1 Introduction

Precise global monitoring of water levels (also known as the stage) in river systems, particularly in regions characterized by complex terrains, is paramount. Water levels and discharges are critical indicators for tracking the availability of freshwater supplies [1]. The stage is a critical component of river hydrology and hydraulics. The water surface elevation (WSE) above a datum determines a river’s or stream’s stage. The water levels can be identified manually using staff gauges or automatically using optical sensors, pressure transducers, and radar sensors [2]. However, regulatory impediments prevent data sharing. Also, ground-based monitoring stations typically have low densities. The limited data availability hampers our ability to observe and predict hydrological phenomena, including flooding and environmentally damaging low flows, particularly in remote regions [3]. This situation is prevalent among many large rivers in the Himalayan region, including the Indus River, having complex terrain and varied river channels. This situation underscores the need for innovative and independent approaches to provide reliable and consistent water level estimation even in challenging environments. One such advanced approach is using satellite radar altimetry data, which has shown its efficacy in monitoring water levels from space [4,5,6].

Radar altimeters onboard satellites, such as those in the Sentinel-3 mission, can measure the distance from the satellite to the Earth’s surface with exceptional precision. The altimeter on a satellite transmits a sequence of pulses toward Earth. Using a time delay between the pulse’s emission and reception, it precisely determines the distance between the satellite and the Earth’s surface [7, 8]. In this way, water surface elevation can be determined above a reference datum. This technology offers a promising solution to the data gap and precision challenges of conventional river monitoring methods, particularly in developing regions like Pakistan, where the Indus River flows through a complex terrain.

Satellite radar altimetry has proven its worth in numerous studies by tracking water levels and river discharges over time [9,10,11]. These measurements have been cross-verified with in-situ observations, highlighting the technology’s significance in flood management through river monitoring. In addition to domestic applications, satellite radar altimetry holds promise in addressing international water disputes. Numerous regions worldwide where water resources are shared face conflicts [12,13,14]. Even within countries, these disputes arise [15, 16]. The advantages of satellite technology lie in its ability to provide impartial information from space, removing the complexities of on-ground politics and mistrust. It offers an objective perspective that can help resolve disagreements between countries and states over water resources by bridging the gap between different parties and promoting cooperation and sustainable water management in a world facing increasing water-related challenges.

With the current measuring network, which is restricted to river mouths, bridges, barrages, weirs, sluices, and dams only, a complete strategy for managing the water resources of the Indus River basin cannot be created [17, 18]. The construction and operation of water features, policy formulation, and a basin-wide integrated water resource management (IWRM) strategy require extensive river flow statistics. In doing so, we might anticipate reducing decision-making uncertainty, especially in a changing climate [19, 20]. The lack of readily available information on river water levels presents a key challenge in monitoring. Numerous applications, including flood modeling, water distribution planning, and the construction of hydraulic structures, are hampered by the lack of data [21, 22]. These tasks sometimes require access to historical river data, which are regrettably scarce.

The water data gaps due to poor gauging density can also be addressed using satellite radar altimetry [23, 24]. Supplementing river gauge records, satellite radar altimetry observations of water have proven to be a significant data source. Satellite radar altimetry has also been used for monitoring barrage and dam gate-opening events in the past [25, 26] Many researchers have used satellite radar altimetry data to address in-land water monitoring techniques [27,28,29]. When traditional river water level data are inadequate or unavailable, satellite radar altimetry can carry out these tasks. Despite data constraints, satellite radar altimetry offers a different way to obtain crucial information about river water levels, enabling researchers and planners to make well-informed choices.

Climate change has badly affected international waters, particularly Pakistan’s resources, necessitating continuous monitoring of water resources to manage them [30, 31]. Pakistan is among the top ten countries facing severe water crises [32, 33]. Climate change and population propulsion have stressed Pakistan’s water resources. Hence, water conservation through an alternate and reliable monitoring technique is necessary to combat this water crisis. Therefore, monitoring Chashma Barrage and other essential river locations is necessary for water management, estimating flood levels, and alarming for disaster preparedness in very low or high flows. Thus, satellite radar altimetry (SRA) is a cutting-edge, space-based, and dependable approach that provides a quick, reliable, and cost-effective means of measuring without the limitations of traditional monitoring techniques [34].

