Revealing urban vegetation dynamics: Comprehensive NDVI time series analysis and forecasting using ARIMA/SARIMA models in Karachi
Abstract
Climate change is increasing the vulnerability of cities, and it is therefore important to evaluate vegetation comprehensively for sustainability and resilience. This research utilizes advanced machine learning and remote sensing methods to examine urban vegetation dynamics in Karachi. Utilizing Google Earth Engine (GEE) for large-scale geospatial analysis, mean monthly NDVI values were obtained from Landsat 8 data from the years 2013 to 2022. Seasonal ARIMA (SARIMA) models were utilized to detect trends and seasonal variations in the NDVI time series. The results were utilized to forecast future NDVI trends, which were subsequently incorporated into conservation and urban planning initiatives following the model's verification of accuracy. The SARIMA (0,1,2)x(0,1,2,12) model was the most appropriate for the description of trend and seasonal behavior. The model provided the best overall fit among the tested configurations. Model diagnostics indicated a strong performance, with good values across standard statistical criteria such as log likelihood, AIC, BIC, and HQIC. Residual analysis further confirmed that the model adequately captured the underlying patterns. Predictions indicated a mean NDVI value of 0.196, with a 95% confidence interval of 0.146 to 0.246. Strong fluctuations in NDVI values over time included the highest values in September 2023 (0.232) and October 2024 (0.253), with lower values in the summer months, indicating clear seasonal fluctuations. This interdisciplinary effort offers a holistic assessment of urban vegetation dynamics with global applicability. By focusing on the importance of green spaces in the mitigation of urban heat island and climate adaptation, the research calls for their incorporation in urban planning. The findings provide practical recommendations toward sustainable development, urban planning, and environmental protection, demonstrating the capacity of sophisticated modeling methods to inform resilient urban ecosystems.