Farmer — overview

Farmer monitors crop phenology (the growth cycle) on CAP-declared agricultural parcels using Sentinel-1 SAR time series. Pick a parcel and the app shows its backscatter curves with automatically-detected crop stages — sowing, canopy closure, growth, senescence, harvest — plus soil-moisture and vegetation indicators.

Typical workflow

1. Project: choose the country/year dataset.
2. Crop groups: filter the map to the crop types you care about.
3. Find a field: jump straight to a parcel by its CAP field ID, or click one on the map.
4. Parcel detail: open the popup and read the tabs — Phenology, Vegetation, Moisture, Insights.

Why SAR

Sentinel-1 radar sees through cloud and works day or night, so it delivers a regular, gap-free time series even in cloudy seasons — ideal for tracking growth stage timing across a whole country's parcels.

New to the radar terms? Start with the Concepts & glossary — CAP, VV/VH, RVI, DPA and Savitzky-Golay are all explained there.

Project

A project is one country-and-year set of CAP parcels (for example, Lithuania 2025).

Selecting a project

Pick one from the Project dropdown. The app then:

• loads the parcel boundaries for that dataset,
• zooms the map to its extent,
• fills the crop-group filter and the parcel list,
• shows summary metadata (total fields, total area, year, source).

Switching project resets the map, filters, and any open parcel popups.

Note: At low zoom the map shows parcel centroids (dots) for speed; zoom in past level 13 to see the full parcel polygons.

Find a field

Use Search Field ID to jump directly to a parcel by its CAP declaration identifier.

Steps

1. Type the field_id into the box.
2. Click Go.
3. On success the map zooms to that parcel, highlights it, and opens its detail popup.

The status line shows a spinner while it looks up, a green confirmation on success, or a red message if the ID isn't found in the current project.

Or just click the map

You don't have to search — click any parcel polygon (or a row in the parcel list) to select it and open the same detail popup.

Tip: Field IDs are specific to the selected project/year. If a lookup fails, check you're in the right project first.

Crop groups

The Crop Groups panel filters the map to the crop types you want to see. Each pill is one group, coloured to match its parcels on the map, and shows the parcel count and area for the current project.

Using the filter

• Click a pill to toggle that crop group on or off.
• ALL / NONE quickly select or clear every group.
• The map, the legend, the parcel list, and the header summary all update to match.

The ten groups

Winter Cereal · Winter Oil · Spring Cereal · Spring Oil · Legume · Row Crop · Forage Grass · Permanent Grass · Fallow · Other.

Tip: Isolating a single crop group (NONE, then one pill) makes it much easier to compare phenology across many parcels of the same crop.

Parcel detail

Selecting a parcel opens a draggable, resizable detail popup. Its header shows the field ID, crop name, and area; the body has six tabs.

The tabs

The Overview tab

• Attributes — the parcel's CAP details in a two-column grid.
• Backscatter chart — VV (red, left axis) and VH (blue, right axis) over time, each with its smoothed curve.
• Weather — optionally overlay temperature, soil temperature, precipitation, and a 0 °C reference line to relate signal changes to conditions.

You can open several parcel popups at once to compare fields side by side. Press Escape or the × to close.

Phenology tab

The PHENOLOGY tab is the heart of the app: the VH/VV time series with the crop's growth stages detected automatically and marked on the timeline.

Controls

• Orbit filter — ALL / ASC / DESC to view all passes or one direction (cleaner curves come from a single direction).
• Legend toggles — show/hide VH/VV, the S-G smoothed curve, dVH/dt (growth rate), and METEO (weather overlay).

The growth-rate band

A coloured band under the chart classifies each period:

Event markers

The event summary lists the detected stages with their dates.

Tip: If the markers look noisy, switch the orbit filter to a single direction and turn on the S-G smoothed curve — stage transitions become much clearer. See Concepts for what dVH/dt and S-G mean.

Vegetation tab

The VEGETATION tab plots the RVI (Radar Vegetation Index, 0–1) over the season, with a moving average and growth-stage background bands.

Reading it

• RVI near 0 = bare soil or very sparse cover.
• RVI rising = the crop is building biomass.
• RVI at its peak = full canopy.
• RVI falling = senescence or harvest.

The background bands shade the chart from light (early/bare) to deep green (peak biomass) so the stage is readable at a glance.

Event dots

Stress and anomaly events are marked as dots — for example frost, drought, or waterlogging signatures detected in the series.

