Error Stats Code ported to Python

[mmaclay]

Geolocation Error Statistics Python Implementation

Overview

Converts MATLAB geolocation error statistics code (developed by Tim Holden) into a production-ready Python implementation that integrates with the CSDS pipeline. This is a first stab at making this error-stats code play nicely with image-matching outputs, and assumes Xarray as the standard format.

Key Changes

MATLAB to Python Conversion

• Translated Tim Holden's original MATLAB algorithm (183262revB_GeolocationStatistics.m) to Python
• Maintains mathematical accuracy and numerical precision from heritage implementation
• Preserves all 13 original test cases for validation against MATLAB reference results (100% agreement)

Basic Class architecture

• Main Class: ErrorStatsProcessor - handles geolocation error processing
• Configuration: GeolocationConfig dataclass with configurable parameters
• Input/Output: Xarray Dataset support for pipeline integration
• Error Handling: Validation and error management

Core Algorithm Functionality (from Matlab)

Processes geolocation errors through all matlab steps:

1. Angular to Distance Conversion - Converts lat/lon errors to meters using Earth radius
2. Coordinate Transformations - Transforms between sensor, Earth-centered, and local coordinate systems
3. View-Plane Projection - Decomposes errors into along-track/cross-track components
4. Nadir-Equivalent Scaling - Applies geometric corrections for off-nadir viewing angles
5. Performance Statistics - Computes pass/fail metrics against 250m threshold (39% requirement)

Testing

• Unit Tests: test_geolocation_error_stats.py using unittest
• Validation Tests: Regression testing against original MATLAB results
• Edge Case Coverage: Input validation, coordinate transformations, statistical calculations
• Integration Tests: End-to-end processing with heritage test data

Performance & Validation

• Numerical Accuracy: Matches MATLAB results within 0.1m tolerance
• Matlab Test Validation: 1203.26m mean error, 61.5% pass rate (8/13 cases < 250m)

Files Added

• geolocation_error_stats.py - Main implementation
• test_geolocation_error_stats.py - Comprehensive unit tests

[medley56]

My only real feedback is that I'm skeptical that this repo is the right place for this code. This feels like a debugging utility script for a specific mission (CLARREO?) and we're trying to really push Curryer towards being a generic SPICE/geolocation utility library rather than a repository for mission code.

[medley56]

This seems very mission-specific. Does this really belong in a generic geolocation library?

[mmaclay]

Adding a comment here to remind myself. I'm going to be pulling out mission-specific code during further development and integration of this with the GCS module - plan for now is that these variables will be named in a config json.