The invasive Asian tiger mosquito (Aedes albopictus) is a major public health concern in Italy, particularly in the Po Valley, where it has contributed to repeated outbreaks of mosquito-borne diseases over the past two decades. In response, the Emilia-Romagna region (Northern Italy) has maintained an intensive surveillance program based on oviposition traps since 2010, generating a rich long-term observational dataset for this invasive species.

This study aims to develop a forecasting model that estimates abundance at both monitored and unmonitored locations, directly supporting regional mosquito control efforts. We implemented a stacked machine learning framework to forecast weekly Ae. albopictus distribution and abundance using ovitrap data and ecologically relevant environmental covariates, with the aim of quantifying spatial and temporal data needs under realistic surveillance constraints. Therefore, we evaluate model performance across a structured set of training-window configurations that vary both the temporal depth and the spatial coverage of the input data.

Our evaluation across configurations reveals that broader spatial coverage can compensate for a shorter recent temporal window. Configurations spanning only two recent years but incorporating more sampling locations matched or exceeded the predictive performance of configurations with longer historical records, suggesting that recency and spatial breadth are important for forecasting under stable environmental conditions. Regardless of the training-window scenario, all models consistently reproduced the seasonal and spatial patterns of Ae. albopictus. These findings offer practical guidance for designing operational mosquito surveillance and forecasting systems and support the use of abundance-based predictions to inform public health planning and vector control strategies.