Three-dimensional quantification method of lateral preferential flow in dry-hot valley based on GPR detection and SSIM analysis

Abstract

This invention discloses a three-dimensional quantification method for subsurface lateral preferential flow in arid and hot valleys based on GPR (Geophysical Permeability Reduction) detection and SSIM (Spatial Infiltration Model) analysis. It integrates time-delayed GPR scanning with artificial infiltration experiments and applies SSIM analysis to depict LPF (Limited Placement Flow) path activation, network structure evolution, and humid zone distribution in a three-dimensional manner. This method achieves, for the first time, three-dimensional non-destructive quantification of the dynamics of the LPF network on arid and hot valley slopes. The established hydrogeophysical data acquisition and post-processing workflow provides a crucial and transferable research method for quantitatively analyzing the interaction between structural connectivity and process connectivity in slope runoff generation and confluence. This method overcomes the limitations of traditional two-dimensional slope LPF analysis by quantitatively characterizing the spatiotemporal dynamics of subsurface three-dimensional connectivity. It provides direct support for improving the water cycle mechanism in arid and hot valleys and improving hydrological models based on physical mechanisms, and provides a scientific basis for sustainable management decisions in water conservation areas of arid and hot valleys under climate change conditions.

Description

Technical Field

[0001] This invention relates to the field of underground lateral priority flow quantification technology, specifically to a three-dimensional quantification method for underground lateral priority flow in hot and dry valleys based on GPR detection and SSIM analysis. Background Technology

[0002] The critical zone of a hillside is a fundamental hydrological response unit of the terrestrial water cycle, regulating runoff generation and water retention. In the face of increasingly frequent extreme hydrological events, understanding groundwater flow distribution is crucial for flood mitigation and water resource management. Hillsides regulate runoff generation and water retention through subsurface lateral preferential flow (LPF), a key driver of hydrological connectivity. In water source catchments, infiltration processes are typically dominated by preferential flow, which bypasses the soil matrix and accelerates the hydrological response. Specifically, the key processes for slope runoff formation are when infiltration encounters low-permeability structures (such as soil-bedrock interfaces or soft rock layers) or enters interconnected pore systems (such as soil conduits or macropores). Quantifying LPF dynamics is essential for improving hydrological models and water resource management.

[0003] Numerous studies have confirmed the prevalence of LPF (Low-Level Fluctuation) across different spatiotemporal scales. Spatially, research ranges from porosity activation at the local plot scale to threshold responses at the hillside scale and the broader watershed-level runoff generation. Temporally, LPF dynamics encompass rapid timescales of single storm events to long-term seasonal wet-dry cycles. To explain these multi-scale observations at the mechanistic level, conceptual models (such as the "overflow model") have proposed threshold-driven initiation mechanisms and lateral diffusion. A key advance lies in identifying two distinct control mechanisms: structural connectivity, involving the static tectonic features of macropores, root channels, and fissures in weathered bedrock, and structural connectivity, where these structures dynamically activate under specific groundwater conditions to guide water flow.

[0004] Despite these advances, directly and non-destructively observing the spatiotemporal evolution of the LPF remains a persistent obstacle. Geophysical methods have emerged as powerful tools to overcome these challenges, with ground-penetrating radar (GPR) attracting particular attention due to its sensitivity to water-induced dielectric changes and its ability to resolve the fine organization of preferred pathways. However, most studies still rely on qualitative interpretation or two-dimensional GPR data analysis, failing to capture the continuous three-dimensional evolution and connectivity of the LPF network. Furthermore, the lack of robust quantitative indicators to resolve time-series signals hinders the integration of geophysical imagery with process-based hydrological models. These limitations underscore the urgent need to establish a workflow capable of quantifying the three-dimensional evolution of LPF connectivity and directly correlating it with slope runoff response.

[0005] The dry-hot valley slopes provide a natural experiment for studying these structure-process interactions. Different canopy interception, root structures, and bioturbation create unique subsurface environments. Specifically, the natural circulation of plant roots forms a macroporous network underground, a key physical channel for groundwater flow, determining the degree of LPF activation. However, the mechanisms by which these different structures regulate the three-dimensional spatiotemporal evolution of LPF remain unclear. Specifically, this duality is central to understanding vegetation-regulated rainfall infiltration: different canopies and root systems simultaneously alter the physical subsurface network (structure) and activate the water input (process) that activates that network, potentially leading to different hydrological behaviors.

[0006] To fill these research gaps, this invention proposes a three-dimensional quantification method for subsurface lateral preferential flow in arid and hot valleys based on GPR detection and SSIM analysis. It integrates time-delayed GPR with artificial infiltration experiments and applies SSIM analysis to map LPF activation, network evolution, and flow distribution in a three-dimensional manner. This invention aims to provide a transferable framework for quantifying structure-process interactions in slope hydrological systems, offering scientific support for optimizing hydrological model representation and water resource management in arid and hot valleys. Summary of the Invention

[0007] This invention provides a three-dimensional quantification method for subsurface lateral preferential flow in hot and dry valleys based on GPR detection and SSIM analysis. It is the first to achieve non-invasive three-dimensional quantification of LPF dynamics on hot and dry valley slopes. The reliability of the geophysical interpretation is verified by conducting paired experiments on adjacent secondary forest and plantation slopes and by synchronous hydrological monitoring.

[0008] According to the first aspect, one embodiment provides a three-dimensional quantification method for subsurface lateral preferential flow in hot and dry valleys based on GPR detection and SSIM analysis, the method comprising: A dry-hot valley slope sample plot was selected in the target area to carry out artificial infiltration experiments and time-delayed GPR scanning. The time-delayed GPR scanning was carried out by laying out multiple parallel GPR measurement lines perpendicular to the slope in the sample plot and collecting time-series GPR data of multiple parallel measurement lines at multiple time points before and after infiltration. The raw time-series GPR data of each parallel survey line at multiple time points before and after infiltration were preprocessed. Based on the preprocessed time-series GPR data of each survey line at multiple time points before and after infiltration, the SSIM difference image of each survey line before and after infiltration at each time point is obtained by calculating the SSIM difference between the GPR images before and after infiltration at each time point. Based on the set LPF activation threshold, the binary segmentation method was used to perform binary segmentation data analysis on the SSIM change difference images of each survey line before and after infiltration at each time node, in order to extract the LPF activation domain. The SSIM variation difference images of each measuring line before and after infiltration at each time node were standardized to a unified Cartesian coordinate system. Radar signal interpolation was performed on the non-radar scan area between adjacent measuring lines. At the same time, combined with LPF activation domain extraction and LPF activation domain volume and proportion calculation, a three-dimensional spatiotemporal dynamic evolution model of the LPF activation domain volume of the entire runoff sample three-dimensional detection area was constructed.

