An improved assessment of forest disturbance using a novel approach of combining a Gaussian mixture model with an EM algorithm

• DGE method detects forest disturbances with GMM and EM. • DGE outperforms traditional methods like ANAC, PGC, BL, and AI. • Broad DGE applications in disturbances, growth-climate relationship and other ecology fields. Forests have experienced unprecedented decline and mortality beyond their historical range in past decades, which is attributed to disturbances like drought, fire, insects and disease. Traditional disturbance detection methods that typically employ a time series perspective to identify discrete disturbance events within continuous tree growth signals and were mainly designed for gap-scale often fail to identify disturbances across populations. To more accurately identify growth suppression and release clusters based on the perspective of forest population dynamics, here we applied a novel method of detecting forest disturbances using a Gaussian mixture model with an expectation maximisation algorithm (DGE), to fit annual distributions of growth indicators, i.e., tree-ring index. We further compared our novel approach of DGE with five traditional methods based on two sets of real tree-ring data and simulated tree-ring data. The results show an improvement of accuracy (35.5 %–48.1 %), sensitivity (36.6 %–58.1 %), precision (21.2 %–51.5 %) and specificity (11.1 %–20.6 %) for the average of two real radial growth datasets and an improvement of accuracy (7.6 %–13.9 %), sensitivity (70 %), precision (100 %) and specificity (9.1 %–9.7 %) for the simulation radial growth dataset, indicating that our DGE approach performs much better than the traditional methods. This study contributes to an improved strategy and understanding for sustainable forest management in the context of climate warming, demonstrating that our DGE approach can be better applied to global forest disturbance detection under global change.