zsplit 1.0.0

Z-Split Normalization: Zero-Preserving Split Normalization for bipolar spectral indices

zsplit — Zero-Preserving Split Normalization

A normalization technique for bipolar spectral indices where zero carries physical meaning as a class boundary (e.g., water vs. non-water in flood detection).

Installation

pip install zsplit

With raster support:

pip install zsplit[raster]

Quick Start

import numpy as np
from zsplit import normalize

# Simulated raw ENDWI values (DN scale — extremely narrow range)
arr = np.array([-0.00041, -0.00020, 0.0, 0.00008, 0.00015])

result = normalize(arr)
print(result)
# [-1.     -0.488   0.      0.533   1.   ]

Raster Support

from zsplit import normalize_raster

normalize_raster('ENDWI_raw.tif', 'ENDWI_zsplit.tif')

Formula

         ⎧  x / max(X⁺)    if x > 0   →  [0,  1]
Z(x)  =  ⎨  0               if x = 0   →  0 (preserved)
         ⎩  x / |min(X⁻)|  if x < 0   →  [-1, 0]

Why Z-Split?

Property	Min-Max	Z-Score	Z-Split
Preserves zero as class boundary	❌	❌	✅
Stable on narrow near-zero ranges	❌	⚠️	✅
Suitable for Otsu thresholding	❌	⚠️	✅
Handles skewed distributions	❌	⚠️	✅
Multi-temporal direction preserved	❌	❌	✅
Output range	[0, 1]	Unbounded	[−1, 1]

Citation

@article{almoadi2026endwi,
  author  = {Almoadi, Abdulrhman},
  title   = {Reducing False Alarms in Urban Flood Detection: An Enhanced NDWI (ENDWI)
             with Hybrid Max Fusion on Sentinel-2 Data},
  journal = {EGUsphere [preprint]},
  year    = {2026},
  doi     = {10.5194/egusphere-2026-672}
}

License

MIT License