# catalog/operations/analysis/dnbr.yaml # # dNBR (delta Normalized Burn Ratio) — MTBS-convention burn severity. # Producer for registry/concepts/burn-severity.yaml. id: analysis_dnbr name: dNBR Burn Severity description: > Compute differenced Normalized Burn Ratio (dNBR) from a pre-fire and post-fire scene pair, emit the scaled (×1000) dNBR raster, and classify pixels into the canonical MTBS / Key & Benson (2006) severity codes. Two input modes are supported: - `from-scenes` (default): inputs are pre/post multispectral scenes with NIR and SWIR2 bands. NBR is computed internally as `NBR = (NIR − SWIR2) / (NIR + SWIR2)` for each scene. - `from-nbr`: inputs are already-computed NBR rasters (e.g. chained from `imagery_radiometric_indices`). The op skips NBR computation and only differences and classifies. dNBR is emitted scaled by 1000 to match the MTBS convention (the same axis used by the `burn-severity` concept class table). The categorical severity raster is always produced as a secondary output. version: 1.0.0 category: analysis type: raster-raster-to-raster default_implementation: native inputs: - name: pre type: raster format: geotiff description: > Pre-fire input. When `mode: from-scenes`, a multispectral scene containing NIR and SWIR2 bands (e.g. Sentinel-2 with B8 + B12). When `mode: from-nbr`, a single-band NBR raster (float, [-1, 1]). required: true - name: post type: raster format: geotiff description: > Post-fire input. Same band/format expectations as `pre`. Must be co-registered to `pre` to sub-pixel accuracy; phenological anniversary-date pairs are recommended to reduce false change. required: true requires: # Op-level band rollup — both inputs need NIR + SWIR2 to compute NBR. # Ignored when `mode: from-nbr` (inputs are already NBR rasters). bands: [nir, swir2] outputs: - name: dnbr type: raster format: cog dtype: float32 unit: scaled_dnbr index: dNBR # formula artifact — `dnbr` is an index (delta NBR), not a # phenomenon. The categorical `severity` output (below) is # the producer of registry/concepts/burn-severity.yaml. # See OPERATIONS.md #outputs-concept-vs-index # NOTE: registry/indices/dnbr.yaml does not yet exist # (upstream Awesome Spectral Indices gap); slug is reserved. description: > dNBR raster, scaled ×1000 to match the MTBS convention (`(NBR_pre − NBR_post) * 1000`). Typical range −2000 to +2000; MTBS class breaks reference this scaled axis. - name: severity type: raster format: cog dtype: int8 concept: burn-severity description: > Categorical MTBS burn-severity raster derived by applying `class_breaks` to the scaled dNBR. Codes follow `registry/concepts/burn-severity.yaml`: -1 = Increased greenness / regrowth (dNBR < -100) 1 = Unburned to low (-100 to +99) 2 = Low severity (+100 to +269) 3 = Moderate severity (+270 to +439) 4 = High severity (+440 to +659) 5 = Increased greenness (post-fire) (≥ +660) NoData is encoded as the int8 sentinel `-128`. int8 is used so the legal class code `-1` (regrowth) can be represented natively. validated_by: - "key-benson-2006" # Key & Benson (2006), Landscape Assessment (FIREMON RMRS-GTR-164) — MTBS dNBR severity breaks (no DOI; citation key) params: mode: type: string enum: [from-scenes, from-nbr] default: from-scenes description: > Input variant. `from-scenes` computes NBR internally from the NIR and SWIR2 bands of `pre` and `post`. `from-nbr` assumes `pre` and `post` are single-band NBR rasters already in [-1, 1]. nir_band: type: string default: B8 description: > Band name (or numeric index as string) for NIR in the pre/post scenes. Defaults to Sentinel-2 `B8`. Use `B5` for Landsat 8/9 OLI, `nir` for the canonical Folia band alias, etc. Ignored when `mode: from-nbr`. swir2_band: type: