H.138 Dataset
=============

Description
-----------
* Target Soil Properties: SOC, pH, Clay
* Groups of Features: MIR
* Sample size: 138
* Number of Features: 2,489
* Coordinates: With coordinates (EPSG: 32649)
* Location: Hubei, China
* Sampling Design: Two sampling designs: (1) adapted latin hypercube sampling taking into account legacy samples, correlation and accessibility and (2) uncertainty guided sampling based on uncertainty predictions from a random forest model (Stumpf et al. 2017)
* Study Area Size: 420 ha
* Geological Setting: Sedimentary rocks, mainly dolomite with silt and limestone formed in the middle and lower Jurassic
* Previous Data Publication: Full dataset published in Wadoux et al. (2024)
* Contact Information:
    * Alexandre M.J.-C- Wadoux (Alexandre.Wadoux@inrae.fr), French National Institute for Agriculture, Food, and Environment (INRAE)
* License: CC BY-SA 4.0
* Publication/Modification Date (d/m/y): 28.02.25, version 1.0
* Changelog:
    * Version 1.0 (28.02.25): Initial release

Details
-------

Dataset
^^^^^^^
The dataset contains the following target soil properties and features:

Target Soil Properties:
""""""""""""""""""""""

SOC - Soil Organic Carbon
    * Code: ``SOC_target``
    * Unit: %
    * Protocol: Determined by the difference of total carbon and inorganic carbon, where total carbon was obtained through elemental analysis by measuring the CO₂ release during dry combustion (DIN ISO 10694) without acid pretreatment and inorganic carbon as 0.12 x the calcium carbonate content, determined by the gas-volumetric Scheibler Method (ISO 10693)
    * Sampling Date: June 2013, May, 2014 and November 2014
    * Sampling Depth: 0 - 20 cm

pH
    * Code: ``pH_target``
    * Unit: Unitless
    * Protocol: Measured in water suspension with a glass electrode with unspecified liquid:soil ratio
    * Sampling Date: June 2013, May, 2014 and November 2014
    * Sampling Depth: 0 - 20 cm

Clay
    * Code: ``Clay_target``
    * Unit: %
    * Protocol: Measured through fractioning the soil into the sand fractions by sieving, and the silt and clay fractions by x-ray sedimentation
    * Sampling Date: June 2013, May, 2014 and November 2014
    * Sampling Depth: 0 - 20 cm

Groups of Features:
"""""""""""""""""

MIR – Mid Infrared Spectroscopy
    * Number of Features: 2,489
    * Code(s): ``wn_5397.9``, ``wn_5396``, ``wn_5394`` ... ``wn_599.8``
    * Unit: % (Reflectance)
    * Sensing: VERTEX 70v FT-IR Spectrometer (Bruker Optik, Ettlingen, Germany), on dried and sieved samples (<2 mm) in the laboratory, spectral range was 7,500 - 370 cm^-1 at 0.4 cm^-1 intervals
    * Processing: Discarding irrelevant spectral data of the spectrum (7,500 - 5,397.9 cm^-1) and noisy edges of the spectrum (599.8 - 370 cm^-1)
    * Sampling Date: June 2013, May, 2014 and November 2014
    * Spectral Information (After Data Processing):
        * Data Representation: Wavenumber (in cm^-1)
        * Spectral Resolution: ~2 cm^-1
        * Spectral Range: 5,397.9 – 599.8 cm^-1

Examples
--------

.. code-block:: python

    from LimeSoDa import load_dataset, split_dataset
    from sklearn.linear_model import LinearRegression
    from sklearn.metrics import r2_score, mean_squared_error
    import numpy as np

    # Load and explore the dataset
    data = load_dataset("H.138")
    dataset = data["Dataset"]
    folds = data["Folds"]
    coords = data["Coordinates"]

    # Split into train/test using fold 1
    X_train, X_test, y_train, y_test = split_dataset(
        data=data,
        fold=1,
        targets=["pH_target", "SOC_target", "Clay_target"]
    )

    # Fit model and get predictions
    model = LinearRegression()
    model.fit(X_train, y_train)
    predictions = model.predict(X_test)

    # Calculate performance metrics
    r2 = r2_score(y_test, predictions)
    rmse = np.sqrt(mean_squared_error(y_test, predictions))
    print(f"R-squared: {r2:.7f}")
    print(f"RMSE: {rmse:.7f}")

References
----------

Wadoux, A. M. J.-C., Stumpf, F., & Scholten, T.. (2024). A catchment-scale dataset of soil properties and their mid-infrared spectra. Zenodo repository. https://doi.org/10.5281/zenodo.14557348

Stumpf, F., Schmidt, K., Goebes, P., Behrens, T., Schönbrodt-Stitt, S., Wadoux, A., Xiang, W. & Scholten, T. (2017). Uncertainty-guided sampling to improve digital soil maps. Catena, 153, 30-38.