A New Method to Monitor Flood Dynamics Using GPS Dual-Frequency CNR and a Strength-Based Threshold Constraint Strategy

Abstract

The strength of a GPS carrier-to-noise ratio (CNR) signal is closely influenced by the multipath effect. This effect becomes more pronounced during flood events, as the reflection coefficient of water is significantly higher than that of dry soil. Consequently, the CNR measurements of GPS signals are impacted by floods. Based on this theory, the fluctuation of the CNR during a flood can be used to accurately monitor the process of a flood from occurrence to recession. Considering that the strength of the CNR largely depends on the satellite and frequency, and the characteristics of the influence of a flood on the CNR are different for each frequency, a new method based on the GPS dual-frequency direct signal CNR and the strength constraint threshold strategy was developed to increase the accuracy of the flood dynamics monitoring process. By using 64 MGEX (Multi-GNSS Experiment) stations distributed globally, an accurate direct-signal CNR threshold model of GPS dual-frequency was established. The threshold model demonstrated that the average difference of the direct-signal CNR, which is larger than 45 dB-Hz, between adjacent days at GPS L1 and L2 frequencies is 0.0659 dB-Hz and 0.0661 dB-Hz, respectively. Moreover, GPS real datasets in Zhengzhou city, China, from DOY (day of year) 199 to DOY 203, 2021, were collected to assess the proposed method. Based on the fluctuation of the direct-signal CNR threshold, the experimental results show that the flood appeared at about 16:04 PM on DOY 200, 2021, reached a peak at approximately 5:05 AM on DOY 202, and totally subsided at about 8:54 AM on DOY 202. Thus, the experiment results reveal that the proposed method accurately monitors the entire process of a flood from occurrence to recession, which provides valuable insights into operational flood dynamics, warning, and monitoring based on the GPS technique.

1. Introduction

In recent years, floods and urban waterlogging have become more frequent due to extreme weather caused by global warming, which seriously threatens the lives and property safety of people [1,2,3]. The existing flood detection and measurement method mainly concentrates on the meteorological model method [4,5] and the remote-sensing satellite method [6,7,8]. However, the accuracy of meteorological methods largely depends on the meteorological numerical model, and the remote-sensing satellite method is influenced by weather conditions and temporal resolution [9,10].

GNSSs (Global Navigation Satellite Systems) can provide seamless and real-time observation conditions with high spatial and temporal resolution. Using datasets collected from GPS (Global Position System) and BDS (BeiDou System) satellites, Cai et al. [11] investigated the influence of the surrounding environment on the GNSS signal. Their experimental results indicated that the multipath signal under a water environment is larger than that under open sky. Michael [12] analyzed the characteristics of the reflection coefficient and attenuation factor of the GPS L1 band, and the results showed that the reflection coefficient of dry soil is only about one-third of that of water. Considering that the reflection and diffraction of GPS signals depend strongly on the surrounding environment, Su et al. [13] proposed using the CNR (carrier-to-noise ratio) to detect and mitigate the multipath error. Furthermore, based on the relationship between the multipath error and the CNR, the influence of floods on the GNSS CNR and multipath error was investigated in depth [14,15]. The results demonstrated that a flood can obviously reduce the amplitude of the CNR and increase the pseudo-range multipath error. After that, Wu et al. [16] investigated the impact of floods on the L4 combination of GPS and GLONASS satellites. Moreover, Tong et al. [17] proposed using the CNR to monitor the whole process of a flood, but the influence of frequency bias was not taken into account.

The characteristics of signals differ between each satellite and each frequency because of the difference in satellite hardware. Larson and Nievinski [18] investigated the effectiveness of snow depth retrieval at various GPS satellite frequencies. Their results demonstrated that the L2C signal yields superior performance compared with the L1 frequency. However, the difference between satellites was not taken into account. Jin et al. [19] showed that, in the context of snow depth retrieval, the GPS L2P signal exhibits lower performance compared with the L2C signal. This result indicates that there are differences between different signal types even at the same frequency. Su et al. [15] showed that the difference in the GPS CNR between different frequencies on the same satellite can fluctuate approximately from 1 to 10 dB-Hz. However, whether this difference can be observed in the environment of a flood and whether it can be used to monitor the process of a flood have not been analyzed. In the context of flood monitoring, these frequency-dependent differences in the CNR become critical. The magnitude of CNR variations induced by flooding is often comparable to the intrinsic differences caused by signal frequency and satellite-specific characteristics. Without explicitly accounting for such frequency-dependent behavior, it is difficult to distinguish whether observed CNR changes originate from flood-induced surface alterations or from inherent signal properties. Therefore, dual-frequency CNR observations provide an essential means to separate flood-related multipath effects from frequency- and satellite-dependent biases, thereby improving the reliability of flood detection.

