nextaire-tools 0.1.0__tar.gz

nextaire_tools-0.1.0/.gitignore

@@ -0,0 +1,56 @@
+# Jupyter
+.ipynb_checkpoints/
+
+# Python cache
+__pycache__/
+*.py[cod]
+*.pyo
+
+# Virtual environment
+.venv/
+venv/
+env/
+
+# Packaging / build
+build/
+dist/
+*.egg-info/
+.eggs/
+
+# Tooling caches
+.mypy_cache/
+.ruff_cache/
+.pytest_cache/
+.coverage
+htmlcov/
+coverage.xml
+
+# Docs build
+site/
+
+# Downloaded reanalysis data (extractors write here)
+data/era5/
+data/cams/
+data/era5_land/
+*.grib
+*.grb
+*.grib2
+*.nc
+
+# Example outputs (generated by examples/end_to_end.py)
+examples/figures/
+examples/sample_station.csv
+
+# Publisher PDFs kept locally for reference (not redistributed)
+papers/*.pdf
+
+# Reproduction outputs (generated by reproductions/*.py)
+reproductions/outputs/
+
+# Notebook run artifacts
+notebooks/station.parquet
+notebooks/*.png
+
+# OS
+.DS_Store
+Thumbs.db

nextaire_tools-0.1.0/CHANGELOG.md

@@ -0,0 +1,70 @@
+# Changelog
+
+All notable changes to **nextaire_tools** are documented in this file.
+
+The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/),
+and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
+
+## [Unreleased]
+
+### Added
+
+- **Paper reproductions** — runnable, offline recipes in `reproductions/` that
+  rebuild the methodology of the three studies `nextaire_tools` is based on (Petrić et al.
+  2024; Jiménez-Navarro et al. 2024; Račić et al. 2026). See `papers/README.md`.
+- **Example notebooks** — `notebooks/` with a quickstart, a preprocessing/feature
+  tour, interactive paper reproductions, and a deep-learning + Prophet forecasting
+  tour; each runs on synthetic data and is committed with executed outputs. Install
+  with `pip install "nextaire_tools[notebooks]"`.
+- **Preprocessing**:
+  - `OutlierHandler` gains a `rolling_sigma` method — time-local winsorisation on
+    a centred rolling window (`window`, `sigma`), for spike removal.
+  - `MissingValueHandler` gains an `iterative` strategy — multivariate
+    (round-robin) imputation via scikit-learn's `IterativeImputer`, with a
+    configurable `estimator` (e.g. Bayesian ridge).
+  - `WindDecomposer` — wind direction → sine/cosine (x/y) and speed from u/v
+    components.
+  - `LagFeatures` — causal lag and rolling-aggregate (e.g. 12-hour median)
+    features.
+  - `CorrelationFilter` — drop features above a pairwise-correlation threshold.
+- **Models**:
+  - `HybridProphetRegressor` / `ProphetFeatures` — Prophet forecasts as features
+    for a downstream regressor (the "hybrid" model).
+  - `NMFApportionment` — rank-k NMF factor analysis for source apportionment.
+  - `permutation_importance_report`, `tree_shap_values`, `shap_importance` —
+    model-interpretation helpers.
+  - `make_regressor` registers `decision_tree` and (optional) `xgboost`.
+- **Metrics** — `regression_metrics` adds `wape`, `nmae`, and `nrmse`
+  (IQR-normalised).
+- **Optional extras** — `nextaire_tools[boost]` (XGBoost) and `nextaire_tools[shap]` (TreeSHAP).
+
+## [0.1.0] — 2026-07-05
+
+Initial public release.
+
+### Added
+
+- **IO** — `load_table` / `save_table` for CSV, Excel, and Parquet with optional
+  datetime-index handling.
+- **Preprocessing** (scikit-learn–compatible, DataFrame-in/DataFrame-out steps):
+  - `MissingValueHandler` — drop / impute / interpolate missing values, missingness
+    indicators, and column dropping by missing fraction.
+  - `OutlierHandler` — IQR, z-score, modified z-score, quantile, and Isolation-Forest
+    detection with clip / drop / NaN / flag strategies.
