Prithvi-EO-2.0: A Versatile Multi-Temporal Foundation Model for Earth Observation Applications

TL;DR

<3-5 sentence high-level summary> Prithvi-EO-2.0 introduces a multi-temporal, transformer-based geospatial foundation model trained on 4.2 million time-series samples at 30m resolution from the HLS archive, with 300M and 600M parameter variants that incorporate explicit temporal and geographic embeddings. Benchmarking on GEO-Bench shows the 600M variant achieving up to an 8% improvement over its predecessor and competitive, often superior performance across resolutions from 0.1m to 15m, while SME-led downstream tasks demonstrate practical impact in disaster response, land-use/crop mapping, and ecosystem monitoring. The work emphasizes Trusted Open Science and broad accessibility through Hugging Face and TerraTorch, and includes a careful dataset design and validation framework to support reproducibility and community adoption. Overall, Prithvi-EO-2.0 showcases data-efficient generalization across diverse EO tasks and highlights the value of end-user collaboration in shaping geospatial AI tools.

Abstract

This technical report presents Prithvi-EO-2.0, a new geospatial foundation model that offers significant improvements over its predecessor, Prithvi-EO-1.0. Trained on 4.2M global time series samples from NASA's Harmonized Landsat and Sentinel-2 data archive at 30m resolution, the new 300M and 600M parameter models incorporate temporal and location embeddings for enhanced performance across various geospatial tasks. Through extensive benchmarking with GEO-Bench, the 600M version outperforms the previous Prithvi-EO model by 8\% across a range of tasks. It also outperforms six other geospatial foundation models when benchmarked on remote sensing tasks from different domains and resolutions (i.e. from 0.1m to 15m). The results demonstrate the versatility of the model in both classical earth observation and high-resolution applications. Early involvement of end-users and subject matter experts (SMEs) are among the key factors that contributed to the project's success. In particular, SME involvement allowed for constant feedback on model and dataset design, as well as successful customization for diverse SME-led applications in disaster response, land use and crop mapping, and ecosystem dynamics monitoring. Prithvi-EO-2.0 is available on Hugging Face and IBM terratorch, with additional resources on GitHub. The project exemplifies the Trusted Open Science approach embraced by all involved organizations.

Figures (16)

• Figure 1: LULC distribution of the training samples in comparison to all land tiles.
• Figure 2: Global HLS dataset distribution visualized on a tile-level. The number of training samples are color-coded in blue to green while validation tiles are visualized in magenta.
• Figure 3: Prithvi architecture and general pretraining framework.
• Figure 4: (a) Classification and (b) segmentation examples in the GEO-Bench datasets GEOBench.
• Figure 5: Aggregated performance across (a) all GEO-Bench datasets, (b) all classification tasks, and (c) all segmentation tasks. The Prithvi-EO-2.0 models are highlighted in blue.