Tip: Compare the RVI peak timing here with the harvest marker on the Phenology tab — a sharp RVI drop should line up with the detected harvest date.

Moisture tab

The MOISTURE tab shows DPA — a soil-moisture estimate (0–1) derived from VV backscatter and normalised by RVI so that vegetation doesn't masquerade as wet soil.

Reading the thresholds

Why RVI normalisation matters

Bare wet soil and dense vegetation can both raise backscatter. DPA weights the moisture reading by the vegetation index so a growing canopy isn't mistaken for rising soil moisture — making it more trustworthy once the crop is established.

Tip: Cross-check spikes against the weather overlay (precipitation) on the Overview or Phenology tab — a DPA rise right after rain is the expected pattern.

Insights tab

The INSIGHTS tab turns the curves into plain-language diagnostics — three reference tables that interpret what the signals mean together.

What it contains

• RVI stage classification — maps RVI ranges to growth stages (bare → establishment → growth → peak → senescence).
• dVH/dt diagnostics — reads the VH rate-of-change to flag rapid growth, stalls, stress, and decline.
• RVI × DPA matrix — combines vegetation and moisture into diagnostic scenarios (e.g. high RVI + low DPA = vigorous crop on drying soil; low RVI + high DPA = bare waterlogged ground).

How to use it

Read the Insights tab after skimming Phenology, Vegetation, and Moisture — it ties those three together and tells you the likely agronomic situation rather than leaving you to combine the charts yourself.

Note: These are SAR-derived indicators, not ground truth. Treat them as a screening aid that points to parcels worth a closer look, not as a final verdict.

Neighbours tab

The NEIGHBOURS tab answers one question: is this field doing what its neighbours are doing? It compares the parcel against every same-crop field within 5 km — up to 100 of them — on every Sentinel-1 acquisition of the season.

Because the comparison is relative, weather, orbit geometry and calibration are shared by the whole neighbourhood and largely cancel out. A divergence therefore points at this parcel — its management, stand or soil — rather than at the season at large.

What you see

• Peer fan — the distribution of the neighbourhood on each date. The dark shade is the 25th–75th percentile, the lighter one the 10th–90th, and the dashed green line the median. A narrow fan means the neighbourhood agrees; a wide one means it does not.
• This field — a single line coloured by its percentile rank among its peers that date, from red (lowest of the neighbourhood) through yellow and green to blue (highest).
• Rank track — the strip under the axis repeats that rank for every acquisition of the year, so the season reads as one fingerprint. Dark cells are dates with no data for this field.
• Episode spans — coloured columns with a solid header bar mark every spell of two or more consecutive acquisitions outside the 10–90% band.
• VH/VV ↔ RVI — switch the index; both are computed in the same request, so the toggle is instant.

The crop development phase

Assessment is confined to this field's own growing period, drawn as the bright band between two dashed boundaries. Dates outside it are dimmed and carry no verdict.

Both ends come from the Phenology detectors, so the two tabs can never disagree:

The label next to the heading always names which one applied — green for a detected event, grey for a calendar estimate.

Nothing after harvest is assessed. Once the crop is off, the parcel has entered its next cropping cycle — stubble, a catch crop, bare ground — and what stands there is unknown to this season's comparison. Those acquisitions stay on the chart, dimmed, and the panel says how many were excluded.

The verdict

The headline is scored on the deficit — the side that threatens yield — across five independent axes, plus two modifiers:

Scoring the shape and not just the average matters: a field that spends a third of the phase under the 10th percentile and another third over the 90th averages out to "typical" while actually having two significant excursions. The score only ever escalates the level reading, never softens it, and the card lists the drivers behind it.

Two deliberate exceptions keep the verdict honest:

• A deficit lying entirely after the peer canopy peak is capped at Watch. That pattern is a maturity-timing difference — this field ripening or being cut ahead of the neighbourhood — not lost biomass.
• Being above the band is reported separately, since vigour is not a deficit. A long spell above it after the peers' canopy peak is flagged as delayed ripening rather than extra yield.

Extreme episodes

Below the chart, each significant spell gets a card explaining what may have happened. Four discriminators drive the reading:

1. Did the peer median move too? Only this field moving means something field-specific. Peers moving away while this field held level is the "peers pulled ahead" signature; peers collapsing while this field stands on usually means they were harvested first.
2. Abrupt step or gradual drift? Biomass cannot change in one revisit, so a step is management or damage — mowing, lodging, tillage, harvest. A drift over several acquisitions is growth.
3. Which crop phase? The same calendar windows Phenology uses, so a drop inside the harvest window reads differently from one during canopy growth.
4. Rain and ET₀ in the fortnight before onset — for a moisture-plausibility note, with the caveat that rain is shared across the neighbourhood, so the field-specific part would be soil depth or drainage.

