole`, `surface_mine`, `other`)
- `sample_number` → confidence score (0-1)
- `sample_true_false` → auto-generated flag (true for "other" class samples)

These names are required for Studio's auto-detection of training fields.

---

## Classes

| ID | Tag | Count | % |
|---|---|---|---|
| 0 | surface_depression | 16,295 | 34.66% |
| 1 | sinkhole | 3,810 | 8.10% |
| 2 | surface_mine | 1,916 | 4.07% |
| 3 | other | 24,999 | 53.17% |

`other` represents non-karst terrain inside the AOI (sampled from non-carbonate bedrock). 191 cave records are included in the sinkhole class and tagged with `merged_from_cave: true`.

---

## AOI Regions

| Region | Longitude | Area | Annotations |
|---|---|---|---|
| `pa_karst_east` | -76.2 to -75.0 | ~12,000 km² | 16,009 |
| `pa_karst_central` | -77.4 to -76.2 | ~16,000 km² | 14,841 |
| `pa_karst_west` | -78.7 to -77.4 | ~16,000 km² | 16,170 |

Covers the full Pennsylvania karst belt across 3 task regions, each within OlmoEarth Studio's 20,000 km² limit.

---

## Per-Label Confidence

Every annotation carries a `confidence` score (0-1) and `quality_flags` list.

**Signals:**

| Signal | Weight | Description |
|---|---|---|
| Bedrock alignment | 0.5 | Karst classes on carbonate; "other" off carbonate |
| Local class density | 0.3 | Same-class neighbors within 500m |
| Source provenance | 0.2 | PaGEODE (0.2) vs auto-generated (0.1) |

**Distribution:**

| Class | Mean | High-confidence (≥0.8) |
|---|---|---|
| surface_depression | 0.83 | 73.9% |
| sinkhole | 0.77 | 28.3% |
| surface_mine | 0.76 | 40.3% |
| other | 0.69 | 26.3% |

**Quality flag values:** `on_carbonate`, `off_carbonate_karst`, `dense_cluster`, `isolated`, `auto_generated`, `source_pageode`, `merged_from_cave`.

**Recommended uses:**
- Filter: train only on `confidence >= 0.8`
- Weight: use `confidence` as per-sample loss weight
- Triage: prioritize manual review of low-confidence labels

---

## Studio Configuration

| Setting | Value |
|---|---|
| Data sources | Sentinel-1 (VV/VH) + Sentinel-2 (12 bands) |
| Time range | 2020-01-01 to 2025-12-31 |
| Grid size | 2 km |
| Window resolution | 10 m |
| Window buffer | 31 px (~310 m) |
| Train / val split | 75% / 25% spatial split (128px grid) |

---

## Caveats

- **"Other" class is auto-generated** from non-carbonate bedrock polygons; pending domain review.
- **Cave merged into sinkhole** — 191 records traceable via `merged_from_cave: true`.
- **2,768 outside-carbonate karst labels retained** pending review (some may be valid non-carbonate features).
- **Class imbalance** — use class weighting or oversampling for sinkhole (8%) and surface_mine (4%) during training.

---

## Sources

All data is public domain.

- **PaGEODE Karst Features** — [PA DCNR](https://gis.dcnr.state.pa.us/pageode/) (last updated 2026-02-24)
- **PA Bedrock Geology** — [PA DCNR](https://gis.dcnr.state.pa.us/) (DCNR_Bedrock_Geology_Pa_Geologic_Units52023)
- **Satellite imagery** — Sentinel-1/2 via OlmoEarth Studio (Microsoft Planetary Computer)

## Related

- [allenai/OlmoEarth-v1-Base](https://huggingface.co/allenai/OlmoEarth-v1-Base)
- [olmoearth_projects](https://github.com/allenai/olmoearth_projects)
- [rslearn](https://github.com/allenai/rslearn)
- [OlmoEarth Studio docs](https://olmoearth.allenai.org/docs)

## Citation

```bibtex
@dataset{qi2026karst,
  author    = {Qi, Ziming and BAI Group},
  title     = {OlmoEarth-v1-FT-Karst-Groundwater-Base: Pennsylvania Karst Features for OlmoEarth Fine-Tuning},
  year      = {2026},
  publisher = {Hugging Face},
  url       = {https://huggingface.co/datasets/2imi9/OlmoEarth-v1-FT-Karst-Groundwater-Base}
}
```

## License

Apache 2.0. PaGEODE and PA Bedrock source data are public domain.


---

## v8 Update (2026-05-23)

Two minimal additions to support per-pixel binary karst regression in OlmoEarth Studio:

### `sample_karst_score` (new property in `studio/import.geojson`)
Binary regression target — `0.0` for `other`, `1.0` for `sinkhole` / `surface_depression` / `surface_mine`. Lets Studio's per-pixel regression head learn karst vs non-karst directly, instead of regressing to the mean of `sample_number` (confidence).

### `oe_labels.karst_score` (new key in `rslearn/annotation_features.geojson`)
Same binary value, in rslearn schema #3. Set `label_property: "karst_score"` in your `olmoearth_run.yaml` for binary regression.

### `task_name` (new column in `data.parquet`)
Explicit AOI assignment (`pa_karst_east` / `pa_karst_central` / `pa_karst_west`) by longitude band, previously implicit.

### `by_aoi/` (new subfolder)
Smaller per-AOI and balanced-combined variants for users who hit Studio's 10,000-record upload cap. Same internal layout as root. See `by_aoi/README.md`.

All original fields (`sample_category`, `sample_number`, `sample_true_false`, `oe_labels.category`, `confidence`, `quality_flags`) are unchanged. Original row count (47,020) is preserved at the root.