ValerianFourel/sgt-bavaria-soc-2002-2023

As of Tuesday, 8th July 2025, we are currently looking at:

base_path_data = '/home/vfourel/SOCProject/SOCmapping/Data'

We have the 2 excel files:

file_path_LUCAS_LFU_Lfl_00to23_Bavaria_OC = f"{base_path_data}/LUCAS_LFU_Lfl_00to23_Bavaria_OC.xlsx"
# and
/lustre/home/vfourel/SOCProject/SOCmapping/Data/LUCAS_LFU_Bavaria_OC_joint_data_modified.xlsx

The full dataset of eligible points (And more !) can be found at: LUCAS_LFU_Lfl_00to23_Bavaria_OC.xlsx

The

Inside ofthe folder OC_LUCAS_LFU_LfL_Coordinates, we have the points coordinates in gps points of the sample location, and the XY coordinates matching them in the cation inside of the RasterTensorData Version, we also have to take into account which files it is in, this has been precomputed.

Raster Data Extraction Script - ./Data/Preprocessing/SamplePoints/samplePoints.py

This script is part of the SOC mapping project's preprocessing pipeline, specifically handling the extraction of environmental and remote sensing data from raster files at sample point locations.

Location in Project Structure

SOCmapping/
└── Data/
    └── Preprocessing/
        └── SamplePoints/
            └── samplePoints.py

Purpose

The samplePoints.py script serves as the critical preprocessing step that bridges raw raster data with machine learning-ready datasets. It extracts values from multiple environmental raster datasets at specific coordinate points, creating the feature matrices necessary for soil organic carbon (SOC) prediction models.

Key Features

• Multi-source Data Processing: Handles various raster datasets including:

  • MODIS Net Primary Productivity (NPP) - yearly
  • Elevation data - static
  • Land Surface Temperature (LST) - yearly and seasonal
  • Leaf Area Index (LAI) - yearly and seasonal
  • Soil and Total Evapotranspiration - yearly and seasonal

• Parallel Processing: Uses ProcessPoolExecutor to leverage multiple CPU cores for faster processing

• Spatial Indexing: Implements KD-trees for efficient spatial queries to find the correct raster tiles

• Multiple Coordinate Systems: Handles coordinate transformation between WGS84 and raster coordinate systems

Main Functions

Core Processing Functions

• build_kdtree_from_npy_file(): Creates spatial index from coordinate arrays
• find_closest_indices(): Finds nearest raster tiles for given coordinates
• get_tif_ID(): Determines which raster tile contains a coordinate point
• get_tif_ArrayPosition(): Converts geographic coordinates to raster array indices

Data Processing Workflows

• process_all_subfolders(): Processes regular coordinate grids (1 million sample points)
• process_all_subfolders_OC_yearly(): Processes LUCAS soil sampling data by year
• process_all_subfolders_OC_seasons(): Processes LUCAS soil sampling data by season

Input Data

1. Coordinate Files:

   • coordinates_Bavaria_1mil.csv: 1 million sample coordinates across Bavaria
   • LUCAS_LFU_Lfl_00to23_Bavaria_OC.xlsx: LUCAS soil sampling locations with temporal data

2. Raster Data: Organized in hierarchical folders by data type, temporal resolution, and time period

Output

The script generates .npy files containing coordinate positions and corresponding raster array indices, organized in a directory structure matching the input raster data organization:

Data/
├── Coordinates1Mil/
│   ├── YearlyValue/
│   │   ├── MODIS_NPP/2000/coordinates.npy
│   │   ├── LST/2001/coordinates.npy
│   │   └── ...
│   └── StaticValue/
│       └── Elevation/coordinates.npy
└── OC_LUCAS_LFU_LfL_Coordinates/
    ├── YearlyValue/
    └── SeasonalValue/

Performance Optimizations

• Parallel Processing: Utilizes all available CPU cores
• Memory Efficient: Processes data in chunks to avoid memory overflow
• Progress Tracking: Uses tqdm for progress monitoring
• Error Handling: Robust error handling for missing or corrupted data

Dependencies

• numpy, pandas: Data manipulation
• rasterio: Raster data processing
• scipy.spatial.cKDTree: Spatial indexing
• multiprocessing, concurrent.futures: Parallel processing
• tqdm: Progress bars

Usage

The script is designed to run as a standalone processor that:

1. Builds spatial indices for all raster datasets
2. Processes coordinate files to extract raster positions
3. Saves results for downstream machine learning applications

This preprocessing step is essential for creating training datasets where environmental variables serve as features for predicting soil organic carbon content.

Integration with SOC Mapping Pipeline

This script is a crucial component in the data preprocessing workflow, preparing the spatial data needed for the machine learning models that will predict soil organic carbon levels across Bavaria. The extracted features at each sample point will be combined with actual SOC measurements to train and validate the prediction models.