[0009] Secondly, artificial infiltration experiments and delayed GPR scans were conducted, specifically including: An infiltration trench is dug at a predetermined distance from the boundary of the upslope sample plot. Deionized water is continuously supplied into the trench through a polyvinyl chloride pipe to promote infiltration into the soil matrix and prevent surface runoff. Multiple GPR lines are set at equal intervals in parallel within the sample plot, and are all perpendicular to the slope aspect, in order to dynamically monitor the path of groundwater flow on the slope.

[0010] Artificial infiltration experiments and delayed GPR scans were conducted, specifically including: Before the infiltration experiment, two baseline GPR scans were collected to characterize the initial conditions. Then, two infiltration experiments were conducted: In the first experiment, 25 liters of deionized water were infiltrated at a rate of 3.57 liters / minute. After the initial 25 liters of infiltration was completed, delayed GPR scans were performed along all measuring lines at 5, 10, and 20 minutes after infiltration. In the second experiment, 25 liters of deionized water were infiltrated at a rate of 1.56 liters / minute. Delayed GPR scans were performed along all measuring lines at 5, 10, 20, 30, and 50 minutes after infiltration.

[0011] Secondly, the raw time-series GPR data of each parallel survey line at multiple time points before and after infiltration are preprocessed, specifically including: Baseline correction: Eliminates DC offset and convolution filtering to suppress constant offset and low-frequency trends; Zero-interval correction: Eliminates the influence of air waves and aligns the start time with the ground coupling; Signal gain: Amplitude compensation is performed using energy attenuation gain to correct radar signal attenuation caused by propagation depth and enhance deep reflection; Frequency filtering: reduces high-frequency random noise and low-frequency background noise; Background elimination: Suppresses horizontal coherence stripes and system noise; Kirchhoff offset correction: corrects geometric distortions caused by velocity nonuniformity and non-perpendicular incidence, thereby improving the geometric fidelity of reflector positioning accuracy and path recognition.

[0012] Furthermore, based on the preprocessed time-series GPR data of each survey line at multiple time points before and after infiltration, the SSIM difference image of each survey line before and after infiltration at each time point is obtained by calculating the SSIM difference image between the GPR images before and after infiltration at each time point, specifically including: Based on the preprocessed time-series GPR data of each survey line before and after infiltration, two-dimensional GPR profile images of each survey line before and after infiltration are generated, and the SSIM between the two-dimensional GPR profile images before and after infiltration is calculated to obtain the SSIM change map of each survey line before and after infiltration. At the same time, the average SSIM value of each survey line is calculated by averaging the SSIM values ​​of each survey line to obtain the average SSIM value of each survey line.

[0013] Secondly, the SSIM between the two-dimensional GPR profile images before and after infiltration is calculated, specifically including:

[0014] in, These represent the sampling time window sequences before and after infiltration, respectively. They represent the first and second digits in the sequence, respectively. Amplitude values ​​at each sampling point; The amplitude is the local average. Let be the standard deviation of the radar amplitude, describing the local amplitude variability; The covariance of radar amplitude before and after infiltration; This represents the number of sampling points within the local window; This is a stability constant to avoid abnormally amplified calculation results or inability to calculate.

[0015] Furthermore, based on the set LPF activation threshold, a binary segmentation method was used to perform binarized segmentation data analysis on the SSIM change difference images before and after infiltration at each time point for each survey line, in order to extract the LPF activation domain, specifically including: If SSIM > 0.8, meaning the structural similarity index before and after infiltration is high and the soil moisture change is limited, it is defined as a non-humid area to quantify background noise; if SSIM < 0.5, meaning the structural similarity index before and after infiltration is low and the soil moisture increase is significant, it is determined to be an LPF activation domain signal.

[0016] Secondly, the SSIM variation difference images of each survey line before and after infiltration at each time point are standardized to a unified Cartesian coordinate system, and radar signal interpolation processing is performed on the non-radar scan area between adjacent survey lines, specifically including: In the Cartesian coordinate system, x: slope aspect; y: survey line spacing; z: subsurface depth; For the non-radar scanning area between adjacent survey lines, a cubic spline interpolation algorithm is used to perform radar signal interpolation and filling processing along the y-axis.

[0017] Then, the method further includes: Paired experiments were conducted using two types of sample plots: typical dry-hot valley secondary forest and plantation forest. The comparative study revealed the regulatory patterns of different vegetation types on the spatiotemporal dynamics, activation mechanisms, and hydrological effects of underground lateral preferential flow.

[0018] In addition, the method also includes: By combining high-resolution monitoring of slope water cycle data such as rainfall, throughfall, soil moisture content, and stratified groundwater runoff, the reconstructed three-dimensional spatiotemporal dynamic evolution model of LPF activation domain volume is subjected to correlation verification and inter-group significance test analysis, thereby achieving cross-validation of hydrogeophysical modeling results.

[0019] This invention provides a three-dimensional quantification method for subsurface lateral preferential flow (LPF) in hot and dry valleys based on geophysical resonant flow (GPR) detection and SSIM analysis. It integrates time-delayed GPR and artificial infiltration experiments, and applies structural similarity index analysis to map LPF activation, network evolution, and flow distribution in three dimensions. This represents the first non-invasive three-dimensional quantification of LPF dynamics on hot and dry valley slopes. Paired experiments conducted on adjacent secondary forests (BF) and plantations (CF) revealed fundamental differences in LPF mechanisms dominated by vegetation type. These geophysical models were robustly validated using independent hydrological data. The introduced workflow provides a crucial and transferable tool for quantifying structure-process interactions in slope hydrology. This method overcomes the limitations of two-dimensional analysis by providing quantitative three-dimensional connectivity indices, offering a direct pathway to improving the physical basis of hydrological models and providing a scientific basis for sustainable water resource management strategies in hot and dry valleys under climate change conditions. Attached Figure Description

