CAPCON Farming Help

CAPCON data algorithms on the inputs prepared in Inputs β€” crops, grasslands, forest.
← All apps

Crop lodging

What this panel is for

Lodging of cereals and rapeseed (group Crops) from the raw Sentinel-1 GRD passes, per relative orbit and per crop (v2). Works on the region selected in the sidebar - the same selection as in Inputs.

Before you start

parcels) with the Crop code (text) mapped.

algorithm** (PROJECTS β†’ Periods). Detect, Early detection and Delivery run on the crops recorded there: Crops is pre-set to them, a run with other crops is refused, and the job names the version it used. The panel shows the record under its description, or what is missing.

How to use it

Dekad pairs takes the older dekad of the first pair and the newer dekad of the last pair; the pairs between them are consecutive and each is confirmed by the next dekad (06d2 β†’ 07d2 gives 06d2β†’06d3, 06d3β†’07d1, 07d1β†’07d2, confirmed by 07d1, 07d2, 07d3).

  1. Prepare - once (hours); passes already in the cache are skipped.
  2. Detect - as often as needed, with other thresholds; give it the same

range as Prepare.

  1. Early detection - both steps for one pair right after it ends, without

a confirmation dekad (less accurate).

  1. Delivery - tick the results to hand over (every dekad pair with

results is listed: early = _er, final = _fi, with the number of crops) and the clients:

_isg_nma.gpkg (the lodged fields: parcel attributes, area_ha, dump_ha, dump_perc; one layer, named after the file);

All crops of a pair are merged into one file; Crops limits which. The name is a template, by default LIT_{year}_{n}crops_{older}_{newer} ({n} = number of crops). Early deliveries go to early/, final ones to final/; running again replaces them.

Output file name (Detect, Early detection) is a template; its default gives the former names. It must contain {crop} (and, in Detect, {older} and {newer}), otherwise the files of one run would overwrite each other. _er (early) or _fi (final) is always added; Delivery finds the results under the default template only.

What it writes

<results>/crop_lodging_v2/: per crop and pair <CROP>_lodging_<year>_<older>_<newer>_fi.tif (Detect) or _er.tif (Early detection), _fi_id.gpkg / _er_id.gpkg (fields) and _fi_patch.gpkg / _er_patch.gpkg (patches). Deliveries: <results>/crop_lodging_v2/final/ and early/, e.g. LIT_2026_8crops_06d2_06d3_isg_nma.tif, _isg_nma.gpkg, _isg_coe.tif. The cache is <inputs>/statistic/<year>/crop_lodging_<parcels file>/ - a new parcels delivery starts a new cache; the old one can be deleted.

When it goes wrong

"choose a period" β€” a Manager records the crops of this algorithm for the period first (PROJECTS β†’ Algorithms sets the defaults).

version recorded with them.

Soil moisture

What this panel is for

The SM index (group Crops) as the GeoTIFFs delivered to NMA: exported from the SM VRT mosaics, masked by KZS and colored.

Before you start

How to use it

  1. Export SM rasters - the SM GeoTIFFs (EPSG:3035) into

results/soil_moisture/; narrow the file pattern to export only some dates.

  1. KZS mask and colors - masks every exported file that has no masked copy

yet; the color table is optional (values are unchanged). Output file name is a template that must contain {name} (the SM file name); with a new template, files are masked again under the new names.

What it writes

<results>/soil_moisture/S1_SM_..._TC3035.tif (exported) and S1_SM_..._TC3035_kzs.tif (masked) in the same folder; the rasterized KZS mask is cached in <inputs>/masks/.

Vegetation indices

What this panel is for

SAR vegetation indices (group Crops) as standalone GeoTIFFs clipped to the region boundary, exported from the index VRT mosaics. For now the panel has RVI4S1; other SAR vegetation indices will be added here as their own export steps.

Before you start

(the RVI4S1 files are named S1_VI_...).

How to use it

  1. Export RVI4S1 rasters - the RVI4S1 GeoTIFFs (EPSG:3035) into

results/vegetation_indices/; narrow the file pattern to export only some dates or quartiles, e.g. S1_VI_D10_Q2_AD_2026_06_*.vrt. Existing outputs overwrite exports every matching mosaic again (needed when a mosaic gained tiles), skip only the new ones.

