文献类型：期刊文献

英文题名：Beyond seasonality: A data-fusion approach reveals extreme climate and qualitative ecosystem traits as key litterfall drivers across diverse forests

作者：Chen, Sihan[1,2] Zhang, Huaiqing[1,2] Yang, Jie[1,2,3] Shi, Yukai[1,2] Tan, Jingwei[1,2] Wang, Fanyu[4] Lei, Hao[1,2] Guo, Menglei[1,2] Liu, Yang[1,2]

第一作者：Chen, Sihan

通信作者：Zhang, HQ[1];Zhang, HQ[2]

机构：[1]Chinese Acad Forestry, Inst Forest Resource Informat Tech, Beijing 100091, Peoples R China;[2]Natl Forestry & Grassland Sci Data Ctr, NFGSDC, Beijing 100091, Peoples R China;[3]Beijing Forestry Univ, Coll Forestry, Beijing 100083, Peoples R China;[4]Beijing Forestry Univ, Sch Technol, Beijing 100083, Peoples R China

年份：2026

卷号：182

外文期刊名：ECOLOGICAL INDICATORS

收录：;EI(收录号：20255019677409);Scopus(收录号：2-s2.0-105024062741);WOS:【SCI-EXPANDED(收录号:WOS:001637772500001)】；

基金：This study was supported by the National Key Research and Development Program of China (Grant number No. 2023YFF1303701). We express our sincere gratitude to the staff of the Chinese Ecosystem Research Network (CERN) for their dedication to long-term observation, rigorous data processing, and selfless sharing of the datasets. We specifically thank the following stations for providing the essential data: Ailao mountain forest station (ALF), Beijing ling mountain forest station (BJF), Xishuangbanna forest station (BNF), Changbai mountain forest station (CBF), Dinghu mountain forest station (DHF), Gongga mountain forest station (GGF), Heshan mountain forest station (HSF), Maoxianforest station (MXF), Puding forest station (PDF), Qingyuan forest station (QYF), Shennongjia forest station (SNF), and Yanting agriculture station (YTA) .

语种：英文

外文关键词：Forest ecosystem; Litterfall production; Extreme climatic; Multi-modal data; Explainable automated machine learning

摘要：Accurate prediction of monthly forest litterfall is vital for understanding ecosystem carbon and nutrient cycling. However, this process is governed by complex drivers that vary substantially across different forest ecosystems, posing significant challenges for predictive modeling, especially at high temporal resolutions. Existing models often overlook the combined influence of extreme climatic events and nuanced, qualitative site-specific characteristics. This study reveals that scale-dependent, monthly extreme climatic variables (e.g., maximum wind speed) and long-term, qualitative ecosystem traits (derived from text) emerge as critical drivers, in some cases surpassing the importance of traditional mean climatic indicators. To uncover these relationships, we developed a novel predictive framework by integrating 29 numerical environmental and temporal factors with unstructured textual data from 27 forest plots across China. We employed explainable automated machine learning approach and pre-trained language models to fuse these multi-modal data sources effectively. The resulting integrated model demonstrated significantly enhanced predictive performance (R2 = 0.791, RMSE = 22.775 g/m2/month, MAE = 15.951 g/m2/month) outperforming all models that relied solely on numerical data. Interpretability analysis confirmed that beyond dominant seasonal rhythms, key predictors for the model include monthly short-term extreme weather, acting as powerful triggers, and long-term, qualitative ecosystem characteristics (embedded features) that create the unique local context governing production patterns. In conclusion, this work provides both a highly accurate, interpretable tool for litterfall production prediction and a powerful, generalizable methodology for integrating qualitative knowledge into ecological processing modeling. Our findings offer critical data-driven support for advancing carbon cycle science and informing precise, climate-adaptive forest management.