[0020] Figure 1 A flowchart of a three-dimensional quantification method for subsurface lateral preferential flow in hot and dry valleys based on GPR detection and SSIM analysis is provided as an embodiment of the present invention; Figure 2 A schematic diagram of the infiltration channels and GPR survey lines of BF and CF in a three-dimensional quantification method for subsurface lateral preferential flow in a dry-hot valley based on GPR detection and SSIM analysis, provided as an embodiment of the present invention. Figure 3 The following is a time-delayed GPR imaging and SSIM analysis result of LPF activation in the BF in a three-dimensional quantification method of subsurface lateral preferential flow in a hot and dry valley based on GPR detection and SSIM analysis, provided as an embodiment of the present invention. Figure 4 The results of time-delayed GPR imaging and SSIM analysis of LPF activation in the CF in a three-dimensional quantification method of subsurface lateral preferential flow in hot and dry valleys based on GPR detection and SSIM analysis are provided in one embodiment of the present invention. Figure 5 This invention provides a quantitative characterization of the evolution of the mean SSIM over time in a three-dimensional quantification method for subsurface lateral preferential flow in hot and dry valleys based on GPR detection and SSIM analysis, which is used to characterize the subsurface wetting process. Figure 6 The three-dimensional spatiotemporal dynamic evolution of the LPF network in a three-dimensional quantification method for subsurface lateral preferential flow in hot and dry valleys based on GPR detection and SSIM analysis is provided in one embodiment of the present invention. Figure 7 Spatiotemporal characteristics of soil moisture response to rainfall events in BF and CF plots in a three-dimensional quantification method for subsurface lateral preferential flow in hot and dry valleys based on GPR detection and SSIM analysis, provided as an embodiment of the present invention; Figure 8 The response of multi-depth subsurface runoff in BF and CF plots to rainfall event (#2) in a three-dimensional quantification method of subsurface lateral preferential flow in hot and dry valleys based on GPR detection and SSIM analysis, as provided in an embodiment of the present invention. Figure 9 A conceptual diagram illustrating different rainfall-runoff mechanisms in a three-dimensional quantification method for subsurface lateral preferential flow in arid and hot valleys based on GPR detection and SSIM analysis, provided as an embodiment of the present invention. Figure 10 This invention provides a three-dimensional quantification method for subsurface lateral preferential flow in hot and dry valleys based on GPR detection and SSIM analysis, which examines the connectivity differences between surface and subsurface structures in the critical zone of hot and dry valleys. Detailed Implementation

[0021] The present invention will now be described in further detail with reference to specific embodiments and accompanying drawings. Similar elements in different embodiments are referred to by associated similar element reference numerals. In the following embodiments, many details are described to facilitate a better understanding of the invention. However, those skilled in the art will readily recognize that some features may be omitted in different situations, or may be replaced by other elements, materials, or methods. In some cases, certain operations related to the present invention are not shown or described in the specification. This is to avoid obscuring the core parts of the invention with excessive description. For those skilled in the art, detailed description of these related operations is not necessary; they can fully understand the related operations based on the description in the specification and general technical knowledge in the art.

[0022] Furthermore, the features, operations, or characteristics described in the specification can be combined in any suitable manner to form various embodiments. At the same time, the steps or actions in the method description can be rearranged or adjusted in a manner obvious to those skilled in the art. Therefore, the various orders in the specification and drawings are only for the clear description of a particular embodiment and do not imply a necessary order, unless otherwise stated that a particular order must be followed.

[0023] The first embodiment of this invention provides a three-dimensional quantification method for subsurface lateral preferential flow in hot and dry valleys based on GPR detection and SSIM analysis. The following is a combination of... Figure 1 Please provide a detailed explanation.

[0024] like Figure 1 As shown, in step S100, a dry-hot valley slope sample plot is selected in the target area to carry out artificial infiltration experiments and delayed GPR scanning. The delayed GPR scanning involves laying multiple parallel GPR lines perpendicular to the slope direction in the sample plot and collecting time-series GPR data on the multiple parallel lines at multiple time points before and after infiltration.

[0025] The above steps specifically include: S110, Selection of Research Site and Experimental Slope Specifically, this study was conducted at the Yanbian Positioning Observation and Research Station in Panzhihua, Sichuan Province, China (101°46′E, 48°48′N), a typical dry-hot valley area of ​​the Jinsha River. The dry-hot valley is a typical representative of the ecologically fragile areas in southwestern my country, characterized by concentrated rainfall, a long dry season, and strong evaporation. Extreme hydrological events are frequent, and the slope hydrological processes are highly sensitive, making it an ideal natural experimental field for quantitative LPF research. The Yanbian Station is located on the western edge of the Yangtze Platform, in the south-central section of the Kang-Dian axis, and on the eastern edge of the Qinghai-Tibet Plateau, with a complex geological structure. The oldest exposed strata in the area can be traced back to the Archean to Paleoproterozoic deep to medium-deep metamorphic rock series, with occasional Neogene and Quaternary sedimentary rocks. The bedrock is mainly composed of metamorphic clastic rocks, calcareous schist, and conglomerate. The climate is a subtropical semi-arid monsoon climate with distinct wet and dry seasons.

[0026] The vegetation exhibits a clear altitudinal gradient: planted forests dominate below 1200 meters, while secondary forests are distributed in mid-altitude areas (1200-1600 meters). To investigate the impact of vegetation type on LPF (Low-Altitude Forest Flooding), this study selected two adjacent slopes as paired experiments: a mid-altitude secondary forest (BF) plot and a low-altitude planted forest (CF) plot. Each plot included a 10m × 20m runoff quadrat. To control for topographic differences, paired plots with similar slopes and orientations were selected. To ensure undisturbed long-term monitoring, the infiltration experimental setup was placed in adjacent parallel locations to the permanent plots, maintaining consistent observation conditions across different slopes.

[0027] S120, Artificial Infiltration Experiment and Delayed GPR Scan: Artificial infiltration experiments were conducted in 1.5m × 1.5m quadrats within each forest plot. In each plot, a 0.3m deep trench was dug 0.5m outward from the uphill plot boundary. Figure 2 As shown, Figure 2 (a) shows a schematic diagram of the infiltration channel and GPR survey line layout of the BF. Figure 2 (b) shows a schematic diagram of the infiltration channel and GPR survey line layout for the CF. Figure 2 (c) is a schematic diagram of the geophysical exploration experiment design. Five parallel GPR lines (spaced 0.3 meters apart) are arranged perpendicular to the slope to capture the path of downhill water flow. Deionized water is continuously supplied to the trench through PVC pipes to promote infiltration into the soil matrix and avoid triggering surface runoff.

[0028] Two infiltration tests were conducted in this embodiment: the first test involved infiltrating 25 liters of deionized water at a rate of 3.57 liters per minute, and the second test involved infiltrating 25 liters of deionized water at a rate of 1.56 liters per minute. No surface runoff occurred, confirming that the water had infiltrated into the ground, which is consistent with the results of other preferential flow studies.

[0029] Prior to the infiltration experiment, two baseline GPR scans were conducted in this embodiment to characterize the initial conditions. After the initial 25-liter infiltration was completed, delayed GPR scans were performed along all survey lines at 5, 10, and 20 minutes post-infiltration. Subsequently, a second 25-liter infiltration was applied, and GPR data were collected at 5, 10, 20, 30, and 50 minutes post-infiltration, as shown in Table 1.

