Background and Objective: Land use and land cover (LULC) change is one of the principal drivers of degradation in the Zagros forests. Identifying the factors responsible for these changes and determining reliable indicators for predicting future trends are essential for sustainable forest management and ecosystem conservation. Persian oak (Quercus brantii Lindl.) forests in the Dena region of Kohgiluyeh and Boyer-Ahmad Province, southwestern Iran, provide a representative example of landscapes highly vulnerable to land use conversion because of their long history of under-canopy cultivation, relatively gentle topography, and increasing human pressure. This study aimed to evaluate the effects of canopy cover (1967 and 2023), topographic variables (slope, aspect, and elevation), and socioeconomic variables (distance to roads and settlements) on the probability of forest conversion to agricultural land between 1967 and 2023 using binary logistic regression. The study is distinguished by integrating visual interpretation of georeferenced historical aerial photographs, multi-temporal Google Earth imagery, and environmental datasets spanning 56 years, allowing simultaneous assessment of ecological, topographic, and anthropogenic drivers of land use change.

Materials and Methods: The study area covers approximately 131,213 ha of the Dena forests, dominated by Persian oak with widespread understory rainfed cultivation on gentle slopes. Land use, land cover, and canopy cover data were extracted through visual interpretation of georeferenced aerial photographs (1967) and Google Earth imagery (2023). A total of 176 circular sample plots (1,000 m²) arranged on a systematic 500 × 500 m grid were analyzed in ArcGIS using synchronized interpretation of historical and recent imagery. Topographic variables were derived from a digital elevation model, while socioeconomic variables were calculated as distances to roads and villages. Quantitative variables were standardized using Z-scores, and aspect was transformed into sine and cosine components. Binary logistic regression with the simultaneous-entry method was performed in SPSS version 26. Multicollinearity was assessed using Variance Inflation Factor (VIF) and Tolerance values, while Moran's I was used to evaluate spatial autocorrelation of model residuals. Model performance was assessed using the Omnibus test, Hosmer–Lemeshow goodness-of-fit test, Cox and Snell and Nagelkerke pseudo-R² values, the area under the receiver operating characteristic curve (AUC), and the classification matrix.

Results: Among the 176 sample plots, 91 plots (51.7%) were converted from forest to agricultural land during the 56-year study period, whereas 85 plots (48.3%) remained unchanged. Within the stable forest plots, 49.7% experienced reduced canopy cover, 14.4% showed increased canopy cover, and 35.7% remained unchanged, demonstrating substantial structural degradation even without land use conversion. Multicollinearity diagnostics indicated no significant collinearity (VIF < 1.723; Tolerance > 0.580). The final logistic regression model significantly improved upon the null model (Omnibus χ² = 27.722, df = 5, p < 0.001). The pseudo-R² values (Cox & Snell = 0.316; Nagelkerke = 0.421) indicated satisfactory explanatory power. The Hosmer–Lemeshow test confirmed good model fit (χ² = 8.738, df = 8, p = 0.365), indicating no significant difference between observed and predicted values. The model also demonstrated high predictive performance, with an AUC of 0.839 (95% CI: 0.779–0.900, p < 0.001) and an overall classification accuracy of 80.1%, including sensitivity of 82.4% and specificity of 77.6%. Moran's I (0.00056, z = 0.30, p = 0.763) confirmed the absence of spatial autocorrelation in the residuals.
Among the explanatory variables, canopy cover in 2023 was the strongest predictor of forest conversion, with a negative regression coefficient (B = −1.519), an odds ratio of 0.219, and high statistical significance (Wald = 26.88, p < 0.001). Its 95% confidence interval remained below one, indicating a stable inverse relationship between canopy cover and land use conversion. Each one-standard-deviation increase in canopy cover substantially reduced the probability of forest conversion. Distance from roads was the second most influential predictor (B = −0.445, Exp(B) = 0.641, Wald = 5.93, p = 0.015), with a 95% confidence interval of 0.448–0.917, showing that increasing distance from roads significantly reduced the likelihood of agricultural expansion. In contrast, canopy cover in 1967, slope, aspect, elevation, and distance from villages were not statistically significant (p > 0.05).

Conclusion: The conversion of more than half of the sampled forest plots over the past 56 years demonstrates the high susceptibility of the Dena forests to land use change and reflects broader regional trends of agricultural expansion as a major driver of Zagros forest degradation. The strong predictive effect of current canopy cover indicates that present forest structural condition is the most reliable early-warning indicator of future land use conversion. Moreover, the considerable reduction in canopy cover within nearly half of the unchanged forest plots suggests that structural degradation often precedes land use conversion and highlights the vulnerability of these forests to future disturbances. The significant influence of road proximity emphasizes the role of transportation infrastructure in facilitating agricultural expansion, whereas the non-significance of distance from villages may reflect the relatively uniform distribution of settlements or overlapping accessibility effects. Unlike many previous studies in the Zagros region, topographic variables were not significant predictors, likely because the study focused on areas already characterized by understory cultivation on gentle to moderate slopes, where human activities have largely overridden natural topographic constraints. Overall, the high AUC and classification accuracy demonstrate that binary logistic regression combined with long-term remotely sensed and environmental data provides an effective framework for analyzing long-term LULC dynamics in the Zagros forests. These findings emphasize that canopy cover reduction, particularly severe canopy decline, should be regarded as an early-warning indicator for prioritizing conservation and preventive forest management interventions.