eo-processor 0.3.0

High-performance Rust UDFs for Earth Observation processing

eo-processor

High-performance Rust UDFs for Earth Observation (EO) processing with Python bindings.

Overview

eo-processor is a framework that provides Rust-based User Defined Functions (UDFs) for common Earth Observation and geospatial computations. These functions can be used within local and remote (Dask/Kubernetes) workflows, leveraging PyO3 to create highly efficient and optimized operations.

The Rust implementation bypasses Python's Global Interpreter Lock (GIL), making it ideal for:

• Long-running computations on large satellite imagery
• Parallel processing with Dask
• XArray apply_ufunc and map_blocks workflows
• CPU-intensive geospatial operations

Features

• High Performance: Rust-accelerated computations that bypass Python's GIL
• Easy Integration: Works seamlessly with NumPy, XArray, and Dask
• Common EO Indices: Pre-implemented functions for NDVI, NDWI, and generic normalized differences
• Type Safe: Full type hints for Python IDE support
• Flexible: Supports both 1D and 2D arrays with automatic dimension detection

Installation

From Source

Requirements:

• Python 3.8+
• Rust toolchain (install from rustup.rs)

# Install maturin for building
pip install maturin

# Build and install the package
maturin develop --release

# Or build wheel for distribution
maturin build --release
pip install target/wheels/*.whl

Usage

Basic Usage

import numpy as np
from eo_processor import ndvi, ndwi, normalized_difference

# Compute NDVI from NIR and Red bands
nir = np.array([0.8, 0.7, 0.6])
red = np.array([0.2, 0.1, 0.3])
ndvi_result = ndvi(nir, red)
print(ndvi_result)  # [0.6, 0.75, 0.33333333]

# Note: the functions in `eo_processor` now return NumPy arrays directly.
# They no longer return a (array, dims) tuple or any dims metadata.

# Works with 2D arrays (images)
nir_image = np.random.rand(1000, 1000)
red_image = np.random.rand(1000, 1000)
ndvi_image = ndvi(nir_image, red_image)

# Compute NDWI (water index)
green = np.array([0.3, 0.4, 0.5])
ndwi_result = ndwi(green, nir)

# Generic normalized difference: (a - b) / (a + b)
custom_index = normalized_difference(band_a, band_b)

XArray Integration

import xarray as xr
from eo_processor import ndvi

# Create XArray DataArrays
nir = xr.DataArray(nir_data, dims=["y", "x"])
red = xr.DataArray(red_data, dims=["y", "x"])

# Apply using xr.apply_ufunc
ndvi_result = xr.apply_ufunc(
    ndvi,
    nir,
    red,
    dask="parallelized",
    output_dtypes=[float],
)

Dask Integration (Parallel Processing)

import dask.array as da
import xarray as xr
from eo_processor import ndvi

# Create large Dask arrays (chunked for parallel processing)
nir_dask = da.random.random((10000, 10000), chunks=(1000, 1000))
red_dask = da.random.random((10000, 10000), chunks=(1000, 1000))

# Wrap in XArray
nir_xr = xr.DataArray(nir_dask, dims=["y", "x"])
red_xr = xr.DataArray(red_dask, dims=["y", "x"])

# Compute NDVI (bypasses GIL, enables true parallelism)
ndvi_result = xr.apply_ufunc(
    ndvi,
    nir_xr,
    red_xr,
    dask="parallelized",
    output_dtypes=[float],
)

# Compute result
ndvi_computed = ndvi_result.compute()

Using map_blocks with Dask

import dask.array as da
from eo_processor import ndvi

nir_dask = da.random.random((5000, 5000), chunks=(500, 500))
red_dask = da.random.random((5000, 5000), chunks=(500, 500))

# Apply to blocks (each block processed independently)
ndvi_result = da.map_blocks(
    ndvi,
    nir_dask,
    red_dask,
    dtype=np.float64,
)

result = ndvi_result.compute()

Available Functions

Normalized Difference Functions

• normalized_difference(a, b): Generic normalized difference (a - b) / (a + b)
• normalized_difference_1d(a, b): 1D version
• normalized_difference_2d(a, b): 2D version

Vegetation Indices

• ndvi(nir, red): Normalized Difference Vegetation Index
• ndvi_1d(nir, red): 1D version
• ndvi_2d(nir, red): 2D version

Water Indices

• ndwi(green, nir): Normalized Difference Water Index
• ndwi_1d(green, nir): 1D version
• ndwi_2d(green, nir): 2D version

Performance

The Rust implementation provides significant performance improvements over pure Python/NumPy, especially for large arrays:

import numpy as np
import time
from eo_processor import ndvi

# Large array
nir = np.random.rand(5000, 5000)
red = np.random.rand(5000, 5000)

# Rust implementation
start = time.time()
result_rust = ndvi(nir, red)
time_rust = time.time() - start

# NumPy implementation
start = time.time()
result_numpy = (nir - red) / (nir + red)
time_numpy = time.time() - start

print(f"Rust: {time_rust:.4f}s")
print(f"NumPy: {time_numpy:.4f}s")
print(f"Speedup: {time_numpy/time_rust:.2f}x")

Development

Building

# Development build
maturin develop

# Release build
maturin develop --release

Testing

# Run Rust tests
cargo test

# Run Python tests (if pytest is installed)
pytest

Running Examples

# Basic usage examples
python examples/basic_usage.py

# XArray/Dask examples (requires: pip install eo-processor[dask])
python examples/xarray_dask_usage.py

Why Rust + PyO3?

1. Performance: Rust provides C-level performance with memory safety
2. GIL-Free: Rust code releases the Python GIL, enabling true parallelism
3. Type Safety: Compile-time guarantees reduce runtime errors
4. Easy Integration: PyO3 makes it seamless to call Rust from Python
5. Modern Tooling: Cargo and maturin provide excellent development experience

Use Cases

• Processing large satellite imagery datasets (Sentinel, Landsat, etc.)
• Real-time vegetation monitoring using NDVI
• Water body detection using NDWI
• Custom spectral indices computation
• Distributed processing on Dask/Kubernetes clusters
• Time-series analysis of Earth Observation data

License

MIT License - see LICENSE file for details

Contributing

Contributions are welcome! Please feel free to submit issues or pull requests.

Citation

If you use this library in your research, please cite:

@software{eo_processor,
  title = {eo-processor: High-performance Rust UDFs for Earth Observation},
  author = {Ben},
  year = {2025},
  url = {https://github.com/BnJam/eo-processor}
}