CZ_ProcessesDriversPredictability

Goodwell et al., 2026
📧 goodwel2@illinois.edu

---

Overview

This repository contains code and datasets associated with the manuscript:

“Detecting regimes of critical zone processes, drivers, and predictability with a data-driven framework”

📄 *Accepted at * AGU Advances (March 2026)

---

Data Sources

Time-series variables in the DATA folder are processed or clipped versions of the original datasets listed below.

• GC Flux Tower Data
  https://www.hydroshare.org/resource/0ef3eda3534f44a6bbd65786d57222ea/

• US-Kon AmeriFlux Data
  https://ameriflux.lbl.gov/sites/siteinfo/US-Kon

• Monticello RiverLab Data
  https://www.hydroshare.org/resource/2c6c1d02c3ec4b97a767c787e1889647/

• Orgeval RiverLab Data
  https://doi.org/10.15454/9PUYPN

• Nebraska Root–Soil Gas Flux Data
  https://www.hydroshare.org/resource/405c8669069147b690c04b7063cad6ce/

---

Methodology

This codebase implements a data-driven framework for identifying and interpreting temporal regimes in multivariate environmental time series:

1. Gaussian Mixture Model (GMM) clustering

   • Identifies distinct temporal regimes in multivariate datasets.

2. Principal Component Analysis (PCA)

   • Characterizes dominant modes of variability within each regime. Explained variances of first two principal components go toward a "potential predictability" metric for each cluster.

3. Information Theory (IT) metrics

   • Applied to PC projections to identify dominant predictors of CZ dynamics in each cluster, or for a given dataset. Dominant predictors are defined as the two time-series variables that jointly reduce the most uncertainty in either principal component projection (which represents the reduced response system dynamics).

Repository Structure

📁 DATA/

• Contains processed datasets used in the analyses.
• The Processed/ subfolder includes outputs generated by DataPrep scripts that assemble and harmonize original data files for each case study.

📁 FIGS/

• Analysis scripts automatically save generated figures to this folder.
• Example figures are included.

---

Code Description

🔧 cluster_funcs.py

• Core module containing all functions for:
  • Clustering
  • Dimensionality reduction
  • Information theory metrics
  • Figure generation
• Imported by all analysis scripts.

📊 DataPrep_*.py

• Case-study–specific data preparation scripts that concatenate, do minor gap-filling, and align time-series data for input into the GMM-PCA-IT analysis codes.

📊 Analysis_*.py

• Case-study–specific analysis scripts, including:
  • Flux tower datasets
  • Stream solute concentration datasets
  • Root–soil gas concentration datasets
  • Associated meteorological and soil drivers
• Outputs:
  • Multiple figures (saved to FIGS/)
  • A .csv file summarizing information theory results