02SentinelChange - Sentinel Change Lab

User manual, version 0.1.1 (2026-11-16). QGIS 3.28 to 4.x. GPL-2.0-or-later. Yusuf Eminoğlu (Geophilo).

1. What the plugin does

02SentinelChange carries one Sentinel-1 or Sentinel-2 change study from an area of interest to a reported map inside QGIS, with no account and nothing to install. It searches two public STAC catalogues, reads only the area of interest from the cloud-optimised GeoTIFFs, masks clouds, builds clear before and after composites on one grid in metres, and computes spectral indices, burn severity, radar flood extent, index change and per-pixel break dates. Every layer is a GeoTIFF or GeoPackage that carries its provenance (scene ids, dates, formula, reference and the Copernicus attribution), and the HTML report lists every method with the equations of this manual.

The same steps are available in two forms: the lab dock (section 5), which keeps one study per session and adds styled layers to the map, and the Processing provider (section 6), whose algorithms chain in the Graphical Modeler and in batch mode. Both call the same analysis core; neither re-implements a method.

Only qgis.core, GDAL and numpy are required; SciPy is used for speed when present and gives identical results.

2. Install

In QGIS: Plugins > Manage and Install Plugins, search 02SentinelChange, Install. The release zip from GitLab installs with Install from ZIP. The plugin adds a toolbar button, a Plugins > 02SentinelChange menu (lab and manual) and the 02SentinelChange group in the Processing Toolbox.

3. Data sources and terms

SourceCollectionAccessUsed for
Element 84 Earth Search v1 (default)sentinel-2-c1-l2aanonymous; public COGs over HTTPSSentinel-2 bands and SCL
Microsoft Planetary Computer (fall-back)sentinel-2-l2aanonymous SAS tokenSentinel-2 bands and SCL
Microsoft Planetary Computersentinel-1-grdanonymous SAS tokenVV / VH and calibration tables
Microsoft Planetary Computersentinel-1-rtcneeds a Planetary Computer subscription keyterrain-corrected backscatter (opt-in)
Microsoft Planetary Computerjrc-gsw occurrenceanonymous SAS tokenpermanent-water mask
Local filesGeoTIFF / COG per bandofflineany analysis (Processing: Scene from local files)

Sentinel-2 dates are sent in full RFC 3339 form. When one Sentinel-2 source fails, the other is tried and the log says so. Planetary Computer tokens are cached per collection until five minutes before they expire. HTTP goes through the QGIS network manager, so the QGIS proxy and SSL settings apply; pixel reads use GDAL /vsicurl/ range requests with GDAL options set only for the duration of a read and restored afterwards.

Reflectance: scale and offset are taken from the STAC raster:bands when present (Earth Search Collection 1: s=10−4, o=−0.1s = 10^{-4},\ o = -0.1); Planetary Computer carries none, so the ESA rule applies (offset −0.1-0.1 from processing baseline 04.00, none before).

Surface reflectance from digital numbers
ρ=s⋅DN+o,s=10−4, o=−0.1 (baseline≥04.00)(1)\rho = s \cdot \mathrm{DN} + o, \qquad s = 10^{-4},\ o = -0.1 \ (\text{baseline} \ge 04.00) \tag{1}

Sentinel data are provided under the Copernicus Sentinel Data Legal Notice (free, full and open). Every output carries the line "Contains modified Copernicus Sentinel data [year]". Sentinel-1 RTC is CC-BY-4.0. JRC Global Surface Water is distributed by the EC JRC and cited as Pekel et al. (2016). Demo outputs are labelled "Synthetic demo data (not Sentinel imagery)" instead.

4. Units, working CRS and memory

All pixel work of a study happens on one grid: a projected CRS in metres, an origin snapped to whole multiples of the resolution rr, and a fixed width and height. The CRS is the UTM zone the scenes share (their proj:epsg) when the area lies in it, otherwise the UTM zone of the area centre (EPSG 326xx north, 327xx south); the Area page and the run log state which and why. Each band is warped onto that grid with gdal.Warp: average when coarsening and bilinear when refining reflectance or backscatter, nearest neighbour for SCL and masks. Areas are pixel counts times r2r^2, never measured in a geographic CRS:

Area from pixel counts
A=npx r2104 ha(20)A = n_{\mathrm{px}}\, \frac{r^{2}}{10^{4}}\ \mathrm{ha} \tag{20}

Limits: 40 million pixels per band and date (for example 4 000 km² at 10 m; larger areas are refused with a request for a coarser resolution), at most 120 dates in a time series, and T×pixels≤2⋅109T \times \text{pixels} \le 2 \cdot 10^{9} for break detection. Composites, series and break detection run in row blocks of at most 512 rows, so the peak array is dates × 512 × width rather than the full cube. Missing values are NaN (floats) or a declared nodata code, never 0 or −1 standing in for "undefined".

5. The lab dock, page by page

The dock sits on the right of the QGIS window with a dark icon rail; Help is pinned at the bottom. Long steps run as QGIS background tasks with a progress bar and a Cancel button, so QGIS never freezes. Results go to a 02SentinelChange layer group with sub-groups per study step.

5.1 Area

Step 1 picks the area: the polygons of a layer (only the selected ones when there is a selection), the canvas extent, or a rectangle dragged on the map; Use this area reads it. Load demo writes the synthetic demo (section 9) and fills everything. Step 2 sets the resolution (10, 20, 30 or 60 m); a badge shows the working CRS and why, a gauge shows the pixel budget, and KPI cards show the area in hectares, the pixel count and the grid size. Step 3 sets the output and cache folders and shows source status chips; Check sources sends one request to each catalogue and turns the chip green, or red with the HTTP status.

5.2 Scenes

Choose Sentinel-2 L2A or Sentinel-1 GRD, the date range, the maximum scene cloud cover (Sentinel-2), the minimum share of the area inside the footprint and the Sentinel-2 source; the optional Planetary Computer key enables Sentinel-1 RTC and is sent only to the token endpoint, never saved. Search scenes fills a timeline strip (one bar per scene) and the scene table. Tick scenes into the Before (B) and After (A) baskets, or use Split at this date. Sentinel-1 rows show orbit direction and relative orbit; flood mapping uses only scenes on the flood scene's orbit.

5.3 Composite

Method chips: median, maximum NDVI or least cloudy (section 7.2). SCL class chips set the masked classes (defaults 0, 1, 3, 8, 9, 10), a spin box the dilation in pixels, chips the bands kept. Build both composites writes both composites and adds a true-colour and a SWIR false-colour layer for each; KPI cards show scenes used, valid share and median clear count.

5.4 Indices

A searchable checklist of the 14 indices (the search matches key, label, formula and reference; the tooltip gives formula, bands and reference). Presets: Vegetation (NDVI, EVI, SAVI, NDRE, NDMI), Water (NDWI, MNDWI, AWEIsh), Built-up (NDBI, BSI, UI), Fire (NBR, NBR2, NDMI). Compute on the Before composite, the After composite or both; each index is added with its colour ramp and range, and a chart compares the means.

5.5 Fire

Pre and post windows (filled from the baskets), the offset mode (unburned ring, fixed value or none), ring width, an optional perimeter layer, the class thresholds (editable; Key & Benson defaults restores −250, −100, 100, 270, 440, 660) and the minimum mapping unit. Map burn severity adds dNBR, RdNBR, RBR, the class map and burned polygons; KPI cards show burned hectares, high-severity share and the offset; a bar chart shows hectares per class in the standard colours.

5.6 Flood

The reference window and the flood scene (same orbit), polarisation (VV is the usual choice for open water), speckle filter, log-ratio threshold, automatic threshold method or a fixed value, the permanent-water limit and the minimum mapping unit. Map the flood adds the backscatter, log-ratio and flood layers and polygons; a histogram of post-event backscatter shows the threshold line; KPI cards show new flood and permanent water in hectares, the threshold and the tiles used.

5.7 Change

Index change uses the baskets: pick the index, difference or relative change, and Otsu or a fixed threshold; Map the change adds the change layer and the decrease / stable / increase classes with their areas. Breaks takes the index, the history and monitoring windows, harmonics, trend, zz, the number of consecutive anomalies and the direction; Detect breaks adds the break-date map styled by month. Probe a pixel then draws, for a clicked pixel, the clear observations, the fitted harmonic curve, the end of the history and the break date.

5.8 Report

Write the report writes the self-contained HTML report of every step run so far (inputs, a per-basket scene summary with the full scene list with dates and ids one click below it, methods with these equations and references, KPI tables, map thumbnails, attribution); Open in browser opens it. GeoPackage of polygons writes burned and flood polygons as tables of one GeoPackage; Folder of GeoTIFFs copies every result raster as a Cloud Optimized GeoTIFF.

5.9 Help

A short guide, the method references, the data terms and attribution, and links to this manual and the repository.

6. Processing algorithms

The provider 02SentinelChange has ten algorithms. Scenes pass between them as a scene table: a polygon layer in EPSG:4326, one footprint per scene, with the fields item_id, source, datetime, date, platform, cloud_pct (real), orbit, rel_orbit (integer), epsg (integer), aoi_cover (real), assets (JSON: band key → file or URL) and scale_off (JSON: band key → [scale, offset]). A scene table can be saved to a GeoPackage, filtered or edited, and given to any analysis algorithm; date windows select the scenes from it. Dates are written YYYY-MM-DD.

