---
id: ch4-copernicus
type: spine
title: "Copernicus in JackDaw: from satellite pixels to usable layers"
readingTime: 2
standalone: true
core: false
teaser: "Sentinel imagery reaches a JackDaw user mostly through derived layers. Direct access to imagery and time series is still being prepared."
voice: explorer
parent: null
diagram: null
recallQ: "How does Copernicus data reach JackDaw users in the Beta, and what is still being prepared?"
recallA: "Mostly through derived layers: none of the Beta tool areas queries Sentinel imagery directly, and land cover comes from OLU_EU statistics. D4.1 describes a cloud-removal and NDVI model on Sentinel-1 and Sentinel-2 and a planned classification suite; D4.2 is preparing STAC/OGC discovery of EO assets and Data Cube access to time series."
highlights:
  - "In the Beta, satellite information arrives through derived layers such as OLU_EU land cover"
  - "D4.1: Sentinel-1 radar plus Sentinel-2 optical, UnCRtainTS network, cloud-free images, then NDVI"
  - "Planned: crop detection, land-cover detection and change mapping from EO imagery"
  - "Prepared in D4.2: STAC/OGC discovery and Data Cubes for time series"
status: accepted
concept: copernicus-data
conceptTitle: "Copernicus Earth observation in PoliRural+"
parents: geoai
lang: en
lens: generic
depth: standard
formalism: none
visuality: text-first
lengthBand: standard
genre: explainer
state: edited
carriers: prose
---

The platform design (D4.1) names **Copernicus** as the authoritative source of satellite imagery and Earth observation (EO) data. A JackDaw user, however, rarely touches a Sentinel image. None of the eighteen Beta tool areas in D4.2 queries imagery directly; land cover, for example, comes as classification statistics from **OLU_EU**, part of the project's Open Land Use database.

## Models that turn imagery into indicators

D4.1 describes a **cloud-removal and index model** for crop monitoring. Clouds block optical imagery, so the model combines **Sentinel-2** optical data with **Sentinel-1** radar, which sees through clouds, and reconstructs cloud-free images with the **UnCRtainTS** neural network. It then computes vegetation indices such as **NDVI**. It was trained and validated on cloud-free frames from 2022 and 2023 agricultural data, measured by MAE and MSE; combining both satellites was more accurate than optical data alone.

D4.1 also plans an **EO classification suite** (crop detection, land-cover detection and change-detection mapping) to feed JackDaw. D4.1 also lists a change dataset, built with OLU, that tracks growth and loss of Corine and HILUCS land classes for 1990, 2000, 2005 and 2010.

## What is being prepared

D4.2 is preparing the platform for **STAC and OGC-based discovery of EO assets**, with machine-actionable metadata that GeoRAG can use, and for **Data Cubes** that serve satellite time series, climate variables and derived indicators. These are the formats for questions about change over time, drought or vegetation condition. Cooperation with the FOCAL project adds cloud-gap mitigation and super-resolution methods.

D4.2's warning applies here most of all: retrieval is only as useful as the metadata behind it. An image without a clear acquisition date, resolution and processing history cannot support a [traceable answer](#c/data-to-decisions).

**Sources:**
- D4.1 PoliRuralPlus platform design, §2.1.1 Data sources (Land Use and Land Cover Change Detection); §2.1.2 Models (Cloud removal and index calculations); §5.2.3 EO image interpretation — [PDF](https://www.poliruralplus.eu/docs/162/D4.1_PoliRuralPlus_platform_design_v1.pdf)
- D4.2 PoliRuralPlus platform Beta, §2.4 Data Spaces, EOSC and external infrastructure alignment; §3.3.1 Tool Ecosystem; §9.7 Cross-project Data Cube, GeoRAG and semantic knowledge cooperation — [PDF](https://www.poliruralplus.eu/docs/421/D4.2_PoliRuralPlus_Platform_Beta.pdf)