Each card also carries a cross-check: what to look at in another tab, or what to scout in the field.

Note: These are candidates ranked by signature, not diagnoses. SAR ranks a field against its peers; it does not tell you why. Episodes need at least two consecutive acquisitions (~12 days) precisely so that one speckly date cannot become an "event".

Performance

The band is computed server-side from a single areal load over the neighbourhood — not one request per field — and it starts only after the field's own VV/VH sampling has finished, so opening a parcel is never slowed down by the peer analysis. Progress appears in the shared tasks panel at the top right, and the result is cached until the popup is closed.

Damage tab

The DAMAGE tab looks for physical crop damage on this parcel: water standing in the field, and a crop flattened by a storm. It is a vertical report — one section per damage type, each with its own evidence — so further categories (drought, frost bite, non-germination) slot in without changing how it reads.

The detectors come from the ESA AGRISK project (Agriculture risk assessment based on Sentinel data), where they run over 10-day country mosaics. Here they run over the per-acquisition time series, which buys exact event dates and a "floating" assessment: the newest 10-day window always describes the present, rather than waiting for a decade to close.

Thresholds are not yet calibrated against known damaged parcels. Treat extents as indicative and scout the field before acting.

Reading a section

Each section is three columns: the report text on the left, and on the right a line graph above a raster thumbnail.

• Line graph — how much of the parcel is affected on every acquisition of the phase. Both curves are median-level or harder; the most permissive class each detector produces is deliberately kept off the graph, because it overstates the picture.
• Raster thumbnail — the classified pixels of the worst window, drawn in true metric proportions with the parcel boundary in light grey. Click it for the full-size map, the share of the parcel in each class, and what the window means.

Flooding & waterlogging

Surface water is read from the polarisation ratio VV/VH together with a direct VH test. At C-band a smooth water surface reflects the signal away from the satellite, so VH collapses below about −23 dB where vegetation is submerged.

The quartiles of the last 10 days then separate how long water stood, not just how far it spread:

The graph plots Q2 and Q1, so what you see is water that recurred or persisted.

The erect-cereal false positive

A high VV/VH ratio has two completely different causes: water, and a crop of vertical stems. Winter wheat and barley before heading give strong VV double-bounce and almost no VH volume scattering — a ratio higher than shallow water produces. The ratio alone cannot separate them; the absolute VV level can, so pixels brighter than about −10 dB in VV are read as canopy structure, not water. Without that gate the panel reported flooding on winter barley into late June.

What this loses: flooded woody vegetation can show enhanced VV from double bounce between the water surface and standing stems. That is mostly a forest and reed effect — herbaceous crops under water go specular and dark.

Storm lodging

A flattened canopy loses its vertical structure, so VV drops sharply between one 10-day median and the next. But a VV drop alone fits four different causes, and the direction of VH separates them:

The cross ratio VH/VV cannot do this on its own: if both fall in proportion, the ratio is unchanged and flooding passes a ratio-only test as though it were lodging. So the gate is on VH itself, and the ratio is reported as secondary evidence.

Three further requirements before a candidate is counted:

• A canopy to flatten — the cross ratio before the event must be above the bare-soil boundary.
• Both orbit directions agree. ASC and DESC see the field at different incidence angles, and a change of orbit mix alone can produce a 1–2 dB step. Lodging is structural and shows from both look directions, so the two are compared separately and the stricter result is taken.
• Irreversibility. A flattened crop stays down until harvest. If VV climbs back within three weeks, it was a wet canopy or a moisture swing, and the candidate is rejected. A fresh event with no follow-up acquisitions yet is reported as unconfirmed rather than hidden.

Nothing before 25 May is assessed or plotted, and the detection floor sits at 15 June: before then the crop is still soft, bending and springing back rather than staying down.

Wild-animal damage

Boar trampling and deer bedding flatten a crop the same way and produce the same VV ↓ / VH ↑ signature, so the radiometry cannot separate them from a storm. The rate can. A storm acts overnight; animals return to a field for weeks, so the decline arrives in steps. When most of the decline lands in a single revisit it is read as a storm; when it accumulates across several on a crop animals favour — potatoes, maize, sugar beet, winter rapeseed — it is reported separately as a candidate, and kept out of the lodging figures.