Analysis algorithms write their GeoTIFFs into the output folder (a temporary folder when left empty) and return them as raster outputs, so they chain in models; the main layer of each is added to the project when the run ends. Outputs are not styled by the algorithms. Help for each algorithm in the Toolbox repeats its method, formulas and references.

6.1 Search Sentinel scenes scenes:search_scenes

Processing id zero2sentinelchange:search_scenes, group Scenes.

Searches Sentinel-2 L2A or Sentinel-1 GRD scenes over the area of interest without an account and writes a scene table. Sentinel-2: Element 84 Earth Search (collection sentinel-2-c1-l2a, reflectance scale and offset from raster:bands) or Microsoft Planetary Computer (sentinel-2-l2a, the ESA processing-baseline rule DN x 1e-4 - 0.1 from baseline 04.00); when one source fails the other is tried. Sentinel-1: Planetary Computer sentinel-1-grd with anonymous SAS tokens (the terrain-corrected RTC collection needs a Planetary Computer subscription key).

Scenes are kept when their cloud cover (Sentinel-2) is at most the maximum and their footprint covers at least the given share of the area, measured in the area's UTM zone. The table is sorted oldest first. Data: Copernicus Sentinel Data Legal Notice; outputs carry the attribution 'Contains modified Copernicus Sentinel data [year]'.

Parameters

NameDescriptionTypeDefault
AOIArea of interest (polygons)vector layer
SENSORSensor
Choices: Sentinel-2 L2A (optical); Sentinel-1 GRD (radar)
choiceSentinel-2 L2A (optical)
STARTStart date (YYYY-MM-DD)text
ENDEnd date (YYYY-MM-DD)text
MAX_CLOUDMaximum scene cloud cover (percent, Sentinel-2)number30.0
MIN_COVERMinimum share of the area covered by the scene (percent)number90.0
S2_SOURCESentinel-2 source
Choices: Earth Search (Element 84); Planetary Computer
choice (advanced)Earth Search (Element 84)
ORBITSentinel-1 orbit direction
Choices: Any; Ascending; Descending
choice (advanced)Any
REL_ORBITSentinel-1 relative orbit (0: any)number (advanced)0
PC_KEYPlanetary Computer subscription key (Sentinel-1 RTC only)text (optional, advanced)
LIMITMaximum number of scenesnumber (advanced)200
OUTPUTScene tablevector output

Outputs

NameDescriptionType
OUTPUTScene tableoutputVector
COUNTNumber of scenesoutputNumber

6.2 Scene from local files scenes:local_scene

Processing id zero2sentinelchange:local_scene, group Scenes.

Adds one scene made of local GeoTIFF or COG files (one file per band) to a scene table, so every analysis also runs offline. Give the Sentinel-2 bands you need (the SCL band enables cloud masking) or the Sentinel-1 VV/VH sigma0 files in linear units on one projected grid.

Scale and offset per band: the file's own GDAL scale/offset when set; for integer reflectance files without it, the product's MTD_MSIL2A.xml when the file sits in a .SAFE folder, else the rule chosen under Integer reflectance files: DN x 1e-4 - 0.1 for processing baseline 04.00 and later (every L2A product since 25 Jan 2022 and the reprocessed collection) or DN x 1e-4 before it. Float files and SCL are used as they are. An existing scene table can be given; its rows are copied and the new scene appended.

Parameters

NameDescriptionTypeDefault
SENSORSensor
Choices: Sentinel-2 L2A (optical); Sentinel-1 (radar, sigma0)
choiceSentinel-2 L2A (optical)
DATEAcquisition date (YYYY-MM-DD)text
BLUEBlue (B02)raster layer (optional)
GREENGreen (B03)raster layer (optional)
REDRed (B04)raster layer (optional)
REDEDGE1Red edge 1 (B05)raster layer (optional)
NIRNear infrared (B08)raster layer (optional)
SWIR16Shortwave infrared 1.6 um (B11)raster layer (optional)
SWIR22Shortwave infrared 2.2 um (B12)raster layer (optional)
SCLScene Classification (SCL)raster layer (optional)
VVVV backscatter (linear sigma0)raster layer (optional)
VHVH backscatter (linear sigma0)raster layer (optional)
ITEM_IDScene id (empty: built from the date and file name)text (optional)
ORBITSentinel-1 orbit direction
Choices: Unknown; Ascending; Descending
choice (advanced)Unknown
REL_ORBITSentinel-1 relative orbit (0: unknown)number (advanced)0
DN_RULEInteger reflectance files without scale metadata
Choices: Processing baseline 04.00 or later: DN x 1e-4 - 0.1; Before processing baseline 04.00: DN x 1e-4
choice (advanced)Processing baseline 04.00 or later: DN x 1e-4 - 0.1
SCENESExisting scene table to append to (optional)vector layer (optional)
OUTPUTScene tablevector output

Outputs

NameDescriptionType
OUTPUTScene tableoutputVector
COUNTNumber of scenesoutputNumber

6.3 Demo data (synthetic scenes) scenes:demo_data

Processing id zero2sentinelchange:demo_data, group Scenes.

Writes the offline demo: 61 synthetic Sentinel-2 L2A dates (every 12 days from 6 Jan 2024, 200 x 200 pixels at 10 m in EPSG:32635, with clouds, shadows and an SCL band), a planted burn scar on 15 Jul 2025, a lake, and four Sentinel-1 VV/VH dates on one orbit with a planted flood on the last date, plus a JRC-style water-occurrence raster. It is synthetic data, not Sentinel imagery, and is labelled so in every output.

Outputs: the Sentinel-2 and Sentinel-1 scene tables, the area of interest, the occurrence raster and the date windows that fit the planted events.

Parameters

NameDescriptionTypeDefault
FOLDERFolder for the demo GeoTIFFsfolder
SEEDRandom seednumber (advanced)7
S2_SCENESSentinel-2 scene tablevector output
S1_SCENESSentinel-1 scene tablevector output
AOIArea of interestvector output

Outputs

NameDescriptionType
FOLDERFolder for the demo GeoTIFFsoutputFolder
S2_SCENESSentinel-2 scene tableoutputVector
S1_SCENESSentinel-1 scene tableoutputVector
AOIArea of interestoutputVector
GSWWater occurrence (percent)outputRaster
S2_COUNTSentinel-2 datesoutputNumber
S1_COUNTSentinel-1 datesoutputNumber
FIRE_PREBefore window (fire, change)outputString
FIRE_POSTAfter window (fire, change)outputString
HISTORYBreak history windowoutputString
MONITORBreak monitoring windowoutputString
FLOOD_PREFlood reference windowoutputString
FLOOD_POSTFlood dateoutputString

6.4 Before and after composites composite:composites

Processing id zero2sentinelchange:composites, group Composites and indices.

Builds a cloud-masked composite of the scenes in the before window and another of the scenes in the after window on one snapped UTM grid. Median: the per-pixel median of the clear observations. Maximum NDVI: per pixel, the observation with the highest NDVI. Least cloudy: the scene with the highest clear share over the area, gaps filled from the next. Every band is written as a reflectance GeoTIFF, plus a true-colour and a SWIR false-colour RGB and a clear-observation count (uint16, nodata 65535 outside the area). Work runs in row blocks of at most 512 rows, so memory stays bounded.

Composites mask clouds with the Sentinel-2 Scene Classification (SCL): the ticked classes are masked, and all of them except 'No data' are grown by the dilation in pixels (square element).

Scenes come from a scene table: the output of Search Sentinel scenes, Scene from local files or Demo data. One footprint polygon per scene with its id, date, orbit, band files (JSON) and scale/offset. Rows can be filtered or deleted before a run. Only the area of interest is read from each cloud-optimised GeoTIFF.

All layers are computed on one grid in metres: the UTM zone the scenes share when the area lies in it, else the UTM zone of the area centre, with the origin snapped to whole multiples of the resolution. Areas are pixel counts x resolution squared, never measured in a geographic CRS. Every GeoTIFF carries its provenance (scene ids, dates, formula, reference, Copernicus attribution) in its metadata.

