AI-Aquatica

Resource information

Format

zip

Original repository

22133099

DOI

10.5281/zenodo.22133099

Credit basis

Creators in Zenodo record metadata

Access

Open files available from the original Zenodo deposit

Last checked

Nov 5, 2026

License

mit-license

Last record update

Nov 5, 2026

About this resource

AI-Aquatica v2.3.0 AI-Aquatica is an open-source Python library for reproducible machine-learning-assisted water-quality and hydrochemical analysis. The software integrates environmental data loading and preprocessing, hydrochemical quality control, classical machine learning, visualization, validation, and automated HTML reporting within a single auditable workflow. Version 2.3.0 is a major methodological and software-quality release prepared in connection with the revised SoftwareX software article. Major changes in v2.3.0 Leakage-safe machine-learning workflows Supervised preprocessing has been redesigned to prevent information leakage between training and validation data. Imputation, scaling, and optional predictive PCA are now fitted exclusively on training data within each validation split. The validation framework now supports: stratified holdout validation; GroupKFold validation; campaign-grouped validation; temporal holdout validation; out-of-fold predictions; bootstrap confidence intervals; per-class classification metrics; confusion matrices; regression baselines. Exploratory PCA is explicitly separated from predictive preprocessing. Improved hydrochemical quality control The ion-balance module has been strengthened with explicit validation of: missing ion concentrations; incomplete ion sets; non-numeric values; infinite values; negative concentrations; charge-balance completeness. Samples with insufficient ion information are classified as indeterminate rather than being assigned an artificial zero charge-balance error. Hydrochemical quality status can now directly control subsequent machine-learning workflows using: warn ; filter ; raise ; ignore . This creates an explicit connection between domain-specific chemical quality control and predictive modelling. Correct nitrogen-unit handling The example water-quality dataset reports nitrate, nitrite, and ammonium concentrations as mg N/L . Version 2.3.0 therefore applies nitrogen-equivalent conversion using: 14.0067 mg N/meq for NO3-N , NO2-N , and NH4-N when calculating ionic charge equivalents. The conversion assumptions are recorded in the reproducibility outputs. Reproducible SoftwareX validation workflow A single executable workflow reproduces the primary analyses reported in the accompanying SoftwareX manuscript: python examples/real_dataset_workflow.py \ --output validation/local_release \ --bootstrap 1000 The workflow generates: hydrochemical charge-balance diagnostics; station-classification results; GroupKFold validation outputs; QC-filtered sensitivity analysis; chlorophyll-a regression results; confusion matrices; observed-versus-predicted plots; machine-readable CSV/JSON validation summaries; a standalone HTML water-quality report. Example dataset The repository contains an original, non-synthetic water-quality dataset comprising: 148 observations ; 4 monitoring locations ; 37 monthly sampling campaigns ; sampling period: January 2020–January 2023 . The monitoring locations are urban aquatic systems in Szczecin, north-western Poland: Lake Głębokie; Lake Słoneczne; inflow to Lake Rusałka; outflow from Lake Rusałka. The dataset contains physicochemical, hydrochemical, nutrient, biological-proxy, and trace-metal variables. Analytical procedures include potentiometric, electrochemical, spectrophotometric, titrimetric, cuvette-based, and atomic absorption spectrometric methods conducted according to applicable EN ISO, ISO, or documented laboratory procedures. The accomp