SmartAG · Plaisia Field Lab
MSc research · 2026–2027 · Ελληνικά
Precision agriculture · Hellenic Open University

Monitoring an experimental rainfed pasture from space and from the field

A 0.8 ha mountain pasture at Plaisia (Malakasiou), North Tzoumerka, Greece, is being set up as a field laboratory to test pasture, tree and crop land uses and to monitor them with satellite vegetation indices.

Plot area8,117 m²
Centre (WGS84)39.4865 N · 20.9690 E
Current useNatural pasture
WaterRainfed
Sentinel-2 pixels62 pure / 82 >50%
01 · Site

The plot and its satellite footprint

The plot is a five-sided parcel of about 77 × 105 m on a rainfed grassland used for grazing. It has no irrigation, sensors or weather station yet, which makes the current season a clean baseline before any treatment is applied.

The plan shows the parcel on the Sentinel-2 10 m pixel grid (UTM zone 34N). Only pixels that fall completely inside the boundary are free of mixing with neighbouring land. The zones will be aligned to this grid so that each treatment is observed by whole pixels.

Boundary held in EGSA87 (EPSG:2100) and WGS84; grid origin 497270 E, 4370820 N.

0 1 2 3 4 20 m N
Parcel boundary (vertices 0–4) over the Sentinel-2 10 m grid. Grid north; UTM 34N.
02 · Experiment

Field zoning and treatments

The field is divided into eight equal zones of about 1,000 m² each. Which pasture or land-use treatment goes to each zone will be decided after the multi-year baseline analysis and a soil survey, so that soil and moisture differences do not bias the comparison. The first treatments are planned for 2027.

CodeCandidate treatmentPurpose
CNatural pasture, untreatedControl and continuation of the baseline series
VOCommon vetch + oat forage mixtureHigh-biomass rainfed forage; nitrogen fixation by the legume
PNatural pasture + phosphorus fertilisationImprovement without tillage
CLClover oversowing into the existing swardLegume enrichment with minimal disturbance
SPSilvopasture: pasture with widely spaced treesShade, fodder and timber while keeping grazing
TTimber trees (e.g. walnut, wild cherry)Long-term timber for private use; subject to forestry service advice
VGSmall vegetable plots (possible)Intensive use on the best soil, if the baseline supports it

Each zone is followed through NDVI, NDRE and NDMI time series, read together with soil analyses and an on-site weather station. A 1,000 m² zone holds only about ten Sentinel-2 pixels (10 m); if data access is granted, PlanetScope (3 m) would give about a hundred, and the two are compared for how well they separate the zones and track the establishment of new pasture and young trees.

03 · Platform

SmartAG: a self-hosted data pipeline

The data for the plot are being brought together in SmartAG, an open-source monitoring stack that runs on the researcher's own server. Each source is collected on a schedule, stored per pixel and per day, and analysed alongside the field records. Some components, such as PlanetScope ingestion and alerts, will be added as the experiment starts.

The data will be followed live in Home Assistant, which runs on a separate machine and connects to the server over an encrypted SSL link, together with an OpenClaw AI agent (ChatGPT via OAuth). The agent sends daily, weekly and monthly reports, and an alert whenever extreme weather is forecast or recorded at the plot.

Sources
  • Sentinel-2 L2A, since 2017
  • Sentinel-1 radar backscatter
  • PlanetScope (requested)
  • Daily weather since 1991 (ERA5 reanalysis) and an on-site weather station (planned)
  • SoilGrids and laboratory soil analyses
  • Field records in farmOS
Collection
  • Daily scheduled collectors
  • Cloud and shadow masking
  • Per-pixel extraction on the 10 m grid
Storage & analysis
  • PostgreSQL database
  • NDVI, NDRE, NDMI series
  • Seasonal climatology and anomalies
  • Within-field temporal stability
  • Zone-level comparisons
Outputs
  • Grafana dashboards
  • Live view in Home Assistant
  • AI reports: daily, weekly, monthly
  • Extreme-weather alerts
  • Treatment comparisons
  • Encrypted daily backups
04 · Data

The field in numbers

The charts are refreshed automatically every night from the SmartAG database. Only open data from the Copernicus programme and the ERA5 reanalysis are shown.

Vegetation indices, 2017 to date

Mean of pure pixels per clear Sentinel-2 acquisition. NDVI tracks vegetation density, NDRE chlorophyll and nitrogen status, and NDMI canopy water content. Drag or scroll to zoom. Seven 2022 acquisitions with physically impossible values, caused by a calibration error, were excluded.

Seasonal course

This year and last year against the 2017–2025 median and interquartile range, by ten-day period.

Within-field variability

Mean relative deviation of each 10 m cell’s NDVI from the field mean, over all clear acquisitions. Hover over a cell for its value.

Annual precipitation and temperature, 1991 to date

The current year is shown faded because it is incomplete. The dashed line is the mean of the climate reference period.

Monthly climate 1991–2020

Mean monthly precipitation, reference evapotranspiration ET₀ (FAO-56) and mean temperature.

Sentinel-1 radar, monthly means

VV and VH backscatter and the radar vegetation index (RVI). Radar sees through cloud and responds to vegetation structure and soil moisture.

Water balance and soil moisture

Annual difference between precipitation and reference evapotranspiration (P − ET₀) and the summer minimum of soil moisture at 7–28 cm, from the ERA5 reanalysis. These are model estimates, not on-site measurements.

Heat stress and frost

Days per year with maximum temperature above 30 °C and frost days, with the linear trend over the period.

Soil texture by depth

Estimate from the SoilGrids 2.0 model (ISRIC) at 250 m resolution. It will be replaced by laboratory analyses of the samples from each zone.
Statistical panel

Quality, uncertainty and trends

These indicators complement the charts with quality checks and exploratory statistics. They do not replace the final analysis by experimental zone.


Sources: Copernicus Sentinel-2 L2A (ESA) · ERA5 (ECMWF / Copernicus Climate Change Service).

05 · Timeline

Schedule

Nov–Dec 2026Baseline: Sentinel-2 and Sentinel-1 history since 2017, weather history, soil data; soil sampling
Jan–Feb 2027Zone layout on the satellite grid, marking with GPS, seed and fertiliser, weather station, soil preparation
Spring 2027Establishment of the first treatments
Apr–Jun 2027Monitoring of the zones: satellite time series, weather station, field photos
Jun–Sep 2027Analysis of the first season, Sentinel-2 vs PlanetScope comparison
06 · Data use

Data and contact

All satellite data are used exclusively for non-commercial academic research and education within the MSc programme Precision Agriculture Applications (EGA) of the Hellenic Open University and the University of Patras. Commercial imagery is not redistributed; only derived indices and results appear in course work and publications.

Thomas Sopilidis

MSc student, Hellenic Open University

Click to reveal

Programme page: eap.gr – Precision Agriculture Applications

NDVI comparison: Planet and Sentinel-2, with rainfall, temperature and soil moisture

Plot-mean NDVI per date from clear (cloud-free) pixels. Rainfall, temperature and soil moisture: ERA5-Land (daily means, 9 km). Only aggregate values are shown.

Planet Team (2026). Planet Application Program Interface: In Space for Life on Earth. San Francisco, CA. https://api.planet.com.