Parameters

NameDescriptionTypeDefault
AOIArea of interest (polygons)vector layer
SCENESSentinel-2 scene tablevector layer
BEFORE_STARTBefore window start (YYYY-MM-DD)text
BEFORE_ENDBefore window end (YYYY-MM-DD)text
AFTER_STARTAfter window start (YYYY-MM-DD)text
AFTER_ENDAfter window end (YYYY-MM-DD)text
RESResolution (m)number10.0
METHODComposite method
Choices: Per-pixel median of clear observations; Observation with the highest NDVI (maximum-value composite); Least cloudy scene, filled from the next
choicePer-pixel median of clear observations
MASKSCL classes masked as unusable
Choices: 0 - No data; 1 - Saturated or defective; 2 - Dark area pixels (topographic shadow); 3 - Cloud shadows; 4 - Vegetation; 5 - Not vegetated; 6 - Water; 7 - Unclassified; 8 - Cloud medium probability; 9 - Cloud high probability; 10 - Thin cirrus; 11 - Snow or ice
choice (optional)0 - No data; 1 - Saturated or defective; 3 - Cloud shadows; 8 - Cloud medium probability; 9 - Cloud high probability; 10 - Thin cirrus
DILATEMask dilation (pixels)number2
MIN_CLEARMinimum clear observations per pixelnumber1
OUTPUT_DIROutput folderfolder

Outputs

NameDescriptionType
OUTPUT_DIROutput folderoutputFolder
FOLDEROutput folderoutputFolder
BEFORE_RGBBefore true colouroutputRaster
BEFORE_SWIRBefore false colour (SWIR2, NIR, red)outputRaster
BEFORE_CLEARBefore clear observation countoutputRaster
BEFORE_VALIDBefore valid share of the areaoutputNumber
BEFORE_SCENESBefore scenes usedoutputNumber
AFTER_RGBAfter true colouroutputRaster
AFTER_SWIRAfter false colour (SWIR2, NIR, red)outputRaster
AFTER_CLEARAfter clear observation countoutputRaster
AFTER_VALIDAfter valid share of the areaoutputNumber
AFTER_SCENESAfter scenes usedoutputNumber

6.5 Spectral indices composite:spectral_indices

Processing id zero2sentinelchange:spectral_indices, group Composites and indices.

Computes spectral indices from surface-reflectance rasters, for example the band GeoTIFFs a composite wrote (before_red.tif, before_nir.tif and so on). All inputs must share one projected grid with square pixels. Values are multiplied by the scale and the offset added first, so digital numbers can be given directly (scale 0.0001, offset -0.1 for Sentinel-2 L2A from baseline 04.00). An index whose bands are missing is refused with a message. Pixels where a band is missing or the denominator is zero are NaN, never 0. Keys and formulas follow the Awesome Spectral Indices catalogue (Montero et al. 2023).

Parameters

NameDescriptionTypeDefault
BLUEBlue reflectanceraster layer (optional)
GREENGreen reflectanceraster layer (optional)
REDRed reflectanceraster layer (optional)
REDEDGE1Red edge 1 reflectanceraster layer (optional)
NIRNear-infrared (B08) reflectanceraster layer (optional)
SWIR16SWIR 1.6 um reflectanceraster layer (optional)
SWIR22SWIR 2.2 um reflectanceraster layer (optional)
INDICESIndices
Choices: NDVI - Normalized Difference Vegetation Index; EVI - Enhanced Vegetation Index; SAVI - Soil-Adjusted Vegetation Index; NDRE - Normalized Difference Red Edge; NDWI - Normalized Difference Water Index; MNDWI - Modified Normalized Difference Water Index; AWEIsh - Automated Water Extraction Index (shadow); NDMI - Normalized Difference Moisture Index; NBR - Normalized Burn Ratio; NBR2 - Normalized Burn Ratio 2; NDBI - Normalized Difference Built-up Index; BSI - Bare Soil Index; NDSI - Normalized Difference Snow Index; UI - Urban Index
choiceNDVI - Normalized Difference Vegetation Index; NBR - Normalized Burn Ratio
SCALEScale applied to the input valuesnumber (advanced)1.0
OFFSETOffset added after scalingnumber (advanced)0.0
PREFIXFile name prefixtext (advanced)index
OUTPUT_DIROutput folderfolder

Outputs

NameDescriptionType
OUTPUT_DIROutput folderoutputFolder
FOLDEROutput folderoutputFolder
NDVINDVIoutputRaster
EVIEVIoutputRaster
SAVISAVIoutputRaster
NDRENDREoutputRaster
NDWINDWIoutputRaster
MNDWIMNDWIoutputRaster
AWEIshAWEIshoutputRaster
NDMINDMIoutputRaster
NBRNBRoutputRaster
NBR2NBR2outputRaster
NDBINDBIoutputRaster
BSIBSIoutputRaster
NDSINDSIoutputRaster
UIUIoutputRaster
COUNTIndices writtenoutputNumber

6.6 Burn severity (dNBR, RdNBR, RBR) fire:burn_severity

Processing id zero2sentinelchange:burn_severity, group Fire.

Pre-fire and post-fire composites of NBR = (NIR - SWIR2) / (NIR + SWIR2) give dNBR, RdNBR and RBR on the x1000 scale. The offset removes the change of unburned land between the two windows: the mean dNBR in a ring of the given width outside the burn perimeter (the perimeter layer, or the pixels of moderate-low severity and above when no layer is given), a fixed value, or none. Severity classes follow USGS FIREMON (Key & Benson 2006) and UN-SPIDER: 1 enhanced regrowth high, 2 enhanced regrowth low, 3 unburned, 4 low, 5 moderate-low, 6 moderate-high, 7 high; a value on a threshold goes to the upper class. Burned polygons are the classes low and above, 8-connected, with regions smaller than the minimum mapping unit removed; each polygon carries its class, label, pixel count and area in hectares.

Scenes come from a scene table: the output of Search Sentinel scenes, Scene from local files or Demo data. One footprint polygon per scene with its id, date, orbit, band files (JSON) and scale/offset. Rows can be filtered or deleted before a run. Only the area of interest is read from each cloud-optimised GeoTIFF.

All layers are computed on one grid in metres: the UTM zone the scenes share when the area lies in it, else the UTM zone of the area centre, with the origin snapped to whole multiples of the resolution. Areas are pixel counts x resolution squared, never measured in a geographic CRS. Every GeoTIFF carries its provenance (scene ids, dates, formula, reference, Copernicus attribution) in its metadata.

Parameters

NameDescriptionTypeDefault
AOIArea of interest (polygons)vector layer
SCENESSentinel-2 scene tablevector layer
PRE_STARTPre-fire window start (YYYY-MM-DD)text
PRE_ENDPre-fire window end (YYYY-MM-DD)text
POST_STARTPost-fire window start (YYYY-MM-DD)text
POST_ENDPost-fire window end (YYYY-MM-DD)text
RESResolution (m)number10.0
OFFSET_MODEdNBR offset
Choices: Mean dNBR in an unburned ring around the burn; No offset; Fixed value
choiceMean dNBR in an unburned ring around the burn
OFFSET_VALUEFixed offset (dNBR x1000)number0.0
RING_MUnburned ring width (m)number500.0
PERIMETERBurn perimeter (optional)vector layer (optional)
THRESHOLDSClass thresholds, dNBR x1000 lower bounds (6 values)text (advanced)-250, -100, 100, 270, 440, 660
MMU_HAMinimum mapping unit (ha)number0.5
METHODComposite method
Choices: Per-pixel median of clear observations; Observation with the highest NDVI (maximum-value composite); Least cloudy scene, filled from the next
choice (advanced)Per-pixel median of clear observations
MASKSCL classes masked as unusable
Choices: 0 - No data; 1 - Saturated or defective; 2 - Dark area pixels (topographic shadow); 3 - Cloud shadows; 4 - Vegetation; 5 - Not vegetated; 6 - Water; 7 - Unclassified; 8 - Cloud medium probability; 9 - Cloud high probability; 10 - Thin cirrus; 11 - Snow or ice
choice (optional, advanced)0 - No data; 1 - Saturated or defective; 3 - Cloud shadows; 8 - Cloud medium probability; 9 - Cloud high probability; 10 - Thin cirrus
DILATEMask dilation (pixels)number (advanced)2
MIN_CLEARMinimum clear observations per pixelnumber (advanced)1
OUTPUT_DIROutput folderfolder
BURNEDBurned area polygonsvector output

Outputs

NameDescriptionType
OUTPUT_DIROutput folderoutputFolder
FOLDEROutput folderoutputFolder
DNBRdNBRoutputRaster
RDNBRRdNBRoutputRaster
RBRRBRoutputRaster
CLASSESSeverity classesoutputRaster
BURNEDBurned area polygonsoutputVector
BURNED_HABurned area (ha)outputNumber
OFFSETOffset used (dNBR x1000)outputNumber
HIGH_SHAREHigh-severity share of the burned areaoutputNumber

6.7 Flood extent (Sentinel-1) flood:flood_extent

Processing id zero2sentinelchange:flood_extent, group Flood.

Maps new open water from Sentinel-1 backscatter. GRD digital numbers are calibrated to sigma0 with the sigmaNought look-up table and warped with the scene GCPs (polynomial order 3; no terrain correction, which the log and metadata state); local files are read as linear sigma0. Speckle is reduced with a Lee filter (equivalent number of looks from the scene, 4.4 by default) or a median filter. The reference is the mean linear sigma0 of the scenes in the reference window on the same orbit direction and relative orbit as the flood scene (others are left out). A pixel is new flood when the log-ratio is below the ratio threshold, its post-event backscatter is below the water threshold, its reference backscatter is not (it was not water before), and its JRC Global Surface Water occurrence is below the permanent-water limit. The water threshold is fixed or automatic: Otsu or Kittler-Illingworth on the merged histogram of at most five bimodal tiles chosen by the split-based rule of Martinis et al. (2009) (coefficient of variation >= 0.7, tile mean / scene mean in 0.4 to 0.9), else on the whole area. Flood codes: 0 dry, 1 new flood, 2 permanent water, 255 nodata. Regions below the minimum mapping unit are removed.