Moreover, to analyze and compare the characteristics of the CNR between two adjacent days, the ground track repeat period of the GPS satellite should be determined. Genrich and Bock [20] noted that the ground track repeat period of the GPS satellites closely approximates a sidereal day, lasting approximately 23 h 56 m 4 s. With the improvement in satellite orbit determination accuracy, Choi et al. [21] demonstrated that the ground track repeat period varies for each individual GPS satellite. Furthermore, Wang et al. [22] indicated that the ground track repeat period of the Galileo satellite is about ten days. Similarly, Su et al. [14] investigated the GLONASS satellite in theory and skyplot and pointed out that the repeat period of GLONASS satellites is eight days. Based on the above analysis, the ground track repeat period is different for each satellite and each system. Thus, when analyzing the factors related to the surrounding environment, the satellite’s ground track repeat period should be taken into account.

Taking into account the difference in the CNR between each satellite and each frequency, the difference in the CNR in the environment of a flood was analyzed in depth. The threshold model of the GPS CNR on the L1 and L2 frequencies between two adjacent days was established, respectively. By combining the threshold model of the CNR and the characteristics of the satellite ground track repeat period, a new method based on the GPS dual-frequency CNR and the strength constraint threshold strategy is proposed to monitor the entire process of a flood from occurrence to recession.

2. Related Theory and Model

The theoretical aspects of the relationship between a flood, multipath errors, and the CNR are discussed. To mitigate the impact of reflected signals, the direct-signal CNR is adopted to quantify the flooding process. This direct-signal CNR is extracted through a third-order polynomial fitting model. Additionally, the model to calculate the ground track repeat period of GPS satellites is given. To further clarify the observational geometry underlying the multipath effects discussed above, a schematic diagram is provided in Figure 1. The figure illustrates the relative geometry between the GNSS transmitter, the receiver, and the water surface, as well as the direct and reflected signal paths contributing to the observed CNR variations.

2.1. Influence of Flood on Multipath and CNR

The multipath effect refers to the phenomenon where a satellite signal reaches the receiver through multiple paths, caused by reflection and diffraction. The multipath effect usually reduces the strength of the received signal and is largely related to the surrounding environment because the reflection and diffraction coefficients vary with the environment. During a flood event, the observation station becomes surrounded by water, creating a reflective environment that differs significantly from that of dry soil. Michael [12] demonstrated that the reflection coefficient of water is roughly three times larger than that of dry soil. Thus, the multipath error in a flood environment is more serious than that without a flood environment.

Based on the theory of GPS signal processing, the relationship between multipath error and the CNR can be described as follows [23]:

$$CN{R}^2=A_d^2+A_m^2+2A_d{A}_m\cos {\gamma }_m$$

(1)

where CNR denotes the strength of the received signal, which is output by the receiver and can be obtained from the RNIEX file; A_d is the amplitude of the direct signal; A_m means the amplitude of the multipath error; and γ_m is the phase shift induced by the multipath.

In conclusion, flooding alters the surrounding environment of the GPS observation station, leading to increased multipath errors. These errors are subsequently manifested in variations in the CNR. Hence, the connection between flooding, multipath effects, and the CNR can be clearly established.

2.2. Third-Order Polynomial Fitting Model

As shown in Equation (1), the original CNR consists primarily of the direct signal component, along with contributions from non-direct signals such as reflections and diffractions. To reduce the impact of these non-direct components, a third-order polynomial fitting model is applied to isolate the direct-signal CNR, which can be expressed as follows:

$$y(x)=a_{1}x+a_2{x}^2+a_3{x}^3$$

(2)

where y(x) means the direct-signal CNR, which should be extracted; a_i denotes the coefficient of the polynomial model; and x is the original CNR and can be obtained from Equation (1).