[0030] GPR data were acquired using a 750 MHz Mala GX system (Mala Geoscience, Sweden), with a fixed transmitter-receiver spacing of 0.14 meters. Acquisition parameters were set as follows: time window of 30.83 nanoseconds, track interval of 0.0114 meters, 296 sampling points per track, and stacking interval of 1.

[0031] To ensure high repeatability, field markers and measurement templates were used to fix the endpoints of the survey line and the GPR scanning path, and GPR data were collected in the same direction to minimize geometric misalignment. The radar wave velocity was calibrated to 0.092 m / nanosecond by using the reflected signals from three vertical steel rods (0.3 m deep) placed perpendicular to the GPR survey line. A uniform velocity was assumed in the time-depth conversion, which is consistent with mainstream GPR application methods.

[0032] Table 1. Sequence and Time Arrangement of Events in the Field Experiment Conducted by CF Forest

[0033] like Figure 1 As shown, in step S200, the original time-series GPR data of each parallel survey line at multiple time points before and after infiltration are preprocessed.

[0034] Specifically, all raw GPR data were standardized using ReflexW v9.5 software (Sandmeier Scientific Software GmbH, Karlsruhe, Germany). The processing flow included: (i) baseline correction (eliminating DC offset and convolution filtering) to suppress constant offset and low-frequency trends; (ii) zero-time correction to eliminate air wave effects and align the start time with ground coupling; (iii) amplitude compensation using energy attenuation gain to correct radar signal attenuation due to propagation depth and enhance deep reflections; (iv) frequency filtering to reduce high-frequency random noise and low-frequency background noise; (v) background removal to suppress horizontal coherence stripes and system noise; and (vi) Kirchhoff offset correction to correct geometric distortions caused by velocity nonuniformity and non-perpendicular incidence, thereby improving reflector positioning accuracy and geometric fidelity for path identification.

[0035] like Figure 1As shown, in step S300, based on the time-series GPR data of each survey line at multiple time points before and after infiltration after preprocessing, the SSIM difference image of each survey line at each time point is obtained by calculating the SSIM difference between the GPR images before and after infiltration at each time point.

[0036] The above steps specifically include: Based on the preprocessed time-series GPR data of each survey line before and after infiltration, two-dimensional GPR profile images of each survey line before and after infiltration are generated, and the SSIM between the two-dimensional GPR profile images before and after infiltration is calculated to obtain the SSIM change map of each survey line before and after infiltration. At the same time, the average SSIM value of each survey line is calculated by averaging the SSIM values ​​of each survey line to obtain the average SSIM value of each survey line.

[0037] Calculating the SSIM between the two-dimensional GPR profile images before and after infiltration, specifically including:

[0038] in, These represent the sampling time window sequences before and after infiltration, respectively; They represent the first and second digits in the sequence, respectively. Amplitude values ​​at each sampling point; The amplitude is the local average. Let be the standard deviation of the radar amplitude, describing the local amplitude variability; The covariance of radar amplitude before and after infiltration; This represents the number of sampling points within the local window; This is a stability constant to avoid abnormally amplified calculation results or inability to calculate.

[0039] like Figure 1 As shown, in step S400, based on the set LPF activation threshold, the binary segmentation method is used to perform binary segmentation data analysis on the SSIM change difference images of each survey line before and after infiltration at each time node, so as to extract the LPF activation domain.

[0040] Specifically, to distinguish LPF activation from the delayed GPR SSIM field, this embodiment employs a binary segmentation method based on site-specific operational thresholds. To consider local heterogeneity and measurement noise, the following criteria are used: if SSIM > 0.8 (high structural similarity index before and after infiltration with limited soil moisture change), it is defined as a non-wet zone to quantify background noise; if SSIM < 0.5 (low structural similarity index before and after infiltration with significant soil moisture increase), it is determined to be an LPF activation domain signal. Sensitivity analysis (SSIM = 0.4-0.6) confirms the robustness of the above criteria in identifying activation patterns, effectively eliminating false positive results caused by non-infiltration. More importantly, the activation time defined by SSIM is highly consistent with independent slope-scale hydrological observations (soil moisture increment and runoff), ensuring consistency in the interpretation and process of the extracted domain.

[0041] Spatiotemporal patterns of LPF activation revealed by time-lapse GPR: Time-lapse GPR imaging combined with SSIM analysis revealed significant spatiotemporal patterns of LPF activation between two vegetation types. Figure 3 and Figure 4 ).

[0042] Figure 3 Delayed GPR imaging and SSIM analysis of LPF activation in BF Figure 3 (a) is a GPR profile before infiltration, where the root reflection signal marked by the red hyperbola is clearly indicated. Figure 3 (b) and (d) are GPR profiles after the first infiltration (TL1-1 and TL1-3). Figure 3 (f), (h), and (j) are GPR profiles (TL2-1, TL2-3, and TL2-5) after the second infiltration. Figure 3 (c), (e), (g), (i), and (k) are corresponding SSIM change plots, quantifying the deviation from the pre-wetting state; the lower the SSIM value, the greater the water-induced change, i.e., LPF activation. The yellow dashed boxes mark areas with significant changes in reflectance patterns in the delayed GPR images. In BF forests, the first infiltration triggers a rapid response at time TL1-1. Five minutes after infiltration (TL1-1, ... Figure 3 c) Significant LPF activation signals (SSIM < 0.5) were detected at depths of 0.6–1.0 meters in GPR lines 1, 2, and 5. However, this initial activation signal exhibited spatial discontinuity among the GPR lines during the initial infiltration period, with only weak signal changes detected in GPR lines 3 and 4 at time point TL1-1. Figure 3 c). By the 20-minute mark (TL1-3, Figure 3 e), a continuous cross-slope LPF path is formed. In the pre-wetting radar map, the radar reflection hyperbola is considered perpendicular to the GPR survey line (marked in red). Figure 3 The root system signal of a) corresponds to the yellow rectangles B, C, and E. Figure 3 Significant LPF activation signals were detected at positions b, d, f, h, and j. Figure 3 c, e, g, i, and k) indicate that these coarse roots triggered the LPF. Other strongly activated regions (such as boxes A, D, F, and G) correspond to areas with higher underground heterogeneity.