All layers are computed on one grid in metres: the UTM zone the scenes share when the area lies in it, else the UTM zone of the area centre, with the origin snapped to whole multiples of the resolution. Areas are pixel counts x resolution squared, never measured in a geographic CRS. Every GeoTIFF carries its provenance (scene ids, dates, formula, reference, Copernicus attribution) in its metadata.

Parameters

NameDescriptionTypeDefault
AOIArea of interest (polygons)vector layer
SCENESSentinel-1 scene tablevector layer
PRE_STARTReference window start (YYYY-MM-DD)text (optional)
PRE_ENDReference window end (YYYY-MM-DD)text (optional)
POST_DATEFlood scene date (YYYY-MM-DD; empty: the latest scene)text (optional)
RESResolution (m)number10.0
POLPolarisation
Choices: VV; VH
choiceVV
SPECKLESpeckle filter
Choices: Lee filter; Median filter; None
choiceLee filter
SPECKLE_SIZESpeckle window (pixels, odd)number (advanced)5
RATIO_DBLog-ratio threshold (dB, a pixel must fall below it)number-3.0
WATERWater threshold on post-event dB
Choices: Otsu (automatic); Kittler-Illingworth (automatic); Fixed value
choiceOtsu (automatic)
WATER_DBFixed water threshold (dB)number-15.0
TILE_PXSplit-based tile size (pixels)number (advanced)200
GSWPermanent-water mask
Choices: Planetary Computer JRC GSW for remote scenes, none for local; No permanent-water mask; Occurrence raster given below
choicePlanetary Computer JRC GSW for remote scenes, none for local
GSW_RASTERWater occurrence raster (percent, optional)raster layer (optional)
GSW_OCCPermanent water above this occurrence (percent)number80.0
MMU_HAMinimum mapping unit (ha)number0.3
OUTPUT_DIROutput folderfolder
POLYGONSFlood polygonsvector output

Outputs

NameDescriptionType
OUTPUT_DIROutput folderoutputFolder
FOLDEROutput folderoutputFolder
PRE_DBReference backscatter (dB)outputRaster
POST_DBFlood-date backscatter (dB)outputRaster
LOG_RATIOLog-ratio (dB)outputRaster
FLOODFlood extent (0 dry, 1 flood, 2 permanent)outputRaster
POLYGONSFlood polygonsoutputVector
FLOODED_HANew flood (ha)outputNumber
PERMANENT_HAPermanent water (ha)outputNumber
THRESHOLD_DBWater threshold (dB)outputNumber
TILESTiles used for the thresholdoutputNumber

6.8 Index change change:index_change

Processing id zero2sentinelchange:index_change, group Change.

Builds before and after composites with the bands the index needs, computes the index on each and its change: the difference post - pre, or the relative change (post - pre) / |pre|. Pixels whose |change| exceeds the threshold are classed as decrease (-1) or increase (1), the rest stable (0); -128 is nodata. The threshold is a fixed value or Otsu's threshold on |change|. Areas of the three classes are returned in hectares.

Scenes come from a scene table: the output of Search Sentinel scenes, Scene from local files or Demo data. One footprint polygon per scene with its id, date, orbit, band files (JSON) and scale/offset. Rows can be filtered or deleted before a run. Only the area of interest is read from each cloud-optimised GeoTIFF.

All layers are computed on one grid in metres: the UTM zone the scenes share when the area lies in it, else the UTM zone of the area centre, with the origin snapped to whole multiples of the resolution. Areas are pixel counts x resolution squared, never measured in a geographic CRS. Every GeoTIFF carries its provenance (scene ids, dates, formula, reference, Copernicus attribution) in its metadata.

Parameters

NameDescriptionTypeDefault
AOIArea of interest (polygons)vector layer
SCENESSentinel-2 scene tablevector layer
BEFORE_STARTBefore window start (YYYY-MM-DD)text
BEFORE_ENDBefore window end (YYYY-MM-DD)text
AFTER_STARTAfter window start (YYYY-MM-DD)text
AFTER_ENDAfter window end (YYYY-MM-DD)text
INDEXIndex
Choices: NDVI - Normalized Difference Vegetation Index; EVI - Enhanced Vegetation Index; SAVI - Soil-Adjusted Vegetation Index; NDRE - Normalized Difference Red Edge; NDWI - Normalized Difference Water Index; MNDWI - Modified Normalized Difference Water Index; AWEIsh - Automated Water Extraction Index (shadow); NDMI - Normalized Difference Moisture Index; NBR - Normalized Burn Ratio; NBR2 - Normalized Burn Ratio 2; NDBI - Normalized Difference Built-up Index; BSI - Bare Soil Index; NDSI - Normalized Difference Snow Index; UI - Urban Index
choiceNDVI - Normalized Difference Vegetation Index
MODEChange
Choices: Difference: post - pre; Relative: (post - pre) / |pre|
choiceDifference: post - pre
THRESHOLD_MODEThreshold on |change|
Choices: Otsu (automatic); Fixed value
choiceOtsu (automatic)
THRESHOLDFixed thresholdnumber0.1
RESResolution (m)number10.0
METHODComposite method
Choices: Per-pixel median of clear observations; Observation with the highest NDVI (maximum-value composite); Least cloudy scene, filled from the next
choice (advanced)Per-pixel median of clear observations
MASKSCL classes masked as unusable
Choices: 0 - No data; 1 - Saturated or defective; 2 - Dark area pixels (topographic shadow); 3 - Cloud shadows; 4 - Vegetation; 5 - Not vegetated; 6 - Water; 7 - Unclassified; 8 - Cloud medium probability; 9 - Cloud high probability; 10 - Thin cirrus; 11 - Snow or ice
choice (optional, advanced)0 - No data; 1 - Saturated or defective; 3 - Cloud shadows; 8 - Cloud medium probability; 9 - Cloud high probability; 10 - Thin cirrus
DILATEMask dilation (pixels)number (advanced)2
MIN_CLEARMinimum clear observations per pixelnumber (advanced)1
OUTPUT_DIROutput folderfolder

Outputs

NameDescriptionType
OUTPUT_DIROutput folderoutputFolder
FOLDEROutput folderoutputFolder
DELTAIndex changeoutputRaster
CLASSESChange classes (-1, 0, 1)outputRaster
THRESHOLD_USEDThreshold usedoutputNumber
DECREASE_HADecrease (ha)outputNumber
STABLE_HAStable (ha)outputNumber
INCREASE_HAIncrease (ha)outputNumber

6.9 Break detection (index time series) change:break_detection

Processing id zero2sentinelchange:break_detection, group Change.

Per pixel, the clear-observation index series of every scene from the history start to the monitoring end is modelled on the history period by ordinary least squares: an intercept, an optional linear trend and 0 to 3 harmonic pairs per year, solved for all pixels at once. sigma is the history RMSE. In the monitoring period an observation is an anomaly when its residual exceeds z sigma in the chosen direction; a break is confirmed at the first of the given number of consecutive anomalies (CCDC-style monitoring after Zhu & Woodcock 2014, framing after BFAST Monitor, Verbesselt et al. 2012; MOSUM/CUSUM boundaries are not used). Outputs: break flag (1/0, 255 nodata), break date (decimal year), magnitude (median residual of the confirming run), history RMSE, and the series as a multi-band GeoTIFF (one band per date). Pixels with fewer clear history observations than the minimum are nodata. With a trend term, a history shorter than about 18 months cannot separate trend from season and gives false breaks; use at least 18 to 24 months.

Scenes come from a scene table: the output of Search Sentinel scenes, Scene from local files or Demo data. One footprint polygon per scene with its id, date, orbit, band files (JSON) and scale/offset. Rows can be filtered or deleted before a run. Only the area of interest is read from each cloud-optimised GeoTIFF.

All layers are computed on one grid in metres: the UTM zone the scenes share when the area lies in it, else the UTM zone of the area centre, with the origin snapped to whole multiples of the resolution. Areas are pixel counts x resolution squared, never measured in a geographic CRS. Every GeoTIFF carries its provenance (scene ids, dates, formula, reference, Copernicus attribution) in its metadata.