2.3. Ground Track Repeat Period Model

Given that multipath errors are strongly influenced by the surrounding environment of the observation station, it is essential to consider the satellite’s ground track repeat period in the analysis. To better understand the ground track repeat period of a satellite, the operational period of a satellite should be introduced first. The operation period of a satellite refers to the time it takes for the satellite to complete one full orbit around Earth. If Earth stays stationary, the ground track of the satellite remains the same as before, and the operational period is the same as the ground track repeat period. However, since the Earth is continuously rotating and revolving rather than remaining stationary, the satellite’s orbital period typically differs from its ground track repeat period in most situations.

By using the third Kepler’s law and broadcast ephemeris, the ground track repeat period of the GPS satellite can be calculated by [24]

$$\left\{\begin{array}{l}n=\sqrt{GM}/a^{3/2}+\Delta n\\ T=4\pi /n\end{array}\right.$$

(3)

where n is the mean motion; Δn denotes the correction to the mean motion of n; a refers to the semi-major axis of the orbit ellipse; a and Δn can be obtained from the broadcast ephemeris; $\sqrt{GM}$ means Earth’s universal gravitational constant, which is 1.996498 × 10⁷ in SI units; and T denotes the ground track repeat period of the satellite.

2.3.1. Proposed Method

Prior to introducing the proposed method, differences in the CNR across various frequencies and satellites are examined. Additionally, the impact of a flood on the CNR is analyzed, and the feasibility of using the fluctuation of the CNR to monitor the process of the flood is discussed. A dual-frequency CNR threshold model is established by using 64 MGEX stations, which are distributed around the world.

2.3.2. Difference in CNR Between Different Frequencies and Different Satellites

The datasets used were collected on day of year (DOY) 201, 2021, at the Zhengzhou station. Observations included both L1 and L2 frequency signals. The data had a sampling interval of 30 s, with a satellite elevation mask angle set to 10 deg. Based on the satellite types, four satellites were used to demonstrate the variation in the CNR across different signal frequencies and individual satellites, which were G17 (Block IIF-M), G06 (Block IIF), G18 (Block III), and G02 (Block IIR). All of these four satellites can transmit L1 and L2 frequencies, and the results are presented in Figure 2. The red and blue dots denote the L1 and L2 frequencies, respectively. MD denotes the average difference between the CNRs of the L1 frequency and the L2 frequency.

As shown in Figure 2, the CNR is different for each satellite and each frequency. For example, the CNR at the L1 frequency is nearly the same as that at the L2 frequency for the G17 satellite. However, for the G02 satellite, the CNR at the L1 frequency is significantly higher than that at the L2 frequency, with an average difference of approximately 8.35 dB-Hz. In terms of the G06 and G18 satellites, the CNR at the L1 frequency is lower than that at the L2 frequency. The average differences are approximately 2.64 dB-Hz for the G06 satellite and 3.61 dB-Hz for the G18 satellite, respectively. This phenomenon can also be found in other GNSS systems, such as BDS [15]. This phenomenon is primarily attributed to differences in satellite types. However, whether this difference can be reflected in the impact of flood on CNR, and whether it can be used to improve the accuracy of flood monitoring, should be further analyzed.

2.3.3. Influence of Flood on CNR

To investigate the impact of flooding on the CNR, datasets collected at the Zhengzhou station from day of year (DOY) 199 to 201 in 2021 were used. The flood first appeared on the afternoon of DOY 200 and was totally flooded on DOY 201. Considering that DOY 199 was without a flood, it can be used to compare and analyze the influence of a flood on the CNR. In addition, in order to ensure the accuracy of the comparative analysis, the elevation and azimuth angles of these two CNR series should be kept identical to each other. Figure 3 demonstrates that the ground track of the G16 satellite on DOY 199 remains the same as that on DOY 201. Thus, if the surrounding environment of the observation station does not change during the flood, the CNR values on these two days would be expected to be nearly the same.

