Seismic signatures and site characterization of an intermittent stream in dry and flood conditions: an implication for soil losses and landslide triggering

Identifying ambient noise-based (ANb) signatures together with the erosion-prone site conditions retrieved from georadar attribute analysis of streams can help in the estimation of their erosive potential (EP) that promotes reverie landslides and soil losses in the fluvial valleys. This is particularly imperative on flooding or rainy days, leading to stronger erosion-prone conditions (colluvium and boulders) of the valley beds. Developing such research direction can benefit the local communities, as is the case with the Cerrado region of Brazil, where these phenomena have high destructive potential with social, economic, and climatic implications. For the present study, a seasonal stream in the Federal District of Brazil was investigated by ANb monitoring supported by ground penetration radar (GPR) for site characterization. The ANb monitoring was conducted (at a safe distance) with a seismometer over several durations of dry and rainy conditions. The power spectral density (PSDs) as a function of several weather conditions (rainfall, wind speed, and pressure), time–frequency spectrograms, and ambient noise displacement root mean square (dRMS) were computed. This analysis also considered the single station horizontal-to-vertical spectral ratio (HVSR), where rain, wind, pressure, river flow and anthropogenic signatures were evident (at selective frequency ranges). Multi-peaks that emerged on the HVSR curve were further analyzed to identify amplitude and frequency changes, with the three peaks shifting on average to a lower position during the rainy period. The GPR amplitude and waveform variation features were attributed to the stratigraphy of the floodplain and regions susceptible to erosion, such as erosion-prone lithological spots, which provide the basis for non-destructive monitoring tools that enable the detection of “seismic signatures” and weak spots of the fluvial channels for improving environmental management.


Introduction
In recent years, floods and debris flows worldwide have increased riverbank and soil erosion and landslides, impacting the downstream communities and causing severe losses in lives, properties, and land functionality (Somos-Valenzuela et al. 2015;Zhang et al. 2022) . Usually, local and small-scale erosions might not seem like serious instability problems. Still, in the long-term and extreme flood events, this situation often develops into landslides on riverbanks. It may also induce several issues, such as river blockage, a shift of river channel position, or flooding from rising riverbeds (Gu et al. 2020). Consequently, it may lead to the collapse of roads and bridges built along the river network (Chmiel et al. 2022) .
With the effect of climate change, intermittent and ephemeral systems are expected to become increasingly common globally. However, these systems are understudied compared to perennial stream flows. These are of particular concerns as they present a unique connection point at the terrestrial-aquatic interface, promoting soil erosion and landslide triggering. Moreover, complex spatial and temporal variations in hydrologic connectivity in non-perennial systems often require a unique interdisciplinary approach to advance the understanding of their function and response to global change. Therefore, monitoring riverbanks and erosive potential (EP) at an early stage is essential to inform the hazard evaluation and improve the risk management of erosion and landslides in such regions.
Fundamentally, EP is a function of various influencing factors, such as rainfall intensity (Rindraharisaona et al. 2022) and the nature of sediment transport and incision in the catchment area (Lawler 1993) . In addition, the local site condition can play a vital role in this process. This condition may include the types and properties of soil, such as density (compaction), porosity, and permeability, and any variation in these properties that may represent potential shear plans (Hamza et al. 2020) , soil thickness, the topography of bedrock, and the presence of rock fragments (Hussain et al. 2022) . The EP assessment has a wide range of applications in landscape evolution, soil erosion, landslide, ecology, water quality, land use management, and civil and river engineering, such as dams and recreational reservoir silting (Giménez et al. 2012;David et al. 2010;Oeurng et al. 2010;Graf et al. 2010;Araujo et al. 2012;Schmandt et al. 2013;Chao et al. 2015;Lai et al. 2018;Marchetti et al. 2019) .
The accurate monitoring and prediction of EP of a river are challenging tasks. For such tasks, there are currently several adopted techniques, including (i) the stream hydrophones or geophones (Turowski et al. 2011) , (ii) the identification of particles with a tracer or radiofrequency (Schneider et al. 2014) , and (iii) calculation via empirical relationships calibrated in the laboratory (Wilcock and Crowe 2003). However, these techniques can be logistically challenging, especially in significant flooding events, making their application cost-prohibitive (Roth et al. 2016) . As river activity and site conditions of the valleys are coupled with physical properties, including the ambient noise and electromagnetic wavefields, among others, hence can be valuable sources for monitoring (time-variant) and characterization (time-invariant) using geophysical techniques. Thus, with the growing demands for remote monitoring of these signals from outside the river channel, geophysical methods are increasingly adopted for monitoring work.