Parameters

NameDescriptionTypeDefault
AOIArea of interest (polygons)vector layer
SCENESSentinel-2 scene tablevector layer
INDEXIndex
Choices: NDVI - Normalized Difference Vegetation Index; EVI - Enhanced Vegetation Index; SAVI - Soil-Adjusted Vegetation Index; NDRE - Normalized Difference Red Edge; NDWI - Normalized Difference Water Index; MNDWI - Modified Normalized Difference Water Index; AWEIsh - Automated Water Extraction Index (shadow); NDMI - Normalized Difference Moisture Index; NBR - Normalized Burn Ratio; NBR2 - Normalized Burn Ratio 2; NDBI - Normalized Difference Built-up Index; BSI - Bare Soil Index; NDSI - Normalized Difference Snow Index; UI - Urban Index
choiceNDVI - Normalized Difference Vegetation Index
HISTORY_STARTHistory window start (YYYY-MM-DD)text
HISTORY_ENDHistory window end (YYYY-MM-DD)text
MONITOR_STARTMonitoring window start (YYYY-MM-DD)text
MONITOR_ENDMonitoring window end (YYYY-MM-DD)text
HARMONICSHarmonic pairs per yearnumber1
TRENDLinear trend termyes/noTrue
ZAnomaly threshold (residual sigmas)number3.0
CONSECUTIVEConsecutive anomalies to confirm a breaknumber3
MIN_HISTORYMinimum clear history observationsnumber (advanced)12
DIRECTIONDirection
Choices: Negative (index drops); Positive (index rises); Both
choiceBoth
RESResolution (m)number10.0
METHODComposite method
Choices: Per-pixel median of clear observations; Observation with the highest NDVI (maximum-value composite); Least cloudy scene, filled from the next
choice (advanced)Per-pixel median of clear observations
MASKSCL classes masked as unusable
Choices: 0 - No data; 1 - Saturated or defective; 2 - Dark area pixels (topographic shadow); 3 - Cloud shadows; 4 - Vegetation; 5 - Not vegetated; 6 - Water; 7 - Unclassified; 8 - Cloud medium probability; 9 - Cloud high probability; 10 - Thin cirrus; 11 - Snow or ice
choice (optional, advanced)0 - No data; 1 - Saturated or defective; 3 - Cloud shadows; 8 - Cloud medium probability; 9 - Cloud high probability; 10 - Thin cirrus
DILATEMask dilation (pixels)number (advanced)2
MIN_CLEARMinimum clear observations per pixelnumber (advanced)1
OUTPUT_DIROutput folderfolder

Outputs

NameDescriptionType
OUTPUT_DIROutput folderoutputFolder
FOLDEROutput folderoutputFolder
HAS_BREAKBreak flagoutputRaster
BREAK_DATEBreak date (decimal year)outputRaster
MAGNITUDEBreak magnitudeoutputRaster
RMSEHistory RMSEoutputRaster
SERIESIndex series (one band per date)outputRaster
BREAK_SHAREShare of evaluated pixels with a breakoutputNumber
DATESDates in the seriesoutputNumber

6.10 Change study (full pipeline and report) study:change_study

Processing id zero2sentinelchange:change_study, group Study and report.

Runs a whole change study with the same staged pipeline as the dock and writes the self-contained HTML report (inputs, scene list, methods with typeset formulas and references, KPI tables, map thumbnails, attribution). Steps: before and after composites with the chosen indices; burn severity (dNBR, RdNBR, RBR with a ring offset and USGS/UN-SPIDER classes); index change; break detection over the history and monitoring windows; Sentinel-1 flood extent. Each step is described in its own algorithm in this provider; this one uses their default settings for anything not listed here. All layers are written to the output folder, polygons to polygons.gpkg.

Scenes come from a scene table: the output of Search Sentinel scenes, Scene from local files or Demo data. One footprint polygon per scene with its id, date, orbit, band files (JSON) and scale/offset. Rows can be filtered or deleted before a run. Only the area of interest is read from each cloud-optimised GeoTIFF.

All layers are computed on one grid in metres: the UTM zone the scenes share when the area lies in it, else the UTM zone of the area centre, with the origin snapped to whole multiples of the resolution. Areas are pixel counts x resolution squared, never measured in a geographic CRS. Every GeoTIFF carries its provenance (scene ids, dates, formula, reference, Copernicus attribution) in its metadata.

Parameters

NameDescriptionTypeDefault
AOIArea of interest (polygons)vector layer
S2_SCENESSentinel-2 scene table (optional)vector layer (optional)
S1_SCENESSentinel-1 scene table (optional)vector layer (optional)
STEPSSteps
Choices: Composites and indices; Burn severity; Index change; Break detection; Flood extent (Sentinel-1)
choiceComposites and indices; Burn severity; Index change
BEFORE_STARTBefore window start (YYYY-MM-DD)text (optional)
BEFORE_ENDBefore window end (YYYY-MM-DD)text (optional)
AFTER_STARTAfter window start (YYYY-MM-DD)text (optional)
AFTER_ENDAfter window end (YYYY-MM-DD)text (optional)
INDICESIndices on the composites
Choices: NDVI - Normalized Difference Vegetation Index; EVI - Enhanced Vegetation Index; SAVI - Soil-Adjusted Vegetation Index; NDRE - Normalized Difference Red Edge; NDWI - Normalized Difference Water Index; MNDWI - Modified Normalized Difference Water Index; AWEIsh - Automated Water Extraction Index (shadow); NDMI - Normalized Difference Moisture Index; NBR - Normalized Burn Ratio; NBR2 - Normalized Burn Ratio 2; NDBI - Normalized Difference Built-up Index; BSI - Bare Soil Index; NDSI - Normalized Difference Snow Index; UI - Urban Index
choice (optional)NDVI - Normalized Difference Vegetation Index; NBR - Normalized Burn Ratio; MNDWI - Modified Normalized Difference Water Index
CHANGE_INDEXIndex for the change step
Choices: NDVI - Normalized Difference Vegetation Index; EVI - Enhanced Vegetation Index; SAVI - Soil-Adjusted Vegetation Index; NDRE - Normalized Difference Red Edge; NDWI - Normalized Difference Water Index; MNDWI - Modified Normalized Difference Water Index; AWEIsh - Automated Water Extraction Index (shadow); NDMI - Normalized Difference Moisture Index; NBR - Normalized Burn Ratio; NBR2 - Normalized Burn Ratio 2; NDBI - Normalized Difference Built-up Index; BSI - Bare Soil Index; NDSI - Normalized Difference Snow Index; UI - Urban Index
choiceNDVI - Normalized Difference Vegetation Index
BREAK_INDEXIndex for break detection
Choices: NDVI - Normalized Difference Vegetation Index; EVI - Enhanced Vegetation Index; SAVI - Soil-Adjusted Vegetation Index; NDRE - Normalized Difference Red Edge; NDWI - Normalized Difference Water Index; MNDWI - Modified Normalized Difference Water Index; AWEIsh - Automated Water Extraction Index (shadow); NDMI - Normalized Difference Moisture Index; NBR - Normalized Burn Ratio; NBR2 - Normalized Burn Ratio 2; NDBI - Normalized Difference Built-up Index; BSI - Bare Soil Index; NDSI - Normalized Difference Snow Index; UI - Urban Index
choiceNDVI - Normalized Difference Vegetation Index
HISTORY_STARTBreak history window start (YYYY-MM-DD)text (optional)
HISTORY_ENDBreak history window end (YYYY-MM-DD)text (optional)
MONITOR_STARTBreak monitoring window start (YYYY-MM-DD)text (optional)
MONITOR_ENDBreak monitoring window end (YYYY-MM-DD)text (optional)
FLOOD_PRE_STARTFlood reference window start (YYYY-MM-DD)text (optional)
FLOOD_PRE_ENDFlood reference window end (YYYY-MM-DD)text (optional)
FLOOD_DATEFlood scene date (YYYY-MM-DD; empty: the latest scene)text (optional)
RESResolution (m)number10.0
METHODComposite method
Choices: Per-pixel median of clear observations; Observation with the highest NDVI (maximum-value composite); Least cloudy scene, filled from the next
choice (advanced)Per-pixel median of clear observations
MASKSCL classes masked as unusable
Choices: 0 - No data; 1 - Saturated or defective; 2 - Dark area pixels (topographic shadow); 3 - Cloud shadows; 4 - Vegetation; 5 - Not vegetated; 6 - Water; 7 - Unclassified; 8 - Cloud medium probability; 9 - Cloud high probability; 10 - Thin cirrus; 11 - Snow or ice
choice (optional, advanced)0 - No data; 1 - Saturated or defective; 3 - Cloud shadows; 8 - Cloud medium probability; 9 - Cloud high probability; 10 - Thin cirrus
DILATEMask dilation (pixels)number (advanced)2
MIN_CLEARMinimum clear observations per pixelnumber (advanced)1
OUTPUT_DIROutput folderfolder
REPORTHTML reportfile

Outputs

NameDescriptionType
OUTPUT_DIROutput folderoutputFolder
FOLDEROutput folderoutputFolder
REPORTHTML reportoutputHtml
POLYGONSBurned and flood polygons (GeoPackage)outputVector
BURNED_HABurned area (ha)outputNumber
FLOODED_HANew flood (ha)outputNumber
BREAK_SHAREShare of pixels with a breakoutputNumber
STEPS_RUNSteps runoutputNumber

7. Methods and equations

Equation numbers are the same in the HTML report.

7.1 Cloud masking

Sentinel-2 L2A Scene Classification (SCL, 20 m, nearest-neighbour resampling). An observation tt of pixel xx is clear, ct(x)=1c_t(x) = 1, when its SCL class is not in the masked set MM. The masked classes other than "no data" are grown by the dilation dd pixels with a square structuring element (one cloud pixel becomes a (2d+1)×(2d+1)(2d+1) \times (2d+1) block); class 0 is masked but not grown, so swath edges keep their pixels (Main-Knorn et al. 2017).

