[Objective]In semi-humid to semi-arid transition zones,water availability is a primary constraint on forest ecosystem functioning.In these regions,stem water storage(Ws)plays a critical role in buffering mismatches between transpirational demand and root water uptake,particularly under fluctuating soil moisture conditions.However,the dynamic responses of Ws to variations in soil moisture at different depths remain poorly quantified.Most existing studies have focused on whole-profile soil moisture or atmospheric drivers while neglecting the layered structure of water supply within the root zone.To address this knowledge gap,this study investigated the response characteristics and regulatory mechanisms of Ws in relation to soil moisture at different depths in a Larix principis-rupprechtii plantation in the Xiangshui River area of the Liupan Mountains,Ningxia,China,a typical semi-humid to semi-arid transition zone.[Methods]The research was conducted during the 2024 growing season,from May 10 to October 10.Four sample trees with different stem diameters were selected within a permanent 30 m×30 m plot.Stem sap flow was continuously measured at two heights—the crown base at 4.0 m and stem base at 0.35 m—using thermal dissipation probes(STDP30),with data recorded at 5-minute intervals.Ws was calculated from the difference in daily cumulative sap flow between the crown base and stem base.Soil volumetric water content was measured at three depths(0-20,20-40,and 40-60 cm)using 5TE sensors,and relative soil water content(RSWC)was calculated from the measured values.Meteorological variables,including solar radiation,air temperature,relative humidity,and precipitation,were recorded using an automatic weather station.Potential evapotranspiration(Ep)was calculated using the Penman-Monteith equation.Leaf area index(LAI)was measured every 7-10 days using an LAI-2200c canopy analyzer.To model the response of Ws to multiple interacting factors,an upper envelope method was first used to derive single-factor response functions for Ep,LAI,and RSWC at each soil depth.These functions were subsequently integrated into a modified Jarvis-Stewart model to simulate daily Ws.Even-numbered days were used for model calibration(n=76),whereas odd-numbered days were used for model validation(n=78).To further investigate nonlinear and asymmetric responses to changes in soil moisture,gradient boosting regression trees(GBRT)combined with k-fold cross-validation were used to improve model stability.Sensitivity analyses were performed by perturbing RSWC in each soil layer by±5%while holding Ep and LAI constant,thereby allowing the response magnitudes among soil layers to be compared.Finally,a factor-substitution method was used to quantify the relative contribution of each soil layer to Ws under dry((RSWC0-60 cm<0.36)and wet(RSWC0-60 cm≥0.36)conditions.[Results]The diurnal dynamics of Ws exhibited a consistent cycle of morning recharge,daytime release,and evening recharge.Daily Ws ranged from 0.16 to 3.69 kg·d-1,with a mean of 1.97 kg·d-1.Single-factor response analysis showed a quadratic relationship between Ws and Ep,with Ws reaching its maximum at an Ep threshold of 3.68 mm and decreasing slightly beyond this threshold.LAI and RSWC at all three soil depths exhibited saturating exponential relationships with Ws.Notably,the RSWC threshold below which soil moisture limited Ws increased with soil depth,from 0.35 at 0-20 cm to 0.36 at 20-40 cm,and 0.37 at 40-60 cm.The modified Jarvis-Stewart model performed well,with R2=0.82 and Nash-Sutcliffe efficiency(NSE)of 0.81 during calibration and R2=0.75 and NSE=0.74 during validation.The simulated cumulative Ws over the growing season was 296.79 mm,slightly lower than the observed value of 303.24 mm.The GBRT model revealed a clear asymmetric response:a 5%increase in RSWC produced a greater increase in Ws than the decrease in Ws caused by a 5%reduction in RSWC.Specifically,the mean increase in Ws following a 5%increase in RSWC was 0.31 kg·d-1,whereas the mean decrease following a 5%reduction in RSWC was only 0.19 kg·d-1.This asymmetry suggests a physiological priority for water recharge over excessive release under variable moisture conditions,thereby contributing to hydraulic safety and embolism avoidance.The sensitivity ranking was consistent across soil layers:0-20 cm>20-40 cm>40-60 cm,indicating that shallow soil moisture exerted the strongest short-term control over Ws dynamics.Contribution analysis showed that under wet conditions,the 0-60 cm layer contributed more to Ws than under dry conditions.The relative contributions of the 0-20,20-40,and 40-60 cm layers were 2.04%,11.88%,and 0.34%,respectively,under dry conditions,and increased to 5.16%,13.94%,and 1.27%,respectively,under wet conditions.The 20-40 cm layer consistently made the largest contribution under both moisture regimes,indicating that this layer represents a relatively stable source of water for Larix principis-rupprechtii.[Conclusion]First,the depth-dependent soil moisture thresholds indicate that water management should target specific root-zones layers rather than treating the entire soil profile uniformly.Information on these layer-specific thresholds can support precision irrigation and forest density regulation.Second,the asymmetric response to soil moisture perturbations implies that short-term wetting events may restore Ws more strongly than equivalent drying events deplete it,which has implications for rainfall-frequency effects and irrigation scheduling.Third,the dominant and stable contribution of the 20-40 cm soil layer highlights the importance of maintaining adequate soil moisture within this zone to support transpiration under both dry and wet conditions.In practical terms,regulating stand density and optimizing stand structure may reduce competition for water and enhance the stability of shallow and intermediate soil moisture,thereby improving forest resilience to climate variability.Nevertheless,this study has several limitations.The conclusions are based on observations from a single growing season characterized by relatively normal climatic conditions and the applicability of the model to years with extreme drought or prolonged wet conditions requires further validation.In addition,although the Jarvis-Stewart model captures the principal driving factors,its instantaneous-response structure and the buffering effects of stem capacitance may cause it to underestimate contributions from deep soil water under extreme conditions.Future studies should integrate multi-year observations and stable-isotope tracing to directly verify shifts in water sources among soil layers.Despite these limitations,the integrated modeling framework provides a robust quantitative approach for disentangling the effects of layered soil moisture on stem water storage and offers a scientific basis for water-oriented forest management in semi-humid to semi-arid transition zones.