For this remote monitoring of river flow, two types of seismic-based geophysical techniques are broadly emerging, where the generated seismic-based (GSb) and ambient noise-based (ANb) activities are measured. The seismic signals emitted by river dynamics can be monitored by non-invasive, cost-effective, long-term, and continuous methods with broad applications (Schmandt et al. 2013;Gimbert et al. 2014) . Significant efforts have been made to quantify the spectral signature of bedload transportation water; however, it has been found that seismic power at low frequency is primarily generated by water discharges rather than bedload transportation (Burtin et al. 2008;Schmandt et al. 2013;Barrière et al. 2015) . A theoretical model of power spectral density (PSD) was calculated from the Rayleigh waves generated by saltating the bedload particles (Tsai et al. 2012) . Another seismic activity produced by the turbulent river flow-based theoretical model was proposed by Gimbert et al. (2014) ; it contributes to SPD variations in response to the seasonal hysteresis magnitude variations and has been reported in many other studies (e.g., Burtin et al. 2008Burtin et al. , 2011Schmandt et al. 2013;Chao et al. 2015;Barrière et al. 2015) . These findings support the use of seismic for the high-resolution monitoring of river bedload and other flow attributes (Roth et al. 2016) . From the previous studies, it can be concluded that for larger fluvial systems, high frequency (> 15 Hz) power excitation is created by high levels of bedload transportation (Schmandt et al. 2013;Roth et al. 2016;Anthony et al. 2018) . In contrast, turbulent flow and discharge modulate lower frequency seismic power (between ~ 1 and 10 Hz) (Burtin et al. 2011;Schmandt et al. 2013;Bartholomaus et al. 2015;Anthony et al. 2018) .
Contrary, in ANb, the river-sponsored variations in the ambient noise wavefield are induced because of the impacts of the bedload on the riverbed and banks, flow noise generated by the water turbulence, and in response to the acoustic waves generated by the interaction of water and atmosphere (Díaz et al. 2014;Chao et al. 2015;Bakker et al. 2020;Lagarde et al. 2021). Monitoring such river-generated noise is consistently reported in hydrologic studies (Díaz et al. 2014) . ANb monitoring techniques can provide high temporal and spatial resolutions of the landscapes, and their demands are increasing in bedload monitoring because of their low cost and non-invasiveness (Roth et al. 2016) . In the past, an increasing number of contributions have been developed using the variations in ambient noise as the basis of the study of the river flow system (Burtin et al. 2008;Goodling et al. 2018;Anthony et al. 2018;Smith and Tape 2019;P.C and Sawazaki 2021) . It was also found that the river discharge vs noise power followed a seasonal hysteresis trend consistent with the regional sediment transport rates in the river (Gabet et al. 2008) .
To better decipher the EP of a river, particularly to better monitor the damage potential during floods episodes, the characterization of riverbeds and valleys can be achieved by GPR application (Arcone et al. 1998;Szuch et al. 2006;Weihermüller et al. 2007;Chalikakis et al. 2011;Fabregat et al. 2019) , which also involved various types of landslides (Rehman et al. 2021;Velayudham et al. 2021;Cook et al. 2022). Overall, soil compaction affects soil dielectric constant and GPR EM signals, as reported in previous studies (Wang et al. 2016) . As in the case of the Cerrado region of Brazil, especially in active geomorphological valleys, such as Ribeirão Contagem valley, erosion by water is a crucial soil threat and a significant cause of riverside landslides (Gomes et al. 2019;Fonseca et al. 2022;Cunha et al. 2022) .
The present study introduces several novel aspects to the existing literature on soil erosion and landslides in active geomorphological valleys, such as Ribeirão Contagem valley in the Cerrado region of Brazil. First, it focuses on the analysis of an intermittent stream that generates small seismic energies related to the sediments and water flows during rainy days in the Federal District of Brazil, a region that has not been previously studied. Second, it employs a non-invasive method to detect weak spots susceptible to erosion and landslides based on geology (soil, boulder, bedrock) and degree of compaction (less compacted, more susceptible to erosion) using GPR attribute analysis. Third, the study explores how the variations of water discharge in a small river can be observed and monitored using ambient noise analysis, a novel monitoring technique that calculates the displacement root mean square (dRMS) amplitudes, PSD, spectrograms, and HVSR curves of ambient noise recorded during dry and rainy days. Finally, the study analyzes the changes in HVSR peak attributes (i.e., amplitude and frequency) in relation to different meteorological factors. The approach and findings of this study are expected to benefit the broader scientific community involved in the environment and geohazard management under future climate change extreme scenarios.