What it writes

<results>/vegetation_indices/S1_VI_..._TC3035.tif, one per exported VRT mosaic (index values x100, UInt16).

Classification

What this panel is for

Pixel-based crop classification (group Crops) from the Sentinel-1 Q2 composites. Works on the region selected in the sidebar - the same selection as in Inputs.

The classification is made seven times a season, once per dekad cut-off: 0601 uses the data of 04-01 to 06-01, 0611 uses 04-01 to 06-11, and so on up to 0801 (04-01 to 08-01). Each gives a classification raster and a probability raster.

Before you start

parcels) with Parcel ID (integer), Crop code (text), Crop code (number) and Area, ha mapped.

Crop statistics of the same parcels layer (Inputs β†’ Field Statistics): the D10 Q2 dekads from 04-01 up to the date and the M01 Q2 months before it.

algorithm** (PROJECTS β†’ Periods). the classification run on the crops recorded there: Classes (crop codes) is pre-set to them, a run with other crops is refused, and the job names the version it used. The panel shows the record under its description, or what is missing.

How to use it

exist; a date whose statistics are incomplete says how many files are missing (it is skipped). The dates that can run and have no rasters yet are preselected.

crop models, or the default set (the 17 field crops) before any model exists. The other crop codes of the parcels layer, most frequent first and with their parcel count, are folded under more - open it and tick a code: it moves up to the ticked ones at once (no separate add); untick one and it goes back under more. Sort: by count (most parcels first) or A–Z, for both lists.

and pixel models; improved+standard also writes the standard (parcel model alone) rasters _clf_s.tif / _proba_s.tif.

resampling). The file names do not say the CRS.

After a new parcels layer (and its statistics): run the dates again with Existing outputs = overwrite. New parcel rasters, models and predictions are made for the new layer; the old models stay.

What it writes

sf_<year>_d_<M>_<DD>_clf.tif (crop number, uint16) and _proba.tif (highest probability x100, uint8), nodata 0.

<REGION>_<year>_<MMDD>_crops_<parcel|pixel_s>_<parcels version>/ - model, features, classes, importance, info.json. (Grasslands β†’ Classification keeps its ..._grassland_... models in the same folder.)

predictions per parcels version (a cache; may be deleted). The parcel rasters and the pixel sample are shared with the grassland classification.

When it goes wrong

"choose a period" β€” a Manager records the crops of this algorithm for the period first (PROJECTS β†’ Algorithms sets the defaults).

version recorded with them.

Mowing

What this panel is for

Grassland mowing v1 (group Grasslands): per S2 date, the median S2 'pic' color of the grassland parcels gives mowed / not-mowed reference parcels, and an ExtraTrees classifier trained on them over the S1 VV/VH dekad statistics gives every grassland parcel the q1/q2/q3 probability of being mowed.

The Parcel catalog step builds PDFs of the S2 images of parcels you name, to look at them by eye.

Before you start

Sentinel-2 RGB maps, <year>/<CC>/ (e.g. 2026/LIT/), with a _pic_thumb.png preview beside each image. The parcel catalog also uses the cloud masks of the S2 store (<CC>_<year>/merged_scl/); without them it shows every image, unchecked for cloud. Only the region's own country is offered: a region named NMA_LIT_2026 sees LIT_2026 only. A region whose name holds no country code sees every country's folder.

How to use it

Mowing is computed per dekad. Choose the dekad (only dekads with S2 images are listed); the previews of its images appear below. Click a preview to see it full size (click again or Esc to close) and judge its quality, then tick the images to use. R = the mowing reference of that date exists, C = it is already classified. All, Not classified yet and Clear set the ticks of the shown dekad.

One date's image is often a single satellite swath. Under each preview the share of the region it has data for is shown; an image with less than 5 % is dimmed as outside the region and left out by All / Not classified yet - it gives no mowing reference. If a ticked date still has too few reference parcels (fewer than 6 mowed or 6 not mowed, e.g. clouded over the region), the classifier skips it and names it at the end of the log; the run fails only if no date could be classified.

Step 1 builds the S2 color reference of each ticked date, Step 2 trains and applies the classifier; the classifier needs the reference of the same date, so run both unless the reference is already there.