The comparison of CNR at the L1 frequency for the GPS G16 satellite is illustrated in Figure 4. Blue and red dots represent the CNR values on DOY 199 (no flood) and DOY 201 (with flood), 2021, respectively. It is evident that the CNR values on DOY 199 are generally higher than those on DOY 201, particularly when the CNR exceeds 45 dB-Hz. The average difference between these two days is approximately 0.75 dB-Hz across the entire CNR series. This discrepancy is mainly attributed to multipath effects, which are typically more pronounced at low elevation angles under normal conditions. As a result, the CNR at lower elevations is less affected by flooding compared with higher elevations. Furthermore, the maximum difference is approximately 3.75 dB-Hz. Thus, it can be concluded that the CNR is influenced by floods, and the CNR will experience an obvious decrease in the environment of a flood. More information can be found in Su et al.’s study [14,15].

To investigate whether the difference in the CNR between different frequencies and different satellites can be reflected in the influence of a flood on the CNR, the four satellites used in Figure 2 were still used in this experiment. CNR values larger than 45 dB-Hz are affected by floods more seriously than other CNR values; thus, only CNR values larger than 45 dB-Hz were adopted to analyze the difference in the influence of a flood on different frequencies. The experimental results are shown in Figure 5. The red and blue lines represent the differences in the L1 and L2 CNR values between DOY 199 and DOY 201, 2021, respectively, highlighting the impact of flooding on each frequency.

The results presented in Figure 5 should be interpreted in conjunction with Figure 2. For the G17 satellite, the impact of flooding on the L1 and L2 frequencies is highly consistent, as indicated by the similar trends of the red and blue lines. The magnitude of MD is merely 0.01 dB-Hz. In contrast, for the G06 and G18 satellites, the L2 frequency exhibits a greater change compared with L1, with MD values of 0.13 dB-Hz and 0.26 dB-Hz, respectively. This suggests that the flood has a more pronounced effect on the L2 frequency for these satellites. Conversely, for the G02 satellite, the L1 frequency is more affected than L2, with an MD of approximately 0.18 dB-Hz. Furthermore, the comparison between these four satellites also reveals variations in CNR responses, indicating satellite-specific differences. Therefore, the above analysis demonstrates that the impact of flooding on the CNR not only varies with signal frequency but also differs among individual satellites.

It should be pointed out that although the difference between L1 and L2 on the G02 satellite is larger than that on the G06 and G18 satellites for the original CNR, which can be observed in Figure 2, the MD of G02 is lower than that of G06 and G08 satellites. The main reason for this phenomenon is that the strength of the original CNR on the G02 satellite is lower than that of the G06 and G18 satellites, and the impact of the flood on the CNR increases with the strength of the CNR. Thus, to accurately monitor the entire process of a flood based on the GPS dual-frequency CNR, the difference between different frequencies and different satellites should be taken into account.

2.3.4. Establish the Threshold Model of CNR

To establish the threshold model of the CNR, real datasets collected on DOY 200 and DOY 201, 2021, from 64 MGEX stations were used. These stations were randomly selected to provide broad coverage across various latitudes and longitudes. The locations of the selected MGEX stations are illustrated in Figure 6. The datasets had a sampling interval of 30 s, with the satellite elevation mask angle set at 10 deg.

Based on the above collected datasets, the difference between all original CNRs between DOY 200 and DOY 2021, 2021, at the L1 frequency was calculated. The result is presented in Figure 7, where the horizontal axis denotes different GPS satellites, and the vertical axis represents the CNR difference in dB-Hz. It can be seen that the difference in the CNR between adjacent days varies for each satellite. The maximum of the difference is the G06 satellite, which is approximately 0.58 dB-Hz. Even for the minimum of the difference, it can also reach approximately 0.39 dB-Hz. Thus, the difference in the CNR between satellites should be considered. In addition, the average difference across all GPS satellites is approximately 0.48 dB-Hz. However, this threshold is largely affected by indirect signal components and random noise, which tend to amplify inter-satellite variability and obscure the true flood-induced signal variations. Consequently, such a threshold derived from original CNR observations cannot meet the accuracy requirement of reliable flood monitoring, further motivating the use of direct-signal selection and strength-constrained processing in the proposed method.

