Facing the challenging problems of data quality, the ambitious targets of greenhouse gas emissions need practical forecasting methods to aid in formulating policy. A forecasting method, which has shown strong potential, is the rolling grey prediction model. To enable accurate forecasting, this study develops an adaptive grey model, which captures the essential features of a developing trend. The novel model is combined with a buffered rolling method to enhance accuracy. Compared with conventional models, the adaptive grey model with a buffered rolling method improves the adaptability in pursuit of data characteristics and it uses a nonlinear programming method to generate a satisfactory time response function for prediction. The proposed approach is constructed to forecast Chinese greenhouse gas emissions from 2017 to 2025 and compared with other benchmark models. Empirical applications show that the proposed model has advantages over others. The forecast data are decomposed by logarithmic mean Divisia index, and suggestions on emission mitigation are put forward based on factors analysis. (C) 2019 Elsevier Ltd. All rights reserved.