Although the initial goal of satellite radar altimetry was to monitor ocean surface during its earlier years, many researchers published their work on rivers [35], wetlands [36], and lakes [37] for monitoring water heights from space. Recent advancements in altimetry data processing using sophisticated retracking algorithms for accurate water level estimation, particularly for inland water have been developed and utilized. Techniques such as OCOG (offset center of gravity), ICE-1/ICE-2, and SAMOSA tracker have been increasingly adopted to improve signal clarity for inland water monitoring. Many studies around the world have used these retracking techniques for river and lake water estimation and found good results, demonstrating their effectiveness [38, 39]. By integrating the SAMOSA + + tracker data, the filtering technique was employed in this study. Overall, Limited work has been done on the Indus River in this regard. The Indus River is a vital transboundary watercourse crucial for the survival of the region’s population, ecosystem, and diverse fauna and flora, which have varying topography and irregular banklines [40, 41]. Notably, studies on the Indus River have addressed its unique challenges, given its varying topography and complex terrain resulting in noisy waveforms, particularly in mountainous areas and narrower river widths [42, 43]. So, the standards used in this case rely on the geophysical parameter for outlier identification.

Several factors can influence the return signals and the shape of the waveform, which in turn affect the precision of height measurements. Thus, it is crucial to flag poor-quality data to maintain accurate estimates. For our study, we specifically selected Misfit and Pulse Peakiness as filtering criteria because they have proven effective in previous research for identifying and filtering out low-quality data. These metrics provide valuable insight into signal accuracy, ensuring that only reliable measurements are used, which is crucial for improving the precision of height estimates in our analysis [44,45,46]. The quality parameter, the Misfit between the L1B waveform and the fitted model, filters out the ‘bad’ measurements. The filter discards measurements when the tracking Misfit (Mf) value exceeds four [45, 47]. We have also used Pulse Peakineess (PP) with a range greater than 0.3. If Pulse Peakineess is high, the returns are specular. High values are expected over still waters.

This research aims to evaluate and identify feasible data filtering criteria (Misfit & Pulse Peakineess) for enhancing water level estimation accuracies using sentinel 3A altimetry data. Noisy waveforms and river width can introduce data errors, requiring sophisticated filtering techniques to enhance accuracy. This study evaluates inland water surface levels derived from Sentinel-3 satellite radar altimetry data against in-situ observations at the middle reach of the Indus River in Pakistan. The selected study area is Chashma Barrage (elevation 193 m MSL). The Chashma Barrage holds great worth in Pakistan, providing essential services such as irrigation, flood control, and power generation [48, 49]. Additionally, it serves as a habitat for various riverine ecosystems [50,51,52].

2 Study area and data

Barrages serve as flow barriers, controlling river water by directing it towards the extensive canal system for agricultural purposes. Chashma Barrage is a prominent water management structure on the Indus River, downstream of Jinnah Barrage in Pakistan, comprising embankments dividing the reservoir into five small lakes, each spanning up to 250 hectares.

Two main canals, the Chashma Right Bank Canal (CRBC) and the Chashma Left Bank Canal (CLBC), arise from the Chashma Barrage in Pakistan. The CRBC primarily serves Bhakkar, Layyah, and Mianwali districts, while the CLBC distributes water to Khushab, Dera Ismail Khan, Tank, and Laki Marwat districts. These canals are instrumental in providing essential irrigation water, sustaining agricultural activities, and bolstering the economic activities in the region. Hence, the Chashma Barrage is crucial in regulating the water flow and facilitating regional irrigation. Altimetry satellite Sentinel-3A’s track#337 passed along the Indus crossing the Barrage, as shown in Fig. 1.

Fig. 1
figure 1

Study area (Chashma Barrage). The red dots indicate the satellite pulses as they pass over the surface of Chashma Barrage

Full size image

2.1 Data acquisition

The details of the dataset used in this study are presented in Table 1. Detailed descriptions are also provided in the subsequent sections.

Table 1 Dataset and its details
Full size table

2.1.1 Satellite altimetry data

Altimetry satellite sentinel 3-A track#337 was found crossing the Chashma Barrage almost parallel to the river flow. The altimetry pulses at this track were acquired from the Earth Console P-PRO service (https://ui-ppro.earthconsole.eu/). This service provides Sentinel-3A and Sentinel-3B altimetry data according to users’ demands based on several data collection options. Some options that were opted for during the data acquisition are presented in Table 2.

