A 100 m annual soil salinization dataset from 1985 to 2024 in the Western Songnen Plain, China

In our study, we identified and classified soil salinization degrees in the Western Songnen Plain at 100 m spatial resolution using ground surveys data and remote sensing imagery, combined with machine learning algorithms over the period 1985 to 2024.

The publicly available data used in this paper, as well as the modeling code, are shown below:

1)        The remote sensing satellite data used in this study are freely available on GEE platform (available at https://code.earthengine.google.com/).

2)        Land cover data can be accessed at https://doi.org/10.5281/zenodo.4417810.

3)        SSSG dataset is available at  https://files.isric.org/public/global_soil_salinity.

4)        LDSS data are available at https://doi.org/10.6084/m9.figshare.13295918.v1.

5)        Yearly identification results of saline soils are available at https://code.earthengine.google.com/3dd47875d7455825297b9a6a8766a8ee.

6)        The data required for model inputs (TIR、SIT、PDI) are available at https://code.earthengine.google.com/7fc0e7ee2cf4cb6b744aaf4706a140c9.

The presented data file contains:

1)        Soil EC sampling points

2)        TIRSITPDI_predicted.mat

3)        Soil_EC_prediction_model.m

4)        Statistical results of soil salinization degree by area (1985-2024)

5)        Yearly degree of soil salinization (1985-2024)

6)        Yearly identification results of saline soils (1985-2024)

7)        Yearly soil EC results (1985-2024)

Samples:

The file, Soil_EC_sampling_points.csv, provides essential information on 942 sample points. Below are the descriptions for each column in the metadata:

•   Municipal: Location where the point is located
•   LnEC: The observed LnEC using a conductivity meter (Light Magnetic, model: DDS-307A).
•   TIR、SIT、PDI: Characteristic parameters in the models.

Models:

We trained on this data using eight models (NNF, GPR, LSBoost, KPLS, Tree, SVM, LR, and SVM Kernel) from the Matlab (R2024a) Toolbox. To optimize the NNF for accurate soil salinity estimation (R² = 0.467; RMSE = 0.729 dS m⁻¹), we trained the model using nested loops and fine-tuned parameters such as the training algorithm, the number of hidden layer nodes, and the learning rate. After 14,000 rounds of cycling, the optimal model TIRSITPDI_predicted.mat is finally obtained. Finally, the characteristic parameters with the model are put into Soil_EC_prediction_model.m to obtain the soil salinity values.

Results:

• Statistical results of soil salinization degree by area (1985-2024) shows statistical results for each class (Slightly, Moderately, Highly, Extremely) as well as total saline  acreage in the Western Sonnen Plain for the years 1985-2024.
• Yearly degree of soil salinization (1985-2024) presents the results of saline soil classification based on the U.S. Salinity Laboratory’s classification system for the Western Sonnen Plain from 1985-2024.
• Yearly identification results of saline soils (1985-2024) displays the results of saline soil identification in the Western Sonnen Plain from 1985-2024.
• Yearly soil EC results (1985-2024) exhibits the results of model-predicted soil EC content in the Western Sonnen Plain for the years 1985-2024.