To mitigate the impact of the indirect signal CNR, the threshold model of the direct-signal CNR was established based on the same datasets. The direct-signal CNR was extracted by using a third-order polynomial fitting model. The threshold values for all GPS satellites at the L1 frequency are shown in Figure 8, where the horizontal axis denotes different GPS satellites, and the vertical axis represents the threshold value in dB-Hz. It is evident that the threshold of the direct-signal CNR is significantly lower than that of the original CNR. The average threshold for the direct signal is approximately 0.07 dB-Hz, demonstrating improved accuracy compared with the 0.48 dB-Hz observed for the original CNR. Therefore, the above results indicate that mitigating the influence of non-direct signal components substantially enhances the accuracy of the threshold estimation. However, the threshold of the direct-signal CNR is still affected by random noise. In addition, it should be noted that the original CNR threshold model and direct signal threshold model, which are presented in Figure 7 and Figure 8, were mainly used to demonstrate that the influence of the indirect-signal CNR and random noise should be considered; thus, only the L1 frequency is presented.

To further improve the accuracy of the direct signal threshold, random noise should be taken into account. Considering that random noise predominantly affects CNR values below 45 dB-Hz, a strength-constrained threshold strategy was employed to reduce the impact of such noise. Therefore, only direct-signal CNR values exceeding 45 dB-Hz were utilized to develop the threshold model. The established threshold model for the L1 and L2 frequencies is demonstrated in Figure 9, where the horizontal axis denotes different GPS satellites, and the vertical axis represents the threshold value in dB-Hz. It can be observed that the threshold at the L1 frequency demonstrates higher accuracy compared with that derived from the overall direct-signal CNR, indicating that excluding low-strength observations effectively suppresses noise-dominated variations. The average thresholds of the L1 and L2 frequencies are approximately 0.0658 dB-Hz and 0.0661 dB-Hz, respectively. Although the average thresholds at L1 and L2 are close to each other, differences between different satellites and between the two frequencies can still be observed, reflecting satellite-dependent signal geometry and frequency-dependent surface interaction effects. Thus, this difference should be considered in flood-monitoring applications. In addition, considering that both L1 and L2 frequencies were used to monitor the flood, the average threshold of all satellites at the L1 and L2 frequencies was adopted as the reference in the proposed method.

2.3.5. Summary of the Proposed Method

In this study, we propose a GPS dual-frequency CNR-based flood dynamics monitoring method that exploits flood-induced multipath effects to detect and track the entire flood process. Based on the above analysis, it is evident that flooding has a significant impact on the CNR, and this effect is strongly dependent on both the satellite and signal frequency. Thus, the difference in the CNR between the flood environment and without flood can be used to monitor the entire process of the flood. Both the L1 and L2 frequencies were used to improve the accuracy of the developed method. Moreover, to mitigate the influence of the indirect-signal CNR and random noise, only the direct-signal CNR, which was larger than 45 dB-Hz, was used.

The proposed flood dynamics monitoring method involves several data models that are tested and validated using real GPS observations. First, a polynomial fitting model is applied to the raw dual-frequency CNR observations to model the direct signal component. By removing the slowly varying trend of the direct signal, the multipath-related variations induced by floodwater are effectively isolated. Second, a statistical flood threshold model is established using long-term non-flood CNR observations from multiple MGEX stations. This model characterizes the normal fluctuation range of CNR differences and serves as a criterion for identifying flood-induced anomalies. Finally, a multi-satellite statistical averaging model is employed to integrate detection results from multiple satellites. This model reduces satellite-specific noise and enhances the robustness and reliability of the flood detection results. These data models are jointly tested within the proposed workflow to validate their effectiveness in monitoring flood occurrence and temporal dynamics.

The proposed method can be summarized in the following steps: (1) Establish the threshold model by using data from globally distributed MGEX stations. (2) Obtain the original CNR from the observation file in RINEX format. (3) Extract the direct-signal CNR by using a third-order polynomial fitting model. (4) Select the frequency based on a comparison of the CNR strength between the L1 and L2 signals. (5) Exclude the satellite by comparing the strength of the CNR; only a CNR larger than 45 dB-Hz is used. (6) Calculate the difference in the CNR between the flood environment and without flood for all effective satellites, and then compare the difference to the established threshold model. Only the satellite whose bias is larger than the threshold is selected. (7) Obtain the final difference by averaging all effective satellites for each epoch. (8) Monitor the entire process of the flood by using the results of the final difference and the established threshold model. Figure 10 illustrates the flood-monitoring workflow of the proposed method.

