Sen4AgriNet is a benchmark dataset that contains Sentinel2 multi-temporal, multi-country labeled data, exploiting the recent opening up of LPIS parcel data. We construct an annotated dataset that consists of 225,000 patches with corresponding pixel-based crop types. Sen4AgriNet provides access to the data and pre-trained models for pixel-based and parcel-based crop segmentation and classification respectively.

Hephaestus is the first large scale Sentinel-1 InSAR dataset that was manually annotated by a team of experts, inspired by global volcanic unrest detection. It is designed to address multiple computer vision tasks related to InSAR interpretation, including image classification, semantic segmentation, and image captioning. It consists of 19,919 individual Sentinel-1 interferograms acquired over 44 different volcanoes globally, which are split into 216,106 InSAR patches.

We provide an implementation for the distributed training on the BigEarthNet dataset, over multiple GPU nodes. EfficientBigEarthNet then benchmarks different state-of-the-art deep learning models, contributing with an exhaustive zoo of 56 trained models. Our benchmark includes standard and scalable Convolution Neural Network architectures, but we also test non-convolutional methods, such as Multi-Layer Perceptrons and Vision Transformers.  We provide access to all trained models.

FireCube provides a collection of analysis-ready, pre-processed and harmonized set of covariates that jointly affect the fire occurrence and spread, such as weather conditions, satellite-derived products, topography features and variables related to human activity. FireCube also provides a set of pre-trained deep learning models for next day fire hazard prediction, which are able to capture the spatio-temporal context of wildfire drivers.

Kuro Siwo is a mutli-temporal, Sentinel-1 GRD based dataset, tailored for flood extent mapping with deep learning. For the pixel-based annotation we used the Copernicus Emergency Management Service Rapid Mapping products.