| Preferential flow(PF)refers to the phenomenon in which fluids pass through porous media by bypassing a large part of the matrix and selecting preferential pathways to flow through the media at a faster rate.It is widely present in different soils and landscapes and has a significant implication for ecohydrological processes.However,the controlling factors of PF are complex,and its occurrence mechanism is unique,making it difficult to determine its spatial distribution and temporal occurrence patterns based on traditional statistical methods.Additionally,due to limited monitoring data and less attention to PF processes in alpine mountainous areas,the occurrence mechanism of PF in such regions is still not well understood.Therefore,it is necessary to analyze the factors influencing PF in alpine mountainous areas based on soil moisture observations and utilize methods such as machine learning to establish the spatial distribution and temporal occurrence patterns of PF.This will enhance our understanding of the occurrence mechanism of PF and the soil hydrological processes in mountainous areas.To achieve this,the study area selected for this research is the Qilian Mountains in the upstream of the Heihe River.A large-scale and long-term soil moisture monitoring network was established based on 32 representative soil moisture observation stations,covering the growing seasons from 2014 to 2019.The monitoring depths include 5 cm,15 cm,25 cm,40 cm,and 60 cm,with measurements taken at 30-minute intervals.By identifying the differences in response time to rainfall between shallow and deep soil moisture,PF events were recognized.Combining the spatial characteristics of the stations and the temporal features of infiltration events,spatial analysis methods,traditional statistical methods,and machine learning techniques were employed to analyze the spatial(vertical and horizontal)and temporal variations of PF.The main spatial and temporal factors controlling the occurrence of PF were identified,and models were developed to depict the spatial distribution and temporal patterns of PF.The major findings of this study are as follows:1.Spatial variation of PFAt the profile scale,as the soil depth increases,the average and standard deviation of the relative proportion of PF in different soil layers gradually decrease,while the coefficient of variation increases.Additionally,the average soil water storage increment of individual PF events also increases gradually.At different sites,as vegetation coverage increases,the overall proportion of PF at the sites gradually rises.The main soil layers where PF is detected also become deeper,consistent with the distribution of vegetation roots,indicating that roots play an important role as preferential pathways for PF.Some sites in the eastern and central parts of the study area exhibit a higher overall proportion of PF,while sites in the western part show a relatively lower proportion.This pattern aligns with the spatial variations in regional rainfall,vegetation,soil organic matter,and clay content.2.Spatial distribution pattern of PFAt the watershed scale,a significant linear relationship was observed between the overall proportion of site PF and soil organic matter content(R~2=0.13)and normalized difference vegetation index(NDVI:R~2=0.11).However,using the Random Forest method,soil saturated hydraulic conductivity and residual water content were identified as the primary factors controlling the spatial occurrence of PF.Multiple regression analysis was further conducted to establish an empirical prediction equation for the overall proportion of PF at the sites in the Qilian Mountains.The results indicate that using soil saturated hydraulic conductivity and residual water content(Adjusted R~2=0.39;RMSE=11.65%)provides a more accurate prediction of the overall proportion of PF at the sites compared to soil organic matter and NDVI(Adjusted R~2=0.2;RMSE=13.36%).Therefore,the Random Forest method can identify the complex controlling effects of spatial factors on PF occurrence,while traditional statistical methods may not yield optimal main controlling factors.Furthermore,the equation for predicting the overall proportion of PF at the sites based on soil saturated hydraulic conductivity and residual water content can effectively reflect the spatial distribution characteristics of PF in the study area.3.Temporal variability and patterns of PF occurrenceDue to the influence of vegetation,rainfall,and seasonal changes in soil moisture,PF exhibits significant seasonal variations.The relative proportions of PF were found to be higher in summer months(June:12.73%;July:16.55%;August:13.78%)than in other months(May:9.22%;September:11.14%;October:4.01%).The impact of initial soil moisture on PF is related to soil sand content.In soil with high sand content,dry initial conditions promote hydrophobicity and facilitate the occurrence of PF.However,in soils with low sand content,wet initial conditions are more favorable for PF.Furthermore,using Classification and Regression tree methods,the time occurrence patterns of PF were established for three typical vegetation types and the entire study area to determine the occurrence conditions of PF.The results indicate that PF mainly occurs when the soil is moist(excluding hydrophobic soils),vegetation is dense,and rainfall amount and intensity are high.Moreover,the above factors need to exceed certain threshold values for PF to occur,and the threshold values for the temporal factors differ among different vegetation types.In summary,spatial factors(soil properties and vegetation)and temporal factors(initial soil moisture and rainfall characteristics)jointly control the occurrence of PF.The study also demonstrates the great potential of machine learning methods in exploring the spatiotemporal occurrence patterns of PF.The findings contribute to understanding the occurrence mechanism of PF,enhancing our understanding of soil hydrological processes in mountainous areas,and providing scientific support and technical tools for regional water resource management. |
PDF Full Text Request