Table 2 Data specifications
Full size table

2.1.2 Observed data

The observed data in this study refers to the gauge readings (in-situ). The in-situ data (2019 to 2022) were obtained from the Pakistan Water and Power Development Authority (WAPDA). WAPDA collects gauge readings by installing gauges at various water bodies, including barrages and bridges across the region. The data collection of WAPDA involves regular monitoring and recording of gauge readings at the designated locations, ensuring a comprehensive understanding of water level dynamics over time. The daily measurements of these data are made in feet with respect to mean sea level. We converted the gauge readings into meters to make comparisons with satellite altimetry data easier. Notably, available observed data are usually limited to specific sites, like barrages, where gauges are installed. Thus, an altimetry track that was close to the gauge location was chosen for this study to validate the satellite data.

As the temporal resolution of Sentinel-3 is 27 days, satellite measurements were available roughly every 27 days. The temporal resolution of satellite altimetry data was substantially low in many aspects. However, the WAPDA provides daily water level data. To compare the water levels between in-situ measurements and altimetry, we selected only the in-situ water level data that matched the exact dates of satellite overpasses. This approach ensures proper synchronization of both datasets for accurate analysis. Aligning the dataset this way allowed us to achieve more accurate and reliable validation of altimetry-derived water levels.

3 Methodology

This section discusses detailed data processing steps to achieve our objective. This study employed an iterative model builder to calculate water levels from satellite radar altimetry data focusing solely on the water surface of the Chashma Barrage. The water levels were analyzed after applying multiple pulse filtering criteria (Mf & PP) to improve their accuracies. Figure 2 illustrates the structured approach used to derive reliable water levels.

Fig. 2
figure 2

Research Methodology. This workflow shows the model builder process of calculating water level from altimetry data for a specified target area and water surface. Water levels are analyzed using pulse filtering criteria (Mf & PP) and validated against in-situ measurement

Full size image

3.1 Altimetry data processing

We applied several processing and post-processing steps on the altimetry dataset to obtain precise water levels, including data acquisition, refining, processing, filtering, and evaluation. The Sentinel-3 altimetry dataset includes numerous parameters far exceeding those utilized in this study. We focused solely on the parameters presented in Eq. 1. It is important to note that not all radar altimetry pulses received from the satellite are usable or of sufficient quality. Various factors can affect the integrity of these pulses, introducing noise, particularly in mountainous regions where signal interference is more prevalent. Therefore, before estimating water levels, it is essential to filter out contaminated pulses and select only those that conform to an acceptable waveform.

Scientists have proposed various filtering mechanisms for this purpose. To identify and isolate noise-free pulses from the dataset, we assessed the applicability of different filters for our study area, including a range of Misfit (Mf) and Pulse Peakiness (PP) values, as mentioned by Dinardo (2020). We used these additional filters to improve our results and ensure the reliability of the water level measurements.

$$\:WL={(Altitude\_20Hz}^{*})-{(Range\_unc\_20Hz}^{*})-{(EGM\_2008\_20Hz}^{*})-(GEO\_Corr\_Land)$$
(1)

Where,

WL = water levels.

Altitude_20Hz = altitude measurement obtained from satellite radar altimetry at a 20 Hz data rate, indicating the height above the Earth’s surface.

Range_unc_20Hz = range uncertainty or errors associated with 20 Hz altimetry data. It reflects the potential error or inaccuracies in the range measurement.

EGM_2008_20Hz = Earth’s gravitational model corrections for 20 Hz data.

GEO_Corr_Land = corrections or adjustments made specifically for the geoid (Earth’s surface) over land areas.

* 20 Hz data were used instead of 80 Hz because higher data rates can introduce more noise in measurements, especially in challenging environments like coastal regions, lakes, and rivers.

3.2 Enhancing altimetry data quality through filtering techniques

3.2.1 Altimetry data filtering: targeting water surface signal through spatial criteria

A single satellite track path is not static but dynamic, shifting with each return pass. We chose passes that fell upstream of the central Chashma Barrage gate. Since the path runs parallel to the river’s flow, extending well beyond the upstream region. To ensure data accuracy, we carefully selected pulses within reasonable proximity to the Barrage, specifically within an upstate distance of 320 m. This distance was chosen based on our expertise and understanding of the area, considering two key factors. First, the river splits into two channels beyond this distance, increasing the likelihood of wave contamination and reducing reliability for water level estimation. Second, the main pond area of the barrage is concentrated within this range, where flow dynamics are more stable, making water level measurements more consistent (Fig. 3).

