Romania/README.md

🇷🇴 Romania - Bark Beetle Detection Dataset

📍 Geographic Overview

A forest area in Romania

⚠️ Research Purpose: This multitemporal satellite dataset has been developed and can be used for research purposes solely.

---

📦 Dataset Structure

Romania/
└── Bark Beetle/
    └── GEE Median/
        ├── 2020-04-01 - 2020-04-30/
        │   └── sentinel_2/
        ├── 2020-05-01 - 2020-05-31/
        │   └── sentinel_2/
        ├── 2020-06-01 - 2020-06-30/
        │   └── sentinel_2/
        ├── 2020-07-01 - 2020-07-31/
        │   └── sentinel_2/
        ├── 2020-08-01 - 2020-08-31/
        │   └── sentinel_2/
        ├── 2020-09-01 - 2020-09-30/
        │   └── sentinel_2/
        └── masks/
            ├── train/              # Training masks
            ├── test/               # Test masks

---

📚 Ground Truth Sources

Resources

The ground truth masks of bark beetle damages used to create these satellite datasets have been derived from the Database of European Forest Insect and Disease Disturbances (DEFID2), https://jeodpp.jrc.ec.europa.eu/ftp/jrc-opendata/FOREST/DISTURBANCES/DEFID2/

Acknowledgment

For the use of DEFID2 polygons, acknowledgement to:

• Forzieri G, Dutrieux LP, Elia A, Eckhardt B, Caudullo G, Taboada FÁ, Andriolo A,Bălacenoiu F, Bastos A, Buzatu A, Castedo Dorado F, Dobrovolný L, Duduman M,Fernandez-Carillo A, Hernández-Clemente R, Hornero A, Ionuț S, Lombardero MJ, Junttila S, Lukeš P, Marianelli L, Mas H, Mlčoušek M, Mugnai F, Nețoiu C,Nikolov C, Olenici N, Olsson P, Paoli F, Paraschiv M, Patočka Z, Pérez-Laorga E,Quero JL, Rüetschi M, Stroheker S, Nardi D, Ferenčík J, Battisti A, Hartmann H, Nistor C, Cescatti A, Beck PSA (2023). The Database of European Forest Insect and Disease Disturbances: DEFID2. Global Change Biology, 29(21), 6040-6065. https://doi.org/10.1111/gcb.16912

🛰️ Data Creation Methodology: GEE Median - Temporal Median (12 bands)

• Period: Bark Beetle: September 2019, April-September 2020; Forest Type: September 2022

• Data Source: Google Earth Engine

• Methodology: Temporal Median

• Bands: 12 bands (B1-B9, B11, B12)

• Please cite the following work if you find the dataset useful for your reasearch

@article{DBLP:journals/ecoi/RecchiaAAIFM26,
  author       = {Vito Recchia and
                  Giuseppina Andresini and
                  Annalisa Appice and
                  Dino Ienco and
                  Giuseppe Fiameni and
                  Donato Malerba},
  title        = {{ULISSE:} Parameter-efficient adaptation of earth vision models to
                  monitor forest disturbance in sentinel-2 time series},
  journal      = {Ecol. Informatics},
  volume       = {94},
  pages        = {103668},
  year         = {2026},
  doi          = {10.1016/J.ECOINF.2026.103668}
}

---

📊 Sentinel-2 Data Details

Sentinel-2 Bands (12 bands at 10m resolution)

Band	Central Wavelength	Resolution	Band Name
B1	443 nm	60m → 10m	Coastal aerosol
B2	490 nm	10m	Blue
B3	560 nm	10m	Green
B4	665 nm	10m	Red
B5	705 nm	20m → 10m	Red edge 1
B6	740 nm	20m → 10m	Red edge 2
B7	783 nm	20m → 10m	Red edge 3
B8	842 nm	10m	Near Infrared
B8A	865 nm	20m → 10m	Narrow NIR
B9	940 nm	60m → 10m	Water vapor
B11	1610 nm	20m → 10m	SWIR 1
B12	2190 nm	20m → 10m	SWIR 2

Note: B10 (SWIR Cirrus) is excluded as reserved for atmospheric corrections.

---

📄 License

CC BY-NC-SA 4.0 - Non-commercial use with attribution

---

The development of these datasets is supported from the SWIFTT Project - Satellites for Wilderness Inspection and Forest Threat Tracking, funded by the European Union under Grant Agreement 101082732

Update: 2026