Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

CIPHER: Scalable Time Series Analysis for Physical Sciences with Application to Solar Wind Phenomena

Jasmine R. Kobayashi, Daniela Martin, Valmir P Moraes Filho, Connor O'Brien, Jinsu Hong, Sudeshna Boro Saikia, Hala Lamdouar, Nathan D. Miles, Marcella Scoczynski, Mavis Stone, Sairam Sundaresan, Anna Jungbluth, Andrés Muñoz-Jaramillo, Evangelia Samara, Joseph Gallego

TL;DR

The paper tackles the shortage of expert-labeled time series in physics and the need for scalable, reliable annotations. It proposes CIPHER, a pipeline that combines iSAX symbolic compression, HDBSCAN clustering, and a human-in-the-loop to label representative windows and propagate annotations across large datasets. Applied to 1-minute OMNI solar wind data, CIPHER successfully recovers meaningful phenomena such as CMEs and SIRs, illustrating scalable, interpretable labeling in a noisy, high-dimensional setting. The approach offers a generalizable strategy for physics domains with sparse annotations and provides public code to support reproducibility and adoption.

Abstract

Labeling or classifying time series is a persistent challenge in the physical sciences, where expert annotations are scarce, costly, and often inconsistent. Yet robust labeling is essential to enable machine learning models for understanding, prediction, and forecasting. We present the \textit{Clustering and Indexation Pipeline with Human Evaluation for Recognition} (CIPHER), a framework designed to accelerate large-scale labeling of complex time series in physics. CIPHER integrates \textit{indexable Symbolic Aggregate approXimation} (iSAX) for interpretable compression and indexing, density-based clustering (HDBSCAN) to group recurring phenomena, and a human-in-the-loop step for efficient expert validation. Representative samples are labeled by domain scientists, and these annotations are propagated across clusters to yield systematic, scalable classifications. We evaluate CIPHER on the task of classifying solar wind phenomena in OMNI data, a central challenge in space weather research, showing that the framework recovers meaningful phenomena such as coronal mass ejections and stream interaction regions. Beyond this case study, CIPHER highlights a general strategy for combining symbolic representations, unsupervised learning, and expert knowledge to address label scarcity in time series across the physical sciences. The code and configuration files used in this study are publicly available to support reproducibility.

CIPHER: Scalable Time Series Analysis for Physical Sciences with Application to Solar Wind Phenomena

TL;DR

The paper tackles the shortage of expert-labeled time series in physics and the need for scalable, reliable annotations. It proposes CIPHER, a pipeline that combines iSAX symbolic compression, HDBSCAN clustering, and a human-in-the-loop to label representative windows and propagate annotations across large datasets. Applied to 1-minute OMNI solar wind data, CIPHER successfully recovers meaningful phenomena such as CMEs and SIRs, illustrating scalable, interpretable labeling in a noisy, high-dimensional setting. The approach offers a generalizable strategy for physics domains with sparse annotations and provides public code to support reproducibility and adoption.

Abstract

Labeling or classifying time series is a persistent challenge in the physical sciences, where expert annotations are scarce, costly, and often inconsistent. Yet robust labeling is essential to enable machine learning models for understanding, prediction, and forecasting. We present the \textit{Clustering and Indexation Pipeline with Human Evaluation for Recognition} (CIPHER), a framework designed to accelerate large-scale labeling of complex time series in physics. CIPHER integrates \textit{indexable Symbolic Aggregate approXimation} (iSAX) for interpretable compression and indexing, density-based clustering (HDBSCAN) to group recurring phenomena, and a human-in-the-loop step for efficient expert validation. Representative samples are labeled by domain scientists, and these annotations are propagated across clusters to yield systematic, scalable classifications. We evaluate CIPHER on the task of classifying solar wind phenomena in OMNI data, a central challenge in space weather research, showing that the framework recovers meaningful phenomena such as coronal mass ejections and stream interaction regions. Beyond this case study, CIPHER highlights a general strategy for combining symbolic representations, unsupervised learning, and expert knowledge to address label scarcity in time series across the physical sciences. The code and configuration files used in this study are publicly available to support reproducibility.
Paper Structure (5 sections, 2 figures)

This paper contains 5 sections, 2 figures.

Table of Contents

  1. Introduction
  2. Model: CIPHER
  3. Experimental Setup
  4. Results and Discussion
  5. Conclusion

Figures (2)

  • Figure 1: a) Structure of a CIPHER cluster built from smoothed and detrended proton density sequences with a 35-hour window. Top: individual preprocessed time series assigned to the cluster. Middle: symbolic representations obtained via iSAX compression, showing the index values that defined the cluster membership. Bottom: summary statistics of the cluster, with shaded confidence intervals (5–95%) in purple and the black curve representing the mean sequence. b) Raw solar wind data of one representative sequence from the cluster (Subfigure a), capturing the March 11–12, 2021 CME Möstl_2020helioforecast_icmecat. The five panels display different OMNI parameters (total magnetic field, magnetic field components, flow speed, proton density, and temperature) observed simultaneously. These complementary signatures were jointly analyzed by the domain expert to confirm the CME classification. Shaded regions highlight the key substructures that guided this expert validation: yellow: forward shock, green: compressed sheath, blue: magnetic ejecta, and red: the trailing solar wind.
  • Figure 2: CIPHER cluster associated with a stream interaction region (SIR), discovered from solar wind flow speed data. Rows: Flow speed (top), proton density (middle), and temperature (bottom). Columns: From left to right: (1) confidence intervals (5–95%) of preprocessed data (smoothed and detrended, with mean in black); (2) individual preprocessed sequences; (3) confidence intervals of the corresponding raw sequences(5–95%; with mean in black); (4) individual raw sequences. The comparison highlights the contrast between the noisy raw measurements and the structured coherence revealed by preprocessing. While raw time series appear too irregular for manual grouping, the narrow confidence intervals of the preprocessed data show that these parameters evolve consistently across events, confirming the cluster’s physical meaning as validated by the domain expert.