2.3.6. Experiments and Analysis

To evaluate the performance of the proposed method, datasets collected at Zhengzhou Station in China from DOY 198 to DOY 203, 2021, were used. The location of Zhengzhou Station is demonstrated in Figure 11. The datasets can be obtained from the following website: http://data.earthquake.cn. The flood appeared from DOY 200 to DOY 202, 2021. The datasets have a sampling interval of 30 s and include observations at both the L1 and L2 frequencies. The receiver and antenna are set on the top of an observation pier, which is about 3 m above the ground. Relevant flood information can be accessed on the following websites: https://gis.ncdc.noaa.gov/ (accessed on 25 November 2024) and www.cma.gov.cn.

The CNR bias under non-flood conditions was first analyzed using datasets collected on DOY 198, DOY 199, and DOY 203, 2021. The results are shown in Figure 12. The dark green line represents the CNR bias between DOY 198 and DOY 199, while the red line indicates the bias between DOY 199 and DOY 203. In this analysis, the bias is computed by subtracting the CNR values of the second day from those of the first, and this convention is consistently applied throughout all figures. The blue dashed line indicates the mean threshold for all satellites across both the L1 and L2 frequencies, as determined by the threshold model. The CNR bias between DOY 198 and DOY 199 remains below this threshold, suggesting that no flood occurred on DOY 199. Similarly, the comparison between DOY 199 and DOY 203 shows minimal difference, indicating that DOY 203 was also unaffected by flooding. Therefore, both DOY 199 and DOY 203 can be considered reliable reference days for monitoring the flood event.

The CNR bias between DOY 199 and DOY 200, 2021, is illustrated in Figure 13. The blue dashed line represents the threshold, and this convention is maintained in all subsequent figures. As shown, the CNR bias remains below the threshold prior to 16:04 PM, consistent with the observations in Figure 12, indicating no flooding during this period. However, after 16:04 PM on DOY 200, the CNR bias begins to increase and eventually surpasses the threshold. Detailed changes can be seen in the magnified inset in Figure 13. Based on the established threshold model and the observed impact of flooding on the CNR, it can be concluded that the flood event likely began at this time.

To further investigate the flood progression, the CNR bias between DOY 200 and DOY 201, 2021, is presented in Figure 14. It is evident that the CNR bias remains above the threshold throughout the entire day on DOY 201, indicating the presence of flooding for the full duration of that day. Moreover, by combining the analyses from Figure 13 and Figure 14, it is clear that although the flood event began at approximately 16:04 PM on DOY 200, the CNR bias between DOY 200 and DOY 201 after this time remains significantly above the threshold. This is highlighted in the region enclosed by the dashed rectangle. These results suggest that the flooding conditions on DOY 201 were more severe than those on DOY 200.

The CNR bias between DOY 201 and DOY 202, 2021, is demonstrated in Figure 15. By analyzing the entire trend of the results, it can be seen that there is a reversal at 8:39 AM on DOY 202. Before 8:39 am, the bias of the CNR between them is larger than zero and even larger than the threshold. The results indicate that the impact of flooding on the CNR for DOY 202 is greater than that observed on DOY 201, and also suggest that the flood conditions intensified on DOY 202 during this stage. Subsequently, the CNR bias drops below zero, implying that the floodwaters began to recede relative to DOY 201. This transition is highlighted within the green dashed box. However, whether this denotes that the flood waters had receded after 9:17 AM should be further analyzed by comparing the bias between DOY 202 and DOY 203, 2021.

From Figure 15, it can be observed that the flood is weaker than on DOY 201 after 8:39 on DOY 202. Moreover, to determine the exact time when the flood began to recede, the CNR bias between DOY 202 and DOY 203, 2021, is analyzed and shown as the red line in Figure 16. Additionally, to identify the peak of the flood event, the CNR bias between DOY 201 and DOY 203 is also included in Figure 16, represented by the green line. From the red line, it can be observed that the bias falls below the threshold at 8:54 AM on DOY 202, indicating that the flood had subsided by that time. After that, the bias performs under the threshold. Moreover, to determine the peak of the flood, the wave crests of both red and green lines are displayed in the dashed ellipse. By comparing the results, it can be concluded that the peak of the flood appears at 5:05 AM on DOY 202, since the bias of CNR can reach approximately −0.77, which is larger than the maximum of DOY 201. Therefore, the proposed method demonstrates high accuracy in capturing both the peak and recession phases of flood events.

