Forecasting mammal population dynamics can assist management to keep pest species at acceptable densities. We used 29-years of monthly monitoring Apodemus agrarius in cropland in Yuqing County, Guizhou Province, China to develop a Bayesian hierarchical model of trap catch data that was composed of observation (capture) and population sub-models. The aim was to empirically understand the effects of environmental variables (rain, temperature, and crop types) on the population dynamics, and then to use estimated parameters to predict population outbreaks at the site level before they occur. Our population models, which include variables of temperature and proportion cover of crop types, were able to forecast A. agrarius outbreaks 3 months in advance. The presence of trees had a negative impact on mouse density. The predictive strength of all models was better than using historical mean monthly trap catch data. The model that included the mean temperature for the 3-month period 4–6 months prior to the current month had the best predictive strength for A. agrarius density. The observation model revealed that the capture probability of mice in a given month increased with increasing rainfall. The forecasts can contribute to planning and deployment of control measures to avoid crop damage. The models indicate the population dynamics of this species could be affected by climate change, changes in the agricultural system, and combinations of both factors, which are consistent with global trends. This generic modelling approach can be adapted to predict likely trends in the density of other species, for which there are long-term monitoring data.