Additionally, data accuracy was ensured by applying a mask along the bank of the Barrage using the global permanent water surface [53]. Proximity to the shoreline can also lead to pulse contamination. To ensure that all selected pulses correspond strictly to the water surface, we eliminated pulses within a 40-meter zone from both the riverbank and the Barrage’s gate location [54]. This precaution helps prevent any potential land interference, enhancing the reliability of our measurements.

3.2.2 Optimizing altimetry data analysis for improved estimations

To enhance the accuracy of water level estimations, we analyzed the altimetry pulses both with and without applying quality waveform filter parameters. Two primary filters were used in different sequences. The first filter is a Misfit (Mf), referring to the quantitative measure of the disparity or deviation between observed satellite altimetry pulses and expected waveform characteristics, indicating the accuracy and agreement of the acquired data with the desired model. The second filter is Pulse Peakineess (PP), representing the peak or maximum amplitude of the satellite pulses obtained through radar altimetry. It indicates the intensity or magnitude of the detected signal.

In this study, we implemented an innovative methodology to enhance altimetry data analysis, significantly improving the monitoring of inland water levels. This approach not only refines existing techniques but also underscores the importance of accurate water level monitoring in managing vital water resources. We incorporated various filter criteria through trial and error based on Misfit (Mf) and Pulse Peakiness (PP) values across six distinct scenarios. These criteria included Mf values of less than 1.5, 2.5, and 3.5, as well as the selection of pulses with PP values greater than 0.2, 0.3, and 0.4, as detailed in Table 3. By systematically manipulating these factors, we aimed to identify the optimal configuration for reducing noise and improving the accuracy of inland water level estimation using altimetry data.

Through a trial-and-error process, we tested various threshold values to identify the point at which we could retain the maximum amount of high-quality data to minimize contamination. If PP is set too low (e.g., PP < 0.3), more noisy or distorted waveforms could be included, which might increase inaccurate or unreliable data. On the other hand, setting PP too high might result in rejecting good-quality data, reducing the overall dataset, and possibly missing important information.

Similarly, for Misfit (Mf), if the threshold were set too high, there is a chance that contaminated or poorly fitted waveforms would remain in the dataset, leading to potential errors in significant wave height (SWH) calculations. If the threshold is set too low, a substantial portion of valid data might be unnecessarily filtered out, diminishing the sample size. A misfit beyond 4 in SAR could imply distortions not linked to oceanographic phenomena but external factors like land interference. The thresholds (4 for SAR) are derived from empirical patterns observed in SAR waveforms, as demonstrated in studies by Cipollini and Calafat (2016) and Dinardo (2020). These studies indicate that misfit values beyond this threshold are strong indicators of waveform contamination, often due to land or specular surface interference. Therefore, Mf < 3.5 ensures that only data with acceptable levels of misfit are retained, thus maintaining a balance between data quality and completeness. The specific number (4) is based on Dinardo’s study (2020).

Table 3 Different scenarios for altimetry data analysis
Full size table
Fig. 3
figure 3

This map presents an elevation profile of a river section carefully chosen to facilitate the selection of satellite radar altimetry pulses. The profile line is strategically placed at the river merge point, allowing for the targeted collection of altimetry data within a 40-meter buffer zone from the riverbanks. This map aids in precise positioning and assessing satellite radar altimetry pulses at this significant and critical confluence location

Full size image

4 Validation

4.1 Correlation coefficient (R)

Pearson correlation coefficients were calculated to determine the relationship between water levels derived from altimetry and those measured on the ground (in-situ). Pearson correlation coefficient between in-situ and altimetry datasets was found using Eq. (2).

$$\:R=\frac{n\sum\:\left({x}_{i}{y}_{i}\right)-(\sum\:{x}_{i})(\sum\:{y}_{i})}{\sqrt{[n\sum\:{{x}_{i}}^{2}-{\left(\varSigma\:{x}_{i}\right)}^{2}}\left]\right[n\sum\:{{y}_{i}}^{2}-{\left(\varSigma\:{y}_{i}\right)}^{2}]}$$
(2)

Where; \(\:{x}_{i}\) = in-situ water levels (m), \(\:{y}_{i}\) = altimetry-derived water levels (m), and n = number of observations.

