Objectives

AgriScienceFM builds the foundational AI infrastructure for agricultural science, open, explainable, and reusable.

Project Mission

Foundational AI infrastructure for agricultural science.

AgriScienceFM aims to create the foundational AI infrastructure for agricultural science by developing open, explainable, and reusable agricultural foundation models.

The project addresses the challenges of conventional AI approaches in agriculture by using foundation models as jump-off points, enabling the fine-tuning of AI models for agricultural science applications in two distinct ways: first, through performance improvements in accuracy and less uncertain decision-making; and second, by lowering the data barriers for developing performant models in new conditions or tasks.

Vision

A future where agricultural scientists, farmers, policymakers, and innovators can access open, trustworthy, and reusable AI tools that improve food security, sustainability, and climate resilience.

Goal

To develop, benchmark, and openly share three agriculture-specific AI foundation models that help researchers and practitioners better understand complex agricultural systems and support data-driven solutions for food security, sustainability, and climate resilience.

Core Objectives

Six objectives shaping the project

01

Develop Agricultural Foundation Models

Create three complementary AI foundation models focused on environmental systems, agricultural management, and plant structure & growth, supporting a wide range of downstream applications.

Environmental systems
Agricultural management
Plant structure & growth

02

Enable Data-Efficient AI

Reduce dependency on large labeled datasets using self-supervised learning and large-scale multimodal pretraining.

Few-shot learning
Zero-shot adaptation
Cross-region generalization

03

Build Open and FAIR Resources

Develop reusable, openly accessible datasets, pipelines, benchmarks, evaluation protocols and models, findable, accessible, interoperable and reusable by design.

Findable
Accessible
Interoperable
Reusable

04

Advance Explainable & Trustworthy AI

Integrate transparency, uncertainty quantification and responsible AI practices for safe, reliable systems.

Transparency
Uncertainty
Bias monitoring
Explainability

05

Create Standardized Benchmarks

Establish comprehensive evaluation benchmarks for agricultural AI across diverse scientific and real-world challenges.

Accuracy
Robustness
Transferability
Domain adaptation

06

Support Climate-Resilient Agriculture

Accelerate innovation for sustainable farming, climate adaptation, efficient resource management and food security.

Sustainable farming
Climate adaptation
Food security

Key Deliverables

What the project delivers

Beyond the three AgriFM foundation models, detailed on the Foundation Models page, the project delivers open, reusable resources for the agricultural research community.

01 Open Benchmarking Suite

A comprehensive evaluation framework for agricultural AI systems.

Standardized datasets
Evaluation protocols
Baseline methods
Domain-specific metrics
Reproducible experiments

02 FAIR-Compliant Datasets

Curated multimodal datasets for the research community.

Satellite & Earth Observation data
Weather and climate data
Soil datasets
Plant imaging datasets
Agronomy literature & advisory materials

03 Reusable AI Pipelines

Open-source tools and pipelines across the model lifecycle.

Data preprocessing
Model training
Fine-tuning
Benchmark evaluation
Model deployment

04 Publications & Dissemination

Scientific contributions and community engagement.

Research publications
Open reports
Workshops and events
Public demonstrations
Community engagement activities

05 Roadmap for Agricultural AI

Guidelines and best practices for future AI development in agriculture.

Explainable AI methods
Responsible AI frameworks
Interoperability standards
Cross-modal AI integration

Long-Term Impact