Change Detection

Persona: Forest ecologist, land manager, conservation analyst Concept: #vegetationDifficulty: Intermediate

Scenario

Detect vegetation change in the Sierra Nevada, California, by comparing Sentinel-2 summer composites from 2020 (baseline) and 2024 (current). Compute NDVI for both periods, calculate the difference, classify change into health zones, and summarize by watershed.

Data Sources

Layer	Source	Type	Resolution
Sentinel-2 2020	Copernicus Sentinel-2 L2A	raster	10m
Sentinel-2 2024	Copernicus Sentinel-2 L2A	raster	10m
Watersheds	USGS WBD HUC-10	vector	N/A
Fire perimeters	MTBS 2020-2024	vector	N/A

Operations Used

Operation	Purpose
imagery_radiometric_indices	Compute NDVI from Sentinel-2 bands
raster_calc	Compute NDVI difference (2024 - 2020)
raster_reclassify	Classify change into loss/stable/gain zones
analysis_zonal_stats	Summarize change per watershed

Complete Workspace

name: sierra-ndvi-change
version: "1.0"
description: >
  NDVI change detection in the Sierra Nevada, California.
  Compares summer 2020 baseline to summer 2024 to identify
  vegetation loss (drought/fire) and recovery areas.

settings:
  default_bbox: [-120.3, 37.9, -119.8, 38.4]
  default_crs: EPSG:4326

layers:

  # --- SOURCE IMAGERY ---

  source/sentinel2-2020:
    uri: cache/sierra-sentinel2-2020.tif
    type: raster
    description: Sentinel-2 L2A summer composite, July-August 2020

  source/sentinel2-2024:
    uri: cache/sierra-sentinel2-2024.tif
    type: raster
    description: Sentinel-2 L2A summer composite, July-August 2024

  source/watersheds:
    uri: cache/sierra-huc10.geojson
    type: vector
    description: HUC-10 watershed boundaries
    style:
      stroke: "#1565C0"
      fill: transparent

  source/fire-perimeters:
    uri: cache/sierra-fires-2020-2024.geojson
    type: vector
    description: MTBS fire perimeters (2020-2024)
    style:
      stroke: "#FF4444"
      fill: "rgba(255,68,68,0.2)"

  # --- NDVI COMPUTATION ---

  analysis/ndvi-2020:
    type: raster
    description: NDVI from 2020 baseline imagery
    compute:
      op: imagery_radiometric_indices
      inputs:
        imagery: { layer: source/sentinel2-2020 }
      params:
        index: ndvi
        red_band: 4
        nir_band: 8
    style:
      palette: ndvi
      rescale: "-0.2,0.9"

  analysis/ndvi-2024:
    type: raster
    description: NDVI from 2024 current imagery
    compute:
      op: imagery_radiometric_indices
      inputs:
        imagery: { layer: source/sentinel2-2024 }
      params:
        index: ndvi
        red_band: 4
        nir_band: 8
    style:
      palette: ndvi
      rescale: "-0.2,0.9"

  # --- CHANGE DETECTION ---

  analysis/ndvi-change:
    type: raster
    description: NDVI difference (2024 - 2020). Negative = vegetation loss.
    compute:
      op: raster_calc
      inputs:
        a: { layer: analysis/ndvi-2024 }
        b: { layer: analysis/ndvi-2020 }
      params:
        expression: "a - b"
    style:
      palette: rdylgn
      rescale: "-0.5,0.5"
      info:
        fields: [ndvi_change]
        section: Vegetation Change

  # --- CLASSIFICATION ---

  analysis/change-classes:
    type: raster
    description: >
      Vegetation change classification:
      1=severe loss, 2=moderate loss, 3=stable, 4=moderate gain, 5=recovery
    compute:
      op: raster_reclassify
      inputs:
        raster: { layer: analysis/ndvi-change }
      params:
        breaks: [-1.0, -0.3, -0.1, 0.1, 0.3, 1.0]
        labels: [1, 2, 3, 4, 5]

  # --- WATERSHED STATISTICS ---

  results/change-by-watershed:
    type: vector
    description: NDVI change statistics per watershed
    compute:
      op: analysis_zonal_stats
      inputs:
        raster: { layer: analysis/ndvi-change }
        zones: { layer: source/watersheds }
      params:
        stats: [mean, min, max, std, count]
    style:
      type: fill
      property: mean
      palette: rdylgn
      table:
        columns:
          - { field: name, label: Watershed }
          - { field: mean, label: "Mean NDVI change", format: ".3f" }
          - { field: min, label: "Min NDVI change", format: ".3f" }
          - { field: count, label: Pixels }

Key Insight

The workspace YAML is a DAG. Each layer references its inputs by path:

source/sentinel2-2020 → analysis/ndvi-2020 ──┐
                                               ├→ analysis/ndvi-change → analysis/change-classes
source/sentinel2-2024 → analysis/ndvi-2024 ──┘                       → results/change-by-watershed

The platform resolves dependencies and executes in the correct order. No imperative script needed.