+  - `TemporalFeatures` — calendar features plus cyclical (sine/cosine) encodings for
+    hour, day-of-week, month, and day-of-year.
+  - `Scaler` — standard / min-max / robust / max-abs scaling with `inverse_transform`.
+  - `Pipeline` / `make_pipeline` — compose steps; build from config.
+- **Extractors** for the Copernicus data stores:
+  - `ERA5Extractor` (Climate Data Store), `CAMSExtractor` (Atmosphere Data Store),
+    `ERA5LandExtractor` (ERA5-Land) with nearest-neighbour point sampling at stations.
+  - `load_stations`, `dms_to_dd`, and GRIB sampling helpers.
+- **Visualization** — a colorblind-safe, light/dark-aware plotting theme with EDA,
+  outlier, and model-evaluation figures.
+- **Models** — time-series cross-validation splitters, air-quality regression metrics
+  (incl. index of agreement and FAC2), a scikit-learn regressor factory, PyTorch
+  deep-learning regressors (MLP / LSTM / CNN), and a Prophet forecasting wrapper.
+- Full MkDocs documentation site, typed API (`py.typed`), and a test suite.
+
+[Unreleased]: https://github.com/NextAIRE-Horizon/nextaire_tools/compare/v0.1.0...HEAD
+[0.1.0]: https://github.com/NextAIRE-Horizon/nextaire_tools/releases/tag/v0.1.0

nextaire_tools-0.1.0/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 Valentino Petric
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

nextaire_tools-0.1.0/PKG-INFO

@@ -0,0 +1,368 @@
+Metadata-Version: 2.5
+Name: nextaire_tools
+Version: 0.1.0
+Summary: Reproducible preprocessing, feature engineering, Copernicus data extraction, visualization and ML/DL modeling for air-quality time series.
+Project-URL: Homepage, https://github.com/NextAIRE-Horizon/nextaire_tools
+Project-URL: Documentation, https://nextaire-tools.readthedocs.io
+Project-URL: Repository, https://github.com/NextAIRE-Horizon/nextaire_tools
+Project-URL: Issues, https://github.com/NextAIRE-Horizon/nextaire_tools/issues
+Project-URL: Changelog, https://github.com/NextAIRE-Horizon/nextaire_tools/blob/main/CHANGELOG.md
+Author-email: Valentino Petrić <valentino.petric@lisboncouncil.net>
+Maintainer-email: Valentino Petrić <valentino.petric@lisboncouncil.net>
+License: MIT
+License-File: LICENSE
+Keywords: CAMS,Copernicus,ERA5,air quality,atmospheric science,deep learning,feature engineering,machine learning,preprocessing,reanalysis,time series
+Classifier: Development Status :: 4 - Beta
+Classifier: Intended Audience :: Science/Research
+Classifier: License :: OSI Approved :: MIT License
+Classifier: Operating System :: OS Independent
+Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3.10
+Classifier: Programming Language :: Python :: 3.11
+Classifier: Programming Language :: Python :: 3.12
+Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
+Classifier: Topic :: Scientific/Engineering :: Atmospheric Science
+Classifier: Typing :: Typed
+Requires-Python: >=3.10
+Requires-Dist: matplotlib>=3.7
+Requires-Dist: numpy>=1.24
+Requires-Dist: openpyxl>=3.1
+Requires-Dist: pandas>=2.1
+Requires-Dist: pyarrow>=14
+Requires-Dist: scikit-learn>=1.4
+Requires-Dist: scipy>=1.10
+Requires-Dist: seaborn>=0.13
+Provides-Extra: all
+Requires-Dist: cdsapi>=0.7.2; extra == 'all'
+Requires-Dist: cfgrib>=0.9.11; extra == 'all'
+Requires-Dist: eccodes>=2.30; extra == 'all'
+Requires-Dist: geopandas>=1.0; extra == 'all'
+Requires-Dist: holidays>=0.50; extra == 'all'
+Requires-Dist: osmnx>=2.0; extra == 'all'
+Requires-Dist: prophet>=1.1; extra == 'all'
+Requires-Dist: shap>=0.44; extra == 'all'
+Requires-Dist: shapely>=2.0; extra == 'all'
+Requires-Dist: torch>=2.0; extra == 'all'
