A Spatio-temporal Framework for Soil Property Prediction with Digital Soil Mapping (DSM)

This new architecture, incorporates spatial information using a base convolutional neural network (CNN) model and spatial attention mechanism, along with climate temporal information using a long short-term memory (LSTM) network.

In case you need more information, feel free to send an email: nafiseh.kakhani@uni-tuebingen.de

Paper: https://arxiv.org/abs/2308.03586

To access LUCAS topsoil dataset: https://esdac.jrc.ec.europa.eu/content/topsoil-physical-properties-europe-based-lucas-topsoil-data

1. Prerequisites:

   • Install Conda: Make sure you have Conda installed on your system.

2. Clone the Repository:

   git clone https://github.com/moienr/SoilNet.git

3. Create the Environments:

    cd SoilNet

   3.1. for training:

   conda env create -f requirements/pytorch_reqs.yml

   3.2. for dataset:

   conda env create -f requirements/geemap_reqs.yml

4. Activate the Environment:

   4.1. for train:

   conda activate pytorch

   4.2. to download the dataset:

   conda activate geemap

1. Run the Code:

   Flags are explained in the next section.

   python train.py -ne 100 -tbs 8 -ne 10 -ca resnet101

   Although to train, you're gonna need to have the .csv files. namely LUCAS dataset under the flag of --lucas_csv and the TerraClimate dataset under the flag of --climate_csv_folder_path.

The output is a Training Plot and a JSON file containing all of the results of the cross-validation. all will be saved in the results/ folder