Abstract:

Forest volume is a crucial indicator for assessing forest resource abundance, ecosystem health, and carbon sequestration capacity. Accurate predictions of forest volume are essential for sustainable forest management and carbon neutrality planning. This study comprehensively compared the performance of four methods for predicting forest volume: the auto regressive moving average(ARMA) model combined with the partial least squares method (ARMAP), the ordinary least squares method (ARMAO), the grey system theory (GST), and the BP neural network (BPNN). Using forest land area and forest volume data from Chengfeng Town, Yongtai County, Fuzhou City, from 1986 to 1999, models were developed and validated with data from 1998 and 1999. The results indicate that the highest prediction accuracy by effectively filtering colored noise, with significantly lower residual variance and prediction errors compared with other models. Further validation using Zhejiang Province's 2020 forest volume data demonstrated the model's strong generalization ability. This study shows that the ARMAP model can achieve high-precision and efficient forest volume predictions under complex disturbance scenarios, providing methodological support for forest resource management and carbon sink evaluation.

Key words: forest volume, ARMA model, partial least squares method