Output file name is a template that must contain {q} and {date}. The default is S1_{parcels_date}_S2_{date}{q}_PV_{n}, e.g. S1_0801_S2_0903_q2_PV_90_pred_proba.csv; PV_{n}{q}_S1_{parcels_date}_S2_{date} gives the former names. The C mark recognises both default forms, not a name of your own.

Parcel catalog: type the parcel codes (the Parcel ID values, space or comma separated), choose the months and the catalog folder name. One PDF per parcel: the S2 crops around the parcel (boundary burned in) by date; images where the parcel is outside the image or more than Max cloud share covered by cloud/shadow (SCL) are left out.

What it writes

(plus ..._PV_<n>_S1_oob_scores.csv and ..._S1_drops.csv); the color statistics and reference files in <inputs>/statistic/<year>/mowing/.

(default folder mowing_catalog).

Grassland classification

What this panel is for

Pixel-based classification of the grassland crops (group Grasslands): grasses, legumes, pastures and the other crops they are told apart from. The method, the inputs and the options are those of Crops β†’ Classification; only the classes, the results folder and the model names differ. Works on the region selected in the sidebar.

Seven classifications a season, once per dekad cut-off: 0601 uses the data of 04-01 to 06-01, 0611 uses 04-01 to 06-11, and so on up to 0801 (04-01 to 08-01). Each gives a classification raster and a probability raster.

Before you start

The same as for Crops β†’ Classification: a parcels layer (kind parcels) with Parcel ID (integer), Crop code (text), Crop code (number) and Area, ha mapped, the S1 Q2 composites and the Crop statistics of that layer.

algorithm** (PROJECTS β†’ Periods). the classification run on the crops recorded there: Classes (crop codes) is pre-set to them, a run with other crops is refused, and the job names the version it used. The panel shows the record under its description, or what is missing.

How to use it

statistics are incomplete says how many files are missing. The dates that can run and have no rasters yet are preselected.

grassland models, or before any model exists the default set DGP, KVΕ½, RAΕ½, KVV, AVV, MIV, Ε½IR, PUP, MIΕ½, PDJ, GRI, DOB, RUΕ½, LIC, GAB, LUB, RAV, FAC, BAR, KMY, KAN, LIN (those found in the parcels layer). The other crop codes of the layer are folded under more - open it and tick a code: it moves up to the ticked ones at once; untick one and it goes back. Sort: by count (most parcels first) or A–Z.

as in Crops β†’ Classification.

After a new parcels layer (and its statistics): run the dates again with Existing outputs = overwrite.

What it writes

sf_<year>_d_<M>_<DD>_clf.tif (crop number, uint16) and _proba.tif (highest probability x100, uint8), nodata 0.

<REGION>_<year>_<MMDD>_grassland_<parcel|pixel_s>_<parcels version>/.

are shared with the crop classification (computed once per parcels version); the tile predictions are in prediction/grassland_<MMDD>_....

When it goes wrong

"choose a period" β€” a Manager records the crops of this algorithm for the period first (PROJECTS β†’ Algorithms sets the defaults).

version recorded with them.

Deforestation

What this panel is for

Forest cuts (group Forest) from the drop of the S1 backscatter between two dekads or two months, inside the forest-stand mask: a 0-100 change score, above 50 usually an actual cut.

Before you start

also VRT mosaics - S1 composites).

How to use it

The period is a small table: year / month / dekad columns, the older period in the first row and the newer one in the second. The thresholds are the former script's defaults.

Output file name is a template; its default gives the former names. Use the placeholders shown under it (e.g. cuts_{year2}{month2}{dekad2}); Default restores it. .tif is added.

What it writes

<results>/deforestation/<period>_<region>_diff_proc_rem<threshold>.tif; the rasterized forest mask is cached in <inputs>/masks/.

Jobs

What this panel is for

Every Farming run of the selected region, newest first - the way back to a run started from a page that has since been closed. (Inputs β†’ Jobs lists them too, with the SAR and statistics runs.)

How to use it

stopped from the console header.

used (PROJECTS β†’ Periods) under its step name; job.json keeps it as record.

What it writes

Each run is a folder <regions base>/<REGION>/logs/pipelines/<job id>/ with job.json, output.log and rc (exit code).