[0043] Figure 4 Delayed GPR imaging and SSIM analysis of LPF activation in CF. Figure 4 (a) is a GPR profile before infiltration, where the root reflection signal marked by the red hyperbola is clearly indicated. Figure 4 (b) and (d) are GPR profiles after the first infiltration (TL1-1 and TL1-3). Figure 4 (f), (h), and (j) are GPR profiles after the second infiltration (TL2-1, TL2-3, and TL2-5). Figure 4 Figures (c), (e), (g), (i), and (k) show the corresponding SSIM variation diagrams, quantifying the deviation from the pre-wetting state. In contrast, the LPF activation signal in CF is weaker and shallower during the first infiltration period. Figure 4 During the TL1-1 phase, the LPF activation signal was mainly confined to survey lines 1 and 2, primarily occurring at depths of 0.2-0.4 meters and 1.0 meters, while survey lines 3, 4, and 5 remained structurally stable. Figure 4 c). During the TL1-3 phase, stronger responses were observed at survey lines 3 and 4 (0.2-0.5 m and 1.0 m depths, respectively). Figure 4 e). Survey line 5 showed only weak shallow activation (0.2 m) after the second infiltration (TL2-1, Figure 4 g), indicating that its lateral flow process is slower and spatially restricted. Similar to BF, the activation signal matches the root characteristics (boxes A, B, E, G), but the signal intensity is consistently low. Figure 4 c, e, g, i, and k).

[0044] Linear average SSIM quantitatively summarizes these initial dynamic characteristics (such as...) Figure 5 ), Figure 5 In the middle (a), BF is represented. Figure 5 (b) represents CF. After the initial infiltration, the average SSIM value of all survey lines in the BF zone decreased significantly faster and at a much greater rate than that in the CF zone. The decrease was most significant in survey line 4 of the BF zone (from 0.769 to 0.422). Figure 5 a), while the CF zone No. 3 measuring line saw the largest decrease (from 0.783 to 0.599, Figure 5b). This indicates that the LPF activation in the BF region is more rapid and spatially distributed, in stark contrast to the weaker and more localized activation characteristics of the CF region.

[0045] like Figure 1 As shown, in step S500, the SSIM change difference images of each measuring line before and after infiltration at each time node are standardized to a unified Cartesian coordinate system, and radar signal interpolation filling is performed on the non-radar scanning area between adjacent measuring lines. At the same time, combined with LPF activation domain extraction and LPF activation domain volume and proportion calculation, a three-dimensional spatiotemporal dynamic evolution model of LPF activation domain volume of the entire runoff sample three-dimensional detection area is constructed.

[0046] Specifically, to analyze the three-dimensional spatial organization of these paths, the classified two-dimensional SSIM profiles of the five parallel survey lines were standardized to a unified Cartesian coordinate system (x: slope aspect; y: survey line spacing; z: subsurface depth). To fill the scanning gaps (0.3-meter spacing) between adjacent GPR survey lines, a cubic spline interpolation algorithm was used along the y-axis. This method ensures the continuity of the first and second derivatives of the flow field, thereby generating a smooth volumetric representation of the subsurface space. This enables the quantitative calculation of the activated LPF volume and visualization of the evolution of network connectivity in different infiltration experiments.

[0047] Three-dimensional evolution comparison of LPF networks among vegetation types: Three-dimensional reconstruction of SSIM activation volumes reveals fundamental differences in LPF network evolution (e.g. Figure 6 ), Figure 6 Figure (a) shows the three-dimensional evolution of the LPF network beneath the BF radar detection area after two infiltration events. The blue area in the pie chart represents the volume percentage of LPF in the subsurface space detected by the radar. A lower SSIM value indicates stronger hydrological connectivity of the LPF network. Figure 6 Figure (b) shows the corresponding data for CF. In the BF region, the activated LPF volume reached 28.74% of the GPR scan area after the first infiltration pulse (TL1-1), significantly increased to 50.85% at TL1-2, and peaked at 58.14% at TL1-3, forming a broad and highly interconnected network. After the second infiltration pulse, the activated volume gradually decreased from 57.33% (TL2-1) to 52.76% (TL2-5). This downward trend after the peak indicates that efficient drainage or redistribution was achieved through the established network.

[0048] CF networks exhibit latency threshold-driven response ( Figure 6(b) The activation volume was initially limited to 14.10% (TL1-1), increasing slightly to 16.99% in the TL1-2 stage. A significant connectivity threshold appeared in the TL1-3 stage, with the volume doubling to 32.41%, marking an improvement in network connectivity. During the second stage of infiltration, the volume increased linearly, eventually reaching a peak of 46.31% (TL2-4). Unlike the rapid saturation of the BF network, the CF network exhibited gradual development, achieving full continuity only after the second infiltration.

[0049] In this embodiment, the method further includes: combining high-resolution monitoring of slope water cycle with synchronous rainfall, throughfall, soil moisture content, and stratified groundwater runoff, to perform correlation verification and inter-group significance test analysis on the reconstructed three-dimensional spatiotemporal dynamic evolution model of LPF activation domain volume, thereby achieving cross-validation of hydrogeophysical modeling results.

[0050] The hydrological monitoring network is structured as follows: 1. Rainfall and Saturated Rainfall Monitoring Atmospheric precipitation was monitored at the meteorological station using a tilting rain gauge (JD-01, resolution 0.1 mm; nominal accuracy ±2%), with a recording interval of 10 minutes. Rainfall was monitored at the BF and CF sample plots using a penetration rain monitoring device (TY2-L-DG; nominal measurement error ≤4%) connected to an automatic recording system, with rainfall monitored at 10-minute intervals.

[0051] Based on common standards in rainfall analysis, a dual-threshold event separation criterion is adopted to divide continuous rainfall records into independent events. This criterion is based on the minimum value of the rainless period. and the minimum rainfall of the event Specifically, if a rainfall event is separated from the next rainfall event by at least 6 hours without precipitation, the event is considered an independent event. To eliminate interference from light rain and focus on events more likely to trigger hillside runoff responses, this embodiment retains rainfall events with a total precipitation >3 mm. Rainfall and seepage data from May 1 to September 30, 2025, were used to quantify the differences in forest precipitation transport by different vegetation types. The number and layout of seepage samplers in each vegetation type plot were consistent. Rainfall events were classified into four intensity levels based on total precipitation (P): light rain (3-10 mm), moderate rain (10-25 mm), heavy rain (25-100 mm), and torrential rain (>100 mm).

[0052] 2. Soil Moisture Dynamics Soil moisture monitoring points were set up at the upslope, middle slope, and downslope of each runoff monitoring plot. Each monitoring point used a TEROS 11 probe (manufactured by Meter Group, Pullman, Washington, USA, with a resolution of 0.001 cm³ / cm²). -³) Volumetric water content (θ) was recorded every 10 minutes. In the BF quadrat, sensors were installed at depths of 10, 20, 30, 40, and 50 cm; in the CF quadrat, the monitoring profile extended to 100 cm (10, 20, 30, 50, 70, and 100 cm). High-frequency records from July 13 to September 30, 2025, were analyzed.