The above analysis confirms that the proposed method is capable of reliably capturing the entire process of flood from occurrence to recession. In addition, the fluctuation of the flood can also be accurately reflected by using the proposed method. However, it is important to note that the identified flood recession time, as defined by this method, indicates the point beyond which surface water ceases to affect the CNR monitoring, rather than implying a complete absence of surface water.

3. Discussion

The GPS carrier-to-noise ratio (CNR)-based flood-monitoring method proposed in this study demonstrates good robustness; however, several limitations should be carefully considered for practical applications. First, CNR measurements are inherently influenced by receiver type, antenna characteristics, and firmware-specific signal-processing strategies. Although ensemble averaging over 64 MGEX stations was employed to mitigate individual receiver biases, the derived threshold values should primarily be regarded as reference levels applicable to geodetic-grade receivers, and receiver-specific calibration may be required for low-cost devices. Second, environmental factors such as snow cover, vegetation growth, wet soil, and surrounding buildings can modify the local multipath environment and may induce CNR variations similar to those caused by flooding. While the dual-frequency strategy and the strength-constrained direct-signal selection partially alleviate these effects, non-flood-related disturbances cannot be completely eliminated in the absence of auxiliary information, such as land cover data or meteorological observations. Third, in non-urban or mountainous regions, the applicability of the proposed approach may be reduced due to sparse GNSS station coverage and strong terrain-induced multipath effects, which complicate the characterization of baseline CNR behavior and may necessitate site-specific thresholds or more conservative decision criteria. In addition, temporary obstructions, construction activities, severe weather conditions, or equipment malfunctions may also lead to false positives. Therefore, operational implementations should incorporate temporal consistency checks, spatial coherence analysis among neighboring stations, or integration with rainfall and hydrological data to enhance detection reliability.

Overall, these limitations indicate that further research is required on receiver-independent normalization techniques, adaptive thresholding strategies, environmental effect modeling, and multi-source data fusion to improve the generalization capability and robustness of the method; nevertheless, they do not diminish the value of GNSS CNR observations as a low-cost and weather-independent complement to existing flood-monitoring approaches.

4. Conclusions

The established relationship between multipath effects, flood dynamics, and CNR observations confirms that the GPS CNR can serve as an effective indicator for monitoring flood processes. This relationship is quantitatively derived and validated using real flood event observations, which provide the empirical basis for applying the proposed method to practical flood monitoring. Compared with traditional optical remote sensing, the proposed approach is not affected by cloud cover or illumination conditions, and compared with SAR-based methods, it does not rely on complex scattering models or dedicated satellite acquisitions. Moreover, unlike conventional hydrological gauge networks that are spatially sparse and vulnerable during extreme events, the method can exploit existing GNSS infrastructure to provide continuous and wide-area flood information.

By incorporating dual-frequency observations and a strength-constrained direct-signal selection strategy, the proposed framework improves the robustness of CNR-based flood detection against random noise and indirect signal contamination. The use of multi-station and dual-frequency GNSS data enables the construction of a stable threshold model and demonstrates the applicability of the proposed framework under real observational conditions. In practice, the threshold model can be integrated into near–real-time GNSS data-processing systems to enable automated flood anomaly detection and flood evolution tracking with minimal additional hardware cost.

Several aspects deserve further investigation. First, the threshold model should be refined using multi-event and multi-region datasets to enhance its generalization capability under different climatic and surface conditions. Second, data fusion with SAR or optical observations could be explored to improve spatial completeness and detection reliability. Third, electromagnetic forward modeling of signal–water–surface interactions may help to better quantify the physical mechanisms underlying frequency-dependent CNR variations. Finally, adaptive thresholding strategies based on machine learning could be developed to account for seasonal variability and long-term environmental changes.

Overall, the proposed GPS-based approach provides a cost-effective and weather-independent complement to existing flood-monitoring techniques and has the potential to become an important component of integrated multi-source flood early warning and disaster assessment systems.