Description of the study site
The Ribeirão Contagem watershed is extended over 146 km 2 in the northern part of the Federal District of Brazil in the Sobradinho administrative unit (Fig. 1). The study area hosts many shallow landslides whose dynamisms are controlled by river erosion during rainy season (Hussain et al. 2019a) . There are alluvial and colluvial materials weathered from the bedrock of the Paranoá group, though bedrock exposures within the channel itself are rare. Most of the Contagem catchment is located on a large erosion complex, where highly active hill slopes supply the channel during summer rainfalls (Ferreira and Uagoda 2015) . The drainage and channel densities of the watershed are 5.7 and 32.9 channels/km 2 , respectively. The climate in the area is semi-humid tropical, with a rainy summer and dry winter. The mean annual precipitation in the area is 1442.5 mm.
The soil analysis included granulometry, and geotechnical tests were performed on the samples collected with an auger and from trenches (Braga et al. 2018) . As a result, six soil types are identified in the Contagem basin, including  (Hussain et al. 2019b) and d photograph of river floodplain in the dry season deep and reddish oxisols on the hilltops, shallow inceptisols on the hillslopes, and ultisols because of the presence of clayey carbonate-rich rocks on valleys and plinthic oxisols on the border of hilltops. The erodibility (K factor) is larger for plinthic oxisols than the oxisols and ultissols, having an average value of 0.005790 ton ha h MJ -1 ha -1 mm -1 and 0.004490 ton ha h MJ1 ha -1 mm -1 , respectively. The clayey soils in steep hillslopes (from 20° upper to 35° in hills and valleys) can control various linear erosional features, such as gullies and landslides, e.g., creeps, translational, and rotational movements. A significant proportion of these erosive features is found in inceptisoil in clay-rich rocks. About 63% of these features occur in close proximity to the river (~ 20 m), highlighting the active role of hydrological processes in erosion (Ferreira and Uagoda 2015) .
The Sobradinho Unit of Votorantim Cimentos Brazil is located in the Ribeirão Contagem Basin, where low-grade metamorphic sediments of the Paranoá and Canastra groups occur. The area is dominated by pelitic rocks, such as gray slates and clayey metasiltites, trapped with limestone layers. The thickness of the unit varies between 120 and 150 m, and the rocks that make up this unit are strongly influenced by the paleogeography of the bottom, marking the end of the deposition of the Paranoá Basin. Due to the composition of this unit, the primary minerals, when in contact with water or subjected to atmospheric conditions, are quickly weathered. The riverbank shows the connection between the colluvial-alluvial material and the alluvial material. The latter has a lateral continuity that varies between ~ 30 and ~ 100 m long from the drainage bed (da Silva Nunes et al. 2019) . In the drainage bed, it was possible to observe that the finergrained alluvial material (with well-selected grains, secondary minerals and low humidity) was superimposed on the colluvial material with poorly selected grains (varying from medium sand to gravel with decimetric boulders), composed of quartz with a high moisture content due to shallow water table. In general, the deposits in the Ribeirão Contagem channel are distributed as follows: soil masses moved by recent rotational landslides; river sediments deposited in recent river bars; sediments deposited in an alluvial fan; colluvial sediments; and alluvial sediments (Fig. 2).
Previously, Ferreira and Uagoda (2015) proposed a classification for estimating EP based on hydrological units, slopes, and forms. Three classes were found in the Contagem basin as low, medium, and high EP. Concave hillslopes higher than 10° were classified as high potential due to subsurface flux and clayey material concentration. While on convex slopes, the critical to a movement was taken as 35°. The map showed the major proportion of known mass movements wherein high potential. Therefore,

Methods
The aim of this study was to assess soil losses and landslide triggering using seismic ambient noise and ground penetrating radar. The collected data were processed using specific algorithms and software to extract the necessary information. This section provides a detailed description of the data acquisition and processing methods employed in this research.

Seismic ambient noise
For hydrodynamic analysis of the river, one Sorcel L-4A-3D short-period seismometer having a natural frequency response of 2 Hz was installed at the bank of the river at the 'Rua do Matto' locality ( Fig. 1). The continuous data for the seasonal impact evaluation was divided into two acquisition campaigns (1) dry from Julian day 101 to 105 of 2017 and (2) rainy period from Julian day 344 to 350 of 2017. The records were performed in a continuous mode and at a sampling rate of 250 samples per second with a DAS-130 RefTEK data logger. Raw ambient noise records for rainy days are presented in Fig. 4.