+Requires-Dist: xarray>=2024.1; extra == 'all'
+Requires-Dist: xgboost>=2.0; extra == 'all'
+Provides-Extra: boost
+Requires-Dist: xgboost>=2.0; extra == 'boost'
+Provides-Extra: deep
+Requires-Dist: torch>=2.0; extra == 'deep'
+Provides-Extra: dev
+Requires-Dist: build>=1.2; extra == 'dev'
+Requires-Dist: mypy>=1.11; extra == 'dev'
+Requires-Dist: pandas-stubs; extra == 'dev'
+Requires-Dist: pytest-cov>=5; extra == 'dev'
+Requires-Dist: pytest>=8; extra == 'dev'
+Requires-Dist: ruff>=0.6; extra == 'dev'
+Provides-Extra: docs
+Requires-Dist: griffe>=1.0; extra == 'docs'
+Requires-Dist: mkdocs-gen-files>=0.5; extra == 'docs'
+Requires-Dist: mkdocs-literate-nav>=0.6; extra == 'docs'
+Requires-Dist: mkdocs-material>=9.5; extra == 'docs'
+Requires-Dist: mkdocs>=1.6; extra == 'docs'
+Requires-Dist: mkdocstrings[python]>=0.26; extra == 'docs'
+Provides-Extra: extract
+Requires-Dist: cdsapi>=0.7.2; extra == 'extract'
+Requires-Dist: cfgrib>=0.9.11; extra == 'extract'
+Requires-Dist: eccodes>=2.30; extra == 'extract'
+Requires-Dist: xarray>=2024.1; extra == 'extract'
+Provides-Extra: forecast
+Requires-Dist: prophet>=1.1; extra == 'forecast'
+Provides-Extra: geo
+Requires-Dist: geopandas>=1.0; extra == 'geo'
+Requires-Dist: osmnx>=2.0; extra == 'geo'
+Requires-Dist: shapely>=2.0; extra == 'geo'
+Provides-Extra: holidays
+Requires-Dist: holidays>=0.50; extra == 'holidays'
+Provides-Extra: notebooks
+Requires-Dist: ipykernel>=6; extra == 'notebooks'
+Requires-Dist: jupyterlab>=4; extra == 'notebooks'
+Requires-Dist: nbconvert>=7; extra == 'notebooks'
+Provides-Extra: shap
+Requires-Dist: shap>=0.44; extra == 'shap'
+Description-Content-Type: text/markdown
+
+# nextaire_tools
+
+Preprocessing, feature engineering, Copernicus data extraction, visualization,
+and ML/DL modeling for air-quality time series.
+
+`nextaire_tools` collects the steps a typical air-quality study repeats by hand — loading
+a table, handling missing values and outliers, building calendar features,
+pulling ERA5/CAMS reanalysis, plotting, and fitting a model with a correct
+time-series split — into a set of composable, tested, scikit-learn-compatible
+building blocks. It implements the methods from three peer-reviewed studies (see
+[Reproducing the papers](#reproducing-the-papers)), so a study can be rebuilt from
+documented, versioned code rather than one-off notebooks.
+
+## Contents
+
+- [Installation](#installation)
+- [Main features](#main-features)
+- [Example data](#example-data)
+- [Quickstart](#quickstart)
+- [Tutorials & notebooks](#tutorials--notebooks)
+- [Reproducing the papers](#reproducing-the-papers)
+- [Documentation](#documentation)
+- [Citing nextaire_tools](#citing-nextaire_tools)
+
+## Installation
+
+```bash
+pip install nextaire_tools                 # core: IO, preprocessing, viz, sklearn models
+pip install "nextaire_tools[deep]"         # + PyTorch (MLP / LSTM / CNN)
+pip install "nextaire_tools[extract]"      # + Copernicus (cdsapi, xarray, cfgrib)
+pip install "nextaire_tools[forecast]"     # + Prophet (and the Prophet→RF hybrid)
+pip install "nextaire_tools[boost]"        # + XGBoost
+pip install "nextaire_tools[shap]"         # + TreeSHAP interpretation
+pip install "nextaire_tools[geo]"          # + geospatial land-use features (geopandas, osmnx)
+pip install "nextaire_tools[notebooks]"    # + JupyterLab to run the example notebooks
+pip install "nextaire_tools[all]"          # everything
+```
+
+## Main features
+
+Everything is a `DataFrame`-in / `DataFrame`-out step, so column names and the
+`DatetimeIndex` survive from raw table to fitted model.