4.2 Root mean square error (RMSE)

Root mean square errors indicate typical variation between in-situ and altimetry-derived water levels. The collective difference among the datasets was calculated using Eq. (3).

$$\:RMSE=\frac{\sqrt{\sum\:_{i}^{n}{\left({x}_{i}-{y}_{i}\right)}^{2}}}{n}$$
(3)

Where xi = in-situ water levels (m), yi = altimetry-derived water levels (m), and n = Number of observations.

4.3 Nash-Sutcliffe efficiency (NSE)

Nush-Sutcliffe efficiency (Eq. 4) was calculated to validate the potential estimation of the water level of satellite radar altimetry with in-situ data. NSE measures one minus the estimation error variance, with an ideal value of one indicating no estimation error. This corresponds to a perfect match between the two datasets being compared. NSE is a commonly used statistical index for evaluating the predictive capability of hydrological models [55].

$$\:NSE=1-\frac{\sum\:_{i}^{n}{\left({x}_{i}-{y}_{i}\right)}^{2}}{\sum\:_{i}^{n}{\left({x}_{i}+{\stackrel{-}{y}}_{i}\right)}^{2}}$$
(4)

5 Results and discussions

5.1 Analysis of filter criteria scenarios

This section presents the results of our study, which focused on enhancing altimetry data processing for inland water monitoring by exploring six scenarios with different filter criteria based on Misfit (Mf) and Pulse Peakiness (PP) values. The results were carefully compared to evaluate each scenario’s efficacy in lowering noise and raising the accuracy of water level estimation (Table 4).

5.1.1 Scenario 1: no filters

In the first scenario, nearly all satellite pulses within the defined boundaries were included for analysis. An RMSE value of 3.42 m and an R-value of 0.08 were found for this baseline scenario, indicating high noise and lesser precision in the water level assessment. Figure 4 shows the relationship between altimetry-derived water levels and in-situ water measurements.

Fig. 4
figure 4

Comparison of water level measurements from in-situ and altimetry data (without any filter) over time. Shaded areas show ± 1 standard deviation for in-situ measurements, while error bars indicate uncertainties in altimetry readings

Full size image

5.1.2 Scenario 2–4: pulse peakineess criteria (PP)

RMSE and R values gradually improved in scenarios 2 and 3, followed by a slight reduction in scenario 4, respectively, representing PP values of > 0.2, > 0.3, and > 0.4. scenario 3 (PP > 0.3) demonstrated the best results, achieving an RMSE of 0.28 m, an R-value of 0.95, and an NSE of 0.85 (close to 1), indicating reduced noise and greater precision. The relationship between altimetry and in-situ water levels using PP > 0.2, 0.3, and 0.4 are shown in Fig. 5a–c, respectively. The chosen threshold (PP > 0.3) strikes a balance by rejecting ambiguous or noisy waveforms while preserving clear, reliable measurements.

Fig. 5
figure 5

Comparison of water level measurements from in-situ and altimetry data (PP > 0.2, 0.3 & 0.4) over time. Shaded areas show ± 1 standard deviation for in-situ measurements, while error bars indicate uncertainties in altimetry reading

Full size image

5.1.3 Scenario 5–7: Misfit Criteria (Mf)

Decreased RMSE and increased R values were also observed in scenarios 5, 6, and 7, which utilized Mf values of less than 1.5, 2.5, and 3.5, respectively. Scenario 5 (Mf < 1.5) produced the best results, achieving an RMSE of 0.27 m, an R-value of 0.93, and an NSE of 0.85. This indicates lower noise and higher precision, like the outcomes observed in scenario 3 based on the PP criteria. The relationship between altimetry water levels and in-situ water levels using Mf < 1.5, 2.5, and 3.5 are shown in Fig. 6a–c respectively.

