Satellite-based remote sensing has revolutionised the way we address global challenges in a rapidly evolving world. Huge quantities of Earth Observation (EO) data are generated by satellite sensors daily, but processing these large datasets for use in ML pipelines is technically and computationally challenging. Specifically, different types of EO data are often hosted on a variety of platforms, withdiffering degrees of availability for Python preprocessing tools. In addition, spatial alignment across data sources and data tiling for easier handling can present significant technical hurdles for novice users. While some preprocessed Earth observation datasets exist, their content is often limited to optical or near-optical wavelength data, which is ineffective at night or in adverse weather conditions.Synthetic Aperture Radar (SAR), an active sensing technique based on microwave length radiation, offers a viable alternative. However, the application of machine learning to SAR has been limited due to a lack of ML-ready data and pipelines, particularly for the full diversity of SAR data, including polarimetry, coherence and interferometry. In this work, we introduce M3LEO, a multi-modal, multi-labelEarth observation dataset that includes polarimetric, interferometric, and coherence SAR data derived from Sentinel-1, alongside multispectral Sentinel-2 imagery and a suite of auxiliary data describing terrain properties such as land use. M3LEO spans approximately 17M data chips, each measuring 4x4 km, across six diverse geographic regions. The dataset is complemented by a flexible PyTorch Lightning framework, with configuration management using Hydra, to accommodate its use across diverse ML applications in Earth observation. Additionally, we provide tools to process any dataset available on popular platforms such as Google Earth Engine for seamless integration with our framework. We show that the distribution shift in self-supervised embeddings is substantial across geographic regions, even when controlling for terrain properties. Data is available at huggingface.co/M3LEO, and code at github.com/spaceml-org/M3LEO.

This work introduces the use of Gaussian processes (GPs) for the estimation and understanding of crop development and yield using multisensor satellite observations and meteorological data. The proposed methodology combines synergistic information on canopy greenness, biomass, soil, and plant water content from optical and microwave sensors with the atmospheric variables typically measured at meteorological stations. A composite covariance is used in the GP model to account for varying scales, nonstationary, and nonlinear processes. The GP model reports noticeable gains in terms of accuracy with respect to other machine learning approaches for the estimation of corn, wheat, and soybean yields consistently for four years of data across continental U.S. (CONUS). Sparse GPs allow obtaining fast and compact solutions up to a limit, where heavy sparsity compromises the credibility of confidence intervals. We further study the GP interpretability by sensitivity analysis, which reveals that remote sensing parameters accounting for soil moisture and greenness mainly drive the model predictions. GPs finally allow us to identify climate extremes and anomalies impacting crop productivity and their associated drivers.

In this work we propose the use of Long Short-Term Memory (LSTM) Recurrent Neural Networks for multivariate time series of satellite data for crop yield estimation. Recurrent nets allow exploiting the temporal dimension efficiently, but interpretability is hampered by the typically overparameterized models. The focus of the study is to understand LSTM models by looking at the hidden units distribution, the impact of increasing network complexity, and the relative importance of the input covariates. We extracted time series of three variables describing the soil-vegetation status in agroe-cosystems -soil moisture, VOD and EVI- from optical and microwave satellites, as well as available in situ surveys on crops across Continental U.S. to perform the experiments. Firstly, the models were validated in error terms. Secondly, the trained models were visualized and, thirdly, some useful statistics were extracted from the hidden unit activation heatmaps, accounting for redundancy and cluttering of activation responses. Results reveal how networks assign most of the relevance to soil moisture and focus on two phenological stages of crop growth.