The ambient noise wavefields of dry and rainy days were compared by applying spectral analysis of the ambient noise recorded two times, including power spectral density estimation, time-frequency analysis by spectrogram, and possible quantification effects on the site response by HVSR. After spectral analysis, the dRMS of ambient noise of two-time series was also calculated using a widely adopted approach This way, the signal's decomposition into a discrete spectrum is achieved. The waveform of dry and flooding conditions is selected, consisting of ground motion records of N-S, E-W and Z components. The steps include (i) subdivision of data into smaller windows of time length 50-60 s each with 10% overlap, (ii) each window is 5% cosine tapered and transformed into Fourier domain, and (iii) each spectrum smoothed prior to the calculation of spectrum for a bandwidth coefficient 40, as the raw signal contained unusual spikes (Singh et al. 2019;Pandey et al. 2020) . (b) The amplitude spectrum is obtained by applying a further transform to the FFT. Power is obtained as a square of the amplitude spectrum. PSDs are gathered by binning periods and powers. These power-period bins are normalized to obtain the probabilistic power spectral density (PPSD). The PSD and its aggregation provide ambi-ent noise energy (power) variations, so the signal strength and distribution are a function of frequency (Díaz et al. 2014;Pandey et al. 2020) . For the validation of spectral analysis, PSD is plotte d as functions of rainfall, wind speed, and pressure from the nearby meteorological station. These data are useful in correlating the variation of amplitudes in seismic noise with the sudden changes brought by rainfall, wind speed and pressure. These meteorological data are divided into ranges, and the changes in PSD are observed accordingly. (c) As the seismic energy of a signal is proportional to the square of its amplitude; therefore, the root means square (RMS) analysis of a continuous record provides another way of highlighting the variations of seismic signals with time (Falanga et al. 2021) . The dRMS of ambient noise records is calculated using a widely adopted methodology in "COVID Seismology" (Lecocq et al. 2020). It first calculates the PSD based on which the dRMS is derived (Condori et al. 2022). (d) Time-frequency spectrograms are calculated using Obspy (a python library) inbuilt functions, which are expressed in units of energy as dB/(m 2 /s 4 ) over different frequencies.

Fig. 4
Seismograms recorded during rainy days. Periods of rainfall and river flow can be clearly seen in the plot (e) Changes in local site response can be an important contributor to soil erosion and landslide, because detached soil blocks along the river bank have their natural period and are excited by the changes in Vs accordingly (Hussain et al. 2019a) . Therefore, following (Goodling et al. 2018;Anthony et al. 2018) , we applied the HVSR method (Nakamura 1989) to the three-component ambient noise records, which provide an ideal scenario for the estimation of site response due to changes in river flow. HVSR provides the response frequency of the loose sedimentary layer over the bedrock if there lies a considerable impedance contrast between them. Using this method, the natural period and the depth of the sedimentary layer are found using Eq. 1. The FFT of the vertical and horizontal ground motions is calculated after applying an energy normalization on each window following the diffuse approach for the HVSR methodology proposed by Sánchez-Sesma et al. (2011) . The spectra of both horizontal components are averaged following the vector summation described by Albarello and Lunedei (2013) and divided by the spectral of the vertical component. In the end, results are smoothened by applying a smooth mean halfwidth 40, and results are plotted. More details about this process are provided elsewhere (Hussain et al. 2020a) . Water infiltration and accumulation within unstable compartments may play a fundamental role in site stability. If the unstable compartment is susceptible to water retention, an increase in water content causes an increase in mass (M) and density (ρ). A decrease in both contact and bulk shear modulus (Gb) is simultaneously expected due to water seepage. A reduction in "fr" and a negative "dV/V" are then expected. Lowering the water table and drying the material generate the opposite effect (Colombero et al. 2021)  where Z is the depth, f r is the natural frequency, and V s is the shear wave velocity. (1)

Ground penetrating radar
The dielectric constant of the soil and the degree of its compaction (bulk density/penetration resistance) are somehow related, as documented in the literature (Wang et al. 2016) . This relationship can be utilized in fluvial seismology for the stratification of soil based on the degree of compaction, Fig. 6 Spectrograms a dry and b rainy days. Different rain related features are evident at different frequency ranges as loose or unconsolidated soil susceptible to erosion and possible related hazards, such as bank erosion. This can also change the river dynamics by increasing the amount of sediment loads and water viscosity, which impacts the seismic (both ANb and ESb). In this regard, two GPR profiles of 180 m and 360 m long were taken along with the riverbank during dry days using a georadar device GPR GSSI SIR 3000 (Geophysical Services Systems, Nashua, NH, USA), with 400 MHz antenna, control unity, and rugged survey car. The authors used the GPR attributes (i.e., coherence, average amplitude and average energy) for the detailed stratigraphy of the river floodplain as well as marks the degree of compaction (weak spots) of different superficial material types and their susceptibility is discussed in terms of their erodibility.