+
+### Data IO
+
+`load_table` reads CSV, Excel, and Parquet into a datetime-indexed `DataFrame`
+(with DMS/decimal coordinate handling for station files); `save_table` writes them
+back. One entry point for messy inputs.
+
+### Cleaning
+
+Scikit-learn transformers for the parts every study gets slightly differently:
+
+- `MissingValueHandler` — drop, statistical / directional fill, time-aware
+  interpolation, or multivariate **iterative** imputation.
+- `OutlierHandler` — IQR, z-score, modified z-score, quantile bounds, a
+  multivariate Isolation Forest, or **`rolling_sigma`** time-local winsorisation
+  that removes short spikes without deleting real pollution events.
+
+### Feature engineering
+
+- `TemporalFeatures` — calendar fields plus sine/cosine encodings (hour 23 is
+  adjacent to hour 0).
+- `WindDecomposer` — wind direction → x/y components, and speed from u/v.
+- `LagFeatures` — causal lags and rolling aggregates (e.g. 12-hour median).
+- `CorrelationFilter` — drop collinear features above a threshold.
+- `Scaler` and `Pipeline` / `make_pipeline` compose it all reproducibly.
+
+### Copernicus extraction
+
+`ERA5Extractor`, `CAMSExtractor`, and `ERA5LandExtractor` download from the
+Climate/Atmosphere Data Stores and sample the grid at your monitoring stations;
+`load_stations` parses the coordinate files.
+
+### Visualization
+
+A colorblind-safe, light/dark-aware Matplotlib theme with EDA
+(`plot_missingness`, `plot_correlation`, `plot_seasonality`, `plot_timeseries`),
+outlier inspection, and evaluation (`plot_predictions`, `plot_residuals`,
+`plot_feature_importance`) figures.
+
+### Modeling & validation
+
+- Leakage-free splitters: `BlockingTimeSeriesSplit`, `SlidingWindowSplit`,
+  `ExpandingWindowSplit`, `temporal_train_test_split`.
+- `make_regressor` — Random Forest, gradient boosting, decision tree, linear
+  family, KNN, SVR, and optional XGBoost by name.
+- PyTorch `MLPRegressor` / `LSTMRegressor` / `CNNRegressor`, a `ProphetForecaster`,
+  and a `HybridProphetRegressor` (Prophet forecasts as features for a regressor).
+- `NMFApportionment` for NMF source apportionment.
+
+### Metrics & interpretation
+
+`regression_metrics` reports the scores atmospheric scientists actually use —
+MAE/RMSE/R², index of agreement, FAC2, and IQR-normalized nMAE/nRMSE plus WAPE —
+and `cross_val_report` tabulates them per fold. `permutation_importance_report`
+and `shap_importance` (TreeSHAP) explain a fitted model.
+
+### Module map
+
+| Subpackage | Key names |
+|------------|-----------|
+| `nextaire_tools.io` | `load_table`, `save_table` |
+| `nextaire_tools.preprocessing` | `MissingValueHandler`, `OutlierHandler`, `TemporalFeatures`, `WindDecomposer`, `LagFeatures`, `CorrelationFilter`, `Scaler`, `Pipeline` |
+| `nextaire_tools.extractors` | `ERA5Extractor`, `CAMSExtractor`, `ERA5LandExtractor`, `load_stations` |
+| `nextaire_tools.viz` | `plot_missingness`, `plot_seasonality`, `plot_predictions`, … |
+| `nextaire_tools.models` | `make_regressor`, `BlockingTimeSeriesSplit`, `cross_val_report`, `regression_metrics`, `LSTMRegressor`, `HybridProphetRegressor`, `NMFApportionment`, `shap_importance` |
+
+## Example data
+
+No downloads are needed to try `nextaire_tools`. The generators in
+[`reproductions/_synthetic.py`](reproductions/_synthetic.py) build data with the
+same columns, frequency, and structure as the papers' real datasets:
+
+- `make_graz_hourly()` / `make_graz_multistation()` — hourly pollutants
+  (NO/NO₂/O₃/PM10) plus ground meteorology and ERA5-style reanalysis, with
+  realistic diurnal/seasonal cycles, gaps, and spikes.
