Zero-Shot Transfer Capabilities of the Sundial Foundation Model for Leaf Area Index Forecasting

2025-11-25

AI Summary

This study investigates the zero-shot transfer capability of general-purpose time-series foundation models (e.g., Sundial) for remote sensing-based leaf area index (LAI) prediction—a critical agricultural monitoring task. Method: Leveraging the HiQ benchmark dataset, we systematically evaluate Sundial’s zero-shot performance against fully supervised LSTM and statistical baselines under multiple evaluation protocols, varying input window lengths and seasonal coverage. Contribution/Results: When the input window spans ≥1 full seasonal cycle, Sundial—without any fine-tuning—achieves significantly higher prediction accuracy than the fully supervised LSTM, establishing, for the first time, “plug-and-play” zero-shot performance of foundation models in agricultural remote sensing. This demonstrates strong cross-domain and few-shot generalization of time-series foundation models, revealing their potential to overcome data scarcity and domain-shift challenges in operational agro-ecological monitoring. The work introduces a new paradigm for remote sensing time-series modeling grounded in foundation model pretraining rather than task-specific supervision.

Technology Category

• Machine Learning: Deep Neural Architectures and Foundation Models
• Computer Vision: Other Foundations of Computer Vision
• Humans and AI: Other Foundations of Human Computation & AI

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Search and Retrieval-Augmented AI: Large language models for search
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications

Abstract

This work investigates the zero-shot forecasting capability of time-series foundation models for Leaf Area Index (LAI) forecasting in agricultural monitoring. Using the HiQ dataset (U.S., 2000-2022), we systematically compare statistical baselines, a fully supervised LSTM, and the Sundial foundation model under multiple evaluation protocols. We find that Sundial, in the zero-shot setting, can outperform a fully trained LSTM provided that the input context window is sufficiently long-specifically, when covering more than one or two full seasonal cycles. This demonstrates, for the first time, that a general-purpose foundation model can surpass specialized supervised models on remote-sensing time series prediction without any task-specific tuning. These results highlight the strong potential of pretrained time-series foundation models to serve as effective plug-and-play forecasters in agricultural and environmental applications.

Problem

• Evaluating zero-shot Leaf Area Index forecasting using time-series foundation models
• Comparing foundation models against specialized supervised models without retraining
• Assessing agricultural monitoring capabilities of pretrained plug-and-play forecasters

Innovation

• Sundial foundation model enables zero-shot forecasting
• Uses long input context window for seasonal cycles
• Pretrained model outperforms specialized supervised models