"Utilizing Otsu Thresholding, Gaussian Mixture Method and a Dual-Season NDVI Framework for Mapping Persistent Vegetation in Kirkuk, Iraq"

This repository contains the Google Earth Engine/Python code and derived datasets for the manuscript "Utilizing Otsu Thresholding, Gaussian Mixture Method and a Dual-Season NDVI Framework for Mapping Persistent Vegetation in Kirkuk, Iraq," prepared for submission to the Journal of the Indian Society of Remote Sensing. Study area: Kirkuk Municipality, northern Iraq (120.86 km²). Analysis period: 1 January–31 December 2025 (Sentinel-2 Level-2A surface reflectance). Methodology summary: Weekly NDVI time series were constructed from Sentinel-2 imagery (SCL-based cloud/shadow masking), with interquartile-range (k=1.5) outlier removal and a minimum two-image-per-week requirement, applied independently to a spring window (1 Feb–15 Jun) and a dry-season window (1 Jul–31 Oct). Seasonal reference dates were selected automatically (spring maximum NDVI: 5 April 2025, NDVI=0.162; dry-season minimum NDVI: 8 July 2025, NDVI=0.076), and ±15-day median composites were built around each date. Otsu's method was applied independently to each seasonal composite (spring threshold=0.3085; dry-season threshold=0.1992). Seasonally persistent vegetation was defined as pixels exceeding both thresholds (logical AND intersection). A Gaussian Mixture Model (k=4, selected via minimum BIC, full covariance) was applied to sub-classify persistent vegetation into four spectral-strength classes. Accuracy was assessed via blind visual interpretation of 250 stratified reference points, with area-weighted correction following Olofsson et al. (2014). Key results: Persistent vegetation = 4.96 km² (4.11% of study area, raw; 3.97% area-weighted, 95% CI 3.85–4.10%). Overall accuracy = 98.4% (raw), 99.87% (area-weighted); Cohen's kappa = 0.968. Contents: Jupyter notebook (Google Earth Engine/Python pipeline), municipal boundary shapefile, sensitivity analysis table, stratified accuracy-assessment points shapefile, derived summary tables (Excel), and Python package requirements. Full methodology details are in the accompanying README. License: CC-BY 4.0.