+- `make_zagreb_daily()` — daily PM10-bound PAHs and metals with meteorology,
+  traffic and heating proxies, and station labels.
+
+```python
+import sys; sys.path.insert(0, "reproductions")
+from _synthetic import make_graz_hourly
+
+df = make_graz_hourly(n_days=180, seed=0)   # hourly, datetime-indexed DataFrame
+```
+
+For real data, replace the generator with `nextaire_tools.load_table("station.csv",
+time_col="timestamp", set_time_index=True)`.
+
+## Quickstart
+
+```python
+import nextaire_tools
+from nextaire_tools import load_table, Pipeline
+from nextaire_tools.preprocessing import MissingValueHandler, OutlierHandler, TemporalFeatures, Scaler
+
+# 1. Load anything (CSV / Excel / Parquet) with a datetime index
+df = load_table("station.csv", time_col="timestamp", set_time_index=True)
+
+# 2. Build a reproducible cleaning + feature pipeline
+pipe = Pipeline([
+    MissingValueHandler(strategy="interpolate", limit=3),
+    OutlierHandler(columns=["no2", "o3", "pm10"], method="iqr", strategy="clip"),
+    TemporalFeatures(
+        add=("hour", "dayofweek", "month", "is_weekend"),
+        cyclical=("hour", "dayofweek", "dayofyear"),   # sin/cos encodings
+    ),
+    Scaler(method="standard"),
+])
+clean = pipe.fit_transform(df)
+```
+
+### Explore before you model
+
+```python
+from nextaire_tools.viz import plot_missingness, plot_correlation, plot_seasonality
+
+plot_missingness(df)
+plot_correlation(df, cluster=True)
+plot_seasonality(df, column="o3", by="hour")
+```
+
+### Fit a model with a correct time-series split
+
+```python
+from nextaire_tools.models import make_regressor, cross_val_report, BlockingTimeSeriesSplit
+
+target = "no2"
+X = clean.drop(columns=[target])
+y = clean[target]
+
+model = make_regressor("random_forest", n_estimators=400)
+report = cross_val_report(model, X, y, cv=BlockingTimeSeriesSplit(n_splits=5))
+print(report)   # per-fold MAE / RMSE / R² / nMAE / nRMSE / IoA / FAC2 + mean & std
+```
+
+### Deep learning (optional `[deep]`)
+
+```python
+from nextaire_tools.models import LSTMRegressor
+
+lstm = LSTMRegressor(window=24, hidden_size=64, epochs=50)
+lstm.fit(X.values, y.values)
+y_hat = lstm.predict(X.values)
+```
+
+### Pull ERA5 meteorology at your stations (optional `[extract]`)
+
+```python
+from nextaire_tools.extractors import ERA5Extractor, load_stations
+
+stations = load_stations("data/Coordinates.xlsx")   # handles DMS or decimal degrees
+era5 = ERA5Extractor(output_dir="data/era5")
+frames = era5.extract_to_frames(
+    stations=stations,
+    variables=["2m_temperature", "10m_u_component_of_wind", "boundary_layer_height"],
+    area=[49.1, 9.53, 46.3, 17.16],   # N, W, S, E
+    start="2024-01-01", end="2024-01-31",
+    save_dir="data/era5",
+)
+```
+
+## Tutorials & notebooks
+
+Runnable Jupyter notebooks in [`notebooks/`](notebooks/) work on the synthetic
+example data — no network required — and are committed with their executed
+outputs, so they render on GitHub and re-run top to bottom.