We used Reflex-win software, version 9.0.5, to do GPR data processing which included: (i) static correction for the time zero setting; (ii) The "energy decay" module was used to compensate for the signal decay; (iii) applying the "background removal" module to suppress or remove the coherent noise or stationary wave noise; (iv) 1D type bandpass frequency filtering for removal of high and low-frequency random noise, cutting intervals were set subjectively according to the frequency spectrum of some traces; (v) The "running average" module was applied to do data smoothing, the average traces was set as three. After the 5th step, the average amplitude, average energy, and coherence attributes were extracted using the C Language we coded. Finally, the attributes were displayed by the Reflex-win software. The average amplitude attribute is extracted from traditional In the next stage, the coherence attributes that measure waveform similarity of neighboring traces are calculated to interpret the features of the river floodplain. There are numerous applications of coherence in GPR signal processing (Gao et al. 2020;Trinks and Hinterleitner 2020) , since its first use for seismic signals by (Bahorich and Farmer 1995) and first applied to GPR data by (Young et al. 1997) . This method achieves coherence using a classical mutual correlation algorithm with 0 and 1 values, and results can be applied to determine valid and invalid signal regions along the GPR profile. Details are provided by (Gao et al. 2020) . In the end, the coherence image is color scaled as white high and black low values. The boundary of these two signal classes (valid and invalid) is interpreted as a lithological boundary or maximum penetrating depth limit. Valid signals are reflection or echo waves that can be attributed to stratigraphy, while invalid signals are the random noise generated by the radar system itself (Jol 2009) .

PSDs
As described earlier, the spectral analysis includes PSDs, and spectrograms of two ambient noise time series recorded during dry and flooding days. In addition, the PSDs were plotted as a function of wind speed and rainfall. The possible imprints of rainfall on the ambient noise may include the noise because of rainfall drop, river discharge, and properties of sediment loads and check on the urban activities in hours of rain. The details of the findings and discussion are documented below.
PSD, as a function of rainfall, wind speed and pressure, are plotted to delineate the effects of these meteorological factors on the power of noise at different frequencies.
The wind speed in the considered time is divided into two ranges (Fig. 5a). Discharge effects on PSD are clearly seen on horizontal and vertical components. The PSD at low frequency is found unaffected by the wind speed at all considered ranges. However, both time series are directly related at high frequency, i.e., PSD is higher at high wind speed and vice versa. Similarly, the rainfall amount during rainy days is divided into two ranges: 0.00-2.47 and 2.47-4.93 mm. Then, PSD is plotted as a function of these ranges. It is interesting to note that, similar to wind speed, the high rainfall affects the ambient noise PSD at higher frequencies (Fig. 6b). The spectral peaks remained the same on all PSD plots.
PSD plots were used to identify variations in power and peak shifting associated with river flow and sediment loads, as discussed by Wenner et al. (2019). These variations are assumed to be excited by discharge and other related phenomena on rainy days. Changes in low-frequency power may be associated with flood-induced changes in riverbank roughness, as explained by Roth et al. (2016) . In rainy conditions, peaks at higher frequencies (Fig. 5c) are associated with variations in ambient noise wavefields resulting from river-related phenomena, as well as the effects of resonant structures such as landslides, local stratigraphy, and other riverine resonant structures related to sediment deposition in the floodplain. Piantini et al. (2021) proposed that these peaks may be associated with the same propagating source during river floods. Another possible explanation for these peaks is the sudden destabilization of debris deposits on slopes and cliffs, which usually result from mass wasting, as suggested by Chmiel et al. (2022) . Overall, the PSD plots provided valuable insights into the complex interplay of various phenomena that influence river flow and sediment transport. Fig. 8 a Averaged HVSR curves of dry (dotted curve) and rainy days (solid curve). b) Zoom to the higher HVSR peaks found between 20 and 100 Hz Figure 5a shows that with the increase in rainfall amount, there is an increase in mean PSD at a broad frequency range. Wind speed has different effects on the noise energy in rainy conditions; as the wind speed rises, there is an increase in PSD at a high frequencies (2-20 Hz), while the band lower frequency remains unaffected (Fig. 5b). At a frequency range from 20 to 100 Hz, this trend does not hold. A similar trend of rising wind speed-increasing PSD during rainy days can be seen in Fig. 5b. The fluvial effects on different frequency bands are documented in previous studies as 15-45 Hz (Schmandt et al. 2013) , 5-15 Hz (Chao et al. 2015) , 10-30 Hz (Schimmel et al. 2018) and ~ 0.1-45 Hz by (Anthony et al. 2018) . The debris flow signature at the 5-10 Hz frequency band was observed by (Lai et al. 2018) . The effects of pressure on ambient seismic noise energy can be seen in Fig. 5c. At lower frequency the energy and pressure are directly related which found inverted at higher frequency ranges.