SCLClassDefault
0No datamasked
1Saturated or defectivemasked
2Dark area pixels (topographic shadow)kept
3Cloud shadowsmasked
4Vegetationkept
5Not vegetatedkept
6Waterkept
7Unclassifiedkept
8Cloud medium probabilitymasked
9Cloud high probabilitymasked
10Thin cirrusmasked
11Snow or icekept

7.2 Composites

For a window WW of scenes, the median composite of band bb and the clear count are

Clear-pixel median composite
ρˉb(x)=median⁡{ ρb,t(x):t∈W, ct(x)=1 }(2)\bar\rho_{b}(x) = \operatorname*{median}\{\, \rho_{b,t}(x) : t \in W,\ c_t(x) = 1 \,\} \tag{2}
Clear observation count
n(x)=∑t∈Wct(x),ct(x)=[ SCLt(x)∉M ](3)n(x) = \sum_{t \in W} c_t(x), \qquad c_t(x) = [\,\mathrm{SCL}_t(x) \notin M\,] \tag{3}

Maximum NDVI takes, per pixel, the clear observation with the highest NDVI (Holben 1986). Least cloudy takes the scene with the highest clear share over the area and fills its gaps from the next one. Pixels with fewer clear observations than the minimum are NaN; the clear count is 65535 outside the area.

7.3 Spectral indices

Band symbols: BB blue (B02), GG green (B03), RR red (B04), RE1RE_1 red edge 1 (B05, 20 m), NN near infrared (B08, 10 m), S1S_1 SWIR 1.6 µm (B11), S2S_2 SWIR 2.2 µm (B12). Keys follow Awesome Spectral Indices (Montero et al. 2023). A pixel where a band is missing or the denominator is zero is NaN.

KeyNameFormulaBandsReference
NDVINormalized Difference Vegetation IndexNDVI=N−RN+R\mathrm{NDVI} = \frac{N - R}{N + R}nir, redRouse, J.W., Haas, R.H., Schell, J.A., Deering, D.W. (1974)
EVIEnhanced Vegetation IndexEVI=G N−RN+C1R−C2B+L,G=2.5, C1=6, C2=7.5, L=1\mathrm{EVI} = G\,\frac{N - R}{N + C_1 R - C_2 B + L},\quad G = 2.5,\ C_1 = 6,\ C_2 = 7.5,\ L = 1nir, red, blueHuete, A., Didan, K., Miura, T., Rodriguez, E.P., Gao, X., Ferreira, L.G. (2002)
SAVISoil-Adjusted Vegetation IndexSAVI=(1+L) N−RN+R+L,L=0.5\mathrm{SAVI} = (1 + L)\,\frac{N - R}{N + R + L},\quad L = 0.5nir, redHuete, A.R. (1988)
NDRENormalized Difference Red EdgeNDRE=N−RE1N+RE1\mathrm{NDRE} = \frac{N - RE_1}{N + RE_1}nir, rededge1Gitelson, A., Merzlyak, M.N. (1994)
NDWINormalized Difference Water IndexNDWI=G−NG+N\mathrm{NDWI} = \frac{G - N}{G + N}green, nirMcFeeters, S.K. (1996)
MNDWIModified Normalized Difference Water IndexMNDWI=G−S1G+S1\mathrm{MNDWI} = \frac{G - S_1}{G + S_1}green, swir16Xu, H. (2006)
AWEIshAutomated Water Extraction Index (shadow)AWEIsh=B+2.5 G−1.5 (N+S1)−0.25 S2\mathrm{AWEI_{sh}} = B + 2.5\,G - 1.5\,(N + S_1) - 0.25\,S_2blue, green, nir, swir16, swir22Feyisa, G.L., Meilby, H., Fensholt, R., Proud, S.R. (2014)
NDMINormalized Difference Moisture IndexNDMI=N−S1N+S1\mathrm{NDMI} = \frac{N - S_1}{N + S_1}nir, swir16Gao, B.-C. (1996)
NBRNormalized Burn RatioNBR=N−S2N+S2\mathrm{NBR} = \frac{N - S_2}{N + S_2}nir, swir22Key, C.H., Benson, N.C. (2006)
NBR2Normalized Burn Ratio 2NBR2=S1−S2S1+S2\mathrm{NBR2} = \frac{S_1 - S_2}{S_1 + S_2}swir16, swir22Key, C.H., Benson, N.C. (2006)
NDBINormalized Difference Built-up IndexNDBI=S1−NS1+N\mathrm{NDBI} = \frac{S_1 - N}{S_1 + N}swir16, nirZha, Y., Gao, J., Ni, S. (2003)
BSIBare Soil IndexBSI=(S1+R)−(N+B)(S1+R)+(N+B)\mathrm{BSI} = \frac{(S_1 + R) - (N + B)}{(S_1 + R) + (N + B)}swir16, red, nir, blueRikimaru, A., Roy, P.S., Miyatake, S. (2002)
NDSINormalized Difference Snow IndexNDSI=G−S1G+S1\mathrm{NDSI} = \frac{G - S_1}{G + S_1}green, swir16Hall, D.K., Riggs, G.A., Salomonson, V.V. (1995)
UIUrban IndexUI=S2−NS2+N\mathrm{UI} = \frac{S_2 - N}{S_2 + N}swir22, nirKawamura, M., Jayamanna, S., Tsujiko, Y. (1996)

7.4 Burn severity

With NBR=(N−S2)/(N+S2)\mathrm{NBR} = (N - S_2)/(N + S_2) from the pre-fire and post-fire composites:

Differenced Normalized Burn Ratio
dNBR=1000 (NBRpre−NBRpost)−δ(4)\mathrm{dNBR} = 1000\,(\mathrm{NBR}_{\mathrm{pre}} - \mathrm{NBR}_{\mathrm{post}}) - \delta \tag{4}
Unburned-ring offset
δ=1∣U∣∑x∈U1000 (NBRpre(x)−NBRpost(x))(5)\delta = \frac{1}{|U|} \sum_{x \in U} 1000\,(\mathrm{NBR}_{\mathrm{pre}}(x) - \mathrm{NBR}_{\mathrm{post}}(x)) \tag{5}

The ring UU is the band of width ring metres outside the perimeter (the perimeter layer, or the pixels of moderate-low severity and above before the offset when no layer is given), Chebyshev distance on the grid (Key 2006; Miller & Thode 2007).

Relativized dNBR
RdNBR=dNBRmax⁡(∣NBRpre∣, 0.001)(6)\mathrm{RdNBR} = \frac{\mathrm{dNBR}}{\sqrt{\max(\lvert \mathrm{NBR}_{\mathrm{pre}} \rvert,\ 0.001)}} \tag{6}
Relativized Burn Ratio
RBR=dNBRNBRpre+1.001(7)\mathrm{RBR} = \frac{\mathrm{dNBR}}{\mathrm{NBR}_{\mathrm{pre}} + 1.001} \tag{7}
Severity class
k(x)=1+#{ j:θj≤dNBR(x) },θ=(−250,−100,100,270,440,660)(8)k(x) = 1 + \#\{\, j : \theta_j \le \mathrm{dNBR}(x) \,\}, \qquad \theta = (-250, -100, 100, 270, 440, 660) \tag{8}

Classes (USGS FIREMON, Key & Benson 2006; UN-SPIDER Recommended Practice); a value on a bound goes to the upper class, 0 is nodata:

ClassLabeldNBR (×1000)
1 Enhanced regrowth, highdNBR < −250
2 Enhanced regrowth, low−250 ≤ dNBR < −100
3 Unburned−100 ≤ dNBR < 100
4 Low severity100 ≤ dNBR < 270
5 Moderate-low severity270 ≤ dNBR < 440
6 Moderate-high severity440 ≤ dNBR < 660
7 High severitydNBR ≥ 660

Burned polygons are the classes 4 to 7, 8-connected, with regions below the minimum mapping unit removed; each polygon carries class, label, pixels and area_ha.

7.5 Flood (Sentinel-1)

GRD digital numbers are read in radar geometry for the window the scene GCPs give for the area, calibrated with the sigmaNought look-up table AσA_{\sigma} interpolated bilinearly (ESA-EOPG-CSCOP-TN-0002), then warped to the grid with a GCP polynomial of order 3. There is no thermal-noise removal or terrain correction. Local radar files are read as linear σ0\sigma^0.

Sentinel-1 radiometric calibration
σi0=DNi2Aσ,i2(9)\sigma^0_{i} = \frac{\mathrm{DN}_i^{2}}{A_{\sigma,i}^{2}} \tag{9}
Lee speckle filter
I^=Iˉ+k (I−Iˉ),k=var⁡(x)var⁡(I),var⁡(x)=var⁡(I)−Iˉ2/L1+1/L(10)\hat I = \bar I + k\,(I - \bar I), \qquad k = \frac{\operatorname{var}(x)}{\operatorname{var}(I)}, \quad \operatorname{var}(x) = \frac{\operatorname{var}(I) - \bar I^{2} / L}{1 + 1/L} \tag{10}

LL is the equivalent number of looks (sar:looks_equivalent_number, 4.4 by default); the median filter is the alternative (Lee 1980). The reference σˉpre0\bar\sigma^0_{\mathrm{pre}} is the mean linear backscatter of the reference scenes on the same orbit direction and relative orbit as the flood scene.

