Contrasting contributions of five factors to wheat yield growth in China by process-based and statistical models

China's wheat growth from 1981 to 2015 was faster than most countries, characterized by notable linear growth and more than 100 % rise in yield. Understanding the reasons for such fast growth is essential for stakeholders in developing countries to gain insight and be able to establish efficient yield-increasing strategies for the future. Here, using a statistical function and a process-based crop model, we dissected, quantified and compared the contributions of five causal factors to China's national wheat yield growth. For past yield growth from 1981 to 2015, the Environmental Policy Integrated Climate (EPIC) model estimated that five drivers have contributed to national wheat yield growth by 57.4 % (fertilizer), 37.9 % (cultivar), 7.9 % (irrigation), -1.0 % (area), and -2.2 % (climate), while the Cobb-Douglas (C-D) production function estimated contributions of 79.8 %, 16.1 %, 8.0 %, -2.6 %, and -1.2 %, respectively. Furthermore, we conducted a temporal two-paired comparison of the yield contributions estimated by the two methods and explored the reasons for such difference. Results indicated that the distinct nature of the two methods and inconsistency of the input data resulted in the divergent estimations in amount and temporal scale. The way to decrease the uncertainty of each factor's contribution in each of the methods was to maintain consistency in input data, clearly understand the methods' assumptions, and include as many interactions as possible. In addition, utilization of an ensemble of multiple models was able to achieve a more robust assessment, especially with the inclusion of both statistical and process-based models. This study has for the first time systematically compared the applicability of the two methods in evaluating the contribution of the five drivers to yield, and provided instructions for improvements. Future researchers should pay attention to developing a consistent manner of integrating multiple approaches, which have the potential partly to offset the weakness of individual methods and thus lead to better estimations.