Colombia damage assessment dataset and processing scripts

datasets/predicted-damage-colombia-2026/README.md

@@ -0,0 +1,108 @@
+# Colombia Damage 2026
+
+## First-time publishing
+
+Validate the STAC collection first and fix any validation errors:
+
+```bash
+pctasks dataset validate-collection datasets/predicted-damage-colombia-2026/collection/template.json
+```
+
+Then submit the collection ingestion:
+
+```bash
+pctasks dataset ingest-collection -d datasets/predicted-damage-colombia-2026/dataset.yaml -s -a registry pccomponents
+```
+
+Take the workflow ID from the output and watch it. It must succeed.
+
+```bash
+pctasks runs status $WORKFLOW_ID --watch
+curl "https://planetarycomputer.microsoft.com/api/stac/v1/collections/predicted-damage-colombia-2026"
+```
+
+## Updating
+
+Simply add `-u` to the command.
+
+```bash
+pctasks dataset ingest-collection -d datasets/predicted-damage-colombia-2026/dataset.yaml -u -s -a registry pccomponents
+```
+
+## Item ingestion
+
+Assets live under `blob://ai4edataeuwest/ai4good/colombia2026/` and are
+organized as `<area>/<YYYY-MM-DD>/`. Chunking lists the `*model-predictions.tif`
+files, so each folder produces exactly one STAC item. The sibling GeoPackages
+and valid-area mask in that folder become its assets.
+
+Print the workflow without submitting anything:
+
+```bash
+pctasks dataset process-items -d datasets/predicted-damage-colombia-2026/dataset.yaml \
+  initial-ingest -a registry pccomponents.azurecr.io
+```
+
+Try a single item first:
+
+```bash
+pctasks dataset process-items -d datasets/predicted-damage-colombia-2026/dataset.yaml \
+  test-ingest -a registry pccomponents.azurecr.io --limit 1 --submit
+```
+
+Then ingest everything:
+
+```bash
+pctasks dataset process-items -d datasets/predicted-damage-colombia-2026/dataset.yaml \
+  initial-ingest -a registry pccomponents.azurecr.io --upsert --submit
+```
+
+`initial-ingest` and `test-ingest` are chunkset IDs. Use a fresh one whenever
+you want to re-list the assets, or pass `-e` to reuse an existing chunkset.
+
+Watch the run and read logs with:
+
+```bash
+pctasks runs status $RUN_ID --watch
+pctasks runs get run-log $RUN_ID
+pctasks runs get task-log $RUN_ID create-splits create-splits -p 0
+```
+
+Verify the items landed:
+
+```bash
+curl "https://planetarycomputer.microsoft.com/api/stac/v1/collections/predicted-damage-colombia-2026/items?limit=10"
+```
+
+## Adding a new area
+
+Upload four files to `colombia2026/<area>/<YYYY-MM-DD>/`:
+
+| File | Becomes |
+| --- | --- |
+| `*model-predictions.tif` | `visual` |
+| `*valid_area_mask.geojson` | `valid-area-mask` |
+| `*overture*.gpkg` | `overture-buildings` |
+| `*google*.gpkg` or `*hdx*.gpkg` | `google-buildings` |
+
+The folder name sets the item ID, so `pereira/2026-08-12` produces
+`pereira-2026-08-12`. Use lowercase, hyphenated area names. Nothing is read
+from the file names except the four patterns above.
+
+Then widen `extent.spatial.bbox` in `collection/template.json` if the new area
+falls outside it, re-run the collection ingestion with `-u`, and run
+`process-items` with a new chunkset ID.
+
+## FAQs
+
+> The workflow failed with "Expected exactly one file matching ...". Why?
+
+Every item needs all four files in its folder, and each pattern must match
+exactly one file. Check for a missing upload or two files matching the same
+pattern.
+
+> Can I reuse a chunkset ID?
+
+Only with `-e`, which skips re-listing the assets. Newly uploaded areas will
+not be picked up that way.
+