Backscatter in decibels
σdB0=10log⁡10σ0(11)\sigma^0_{\mathrm{dB}} = 10 \log_{10} \sigma^0 \tag{11}
Log-ratio change
LR=10log⁡10σpost0σˉpre0(12)\mathrm{LR} = 10 \log_{10} \frac{\sigma^0_{\mathrm{post}}}{\bar\sigma^0_{\mathrm{pre}}} \tag{12}

The water threshold t∗t^{*} on post-event dB is fixed or automatic. Automatic thresholds use the split-based rule of Martinis et al. (2009): the area is cut into tiles of tile pixels; a tile is a candidate when its coefficient of variation is at least 0.7 and its mean over the scene mean lies in 0.4 to 0.9; at most five candidates closest to the mean are kept and the threshold is found on their merged histogram, or on the whole area when no tile qualifies.

Otsu threshold
t∗=arg⁡max⁡t(μT ω(t)−μ(t))2ω(t) (1−ω(t))(13)t^{*} = \arg\max_{t} \frac{\left(\mu_T\,\omega(t) - \mu(t)\right)^{2}}{\omega(t)\,\left(1 - \omega(t)\right)} \tag{13}
Kittler-Illingworth threshold
t∗=arg⁡min⁡t 1+2[P1ln⁡σ1+P2ln⁡σ2]−2[P1ln⁡P1+P2ln⁡P2](14)t^{*} = \arg\min_{t}\ 1 + 2\left[P_1 \ln \sigma_1 + P_2 \ln \sigma_2\right] - 2\left[P_1 \ln P_1 + P_2 \ln P_2\right] \tag{14}

Here ω(t)\omega(t) is the share of values below tt, μ(t)\mu(t) their partial mean, μT\mu_T the total mean (Otsu 1979), and PiP_i, σi\sigma_i the share and standard deviation of each side (Kittler & Illingworth 1986). When the histogram has an empty gap between the modes, the centre of the flat optimum is returned.

New flood rule
F(x)=[ LR(x)<r ]∧[ σdB,post0(x)<t∗ ]∧[ σˉdB,pre0(x)≥t∗ ]∧[ O(x)<o ](15)F(x) = [\,\mathrm{LR}(x) < r\,] \wedge [\,\sigma^0_{\mathrm{dB,post}}(x) < t^{*}\,] \wedge [\,\bar\sigma^0_{\mathrm{dB,pre}}(x) \ge t^{*}\,] \wedge [\,O(x) < o\,] \tag{15}

rr is the log-ratio threshold (−3 dB by default) and O(x)O(x) the JRC occurrence with limit oo (80 %). Pixels that are water after the event and either above the occurrence limit or already water in the reference are permanent water (code 2), reported separately. Regions below the minimum mapping unit are removed. Framing: Twele et al. (2016); UN-SPIDER Recommended Practice for Sentinel-1 flood mapping.

7.6 Break detection

Per pixel, on the clear-observation index series with tt in decimal years, the model is fitted on the history period HH by ordinary least squares, solved for all pixels at once with per-pixel masked normal equations:

Harmonic regression model
y^(t)=a0+a1t+∑k=1K(bkcos⁡2πkt+cksin⁡2πkt)(16)\hat y(t) = a_0 + a_1 t + \sum_{k=1}^{K} \left( b_k \cos 2\pi k t + c_k \sin 2\pi k t \right) \tag{16}
History root mean square error
σ=1n−p∑t∈H(yt−y^(t))2(17)\sigma = \sqrt{\frac{1}{n - p} \sum_{t \in H} \left( y_t - \hat y(t) \right)^{2}} \tag{17}

KK is the number of harmonic pairs (0 to 3), pp the number of coefficients, and the trend term a1ta_1 t is optional. In the monitoring period:

Anomaly and break rule
∣yt−y^(t)∣>z σ on m consecutive clear observations(18)\lvert y_t - \hat y(t) \rvert > z\,\sigma \ \text{on } m \text{ consecutive clear observations} \tag{18}

The direction setting keeps only negative residuals (index drops), only positive ones, or both. The break date is the first of the mm consecutive anomalies; the magnitude is the median residual of that run. Pixels with fewer clear history observations than the minimum are nodata (Zhu & Woodcock 2014, CCDC; monitoring framing after Verbesselt et al. 2012, BFAST Monitor). MOSUM / CUSUM significance boundaries are not used.

With a trend term, a history shorter than about 18 months cannot separate trend from season. In testing, one year of history gave 16 % false breaks and 21 months gave none. Use 18 to 24 months of history, or switch the trend off.

7.7 Index change

Index change
Δ=Ipost−Ipre,Δrel=Ipost−Ipre∣Ipre∣(19)\Delta = I_{\mathrm{post}} - I_{\mathrm{pre}}, \qquad \Delta_{\mathrm{rel}} = \frac{I_{\mathrm{post}} - I_{\mathrm{pre}}}{\lvert I_{\mathrm{pre}} \rvert} \tag{19}

A pixel is a decrease (−1-1) when Δ<−τ\Delta < -\tau, an increase (11) when Δ>τ\Delta > \tau and stable (00) otherwise; τ\tau is fixed or Otsu's threshold on ∣Δ∣\lvert\Delta\rvert.

8. Output files and codes

FileContentNodata
before_<band>.tif, after_<band>.tifcomposite surface reflectance, Float32NaN
before_rgb.tif, before_swir.tif (and after)true colour (red, green, blue); false colour (SWIR2, NIR, red)NaN
before_clear_count.tif (and after)clear observations, UInt1665535
<tag>_<INDEX>.tifspectral index, Float32NaN
fire_dnbr.tif, fire_rdnbr.tif, fire_rbr.tif×1000 scale, Float32NaN
fire_severity_classes.tifclasses 1 to 7, Byte0
flood_pre_db.tif, flood_post_db.tif, flood_log_ratio.tifdB, Float32NaN
flood_extent.tif0 dry, 1 new flood, 2 permanent water, Byte255
change_<INDEX>_<mode>.tif, change_<INDEX>_classes.tifchange; classes −1 / 0 / 1 (Int8)NaN; −128
series_<INDEX>.tifindex series, one band per date (band description = date)NaN
breaks_has_break.tif1 break, 0 none, Byte255
breaks_break_date.tif, breaks_magnitude.tif, breaks_rmse.tifdecimal year; median residual; history RMSENaN
polygons.gpkgtables burned_area and flood_extent: class, label, pixels, area_ha
report.htmlself-contained HTML report (KaTeX bundled, no external request)

Every GeoTIFF is a Cloud Optimized GeoTIFF (DEFLATE) whose metadata hold product, software, scenes, dates, sources, formula, reference, attribution and the step settings (thresholds, offset, windows), readable with gdalinfo or the QGIS layer properties.

9. Worked example: the demo

The demo is synthetic data, not Sentinel imagery: a 200 × 200 pixel grid at 10 m in EPSG:32635 whose top-left corner is at (500 000, 4 252 000), far from the projection origin. It holds 61 Sentinel-2 dates every 12 days from 6 January 2024, with a seasonal red-band cycle, sensor noise, cloud blobs with shadows and a few fully cloudy dates (all coded in SCL), a lake, and a burn scar on 15 July 2025 whose NBR drop is largest at its centre. Unburned land is given a planted NBR bias of 0.04 after the fire, that is 40 dNBR units, which the ring offset must recover. Four Sentinel-1 VV/VH dates on ascending orbit 131 (2, 14 and 26 July, 7 August 2025) have dry land at −8 dB and water at −20 dB with gamma speckle (L=4.4L = 4.4); the last date adds a flood rectangle and two 2-pixel specks. A water-occurrence raster gives the lake 95 %.

9.1 In the dock

  1. Area > Load demo. The baskets are filled with the fire windows: Before 2025-05-01 to 2025-07-14 (6 scenes), After 2025-07-16 to 2025-09-30.
  2. Composite > Build both composites (median, default mask).
  3. Fire > Map burn severity. The offset is 39.98 (planted 40); the burned area is 37.61 ha, matching the planted scar pixel for pixel; about 47 % of it is high severity.
  4. Flood > reference window 2025-07-01 to 2025-07-31, flood date 2025-08-07, Map the flood. Otsu on the split-based tiles gives about −13.7 dB; new flood is 23.91 ha (IoU 0.996 with the planted rectangle); the lake (7.5 ha) is permanent water; the specks are removed by the minimum mapping unit.
  5. Change > NBR difference with Otsu: the decrease class is the burn scar. Breaks on NDVI with history 2024-01-01 to 2025-06-30 and monitoring 2025-07-01 to 2025-12-31 find a break in every burned pixel and in 0.02 % of the others, dated at the first post-fire observation; Probe a pixel inside the scar shows the seasonal fit and the break line.
  6. Report > Write the report.

