Refactor examples to new workflow structure (#15)

* Remove outdated workflows

* Remove outdated workflows

* Update to geospatial task inputs

Author: lukasbindreiter

README.md

@@ -23,13 +23,17 @@ List of examples:
 - [Workflows Hello World, Python](/workflows-hello-world-py/)  
 A simple example that demonstrates how to use the Tilebox SDKs to submit a job and run a worker.
 - [Workflows Cron Automation, Python](/workflows-cron-automation-py/)  
-A simple example that demonstrates how to use the Tilebox SDKs to create a Workflows Cron Automation, a workflow that runs on a schedule, and showcases how to filter timeseries Datasets based on spatial and temporal criteria.
+A release-based async Cron Automation that queries Sentinel-2 statistics for native Shapely areas.
 - [Multi-Language Workflows, Python and Go](/workflows-multilang/)
-An example that demonstrates how to use the Tilebox SDKs to create a Workflows that uses tasks implemented in different languages, Python and Go. Go is used to submit the job, while Python is used to execute it.
+An example that submits native ODC Geo task inputs from Go for execution by an async Python task.
 
 **Mixed**
 
 - [Download S2 Data for Points of Interest, Python](/workflows-download-s2-for-aois/)  
 A more complex example that demonstrates how to use the Tilebox SDKs to create a Workflows to find Sentinel-2 data for a set of points of interest (POIs), filter the data to be as cloud-free as possible and finally download the data.
 - [Sentinel-2 Cloud-free Mosaic](/s2-cloudfree-mosaic/)
-This workflow reads Sentinel-2 data from the Copernicus archive, and writes a cloudfree mosaic to a Zarr datacube.
+This workflow window-reads Sentinel-2 COGs from AWS Earth Search and writes a cloud-free mosaic to Zarr.
+- [Sentinel-2 Clay Change Detection](/s2-clay/)
+This workflow composites two Sentinel-2 periods, runs tiled Clay inference, and writes embedding change distances.
+- [Wyvern Hyperspectral PCA](/wyvern-hyperspectral-pca/)
+This workflow computes distributed PCA with ODC geospatial tiling and a bounded statistics reduction tree.

s2-clay/.env.example

@@ -1,10 +1,4 @@
-RUNNER_NAME=lukas-notebook
-AXIOM_API_KEY=
 TILEBOX_API_KEY=
 AXIOM_TRACES_DATASET=workflow-traces
-AXIOM_LOGS_DATASET=workflow-logs
 OTC_ACCESS_KEY_ID=
 OTC_SECRET_ACCESS_KEY=
-COPERNICUS_ACCESS_KEY_ID=
-COPERNICUS_SECRET_ACCESS_KEY=
-

s2-clay/README.md

@@ -1,2 +1,58 @@
-# Sentinel-2 Change Detection
+# Sentinel-2 Clay Change Detection
 
+This workflow compares two Sentinel-2 time periods with the
+[Clay](https://github.com/Clay-foundation/model) foundation model and writes a patch-level change map to Zarr.
+
+## Design
+
+`DetectChanges` accepts native ODC `Geometry`, `CRS`, and `Resolution` values plus two Tilebox `TimeInterval` values.
+It queries credentials-free Sentinel-2 L2A COGs from AWS Earth Search and divides the exact output `GeoBox` into
+256 × 256 pixel model tiles.
+
+For every tile and period, `ComputeClayTile` reads only the required COG windows, masks clouds with the scene
+classification layer, creates a 25th-percentile composite in memory, and runs Clay. `ComputeChangeTile` then computes
+the cosine distance between corresponding embeddings. No full products, time-series cubes, pickled grids, or
+intermediate mosaics are written.
+
+The Zarr group is written to `change-detection/<job-id>` in the configured `s2-clay` bucket and contains `before` and
+`after` Clay embeddings and the resulting `change` distance array.
+
+## Run the workflow
+
+1. Copy `.env.example` to `.env` and configure the Tilebox and Open Telekom Cloud credentials.
+2. Validate and publish the release:
+
+```bash
+tilebox workflow build-release --debug --json
+tilebox workflow publish-release --json
+tilebox workflow deploy-release --latest --cluster <cluster-slug> --json
+tilebox runner start --cluster <cluster-slug>
+```
+
+The Clay v1.5 checkpoint is downloaded and cached by `huggingface_hub` on each runner when first needed.
+
+## Submit a job
+
+```python
+from datetime import UTC, datetime
+
+from odc.geo import CRS, Geometry, Resolution
+from shapely import box
+from tilebox.datasets.query import TimeInterval
+from tilebox.workflows import Client
+
+from s2_clay import DetectChanges
+
+client = Client()
+client.jobs().submit(
+    "vienna-change-detection",
+    DetectChanges(
+        area=Geometry(box(16.2, 48.1, 16.5, 48.3), "EPSG:4326"),
+        before=TimeInterval(datetime(2024, 6, 1, tzinfo=UTC), datetime(2024, 7, 1, tzinfo=UTC)),
+        after=TimeInterval(datetime(2025, 6, 1, tzinfo=UTC), datetime(2025, 7, 1, tzinfo=UTC)),
+        output_crs=CRS("EPSG:32633"),
+        resolution=Resolution(x=10, y=-10),
+        max_cloud_cover=20,
+    ),
+)
+```