Two pieces are still missing, and the panel says so: distance to broad-leaved or mixed forest, the strongest cue, needs a forest layer the platform does not hold; and winter rapeseed is grazed through the winter months, entirely outside the crop-development phase assessed here.

The assessed period

Like Neighbours, assessment is confined to this field's own growing period — canopy closure to detected harvest, from the Phenology detectors. Nothing after harvest is graded: the parcel has entered its next cropping cycle.

One consequence worth knowing: the mapped window is the one with the most classified area, which is not always a counted event. Where the largest area falls in a window that failed one of the tests above, the map still shows it — that is the raster you need to see — and the full-size view states which test it failed and why it is not counted.

Limits

• 10 m pixels. A 9 ha field is only about 30 × 30 pixels; one pixel is 0.01 ha. Class edges are drawn hard rather than smoothed, because interpolating between classes would invent readings.
• One parcel at a time. The scan is per-parcel by design.
• Damage is detected, not attributed. Lodging and animal damage are separated by rate, not by evidence of a storm; hail that both floods and flattens a soft crop is reported as waterlogging plus a VV drop.

Maps tab

The MAPS tab paints two false-colour pictures of this field's spring from one server-side read of the Sentinel-1 archive, and gives you a full-size viewer with a supervised classification tool for splitting the field into thematic zones.

Both pictures use a country-wide, agriculture-masked colour balance (a GRASS i.colors.enhance stretch pooled over all Lithuanian fields, applied unchanged to every parcel). That means a hue means the same thing on any field — winter cereals read their true dark blue everywhere, not a per-field relative colour.

The two maps

The white line is the declared parcel boundary. Under each map, a table lists the per-channel scene counts and the stretch range actually on screen.

Controls (top of the tab):

• Polarisation — VV or VH. VV carries surface structure; VH carries volume scattering from the canopy.
• Filter — original or a 3×3 median smoothing, applied before the colour stretch to quiet speckle.

The 6-5-4 map needs a median for all three months; if one month has no usable coverage the map is withheld (the intensity map, which pools every date, still renders).

The full-size viewer

Click any map to open it full size. The controls at the top-right of the picture:

• + / − zoom, the arrow pad to pan, ⌂ to reset the view.

Zoom and pan only change what you see — classification always works on the whole field, including pixels currently off-screen.

Supervised classification

The tool splits the field into thematic classes by the on-screen brightness of the three composite bands. Every in-field pixel belongs to exactly one class (or none); classes are shown in a legend on the right, each with its colour, name, and its share of the field as % and hectares.

Automatic classification

1. Press Auto classify. The field is banded by value into classes, coloured on a health ramp — the largest class bright green (healthy), the smallest brown (wasteland) — with any class covering under 5 % of the field folded into a grey Other bucket pinned to the bottom.
2. Drag Auto width to change the detail, live:

• Narrower → finer classes. Each class subdivides; the extra pieces appear as lighter-shaded sub-classes indented beneath their parent.
• Wider → coarser classes. The smallest class is absorbed into its nearest larger neighbour, then the next smallest, and so on. Survivors keep their exact colour and name — classes only ever disappear by merging, never swap colours.

1. Press Apply to keep the result (the slider stops rebuilding it), or Undo to reverse. Apply and Undo are one button that flips between the two.

Refining a class

• Add class splits the selected class into a dense core and a sparser residual sub-class. Adjust where the split falls with Auto width, then Apply to keep both sub-classes or Undo to cancel.
• Click a pixel to assign every similar in-field pixel to the active class — a manual override of the automatic result.
• Each legend row lets you pick a colour, rename the class, ⊘ clear just its pixels, or Delete class (its pixels fall back into Other).
• Show hides or shows the whole class layer so you can check it against the imagery underneath.

Exporting

Export PNG renders the whole field — composite, class overlay, boundary, and a legend of class names with their area and percentage — into one image and downloads it. The classes live only in this window; export before you close it if you want to keep them.

Classification runs on the displayed composite, so it inherits that map's polarisation, smoothing and colour balance. Switch VV/VH or the filter before classifying to work from the picture you want.

Concepts & glossary

A short reference for the terms used throughout Farmer.

A note on accuracy

Everything Farmer shows is derived from C-band SAR — excellent for timing and relative trends across the season, but not a calibrated physical measurement. Read the shapes and the dates, and use ground knowledge to confirm.