Spectograms
The short-period spectrograms of dry and rainy days are shown in Fig. 6a, b, which present four different frequency bands (continuous or discontinuous). The spectrograms of two times show low-frequency ambient noise, query blasts, and high-frequency noise. The significant energies induced by the river flow below 10 Hz and between 10 and 60 Hz can also be seen in Fig. 6b. These are combined effects of rainfall, river flow, and sediment loads. The sedimentary signals can be seen in Fig. 6b as high-frequency events. The frequency band below 2 Hz is the instrumental noise; therefore, it was not possible to see the effects of high wind speed at low frequency. The other high-energy noise band is 2-12 Hz which can be seen on both dry and rainy plots to mark the effects of cultural noise. It also shows a diurnal pattern, another attribute that confirms the presence of cultural noise. The higher frequency band, 20-50 Hz, is only prominent on rainy days, showing river-related processes.
Interestingly, the quarry blast from the nearby mining can be observed on both spectrograms. The spectrograms of dry days do not show any energy at the frequency band where river influence emerged during rainy days. The quiet periods associated with lunch break hours (12-14) at frequencies below 10 Hz can be seen as small windows during rainy days. In literature, the river flow has been observed over different frequency ranges, which depend on the conditions of the river (discharge amount, roughness of riverbed) and its site (soil conditions). Polvi et al. (2020) reported stream flow and sediment transportation the frequency Fig. 9 Averaged HVSR variations in terms of a) peak frequency and b) peak amplitude for the dry (April 2017) and rainy periods (December 2017) measured for the fundamental and secondary peaks observed.

Error bars show the standard deviation
Page 13 of 20 295 ranges as ~ 1-20 Hz and ~ 15-100 Hz, respectively. Very high energy peaks of the rainy spectrogram may be associated with the signatures of rain plus wind, as reported by (Rindraharisaona et al. 2022) .

RMS amplitude and energy release
The ambient noise dRMS shows variations with hours of the day at different frequencies (2)(3)(4)(5)(30)(31)(32)(33)(34)(35)(36)(37)(38)(39)(40) (Fig. 7). It shows stability in ambient noise levels at low frequencies for both the considered time scales. At frequency bands of 30-40 Hz and 50-70 Hz, a decrease in RMS values is observed on rainy days, possibly associated with a break in anthropogenic activities because of rainfall. An increase in dRMS is found at frequency ranges of 2-5 Hz and 70-125 Hz. These are the possible frequency ranges where the combined effects (solid-fluid mixture) can be seen. However, the diurnal patterns can be seen in both series at higher frequencies. This may be associated with human activities and the influence of floods in the river at different frequencies; however, as the river is small enough, the influences are less prominent on RMS plots. Under these limited data availability conditions, it is difficult to separate the urban noise and noise generated by the river.

HVSR curves
The HVSR curves obtained for the studied periods (rainy and dry) are characterized both by a condition of multiple-peak (Fig. 8a). The fundamental HVSR peak identified by (Hussain et al. 2019a) as the contact between the low-velocity soft deposits and the Paranoá bedrock is also observed during both periods at 1.3 Hz. Moreover, the secondary peaks attributed to the landslide surface observed by (Hussain et al. 2019a) are also identified during the dry season at 4 and 9 Hz showing both significant HVSR amplitudes (SESAME 2004) . This behavior agrees with the findings by (Hussain et al. 2019a) for these secondary peaks, which tend to decrease in amplitude or totally disappear during rainy days. On the other hand, the double peak found above 20 Hz is present in both periods (Fig. 8b). It seems to slightly shift the frequency positions on average of each peak and vary the HVSR amplitude when comparing both periods.
To compare the differences between the two periods studied, Fig. 9 gathers the HVSR variations, in terms of frequency and amplitude, for the two persistent peaks observed: the fundamental one at 1.3 Hz and the higher double peak between 20 and 100 Hz. During the rainy period, a wider variation is experienced in the two frequency bands, revealing a period of higher instability in the HVSR curves. Regarding frequency position, the three peaks shift on average to a lower position during the rainy period (Fig. 9a). Such observation agrees with the findings by Stevens and James (2022) , which revealed that the drops in f 0 coincide with sharp increases in water content. This is coherent with the overall soil moisture increase expected during heavy rainfalls in the study area. The daily variations show how a semi-diurnal modulation on the HVSR amplitude exists for the highest secondary peaks, at 40 and 60 Hz, that is kept in both periods (Fig. 10b, c).
