Cotton Phenology Dataset 🌱

The dataset contains labels about the phenological stages of cotton at the parcel level, generated by in-situ inspections in the region of Orchomenos, Greece, during the cotton cultivation period of 2021 (May-October). We collected 1,285 crop growth ground observations. We introduce a new collection protocol, assigning up to two phenology labels that represent the primary and secondary growth stage in the field and thus indicate when stages are transitioning. More information about the annotation procedure can be found in the relevant publication "Fuzzy clustering for the within-season estimation of cotton phenology". Specifically the dataset is composed of:

• a csv file which includes:

  • the main and secondary (if exist) phenological stage and its percentage for each in-situ visit in parcel level.
  • the metalabel for each in-situ visit (as they are described in the paper).
  • the date of the in-situ visit
  • the nearest sentinel-2 acquistion date for each in-situ visit.
  • the sowing and the haverst (if available) dates for each parcel

• a geojson file containing the geometries of all the parcels.

• the nearest sentinel-2 acquistion Each image is named in the following format: {date}_{id}-{type}.jpg, where date is in the format {year}{month}{day} and refes to the exact date that the image was captured, id is the unique id of a parcel and type is one of O, A or B. O refers to a panoramic photo of the entire field, A to a photo that is representative of the majority of the plants in the field and B to one representative of a minority of plants in the field. For each date and each unique parcel the first two exists always while the latter close-up photo had be captured only when the percentage of the minority class, in terms of area, was deemed significant.

Citation

If you use this dataset please cite the publication below

@article{10.1371/journal.pone.0282364,
    doi = {10.1371/journal.pone.0282364},
    author = {Sitokonstantinou, Vasileios AND Koukos, Alkiviadis AND Tsoumas, Ilias AND Bartsotas, Nikolaos S. AND Kontoes, Charalampos AND Karathanassi, Vassilia},
    journal = {PLOS ONE},
    title = {Fuzzy clustering for the within-season estimation of cotton phenology},
    year = {2023},
    url = {https://doi.org/10.1371/journal.pone.0282364},
    volume = {18},
    number = {3},
}

Downloading the phenology dataset through direct links

To download the data, you can use the following links, taken from :

• Phenology ground observations link
• Parcels' geometries link
• Field images link
  • _file size: ~12.3GB
    
    

Downloading phenology dataset through the provided download script

An alternative way to download the dataset, is to use the download.sh script included in this repository. The instructions for the script usage are as follows:

1. download the .csv with the ground observations and the corresponing geojson file of the parcels' geometries to data/

bash download.sh phenology

2. download field images to data/images (~ 12.3GB)

bash download.sh images

or download everything with bash download.sh all

Overview

1286 Unique Visits in 80 Parcels

Overall Statistics

Phenological stage	# Main	# Secondary	Code1
Root Establishment	75	4	1
Leaf Development	422	20	2
Squaring	212	6	3
Flowering	229	149	4
Boll Development	251	313	5
Boll Opening	97	177	6
Total2	1286	668	-

The field visits were appropriately scheduled in order to have minimal differences between ground and satellite observations. Thus, difference between the ground and the Sentinel-2 observation was mean = 0.86 days with standard deviation = 0.89 days .

Empirical cumulative distribution functions

of main & secondary stages as recorded by inspector

Photos

Main stage	Secondary stage	Panoramic

Acknowledgements

The polygons of parcels stem from Agriculture Cooperative of Orchomenos

This work is part of e-shape's³ pilot S1P2 EU CAP Support by the Beyond Center of Excellence of the National Observatory of Athens.

Researchers: Alkiviadis Marios Koukos ([email protected]), Vasileios Sitokonstantinou ([email protected]), Ilias Tsoumas ([email protected])

Footnotes

1. A code=0 placed inside the cells of .csv in the cases of absence of secondary phenological stage. ↩

2. Given that some in-situ visits differed the same number of days from two Sentinel-2 acquisitions we have in total 1524 Sentinel-2 acquisitions in which are assigned in-situ visit. ↩

3. The e-shape project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement 820852 ↩