[0053] To minimize diurnal temperature fluctuations and instrument drift noise while identifying soil moisture response to rainfall, this embodiment adopted the standard proposed by Li et al. The triggering condition for the response was: an increase in θ by ≥1 volume percentage (depth ≤ 0.2 m) or ≥0.5 volume percentage (depth > 0.2 m) within 10 minutes. This embodiment calculated three indicators: (1) response frequency, i.e., the proportion of rainfall events that trigger a response at a specific depth; (2) response amplitude, i.e., the maximum recharge; and (3) rise time, i.e., the time interval from the start of the response to the peak value of θ, as a proxy indicator for infiltration rate.

[0054] 3. Runoff Collection Depth-stratified runoff collection systems were installed at the downslope boundaries of each sample site. Stainless steel diversion channels intercepted the flow at depths of 10, 60, and 100 cm. The water flow was piped to the TY2-L-DG system and recorded at 10-minute intervals using a CR1000X data logger. Representative rainfall events during the study period were selected to characterize natural runoff generation patterns.

[0055] Verification of soil moisture dynamics and runoff response: Independent hydrological monitoring during natural rainfall events validated two distinct LPF mechanisms ( Figure 7 and Figure 8 ).

[0056] Figure 7 The spatiotemporal characteristics of soil moisture response to rainfall events in BF and CF plots are shown. Figure 7 (a) and (b) show the frequency distribution of soil moisture response at different depths and slope positions of BF and CF, respectively. Figure 7 (c) and (d) compare the response amplitude and rise time at different depths in the two regions. The median line represents the median, the upper and lower limits of the box represent the interquartile range, the bar extends to 1.5 times the interquartile range, hollow dots represent outliers, and asterisks indicate significant differences between the two regions as determined by the Mann-Whitney U test (* P < 0.05, ** P < 0.01, *** P < 0.001).

[0057] Analysis of 37 rainfall events (May to September 2025) revealed significant differences in soil moisture response patterns. The basal fraction (BF) maintained a high response frequency across the entire soil profile, with a domain-average frequency of 0.776. Figure 7 a). CF exhibits a clear vertical stratification characteristic, with the response frequency dropping sharply below a depth of 20 cm, resulting in a significant decrease in the domain average frequency to 0.704 ( Figure 7 b). BF produces a sustained, larger soil moisture pulse (Δθ) at all depths. Figure 7 c). Rise time (Trise) showed depth-dependent differences: CF had a shorter Trise at a depth of 10 cm, while BF had a significantly faster response at a depth of 50 cm. Figure 7 d), which is consistent with its ability to rapidly penetrate deep layers.

[0058] Figure 8 The response of subsurface runoff at multiple depths in the BF and CF regions to rainfall event (#2) is shown. Blue bars represent precipitation, and the black curve represents total subsurface runoff. The figures show the subsurface runoff response at depths of 10, 60, and 100 cm on the BF slope (ad) and CF slope (eh). Different vegetation types exhibit fundamental differences in runoff generation mechanisms. In the typical heavy rainfall event (Event #2; total precipitation: 58.7 mm; duration: 19.7 hours), the total runoff in the BF region was significantly higher (…). Figure 8 a). The runoff in this area is mainly dominated by continuous flow from deep layers (0.6 meters and 1.0 meters), with peak flows of 4.18 mm and 3.28 mm, respectively. Figure 8 (c, d). In contrast, runoff in the CF region is predominantly shallow, concentrated at a depth of 0.1 meters (peak: 3.56 mm). Figure 8 f), producing only weak, intermittent pulses from a depth of 1.0 meter (peak value: 2.02 mm, Figure 8 h). This runoff depth distribution perfectly matches the LPF activation depth determined by GPR ( Figure 3 and Figure 4 ).

[0059] This study advances the quantitative research of LPF by integrating delayed GPR and SSIM analysis. This workflow breaks through the common qualitative two-dimensional interpretation in GPR research, achieving objective identification and spatial mapping of LPF activation patterns. Figure 3 and Figure 4 More importantly, it achieves the transformation from line-scale observation to a fully quantized three-dimensional representation of LPF network evolution. Figure 6The resulting 3D reconstruction model captures event-driven activation and expansion dynamics in a directly comparable manner, both temporally and across forest types, thus establishing a tangible link between imaged subsurface connectivity and measured hillside runoff.

[0060] From a methodological perspective, the SSIM-based approach addresses a significant limitation of traditional time-delayed GPR analysis. While time-delayed GPR techniques are widely recognized for revealing preferential flow characteristics, many studies still rely on subjective interpretations of two-dimensional radar amplitude changes, which are susceptible to noise and time migration. The SSIM index provides a standardized metric that objectively distinguishes between water-induced structural changes and background subsurface heterogeneity. This capability helps bridge the observational scale gap between plot-based experiments and watershed hydrological processes.

[0061] However, the robustness and transferability of this workflow still have several limitations. First, the binary segmentation of LPF activation requires setting an operational SSIM threshold. Since the absolute value depends on the heterogeneity of a specific site, soil conditions, and acquisition parameters, there is currently no universal threshold. This necessitates site-specific calibration for stable control areas and comprehensive sensitivity analysis. Second, while this workflow effectively images three-dimensional lateral connectivity, its ability to monitor vertically preferential flow components remains weak, potentially limiting the complete resolution of flow initiation sources and solute pathways. Future implementations could employ multi-offset GPR acquisition techniques or auxiliary sensors to capture this dimension. Finally, although independent hydrological data validated the pattern inferred from geophysical methods in this embodiment, direct in-situ visualization techniques (such as dye tracing) were not employed. Integrating such direct evidence in future work would help further strengthen the mechanistic interpretation of the imaging LPF network.

[0062] The controlling effect of vegetation type on slope runoff generation: By combining the delayed GPR characteristics of LPF dynamics with precipitation, soil moisture, and runoff monitoring, this comprehensive evidence suggests significant differences in the precipitation-runoff mechanism between BF and CF. Based on this evidence, this paper proposes a conceptual model that summarizes the flow paths and connectivity controls that dominate the event responses of each system. Figure 9 (a) is the BF slope, and (b) is the CF slope.