On the contrary, the amplitude variations of the fundamental peak have a daily modulation which loses clarity during the rainy period (Fig. 10a). While this daily modulation can be related to the well-known pattern of the cultural noise, the semi-diurnal behavior observed in the amplitude modulations of the highest peaks could show a tidal modulation likely induced by the river flow. This would support the hypothesis of the river origin behind these two peaks.
Furthermore, when observing the frequency changes on a daily basis during the two periods (Fig. 11), there is no evidence of daily modulations on any of the three HVSR peaks investigated. Comparison with atmospheric data does not show any evidence of a correlation between the nonperiodic HVSR frequency fluctuations observed (Fig. 11) and the windspeed and atmospheric pressure series for the two periods studied. We assume that they might have been excited by the river discharge and other related fluvial source mechanisms that can be delineated by detailed seismic studies together with UAV monitoring of the river. According to Walsh et al. (2020) , the frequency signature is a function of turbulence, velocity, viscosity, and density collectively called a solid-fluid mixture. This relationship has also been documented elsewhere (Coviello et al. 2019) . If the flow velocity increases, the signal will emerge as low frequency, while the viscosity damps the high-frequency signatures (Huang et al. 2004) . The other factors that affect the frequency spectra by impacting the ground vibrations are the properties of the riverbed (geometry, composition, and wetted perimeter) (Kean et al. 2015) . There is an increase in the frequency of the seismic signal because of the smoothness and softness of bedrock compared to gravelly or fine sediment composed and dense bedrock, as observed by Huang et al. (2007) . Natural period as a function of changes in Vs can be a possible indicator of loose sediments subjected to detachment in response to EP of the river. The gravel bars in the river floodplain have their natural period and are a source of possible resonance peaks at the HVSR curve. This way, the riverbank's compaction and material type can be linked with sediment quantification through natural frequency estimation. These similar features have been delineated based

Site characterization by GPR
The GPR coherence attribute of GPR provides additional evidence for characterizing the subsurface of the stream bed in the floodplain. Materials such as boulders, pebbles, and compacted soil exhibit higher coherence values and reflection interfaces. Coherence attributes can delineate the boundaries between valid and invalid signals associated with the bedrock interface. We used the "manual pick" mode of ReflexW software to plot the boundary based on the information reflected by the coherence attributes (see Fig. 12). GPR images are naturally divided into valid and invalid signal regions, and the green line is also shown in other GPR images (Figs. 12,13 and 14). In the valid region of the GPR data profile, the reflection wave signals are dominated. In contrast, the invalid region, does not show reflection and cannot be used to interpret geological units. There are two possible explanations for marking the depth boundary: first, the electrical conductivity of the subsurface determines the penetrating depth of the EM wave (Campbell 1990), the power of GPR pulse signal decays to zero or less than the minimum received level upon reaching the depth marked by the green line (see Fig. 12). Second, the presence of a compacted bedrock interface without a reflector (no reflection wave goes back) may cause GPR can only to record random noise signals.
Contrary to the traditional GPR image, where the boundary is considered the termination position or bottom of a dense, strong amplitude region (Al-fares et al. 2002) , the coherence attribute can reduce the ambiguity and subjectivity of the boundary interpretation. Figure 13 shows the black dashed line marking the different parts of the probable boundary interpreted using the traditional method, with the boundary (green line) interpreted using the coherence attribute. The traditional interpretation method may overlook the weak amplitude signals that contain effective reflection signals, such as the area between the black dashed line and the solid green line in Fig. 13. The coherence attribute removes the amplitude information and only reflects signal waveforms, which reducing the ambiguity in the boundary interpretation. To the best of authors' knowledge, current studies on geohazards/landslide surveys mainly used traditional GPR images for stratigraphy interpretation, without discussing the significance of the boundary between valid/ invalid signals in the GPR image (Rehman et al. 2021;Velayudham et al. 2021;Cook et al. 2022;).
Interestingly, we found that the GPR attributes can imply the possible depth and size of buried pebbles or boulders (e.g., energy attribute) and aid in stratigraphy analysis (e.g., average amplitude attribute).
