What does this research mean for the field?

The Transformer model enables accurate estimation of sugarcane cycle age from satellite imagery, achieving a cross-region precision of 97.7%. Novelty: ClaimNovelty.NOVEL_FINDING. Consensus alignment: ConsensusAlignment.NEUTRAL.

What question did this study set out to answer?

The study aims to develop a framework for estimating sugarcane cycle age using satellite imagery and deep learning methods.

March 3, 2026Open Access

Temporal Deep Learning for Satellite-Based Sugarcane Monitoring and Cycle Age Estimation

Key Points

The study aims to develop a framework for estimating sugarcane cycle age using satellite imagery and deep learning methods.
Monitored sugarcane fields using Sentinel-2 satellite time series data from March to November.
Employed object-based segmentation and extracted various descriptors including spectral and textural features.
Integrated indices like NDVI, EVI, and MSAVI to track cut-regrowth and phenological events.
Evaluated multiple deep learning architectures including LSTM, GRU, TCN, TSMixer, and Transformer.
Used Optuna for model optimization and ensured class balance through a stratified window-level split.
The Transformer model achieved the highest cross-region precision at 97.7%.
TCN and TSMixer followed with 96.6% and 96.4% precision, respectively.
The models effectively captured phenological dynamics for accurate age estimation of sugarcane.

Abstract

• Multi-year monitoring of sugarcane fields using Sentinel-2 satellite time series (March–November). • Object-based segmentation with extraction of spectral, textural, and morphological descriptors. • Integration of NDVI, EVI, and MSAVI indices to detect cut–regrowth events and phenological patterns. • Evaluation of temporal deep learning models (Transformer, TCN, TSMixer, GRU, and LSTM) for field-scale age inference. • Automated classification of sugarcane age into nine intervals (0–2, 3–5, … ≥ 24 months) across multiple regions. • The Transformer model achieved the best cross-region generalization, highlighting its robustness for operational sugarcane age estimation. Accurate monitoring of sugarcane growth stages and field age is essential for optimizing yield estimation and harvest logistics within the sugar–energy sector. This study proposes a temporal deep learning framework for inferring sugarcane cycle age from multitemporal Sentinel-2 imagery. The dataset comprises 30,282 object-level temporal windows extracted from major sugarcane-producing regions in Brazil over two agricultural seasons (2023–2024). Each sample corresponds to an eight-month consecutive sequence (March–October), from which spectral, textural, morphological, and phenological features were derived. Five temporal deep learning architectures—Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), Temporal Convolutional Network (TCN), Temporal Shift Mixer (TSMixer), and Transformer—were trained and optimized using the Optuna framework. A stratified window-level split was adopted to preserve class balance and prevent temporal leakage, while model evaluation emphasized spatial generalization across regions using accuracy and F1-macro metrics. Among the evaluated approaches, the Transformer achieved the highest cross-region precision (97.7%), followed by TCN (96.6%) and TSMixer (96.4%). The results indicate that attention- and convolution-based temporal architectures effectively capture crop phenological dynamics, enabling accurate, scalable, and operational estimation of sugarcane cycle age from satellite time series.

Read Full Paperexternally

Bookmark

View Full Paper

Cite This Study

Silva et al. (Sun,) studied this question.

synapsesocial.com/papers/69a67f06f353c071a6f0ac43 https://doi.org/https://doi.org/10.1016/j.atech.2026.101908

Bookmark

View Full Paper