datasets/predicted-damage-colombia-2026/collection/config.json

@@ -0,0 +1,65 @@
+{
+    "render_config": {
+        "render_params": {
+            "tile_format": "png",
+            "colormap": "{\"2\":[0,255,0,255],\"3\":[255,0,0,255],\"4\":[128,128,128,255]}",
+            "nodata": 0
+        },
+        "minzoom": 11,
+        "assets": [
+            "visual"
+        ],
+        "maxzoom": 18,
+        "create_links": true,
+        "has_mosaic": false,
+        "mosaic_preview_zoom": 13,
+        "mosaic_preview_coords": [
+            3.4803,
+            -76.5367
+        ],
+        "requires_token": true,
+        "max_items_per_tile": null,
+        "vector_tilesets": null,
+        "hidden": false
+    },
+    "mosaic_info": {
+        "mosaics": [
+            {
+                "name": "Most recent available",
+                "description": null,
+                "cql": []
+            }
+        ],
+        "render_options": [
+            {
+                "name": "Predicted building damage",
+                "description": "Per-building damage predictions from the Microsoft AI for Good Lab damage assessment model.",
+                "type": "raster-tile",
+                "options": "assets=visual&colormap=%7B%222%22%3A%5B0%2C255%2C0%2C255%5D%2C%223%22%3A%5B255%2C0%2C0%2C255%5D%2C%224%22%3A%5B128%2C128%2C128%2C255%5D%7D&nodata=0",
+                "vector_options": null,
+                "min_zoom": 11,
+                "legend": {
+                    "type": "classmap",
+                    "labels": [
+                        "Building (no damage)",
+                        "Damaged",
+                        "Unknown"
+                    ],
+                    "trim_start": null,
+                    "trim_end": null,
+                    "scale_factor": null
+                },
+                "conditions": null
+            }
+        ],
+        "default_location": {
+            "zoom": 13,
+            "coordinates": [
+                3.4803,
+                -76.5367
+            ]
+        },
+        "default_custom_query": null,
+        "animation_hint": null
+    }
+}

datasets/predicted-damage-colombia-2026/collection/description.md

@@ -0,0 +1,29 @@
+Creamos evaluaciones de daños a nivel de edificio tras el terremoto en Colombia
+mediante el entrenamiento y la posterior ejecución de un modelo de inteligencia
+artificial sobre imágenes satelitales adquiridas después del desastre. El modelo
+de IA clasifica cada edificio identificado en las imágenes como "sin daños",
+"afectado" o "desconocido". La categoría "desconocido" se utiliza cuando el
+edificio no puede ser evaluado adecuadamente, por ejemplo, debido a la presencia
+de nubes. Utilizamos los polígonos de edificios de Overture Maps, que representan
+el estado de las edificaciones sobre el terreno antes del evento. Los resultados
+se distribuyen como un archivo vectorial en formato GeoPackage, con los
+siguientes atributos para cada edificio:
+
+- `id` – identificador único de Overture Maps y Google para cada edificio.
+- `damaged` – valor 1 si el edificio está dañado; de lo contrario, 0.
+- `unknown` – valor 1 si el edificio está cubierto por nubes, neblina, humo o
+  si, por alguna otra razón, no fue posible clasificarlo; de lo contrario, 0.
+- `area` – área del edificio en metros cuadrados.
+
+We create building level damage assessments by training and then running an AI
+model on the post-disaster imagery. The AI model predicts whether each footprint
+in the imagery is "building", "damaged", or "unknown" (i.e. cloudy). We use
+Overture Maps and Google building footprints which represent the state on the
+ground pre-event and distribute the resulting data as a vector file GeoPackage
+with the following per-footprint attributes:
+
+- `id` – the Overture Maps unique ID for each footprint.
+- `damaged` – 1 if the building is damaged, else 0.
+- `unknown` – 1 if the building was covered by clouds/haze/smoke or otherwise
+  unable to be classified, else 0.
+- `area` – area of the building in sq meters.

