A toolkit for transforming, aggregating, and upscaling various types of vegetation data. Currently used to transform CNN derived canopy height predictions into green volume and upscale them to FORCE Datacube Grid for further use in satellite deep learning models.
- Input: High-resolution canopy height data from CNN predictions
- Outputs: Green volume data, binary canopy data (>2.5m = canopy), original canopy height data
- Code: canopy_height_upscaling.py
- Input: Three raster products from step 1
- Output: For each input, mean, SD, and sum per 10m pixel aligned to FORCE Grid
- Code: canopy_height_upscaling.py
- Input: Vector data on damaged vegetation
- Output: Raster data of vegetation damage at 10m resolution, FORCE grid-aligned
- Code: vector_to_raster_damage_analysis.py
This versatile toolkit handles data type transformations and resolution changes for various ecological and forestry applications.
The following table describes how green volume data is derived from the canopy height prediction. The constants for sealed surface, grassland and cropland in m³/m² are the same in m for the CNN based vegetation height predictions. The constants of 10 % and 25 % substracted for high vegetation represent estimations of missing vegetation volume around tree stems.
| Land Cover Type | Green Volume |
|---|---|
| Sealed surface & Water | 0 m³/m² |
| Grassland | 0.5 m³/m² |
| Cropland | 1.0 m³/m² |
| Shrubs (< 5m) | Pixelsize x canopy height |
| Shrubs and Trees (5 - 7m) | Pixelsize x canopy height - 10% |
| Trees (> 7m) | Pixelsize x canopy height - 25% |
- GDAL, Geopandas, Geowombat, xarray
- Anaconda download
- developed on Windows
- Create a virtual Conda environment with the required Python version and requirements file:
conda create --name gwenv python=3.8
conda activate gwenv
conda install -c conda-forge gdal==3.4.3
conda config --env --add channels conda-forge
conda config --env --set channel_priority strict
cd ../environment
pip install -r requirements.txt- Sebastian Lehmler- Core functionality and canopy height upscaling use case
- Shadi Ghantous - Code improvements and damage analysis use case