MRTS-Boosting: A Quality-Aware Multivariate Time Series Classification Framework for Robust Rice Detection Under Cloud Contamination

Read full text Summarise with VIRAT
Authors: B. Suseno, Guilhem Brunel, Hari Wijayanto, Kusman Sadik, Farit Mochamad Afendi, Bruno Tisseyre
Cited by: 0
References: 63

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Accurate rice detection is essential for food security, sustainable agriculture, and environmental monitoring. Satellite time series observations provide scalable capabilities for rice detection; however, their application in tropical regions is challenged by persistent cloud contamination, asynchronous crop development cycles, and temporal misalignment among multisensor observations, which reduce classification reliability. This study introduces Multivariate Robust Time Series Boosting (MRTS-Boosting), a quality-aware framework for multivariate time series classification (TSC) designed to improve robustness under noisy and irregular observational conditions. The framework integrates quality-weighted feature construction, joint extraction of full-series and interval-based temporal features, and a flexible multivariate formulation that accommodates heterogeneous satellite inputs without strict temporal alignment. Performance was evaluated using synthetic datasets with controlled cloud contamination, 103 benchmark datasets from the University of California, Riverside (UCR) TSC Archive, and 3261 real-world rice field observations from Indonesia. Comparisons were conducted against representative whole-series, interval-based, shapelet-based, kernel-based, and ensemble classifiers. MRTS-Boosting achieved up to 87% accuracy under severe cloud contamination, an average rank of 2.7 on noise-augmented UCR datasets, and 93% accuracy with Cohen’s kappa of 0.76 for Indonesian rice detection, while maintaining moderate computational cost. These results demonstrate that MRTS-Boosting provides a robust, scalable, and computationally efficient framework for satellite-based rice detection. The framework remains competitive in univariate settings while benefiting from multisensor integration, indicating that performance gains arise from both methodological design and the effective use of heterogeneous data. MRTS-Boosting is therefore well-suited for precision agriculture applications under challenging observational conditions.

Fields of Study & Subjects

Remote Sensing in Agriculture Time Series Analysis and Forecasting Smart Agriculture and AI

Publication details

Journal / source
Remote Sensing
Volume
18
Issue
7
Pages
1025–1025
Publication date
29 March 2026
Language
EN