Abstract: Aiming for clarifying the potential distribution characteristics of canopy rainfall partitioning of the loess area, we explored the process of rainfall partitioning across eight typical forest stands (Pinus tabuliformis forest, Robinia pseudoacacia forest, Platycladus orientalis forest, mixed forest of Robinia pseudoacacia-Pinus tabuliformis, mixed forest of Platycladus orientalis-Robinia pseudoacacia, Quercus wutaishanica forest, Populus davidiana forest, mixed forest of Quercus wutaishanica-Populus davidiana), and used boosted regression trees (BRT) to quantify the relative influences of stand structures and meteorological environment factors. We established multiple regression relationships according to the most influential factors extracted by BRT, and applied to the dataset of mining to verify the performance of the BRT-derived predictive model. The results showed that the percentages of throughfall (TF), stemflow (SF), and canopy interception (I_c) in total precipitation were 24.5%-95.1%, 0-13.6%, and 0.7%-55.7% among eight typical forest stands, respectively. For the individual rainfall threshold of TF, coniferous forest (3.06±1.21 mm) was significantly higher than broad-leaved forest (1.97±0.52 mm), but there was no significant difference between coniferous forest and broad-leaved mixed forest (3.01±0.98 mm). There was no significant difference in the individual rainfall threshold of SF among different composition stands. BRT analysis showed that stand structure factors accounted for a relatively small proportion for TF and SF, respectively. By contrast, stand structure factors dominated the I_c. Rainfall was the most important factor in determining TF and SF. Tree height was the most important factor in determining I_c, followed by rainfall, canopy area, diameter at breast height, and stand density. Compared with the general linear function and the power function, the prediction effect of BRT prediction model constructed here on TF and SF had been further improved, and the prediction of canopy interception still needed to explore. In conclusion, the BRT model could better quantitatively evaluate the effects of stand structure and meteorological environmental factors on rainfall partitioning components, and the performance of the BRT predictive model could satisfy and lay the foundation for the optimization strategy for stand configuration.

Key words: throughfall, stemflow, canopy interception, boosted regression tree, relative influence