datasets/predicted-damage-colombia-2026/collection/template.json

@@ -0,0 +1,184 @@
+{
+    "type": "Collection",
+    "id": "predicted-damage-colombia-2026",
+    "stac_version": "1.0.0",
+    "description": "{{ collection.description }}",
+    "links": [
+        {
+            "rel": "cite-as",
+            "href": "https://www.microsoft.com/en-us/research/group/ai-for-good-research-lab/",
+            "type": "text/html",
+            "title": "AI For Good Lab Harnessing AI to help solve some of the world’s greatest challenges"
+        },
+        {
+            "rel": "license",
+            "href": "https://creativecommons.org/licenses/by/4.0/",
+            "type": "text/html",
+            "title": "Creative Commons Attribution 4.0 International"
+        },
+        {
+            "rel": "describedby",
+            "href": "https://planetarycomputer.microsoft.com/dataset/predicted-damage-colombia-2026",
+            "type": "text/html",
+            "title": "Human readable dataset overview and reference"
+        }
+    ],
+    "stac_extensions": [
+        "https://stac-extensions.github.io/table/v1.2.0/schema.json",
+        "https://stac-extensions.github.io/storage/v2.0.0/schema.json"
+    ],
+    "item_assets": {
+        "visual": {
+            "type": "image/tiff; application=geotiff; profile=cloud-optimized",
+            "roles": [
+                "data",
+                "visual"
+            ],
+            "title": "Model prediction imagery",
+            "description": "Cloud-optimized GeoTIFF of the post-event imagery that the damage model was run on."
+        },
+        "overture-buildings": {
+            "type": "application/geopackage+sqlite3",
+            "roles": [
+                "data"
+            ],
+            "title": "Overture Maps building damage footprints",
+            "description": "GeoPackage of Overture Maps building footprints with predicted earthquake damage.",
+            "table:storage_options": {
+                "account_name": "ai4edataeuwest"
+            }
+        },
+        "google-buildings": {
+            "type": "application/geopackage+sqlite3",
+            "roles": [
+                "data"
+            ],
+            "title": "Google building damage footprints",
+            "description": "GeoPackage of Google building footprints with predicted earthquake damage.",
+            "table:storage_options": {
+                "account_name": "ai4edataeuwest"
+            }
+        },
+        "valid-area-mask": {
+            "type": "application/geo+json",
+            "roles": [
+                "metadata"
+            ],
+            "title": "Valid area mask",
+            "description": "GeoJSON mask delineating the area covered by the damage assessment."
+        }
+    },
+    "msft:region": "westeurope",
+    "table:columns": [
+        {
+            "name": "geometry",
+            "type": "byte_array",
+            "description": "Building footprint polygons"
+        },
+        {
+            "name": "id",
+            "type": "string",
+            "description": "Overture Maps unique ID for each footprint"
+        },
+        {
+            "name": "damaged",
+            "type": "int64",
+            "description": "1 if the building is damaged, else 0"
+        },
+        {
+            "name": "unknown",
+            "type": "int64",
+            "description": "1 if the building was covered by clouds/haze/smoke or otherwise unable to be classified, else 0"
+        },
+        {
+            "name": "area",
+            "type": "double",
+            "description": "Area of the building in square meters",
+            "unit": "m2"
+        }
+    ],
+    "msft:container": "ai4good",
+    "msft:storage_account": "ai4edataeuwest",
+    "msft:short_description": "AI-predicted building damage footprints for areas of Colombia affected by the August 2026 earthquake.",
+    "title": "Predicted Building Damage: Colombia Earthquake 2026",
+    "extent": {
+        "spatial": {
+            "bbox": [
+                [
+                    -76.6146,
+                    3.4127,
+                    -75.6922,
+                    4.8433
+                ],
+                [
+                    -76.6146,
+                    3.4127,
+                    -76.4588,
+                    3.5479
+                ],
+                [
+                    -75.7856,
+                    4.7758,
+                    -75.6922,
+                    4.8433
+                ]
+            ]
+        },
+        "temporal": {
+            "interval": [
+                [
+                    "2026-08-08T00:00:00Z",
+                    null
+                ]
+            ]
+        }
+    },
+    "license": "CC-BY-4.0",
+    "keywords": [
+        "Colombia",
+        "Cali",
+        "Pereira",
+        "Earthquake",
+        "Building damage",
+        "Damage assessment",
+        "Buildings",
+        "Microsoft AI for Good Lab"
+    ],
+    "providers": [
+        {
+            "name": "Microsoft AI for Good Lab",
+            "roles": [
+                "producer",
+                "licensor"
+            ],
+            "url": "https://www.microsoft.com/en-us/research/group/ai-for-good-research-lab/"
+        },
+        {
+            "name": "Microsoft",
+            "roles": [
+                "host",
+                "processor"
+            ],
+            "url": "https://planetarycomputer.microsoft.com"
+        }
+    ],
+    "assets": {
+        "thumbnail": {
+            "href": "https://ai4edatasetspublicassets.blob.core.windows.net/assets/pc_thumbnails/colombia2026.png",
+            "type": "image/png",
+            "title": "Predicted Building Damage Thumbnail",
+            "roles": [
+                "thumbnail"
+            ]
+        }
+    },
+    "storage:schemes": {
+        "azure": {
+            "type": "ms-azure",
+            "platform": "https://{account}.blob.core.windows.net",
+            "account": "ai4edataeuwest",
+            "container": "ai4good",
+            "region": "westeurope"
+        }
+    }
+}