+
+| Notebook | What it covers |
+|----------|----------------|
+| [`01_quickstart.ipynb`](notebooks/01_quickstart.ipynb) | Load/save, EDA plots, a cleaning + feature `Pipeline`, and a Random Forest with a leakage-free split. |
+| [`02_preprocessing_and_features.ipynb`](notebooks/02_preprocessing_and_features.ipynb) | Every preprocessing step in turn, then composed into one pipeline. |
+| [`03_reproduce_papers.ipynb`](notebooks/03_reproduce_papers.ipynb) | Compact interactive versions of the three paper recipes. |
+| [`04_deep_learning_and_forecasting.ipynb`](notebooks/04_deep_learning_and_forecasting.ipynb) | PyTorch MLP/LSTM/CNN regressors and Prophet / hybrid Prophet+RF forecasting. |
+
+```bash
+pip install "nextaire_tools[notebooks]"
+jupyter lab                              # open a notebook, or run headless:
+jupyter nbconvert --to notebook --execute --inplace notebooks/01_quickstart.ipynb
+```
+
+There is also a plain-script walkthrough in
+[`examples/end_to_end.py`](examples/end_to_end.py) and a prose tutorial in the
+[documentation](https://nextaire-tools.readthedocs.io).
+
+## Reproducing the papers
+
+`nextaire_tools` packages the methodology of three peer-reviewed air-quality ML studies.
+Each has a runnable recipe in [`reproductions/`](reproductions/) that rebuilds its
+data construction, preprocessing, cross-validation, models, and metrics — offline,
+on synthetic data of the same shape, degrading gracefully when an optional
+dependency is missing. Swap the synthetic generator for `load_table(...)` of the
+real data to reproduce the published numbers.
+
+```bash
+pip install "nextaire_tools[all]"
+python reproductions/paper1_petric2024_aaqr.py            # Petrić et al. 2024 (AAQR)
+python reproductions/paper2_jimenez2024_multitarget.py   # Jiménez-Navarro et al. 2024 (Results in Eng.)
+python reproductions/paper3_racic2026_source_apportionment.py  # Račić et al. 2026 (Atmos. Env. X)
+```
+
+See [`reproductions/README.md`](reproductions/README.md) for the full paper →
+API map, and [`papers/README.md`](papers/README.md) for citations and data
+sources.
+
+## Documentation
+
+Full documentation — user guide, API reference, and tutorials — lives at
+**[nextaire-tools.readthedocs.io](https://nextaire-tools.readthedocs.io)** and is built from the
+Markdown sources in [`docs/`](docs/).
+
+Build it locally:
+
+```bash
+pip install "nextaire_tools[docs]"
+mkdocs serve   # http://127.0.0.1:8000
+```
+
+## Citing nextaire_tools
+
+If you use `nextaire_tools` in academic work, please cite the relevant methodological
+paper(s) it is based on:
+
+> Petrić, V., Hussain, H., Časni, K., et al. (2024). *Ensemble Machine Learning,
+> Deep Learning, and Time Series Forecasting: Improving Prediction Accuracy for
+> Hourly Concentrations of Ambient Air Pollutants.* Aerosol and Air Quality Research,
+> 24, 230317. doi:10.4209/aaqr.230317
+
+> Jiménez-Navarro, M. J., Lovrić, M., Kecorius, S., Nyarko, E. K.,
+> Martínez-Ballesteros, M. (2024). *Explainable deep learning on multi-target time
+> series forecasting: An air pollution use case.* Results in Engineering, 24,
+> 103290. doi:10.1016/j.rineng.2024.103290
+
+> Račić, N., Ružičić, S., Petrić, V., et al. (2026). *Assessment of contributors to
+> airborne PAHs and heavy metals in PM₁₀ using temporal, spatial, traffic and
+> heating data in explainable machine learning models.* Atmospheric Environment: X,
+> 29, 100413. doi:10.1016/j.aeaoa.2026.100413
+
+## License
+
+`nextaire_tools` is released under the [MIT License](LICENSE).