string default: B12 description: > Band name for SWIR2 in the pre/post scenes. Defaults to Sentinel-2 `B12`. Use `B7` for Landsat 8/9 OLI, `swir2` for the canonical Folia band alias. Ignored when `mode: from-nbr`. class_breaks: type: object description: > Override the MTBS class break thresholds (on the scaled ×1000 dNBR axis). Defaults match Key & Benson (2006) / MTBS canonical values referenced by `registry/concepts/burn-severity.yaml`. default: regrowth_max: -100 # dNBR < -100 -> class -1 unburned_max: 99 # -100 to 99 -> class 1 low_max: 269 # 100 to 269 -> class 2 moderate_max: 439 # 270 to 439 -> class 3 high_max: 659 # 440 to 659 -> class 4 # >= 660 -> class 5 granularity: min_resolution_m: 10 max_resolution_m: 500 recommended_resolution_m: 20 output_resolution: inherit minimum_bands: 1 requires_continuous: true suitable_aoi_km2: min: 0.1 max: 100000 realtime_max_km2: 100 uncertainty: error_model: propagated factors: - name: registration_error description: "Misalignment between pre and post scenes produces false dNBR signal" severity: high mitigations: - "Ensure both inputs are co-registered to sub-pixel accuracy" - "Use Sen2Cor / ARD products where geometric correction is enforced" - name: phenological_variation description: "Seasonal NBR variation can be confused with burn signal" severity: high mitigations: - "Use anniversary-date pre/post pairs where possible" - "Prefer composites over single-date scenes for the pre-fire reference" - name: cloud_and_shadow_contamination description: "Cloud/cloud-shadow pixels invalidate NBR and propagate to dNBR" severity: medium mitigations: - "Pre-mask with `analysis_cloud_mask` or scene-level QA bands" - "Reject pre/post pairs above a cloud-fraction threshold" - name: class_break_sensitivity description: "MTBS breaks are calibrated on Landsat / US ecoregions; transferability varies" severity: medium mitigations: - "Override `class_breaks` for non-US / Sentinel-2 calibrated thresholds" - "Validate against field burn-severity plots where available" limitations: - "Assumes radiometrically comparable inputs (surface reflectance, same sensor preferred)" - "MTBS class breaks are calibrated on Landsat dNBR; Sentinel-2 dNBR values are similar but not identical" execution: realtime_max_km2: 100 cost_per_km2: 0.001 time_per_km2_sec: 0.3 memory_profile: low profile: cpu: low memory: low io: medium scaling: model: linear parallelizable: true min_chunk_km2: 1 cache_policy: ttl_days: 30 invalidate_on: [source_update] # backends: audited 2026-08-14 (defect 50). NOTHING dispatches this op: no arm in folia-engine # `dispatch_op`, no `registerOp`/OP_TABLE entry in packages/compute, no `_BUILTIN_OP_MAP` # key in folia/compute.py, no backend manifest. Declared EMPTY on purpose — an absent # block would be indistinguishable from one nobody ever wrote. backends: {} display_hints: map: renderer: maplibre type: raster palette: burn-severity colors: ["#1a9850", "#f7f7f7", "#fee08b", "#fdae61", "#f46d43", "#a50026"] labels: [regrowth, unburned, low, moderate, high, very-high] info: fields: [dnbr_value, severity_class] format: "{value}" section: Fire ui: icon: flame color: "#D32F2F" examples: - name: dNBR from Sentinel-2 pre/post scenes description: > Compute scaled dNBR + MTBS severity from a Sentinel-2 L2A pre/post pair (NIR = B8, SWIR2 = B12). inputs: pre: "s2_l2a_prefire_2023-06-15.tif" post: "s2_l2a_postfire_2023-08-20.tif" params: mode: from-scenes nir_band: B8 swir2_band: B12 - name: dNBR from pre-computed NBR rasters description: > Chain after `imagery_radiometric_indices` (or equivalent) to skip internal NBR computation. Use when NBR has already been cached. inputs: pre: "nbr_pre.tif" post: "nbr_post.tif" params: mode: from-nbr