[0063] Figure 10 It demonstrates the differences in connectivity between surface and subsurface structures on the slopes of hot and dry river valleys. Figure 10 (a) shows the relationship between event-scale penetration rainfall and precipitation for BF and CF. The dashed line represents least-squares linear regression, and the fitted equation and coefficient of determination (R²) are labeled in the legend. Figure 10 (b) is a soil structure profile of the BF soil. Figure 10 (c) shows the soil structure profile of the CF system. The BF system is characterized by a rapid, intense, and efficient drainage response. This "rapid-short-term-intense" pattern stems from the synergistic effect between canopy and subsurface properties. Aboveground, the BF canopy has a lower retention capacity (0.16, compared to 0.31 for CF), resulting in higher flux intensity and a generally wetter soil profile. Figure 10 a). Underground, a well-developed, coarse root system and its associated large pores form a ubiquitous, pre-existing network of vessels ( Figure 10 b). This structure promotes structure-dominated connectivity; water can rapidly activate these highly infiltrative pathways without requiring complete matrix saturation, thus enabling rapid LPF initiation and efficient transfer to deeper soil layers. Therefore, runoff is primarily dominated by continuous groundwater flow from deeper layers (e.g., 0.6 meters), maximizing slope drainage efficiency. Figure 8 ).

[0064] In contrast, the CF system exhibits a slow, decaying, and predominantly shallow response characteristic. This "slow-persistent-weak" response pattern is dominated by a process-based strong connection threshold. Figure 9 b). In this system, runoff formation appears to be impeded. First, the dense canopy and litter layer result in significant interception losses and low soil moisture levels ( Figure 10 a) Secondly, in the underground environment, a denser soil matrix and a finer root network enhance water retention capacity, creating a "sponge effect," thereby inhibiting rapid vertical infiltration. Figure 10 c). This phenomenon only occurs after the soil moisture deficit threshold is met, typically requiring continuous rainfall or infiltration pulses. This mechanism usually confines runoff generation to the shallow, organic-rich layer during most events, thus attenuating and delaying the hydrological response. Figure 8 Deep infiltration and LPF activation occur only when precipitation exceeds the matrix saturation threshold. This measure can protect the watershed from significant runoff peaks, but may also limit deep groundwater recharge.

[0065] Establish a comprehensive framework for the connectivity characteristics of slopes: This invention employs time-delayed geophysical resonant filtration (GPR) technology in controlled infiltration experiments to obtain high-resolution, event-scale snapshots of LPF dynamics. The main limitation is the difficulty in conducting long-term, in-situ field observations, making it impossible to continuously track LPF dynamics over extended timescales. Furthermore, the GPR method is difficult to implement under surface water catchment conditions, limiting its application in monitoring subsurface runoff processes during natural rainfall. Future feasible approaches would be to shift from single GPR detection to combined monitoring using multiple geophysical techniques.

[0066] Time-delayed gravitational resonant imaging (GPR) is well-suited for detecting rapid, shallow wetting phenomena and can clearly reveal the geometry of LPF pathways during stormwater runoff events, providing strong support for analyzing their formation and early evolution. However, event-scale GPR detection alone is insufficient for effectively identifying long-term persistence and deep water redistribution. Time-delayed resistivity tomography (ERT) can fill this gap. This technique enables long-term, fixed monitoring, tracking changes and persistence of water content near deep interfaces, although its spatial resolution is relatively low. Furthermore, time-delayed resistivity tomography (TDR) provides a precise point-scale water content benchmark for calibrating the water identification accuracy of GPR and ERT, aiding in LPF threshold definition and cross-validation. Combining hydrological monitoring and response data (e.g., using precipitation, throughfall, and runoff as co-validation data) can support four-dimensional characterization of subsurface LPF, specifically including: quantification of the initial threshold, evolutionary spatial trajectory, duration of persistence, and hydraulic intensity of the subsurface network.

[0067] By integrating multiple geophysical and hydrological measurement techniques, this embodiment achieves a more reliable LPF characterization than a single method. The complementary nature of different measurement methods across spatial and temporal scales helps to accurately depict groundwater flow movement and improve interpretation accuracy. Cooperative inversion and joint constraint strategies are increasingly used for the hydrological interpretation of stable geophysical time-shifted signals. Furthermore, this integrated multi-method framework supports extending LPF understanding from point scales to watershed scales by correlating groundwater connectivity signals with rainfall-runoff responses. This enables this embodiment to more accurately identify slope hydrological processes and provides more reliable evidence for assessing runoff risk and guiding water storage management in water conservation areas.

[0068] Summary: This study developed and applied a time-delayed GPR workflow based on SSIM, achieving for the first time non-destructive three-dimensional quantification of LPF dynamics on hot-dry valley slopes. Paired experiments conducted in this embodiment on adjacent secondary forests (BF) and plantations (CF) revealed fundamental differences in LPF mechanisms dominated by vegetation type. BF slopes exhibited structure-dominated rapid LPF activation, with network volume reaching 58.14% after the initial infiltration pulse, primarily occurring at depths of 0.6–1.0 meters. In contrast, CF slopes showed a process-dominated shallow response (0.2–0.4 meters), characterized by a defined wetting threshold, with peak activation (46.31%) occurring only after the second pulse. These geophysical models were robustly validated using independent hydrological data: BF generates continuous and deep subsurface runoff, while CF runoff is primarily shallow and attenuating. The results indicate that vegetation type modulates rainfall-runoff patterns by synergistically controlling effective water input (through canopy interception) and subsurface structural connectivity. The introduced workflow provides a crucial and transferable tool for quantifying structure-process interactions in slope hydrology. This method overcomes the limitations of two-dimensional analysis by providing quantitative three-dimensional connectivity indices, offering a direct pathway to improving the physical basis of hydrological models and providing a scientific basis for sustainable water resource management strategies in hot and dry valleys under climate change conditions.

[0069] The above examples illustrate the present invention only to aid in understanding it and are not intended to limit the scope of the invention. Those skilled in the art can make various simple deductions, modifications, or substitutions based on the principles of this invention.

Claims

1. A three-dimensional quantification method for subsurface lateral preferential flow in hot and dry valleys based on GPR detection and SSIM analysis, characterized in that, The method includes: Slope sample plots were selected in the target area to conduct artificial infiltration experiments and delayed GPR scanning. The delayed GPR scanning involved laying out multiple parallel survey lines perpendicular to the slope within the sample plot as GPR scanning lines, and collecting time-series GPR data from all survey lines at multiple time points before and after infiltration. The original time-series GPR images of each parallel survey line before and after infiltration were preprocessed. Based on the preprocessed time-series GPR images, by calculating the SSIM between the GPR images before and after infiltration at each time point, the difference images of SSIM changes before and after infiltration for each survey line and at each time point are obtained. Based on the set LPF activation threshold, the binary segmentation method was used to perform binary segmentation data analysis on the SSIM change difference images of each survey line before and after infiltration at each time node in order to extract the LPF activation domain. The SSIM variation difference images of each survey line before and after infiltration at each time point are standardized to a unified Cartesian coordinate system. Radar data interpolation is performed to fill the non-radar scan area between adjacent survey lines. Combined with the extraction results of the LPF activation domain and its volume and proportion calculation, the three-dimensional spatiotemporal dynamic evolution model of the LPF activation domain volume of the entire sample plot three-dimensional detection area is reconstructed.