9.2 In Processing

  1. Demo data: returns the Sentinel-2 and Sentinel-1 scene tables, the area (400 ha), the occurrence raster and the windows FIRE_PRE 2025-05-01/2025-07-14, FIRE_POST 2025-07-16/2025-09-30, HISTORY, MONITOR, FLOOD_PRE and FLOOD_POST.
  2. Burn severity with the area, the Sentinel-2 table and the fire windows: OFFSET 39.98, BURNED_HA 37.61.
  3. Flood extent with the Sentinel-1 table, the flood windows and the occurrence raster: FLOODED_HA 23.91, PERMANENT_HA 7.47.
  4. Change study with all five steps and a report path runs the same chain through the staged pipeline; its burned area equals the single Burn severity run.

The figures above are those the test suites assert on QGIS 3.44 LTR and 4.2.

10. Limitations

11. References

Methods and data

  1. Bazi, Y., Bruzzone, L., Melgani, F. (2005) An unsupervised approach based on the generalized Gaussian model to automatic change detection in multitemporal SAR images. IEEE TGRS 43(4), 874-887. doi:10.1109/TGRS.2005.845626
  2. ESA (2015) Radiometric Calibration of S-1 Level-1 Products Generated by the S-1 IPF. ESA-EOPG-CSCOP-TN-0002.
  3. Flood, N. (2013) Seasonal composite Landsat TM/ETM+ images using the medoid. Remote Sensing 5(12), 6481-6500. doi:10.3390/rs5126481
  4. Holben, B.N. (1986) Characteristics of maximum-value composite images from temporal AVHRR data. IJRS 7(11), 1417-1434. doi:10.1080/01431168608948945
  5. Key, C.H., Benson, N.C. (2006) Landscape Assessment (LA). FIREMON, USDA Forest Service RMRS-GTR-164-CD, LA 1-55.
  6. Kittler, J., Illingworth, J. (1986) Minimum error thresholding. Pattern Recognition 19(1), 41-47. doi:10.1016/0031-3203(86)90030-0
  7. Lee, J.-S. (1980) Digital image enhancement and noise filtering by use of local statistics. IEEE TPAMI 2(2), 165-168. doi:10.1109/TPAMI.1980.4766994
  8. Main-Knorn, M. et al. (2017) Sen2Cor for Sentinel-2. Proc. SPIE 10427, 1042704. doi:10.1117/12.2278218
  9. Martinis, S., Twele, A., Voigt, S. (2009) Towards operational near real-time flood detection using a split-based automatic thresholding procedure on high resolution TerraSAR-X data. NHESS 9, 303-314. doi:10.5194/nhess-9-303-2009
  10. Miller, J.D., Thode, A.E. (2007) Quantifying burn severity in a heterogeneous landscape with a relative version of the delta Normalized Burn Ratio (dNBR). RSE 109, 66-80. doi:10.1016/j.rse.2006.12.006
  11. Montero, D. et al. (2023) A standardized catalogue of spectral indices to advance the use of remote sensing in Earth system research. Scientific Data 10, 197. doi:10.1038/s41597-023-02096-0
  12. Otsu, N. (1979) A threshold selection method from gray-level histograms. IEEE TSMC 9(1), 62-66. doi:10.1109/TSMC.1979.4310076
  13. Parks, S.A., Dillon, G.K., Miller, C. (2014) A new metric for quantifying burn severity: the Relativized Burn Ratio. Remote Sensing 6(3), 1827-1844. doi:10.3390/rs6031827
  14. Pekel, J.-F., Cottam, A., Gorelick, N., Belward, A.S. (2016) High-resolution mapping of global surface water and its long-term changes. Nature 540, 418-422. doi:10.1038/nature20584
  15. Twele, A., Cao, W., Plank, S., Martinis, S. (2016) Sentinel-1-based flood mapping: a fully automated processing chain. IJRS 37(13), 2990-3004. doi:10.1080/01431161.2016.1192304
  16. UN-SPIDER Knowledge Portal, Recommended Practices: Burn severity mapping; Flood mapping with Sentinel-1.
  17. Verbesselt, J., Zeileis, A., Herold, M. (2012) Near real-time disturbance detection using satellite image time series. RSE 123, 98-108. doi:10.1016/j.rse.2012.02.022
  18. Zhu, Z., Woodcock, C.E. (2014) Continuous change detection and classification of land cover using all available Landsat data. RSE 144, 152-171. doi:10.1016/j.rse.2014.01.011

Spectral indices

  1. Rouse, J.W., Haas, R.H., Schell, J.A., Deering, D.W. (1974) Monitoring vegetation systems in the Great Plains with ERTS. Third ERTS Symposium, NASA SP-351, 309-317. https://ntrs.nasa.gov/citations/19740022614
  2. Huete, A., Didan, K., Miura, T., Rodriguez, E.P., Gao, X., Ferreira, L.G. (2002) Overview of the radiometric and biophysical performance of the MODIS vegetation indices. Remote Sensing of Environment 83, 195-213. 10.1016/S0034-4257(02)00096-2
  3. Huete, A.R. (1988) A soil-adjusted vegetation index (SAVI). Remote Sensing of Environment 25, 295-309. 10.1016/0034-4257(88)90106-X
  4. Gitelson, A., Merzlyak, M.N. (1994) Spectral reflectance changes associated with autumn senescence of Aesculus hippocastanum L. and Acer platanoides L. leaves. Journal of Plant Physiology 143, 286-292. 10.1016/S0176-1617(11)81633-0
  5. McFeeters, S.K. (1996) The use of the Normalized Difference Water Index (NDWI) in the delineation of open water features. International Journal of Remote Sensing 17, 1425-1432. 10.1080/01431169608948714
  6. Xu, H. (2006) Modification of normalised difference water index (NDWI) to enhance open water features in remotely sensed imagery. International Journal of Remote Sensing 27, 3025-3033. 10.1080/01431160600589179
  7. Feyisa, G.L., Meilby, H., Fensholt, R., Proud, S.R. (2014) Automated Water Extraction Index: a new technique for surface water mapping using Landsat imagery. Remote Sensing of Environment 140, 23-35. 10.1016/j.rse.2013.08.029
  8. Gao, B.-C. (1996) NDWI - a normalized difference water index for remote sensing of vegetation liquid water from space. Remote Sensing of Environment 58, 257-266; Wilson, E.H., Sader, S.A. (2002) Detection of forest harvest type using multiple dates of Landsat TM imagery. Remote Sensing of Environment 80, 385-396. 10.1016/S0034-4257(96)00067-3
  9. Key, C.H., Benson, N.C. (2006) Landscape Assessment (LA): sampling and analysis methods. In: Lutes, D.C. et al., FIREMON: Fire Effects Monitoring and Inventory System. USDA Forest Service RMRS-GTR-164-CD, LA 1-55. https://www.fs.usda.gov/research/treesearch/24066
  10. Key, C.H., Benson, N.C. (2006) FIREMON Landscape Assessment; USGS Landsat Normalized Burn Ratio 2 product guide. https://www.usgs.gov/landsat-missions/landsat-normalized-burn-ratio-2
  11. Zha, Y., Gao, J., Ni, S. (2003) Use of normalized difference built-up index in automatically mapping urban areas from TM imagery. International Journal of Remote Sensing 24, 583-594. 10.1080/01431160304987
  12. Rikimaru, A., Roy, P.S., Miyatake, S. (2002) Tropical forest cover density mapping. Tropical Ecology 43, 39-47. http://tropecol.com/pdf/open/PDF_43_1/43104.pdf
  13. Hall, D.K., Riggs, G.A., Salomonson, V.V. (1995) Development of methods for mapping global snow cover using moderate resolution imaging spectroradiometer data. Remote Sensing of Environment 54, 127-140. 10.1016/0034-4257(95)00137-P
  14. Kawamura, M., Jayamanna, S., Tsujiko, Y. (1996) Relation between social and environmental conditions in Colombo, Sri Lanka and the urban index estimated by satellite remote sensing data. International Archives of Photogrammetry and Remote Sensing 31(B7), 321-326. https://www.isprs.org/proceedings/XXXI/congress/part7/321_XXXI-part7.pdf

12. Citation, licence, links

Eminoğlu, Y. (2026). 02SentinelChange - Sentinel Change Lab (version 0.1.1) [QGIS plugin]. https://gitlab.com/geophilo1/zero2sentinelchange. https://doi.org/10.5281/zenodo.24263656. The DOI 10.5281/zenodo.24263656 always resolves to the latest version; version 0.1.1 has DOI https://doi.org/10.5281/zenodo.24263657. ORCID 0009-0005-6000-2934, Dokuz Eylül University. Machine-readable metadata: CITATION.cff.

Licence: GPL-2.0-or-later. KaTeX 0.16.21 (MIT) typesets the equations of this manual and of the report.

Repository: gitlab.com/geophilo1/zero2sentinelchange · Issues: work items · Homepage: geophilo.com/zero2sentinelchange