Drought monitoring in Greece

Description

This research project analyzes drought impacts on Greek mountain ecosystems, focusing on fir forest (Abies cephalonica) die-off in regions such as Chelmos, Mainalo, Taygetos, Parnonas, and Epirus. The work includes environmental monitoring, GIS mapping, climate data analysis, and the development of strategies for resilience and restoration. [source: dasarxeio.com]

The methodology is based on lessons learned from our work with Copernicus Emergency Management Service (EMS), in activation EMSN217 - Fire risk assessment in East Sardinia, Italy.

EODC Development

Based on Open Data Cube. For the EODC, the following steps have been completed:

1. Created a database and credentials
2. Created a Python environment.yaml
3. Initialized the datacube schema
4. Generated a tiling grid schema covering Natura sites
5. Defined EO products
6. Verified dataset storage, indexing, and retrieval from NAS using EO3 datasets and EODC

Tiling schema

The study area is divided into 21 tiles, each measuring 48 × 48 km. Tile naming follows a x00y00 format without negative indices. In 20-m imagery this translates to time×variables×2400×2400 pixel time series. A 2400×2400 size was found appropriate for an effective processing in a 32GB - 8 CPU configuration, with an average product delivery time of 7 minutes.

Composite generation

The composite pipeline automates the creation of 20-m monthly median mosaics of Sentinel-2 L2A data using the Planetary Computer via the STAC catalog.

Key points

• Input: GeoJSON configuration files define the year–month and tile code to process.
• Data access: Imagery is retrieved from the STAC catalog using odc-stac, filtered by cloud cover (≤70%) and data quality (nodata < 33%).
• Native resolutions: Sentinel-2 provides bands both at 10 m and 20 m resolutions. Since 10 m bands are not natively available at 20 m, we apply the rasterio.enums.Resampling.average method to bin the resolution to 20m.
  • Implementation note: If it is applied on a subset the values may vary (due to starting pixel and shape of the raster), but if it applied on a whole granule the values are identical. Sen2Cor uses skimage.measure.block_reduce (and from docs), as it can be found in S2 MPC L2A ATBD, to perform 2×2 mean aggregation (binning), as it can be found in the L2A_Tables.py module. This is also available in xarray's coarsen function, which is a replica of this function as stated in this issue. This ensures consistency with ESA’s Sen2Cor processing chain.
• Baseline change: Working with S2-L2A time series from MS PC requires a baseline change to Sen2Cor 4.00 . For scenes after 2022-01-05 (January 25th, 2022), an offset of -1000 is applied.
• Masking: Cloud, shadow, cirrus, and snow/ice pixels are masked using the Scene Classification Layer (SCL).
• Spectral indices: NDVI, EVI, and PSRI2 are calculated per timestamp before median temporal reduction.
• Reprojection: Before reduction, the data cubes are aligned and reprojected in ETRS89-extended / LAEA Europe (EPSG:3035) with odc-geo, using rasterio's bilinear resampling. If more than two UTM zones are included, a mosaic is generated.
• Compositing: A median temporal composite is produced per tile and month.
• Output: Results are stored as Cloud-Optimized GeoTIFFs (COGs), with metadata (dataset definitions) recorded in both EO3 YAML and STAC JSON formats for datacube indexing. The assets/measurements of the bands are recorded in paths relevant to the metadata document location. To do so on a local machine, raster information were referenced by explicitly providing geo and pixel information to the metadata preparation module. To load the relevant paths of raster images from NAS into xarray, url patching is required.

Time series z-normalization

Classification

OWS

To serve data indeced in the EODC as visualizations, datacube-ows provides the WMS web service endpoint to (in our case) a TerriaJS web map client, by configuring the OWS

• Styles: Styles and Layers are configured in the ows configuration module
• Update: Periodically withing the EO pipelines, or at the end of each one, the OWS database is triggered automatically to be updated.
• Reading: The images are loaded from NAS by patching the URL to the mounted volume inside the Docker container.
• Vizualization: The WMS of the data indexed is provided to a TerriaJS client, and are available via a NGROK app (https://emt-datacube-viewer.ngrok.app/).

Implementation note:

• Dont' forget to allow fire sharing in Docker
• Initialization:

1. docker exec -it drought-ows-drought_ows-1 bash
2. datacube system check - Valid connection: YES
3. datacube-ows-update --schema --write-role odc_phd --read-role odc_phd
4. datacube-ows-update --views
5. datacube-ows-update

References

Hislop, S., Stone, C., Gibson, R.K., Roff, A., Choat, B., Nolan, R.H., Nguyen, T.H. and Carnegie, A.J., 2023. Using dense Sentinel-2 time series to explore combined fire and drought impacts in eucalypt forests. Frontiers in Forests and Global Change, 6, p.1018936.