T³S: Think in Thermal Time for Generalizable Crop Mapping from Satellite Image Time Series

2025/06/15 by Mehmet Özgür Türkoglu, Sélène Ledain, Turkoglu, Mehmet Ozgur +5
Agricultural and Biological Sciences · Environmental Science · #Smart Agriculture and AI #Remote Sensing in Agriculture #Greenhouse Technology and Climate Control

paper · pdf · doi:10.48550/arxiv.2506.12885

Abstract

Crop type classification from optical satellite time series remains limited in its ability to generalize across growing seasons, particularly when crop phenology shifts due to inter-annual weather variability. This hampers deployment in operational settings where current-year labels are unavailable. In addition, uncertainty quantification is often overlooked, reducing the reliability of such approaches for practical crop monitoring. Inspired by ecophysiological principles, we introduce Thermal Time-based Temporal Sampling (T³S), a simple, model-agnostic method that replaces calendar time with thermal time. By re-indexing satellite observations by cumulative growing degree days, T³S aligns phenologically equivalent growth stages across years, reducing temporal redundancy while concentrating on the most biologically informative periods. We evaluate T³S across three architecturally distinct backbones on (i) SwissCrop, a new country-scale, multi-year Sentinel-2 dataset with paired temperature data that we publicly release, and (ii) the cross-region TimeMatch benchmark spanning Denmark and France. Across these settings, T³S consistently improves cross-year and cross-region crop classification over several state-of-the-art baselines, including thermal positional encoding, with particularly strong gains in uncertainty calibration, robustness under label scarcity, and early-season prediction, while requiring no architectural modification.

Citations

• From General to Specialized: The Need for Foundational Models in Agriculture
• TESSERA: Temporal Embeddings of Surface Spectra for Earth Representation and Analysis
• AI4SNOW SnowGalileo Datasets and Model Checkpoints
• SITSMamba for Crop Classification based on Satellite Image Time Series
• Generalization Enhancement Strategies to Enable Cross-year Cropland Mapping with Convolutional Neural Networks Trained Using Historical Samples
• LoRA-Ensemble: Efficient Uncertainty Modelling for Self-Attention Networks
• Boosting Crop Classification by Hierarchically Fusing Satellite, Rotational, and Contextual Data
• Pixel-wise Agricultural Image Time Series Classification: Comparisons and a Deformable Prototype-based Approach
• ViTs for SITS: Vision Transformers for Satellite Image Time Series
• FiLM-Ensemble: Probabilistic Deep Learning via Feature-wise Linear Modulation
• Generalized Classification of Satellite Image Time Series with Thermal\n Positional Encoding
• Panoptic Segmentation of Satellite Image Time Series with Convolutional Temporal Attention Networks
• Crop mapping from image time series: deep learning with multi-scale\n label hierarchies
• Crop Classification under Varying Cloud Cover with Neural Ordinary Differential Equations
• Leveraging Class Hierarchies with Metric-Guided Prototype Learning
• Lightweight Temporal Self-Attention for Classifying Satellite Image Time Series
• Satellite Image Time Series Classification with Pixel-Set Encoders and\n Temporal Self-Attention
• BreizhCrops: A Time Series Dataset for Crop Type Mapping
• Attention Is All You Need
• Snapshot Ensembles: Train 1, get M for free
• Simple and Scalable Predictive Uncertainty Estimation using Deep Ensembles
• Dropout as a Bayesian Approximation: Representing Model Uncertainty in Deep Learning
• U-Net: Convolutional Networks for Biomedical Image Segmentation

Related

• Generalized Classification of Satellite Image Time Series with Thermal\n Positional Encoding
• Combining Satellite and Weather Data for Crop Type Mapping: An Inverse Modelling Approach
• Boosting Crop Classification by Hierarchically Fusing Satellite, Rotational, and Contextual Data
• End-to-End Learned Early Classification of Time Series for In-Season Crop Type Mapping
• Resilient In-Season Crop Type Classification in Multispectral Satellite Observations using Growth Stage Normalization