The GPR data we acquired in the dry season had lower soil water content and conductivity compared to the rainy season, allowing for deeper penetration of GPR signals into the soil. The invalid signal region corresponds to highly compacted bedrock, leading to a reduction in the emergence of the reflection interface. In the valid region, a low coherence value indicates the change in subsurface media or structure, such as the presence of rock fragments, pebbles, soil voids, small faults or other related discontinuities (McClymont et al. 2008). The parts of the valid region with high coherence values (white color) are more compacted than the parts with low coherence values (black color). Therefore, we consider this position reflected by the black low coherence value in the valid region as the susceptible points to erosion. For instance, soil voids or discontinuity can provide permeable paths for rainwater to infiltrate the subsurface, potentially causing instability by building pore water pressure (Hussain et al. 2020b) .
The average energy attribute can be utilized to identify the presence and geometry of buried pebbles or boulders due to the ability of the rock-soil interface to reflect high-energy waves with significant differences in electrical properties between them. As a result, the locations of high-energy values may correspond to the positions with subsurface boulders. Information about their sizes can be inferred from the dimensions of the purple color regions on the energy attribute image (Fig. 14).
To observe the stratigraphic features, the traditional GPR images in Fig. 13 use purple and blue colors to display the strong positive and negative signals. However, Fig. 15 indicates that the subsurface layers generally have the distribution characteristics of approximately horizontal deposition, with some tilted deposit layers evident, especially in Fig. 15a. Layer depth interpretation using the average amplitude attribute is only applicable in the valid signal region (Fig. 15), which is more convenient and visually appealing.

Conclusions
This study presents seismic monitoring and analysis of the river sediment carrying capacity in the Ribeirão Contagem watershed (of the Federal district of Brazil) on rainy and dry days. The GPR-based local site conditions and their possible impacts on the erosion and sediment load of the river are discussed. The ambient noise wavefields during dry and rainy days are measured and compared. This comparison was conducted by applying spectral analysis of the ambient  noise (recorded two times), including power spectral density estimation, time-frequency analysis by spectrogram, and possible quantification effects on the site response by HVSR curves. In addition, the GPR attributes are studied for the delineation of erosion-prone sites as a major source of geohazard and sediment quality in the river of the fluvial valley. Based on the findings, we may conclude the followings: (i) Ambient noise-based analysis showed variations in the PSD on rainy and dry days. An apparent change in the spectrogram of rainy days was identified, attributed to rainfall and river dynamics. PSD vs meteorological agent (rainfall and wind speed) plots showed changes in PSD during rainy days. (ii) Based on the spectrograms of the rainy days (the signals and their characteristics), it can be suggested that one of the signals dominantly results from bedload transport and the other two from fluid transport processes. (iii) The HVSR plots showed an increase in amplitude starting from 10 Hz. During rainy days, a shift to low frequency and a semi-diurnal modulation on the higher peaks are observed. (iv) Ambient noise dRMS plots evidently showed variations in displacement during daytime.
Based on the variations and emergence of new typologies on different frequency bands, we assume that they might have been excited by the river discharge and other related fluvial source mechanisms that can be delineated by detailed seismic studies together with UAV monitoring of the river.
(xxii) GPR average amplitude attribute profiles showed the detailed riverbank and floodplain stratigraphy, including depth and topography of bedrock. The presence of boulders of various sizes is identified by the GPR energy attributes. In contrast, the coherence attribute helped in evaluating the degree of compaction and a possible indication of the region as susceptible to erosion (weak spots). The invalid region corresponds to a high degree of compaction area in the subsurface, because few reflection interfaces reflect waves, and GPR can only record random noise. GPR attributes can help mine more valuable information about landslide surveys than the interpretation obtained from traditional GPR images.
From our seismic noise analysis performed in Brasilia, it is evident that a seismometer can be an efficient tool to characterize the seismic signature of the river (at high frequency) to quantify better the river activity, particularly the bed load transport during floods. In addition, the presence of fluvial lithologies (boulder and colluvial sediments) and the slope steepness, characteristics of fluvial valleys, such as Ribeirão Contagem, make it vulnerable to natural hazards (landslides and erosion). This is closely related to rainfall and lithologies, motivating a better understanding of the phenomena to help decipher such destructive environmental processes. This study is a preliminary step in integrating stream bed conditions and flood levels using remote, cheaper and noninvasive geophysical applications. The present study will allow further field investigation to improve the estimation of sediment transport during flood events by deploying seismometers along rivers in fluvial valleys.
Author contribution Conceptualization, YH; methodology, YH; software, YH, HS, QG and SM; formal analysis, YH; investigation, YH, and RU; writing-original draft preparation, YH; writing-review and editing, OH, and WB. All authors have read and agreed to the published version of the manuscript.