ARCHI

ARCHI imputes missing data in regionally correlated hydrologic records by linear regression using an iterative algorithm that allows each site to serve as either target or reference, expanding the pool of complete references with each imputed record until viable gap-filling ceases.

Key Features:

• Automated Iterative Imputation: Implements a novel iterative algorithm that enables each site to be considered a target or reference for regression, growing the pool of complete references with each imputed record.
• Reference Selection Filters: Allows users to apply geospatial, correlation, and group-based filters to control reference selection and limit artifacts from spurious correlations.
• Model-Acceptance Criteria: Permits specification of criteria to control which regression models are accepted for imputation.
• Data Aggregation: Provides functions for aggregating hydrologic records.
• Clustering: Clusters records with similar correlation structures.
• Visualization: Provides functions for visualizing imputation results.
• Holdout Evaluation: Evaluates model performance using holdout data.

Scientific Applications:

• Groundwater-Level Dataset Imputation: Fills missing data in large regional groundwater-level datasets, as demonstrated on datasets from the Central Valley, CA and Long Island, NY.
• Hydrologic Resource Assessment: Enables resource assessment that would otherwise be limited by missing data in hydrological records.
• Process Understanding: Supports process understanding in hydrology by providing complete records for analysis.
• Predictive Modeling: Facilitates predictive modeling by imputing gaps in hydrologic time series data.
• Multivariate Imputation Benchmarking: Evaluated alongside widely used multivariate imputation software for computational efficiency, imputation accuracy, and model transparency.

Methodology:

ARCHI imputes missing data in target records by linear regression using more complete reference records as predictors, with automated imputation implemented via an iterative algorithm that allows each site to be considered a target or reference for regression.