2. The three-dimensional quantification method for subsurface lateral preferential flow in hot and dry valleys based on GPR detection and SSIM analysis as described in claim 1, characterized in that, Artificial infiltration experiments and delayed GPR scans were conducted, specifically including: An infiltration trench is dug at a predetermined distance from the boundary of the upslope sample plot. Deionized water is continuously supplied into the trench through a polyvinyl chloride pipe to promote infiltration into the soil matrix and prevent surface runoff. Multiple GPR lines are set up at equal intervals in the sample plot and are all laid out perpendicular to the slope aspect to dynamically monitor the path of groundwater flow on the slope.

3. A three-dimensional quantification method for subsurface lateral preferential flow in hot and dry valleys based on GPR detection and SSIM analysis as described in claim 1, characterized in that, The integrated artificial infiltration experiment and delayed GPR scan specifically include: Before the infiltration experiment, two GPR baseline images were acquired to characterize the background conditions before infiltration. Then, two infiltration experiments were conducted: In the first experiment, 25 liters of deionized water were infiltrated at a rate of 3.57 liters / minute. After the initial 25 liters of infiltration was completed, delayed GPR scans were performed along all measuring lines at 5, 10, and 20 minutes after infiltration. In the second experiment, 25 liters of deionized water were infiltrated at a rate of 1.56 liters / minute. Delayed GPR scans were performed along all measuring lines at 5, 10, 20, 30, and 50 minutes after infiltration.

4. A three-dimensional quantification method for subsurface lateral preferential flow in hot and dry valleys based on GPR detection and SSIM analysis as described in claim 1, characterized in that, The raw time-series GPR data collected at multiple time points before and after infiltration are preprocessed, and the specific steps include: Baseline correction: Eliminates DC offset and convolution filtering to suppress constant offset and low-frequency trends; Zero-time correction: Eliminates the influence of air waves and aligns the start time with the ground coupling; Signal gain: Amplitude compensation is performed using energy attenuation gain to correct radar signal attenuation caused by propagation depth and enhance deep reflection; Frequency filtering: reduces high-frequency random noise and low-frequency background noise; Background elimination: Suppresses horizontal coherence stripes and system noise; Kirchhoff offset correction: corrects geometric distortions caused by velocity nonuniformity and non-perpendicular incidence, thereby improving the geometric fidelity of reflector positioning accuracy and path recognition.

5. A three-dimensional quantification method for subsurface lateral preferential flow in hot and dry valleys based on GPR detection and SSIM analysis as described in claim 1, characterized in that, Based on the preprocessed time-series GPR data of each survey line, the SSIM difference image of each survey line before and after infiltration is obtained by calculating the SSIM difference between the GPR images before and after infiltration at each time point. Specifically, it includes: Based on the preprocessed time-series GPR data of each survey line before and after infiltration, two-dimensional GPR profile images of each survey line before and after infiltration are generated, and the SSIM between the two-dimensional GPR profile images before and after infiltration is calculated to obtain the SSIM change map of each survey line before and after infiltration. At the same time, the average SSIM value of each survey line is calculated by averaging the SSIM values ​​of each survey line to obtain the average SSIM value of each survey line.

6. A three-dimensional quantification method for subsurface lateral preferential flow in hot and dry valleys based on GPR detection and SSIM analysis as described in claim 5, characterized in that, Calculating the SSIM between the two-dimensional GPR profile images before and after infiltration, specifically including: in, These represent the sampling time window sequences before and after infiltration, respectively. They represent the first and second digits in the sequence, respectively. Amplitude values ​​at each sampling point; The amplitude is the local average. Let be the standard deviation of the radar amplitude, describing the local amplitude variability; The covariance of radar amplitude before and after infiltration; This represents the number of sampling points within the local window; This is a stability constant to avoid abnormally amplified calculation results or inability to calculate.

7. A three-dimensional quantification method for subsurface lateral preferential flow in hot and dry valleys based on GPR detection and SSIM analysis as described in claim 1, characterized in that, Based on the set LPF activation threshold, a binary segmentation method is used to perform binarized segmentation data analysis on the SSIM variation difference images of each survey line at each time node to extract the LPF activation domain, specifically including: If SSIM > 0.8, meaning the structural similarity index before and after infiltration is high and the soil moisture change is limited, it is defined as a non-humid area to quantify background noise; if SSIM < 0.5, meaning the structural similarity index before and after infiltration is low and the soil moisture increase is significant, it is determined to be an LPF activation domain signal.

8. A three-dimensional quantification method for subsurface lateral preferential flow in hot and dry valleys based on GPR detection and SSIM analysis as described in claim 1, characterized in that, For the SSIM variation difference images of each survey line before and after infiltration at each time point, they were standardized to a unified Cartesian coordinate system, and radar signal interpolation processing was performed on the non-radar scan area between adjacent survey lines, specifically including: In the Cartesian coordinate system, x: slope aspect; y: survey line spacing; z: subsurface depth; For the non-radar scanning area between adjacent survey lines, a cubic spline interpolation algorithm is used to interpolate radar signals along the y-axis.

9. A three-dimensional quantification method for subsurface lateral preferential flow in hot and dry valleys based on GPR detection and SSIM analysis as described in claim 1, characterized in that, The method further includes: Paired experiments were conducted using two types of sample plots: typical dry-hot valley secondary forest and plantation forest. The comparative study revealed the regulatory patterns of different vegetation types on the spatiotemporal dynamics, activation mechanisms, and hydrological effects of underground lateral preferential flow.

10. A three-dimensional quantification method for subsurface lateral preferential flow in hot and dry valleys based on GPR detection and SSIM analysis as described in claim 1, characterized in that, The method further includes: By combining high-resolution monitoring of slope water cycle, including synchronous rainfall, throughfall, soil moisture content, and stratified groundwater runoff, the reconstructed three-dimensional spatiotemporal dynamic evolution model of LPF activation domain volume is subjected to correlation verification and inter-group significance test analysis, thereby achieving cross-validation of hydrogeophysical modeling results.

Images