Work

Forest Loss or Stability? A Binary Analysis of Land Cover in Ferizaj (1985–2025)

Year: 2025
Type: conference-paper
Abstract: Forest cover mapping is used for characterizing environmental processes and for sustainable land use planning. This study monitored long-term forest cover changes for Ferizaj municipality (Kosova) from 1985 to 2025 through a binary classifier of non-forest versus forest. Multitemporal RapidEye images from the years 2014, 2020, 2022, and 2025 were processed through the Random Forest(RF) classifier of the Google Earth Engine (GEE), while the historical cadastral register was used for comparison for the year 1985. Performance of classification was measured based on accuracy indicators like Precision, Recall, F1-score, Balanced Accuracy, and 95% Confidence Intervals of Overall Accuracy. Performance was found to be high for all the years without exception, with F1-scores greater than 0.96 and Balanced Accuracy ranging from 94.8% to 99.0%. To discern temporal trends of change of forest cover, we employed the non-parametric Mann–Kendall trend test and Sen’s Slope Estimator through Python. It was observed that there existed a statistically significant downward trend (Z = −2.20, p = 0.027), which was characterised by a mean annual loss of approximately 0.35 km²/year. In total, the area of the forests was observed to reduce from 126 km² recorded in 1985 to 109.1 km² recorded in 2025, which represented a cumulative loss of nearly 17 km². These outcomes illustrate the importance of long-term monitoring to quantify deforestation and derive insight about landscape change. This integrated methodology of remote sensing, cadastral data, cloud processing, and statistical analysis forms a repeatable framework for assessing rapidly shifting land covers.
Author: Zariqi P.
Language: English
Open Access status: green
OpenAlex ID: W7209070953
FWCI:
Cites: 299
Cited by:
Related to: 10