Time-series spectral dataset for croplands in France (2006-2017)

Area of Science:

• Agricultural Science
• Remote Sensing
• Geographic Information Systems (GIS)

Background:

• Satellite-derived decadal time-series are valuable for crop discrimination and succession identification at broad scales.
• Coarse spatial resolution of satellite data leads to mixed pixels, impacting crop modeling accuracy and requiring extensive field data for calibration.

Purpose of the Study:

• To create a high-quality dataset of pure satellite pixels for crop modeling.
• To enable development of advanced crop classification and succession monitoring methods.

Main Methods:

• Utilized cloud-free MODIS Terra (MOD13Q1) and Aqua (MYD13Q1) Vegetation Index products from 2006-2017 in France.
• Developed a GIS workflow in R to combine satellite data and Land Parcel Information System (LPIS) data.
• Selected pure MODIS pixels within single-crop parcels to generate a spectral time-series dataset.

Main Results:

• Generated a dataset of 21,129 reference plots with spectral time-series (red, near-infrared, NDVI, EVI) and annual crop types at an 8-day time step.
• The dataset covers a 12-year period (2006-2017) for France.
• Successfully identified and isolated 'pure' pixels, mitigating mixed-pixel challenges.

Conclusions:

• The generated dataset addresses limitations of coarse spatial resolution in satellite data for crop modeling.
• This dataset is suitable for developing novel deep learning-based classification methods and for crop succession monitoring and prediction.