Fig. 6
figure 6

Comparison of water level measurements from in-situ and altimetry data (Mf < 1.5, 2.5, & 3.5) over the study period. Shaded areas show ± 1 standard deviation for in-situ measurements, while error bars indicate uncertainties in altimetry readings

Full size image

5.1.4 Combining optimal filter criteria

We determined the best set of filtering criteria to refine altimetry data processing in inland water monitoring after examining the outcomes of each scenario. Of all the scenarios, the combination of PP > 0.3 and Mf < 1.5 proved to be the most effective, exhibiting the lowest RMSE, greatest R, and close to one NSE. The relationship between data quality, data quantity, and different scenarios (Mf, PP) is illustrated in Fig. 7.

Table 4 Statistics of all scenarios
Full size table
Fig. 7
figure 7

The heatmap illustrates the relationship between data quality and data quantity under different pulse peakiness (PP) and Misfit (Mf) filtering criteria. The left vertical axis shows the variations in data quantity and the right vertical color scale represents the variation in data quality, measured in terms of RMSE. Red areas show the highest data quantity but the lowest data quality. Whereas dark blue areas with (Mf < 1.5 & PP > 0.3) represent the lowest data quantity but the best data quality

Full size image

5.1.5 Hexbin analysis of water level scenarios

The hexagon plot in Fig. 8 illustrates the distribution and density of water level data points across all seven scenarios, each utilizing varying data filtering criteria. In these plots, in-situ water levels are plotted against altimetry-derived water levels with color intensity indicating the data point density in each area. Higher data density in certain regions reveals clusters while deviation from the diagonal suggests discrepancies between the altimetry and in-situ measurements. In Scenario 1 where no filters were applied, the plot displays a broad distribution showing a raw variance in the data.

Scenarios 2–4 apply progressively higher Pulse Peakiness (PP) thresholds (PP > 0.2, PP > 0.3 & PP > 0.4). With these thresholds, data quality improves as higher-quality data are retained, and outliers are reduced. Scanerio 3 (PP > 0.3) yields the best alignment with the 1:1 trend line, as the statistical analysis confirms a stronger correlation and lower RMSE value. This shows that the PP > 0.3 filter effectively balances noise reduction without excessive data loss. However, Scenario 4 (PP > 0.4) shows a minor deviation from this trend, suggesting that further filtering could lead to over-exclusion, sacrificing some useful data points. Scenarios 5–7 apply increasing Mf threshold (Mf < 1.5, Mf < 2.5 & Mf < 3.5). Scenario 5 (Mf < 1.5) demonstrates the closest alignment between in-situ and altimetry datasets, suggesting that a lower Misfit threshold provides more precise water level estimates. As the Misfit threshold increases in Scenarios 6–7, data variance and noise appear to increase reducing the clustering around the 1:1 and potentially affecting measurement accuracy.

Fig. 8
figure 8

Hexbin plots for water levels across different scenarios

Full size image

5.2 Discussion

The results of this study highlight the critical importance of applying appropriate filter criteria when analyzing Sentinel-3 altimetry data for inland water monitoring, particularly in regions prone to contaminated waveforms. The rationale for applying these thresholds is to filter out poor-quality measurements that are prone to errors, ensuring data reliability. However, since filters reduce data quantity, this study aimed to id entify the optimal threshold values for each filter. The main intention was to reject contaminated data where the waveform misfit is high, as it indicates significant deviations from expected values. The SAR waveforms can be distorted due to various factors, including land contamination or reflective surfaces. When such distortions occur, they cause misfits to exceed typical values. Hence, a threshold for rejecting such data becomes necessary.

Employing the right filters significantly improves data accuracy and reliability, enabling more precise water level estimations. The scenarios that showed the lowest RMSE values, highest R-values, and NSE scores closest to one—scenario 3 (PP > 0.3) and scenario 5 (Mf < 1.5)—are the most effective in reducing noise and enhancing precision.

The statistical outcomes also revealed that the traditional method, which includes all radar pulses without or with minimal noise filtering, is not capable of delivering favorable results in the study region with challenging terrains. According to Zhang [14] the traditional method, although commonly used and practiced in the past, has proven ineffective in reducing the impact of noisy pulses on the precision of water level calculations.

Various studies have proposed different techniques for estimating water levels using a range of trackers, such as the ocean retracker, offset center of gravity (OCOG), threshold, modified threshold, ice-1, and ice-2 [56,57,58]. While various methods have been suggested to estimate water levels with high accuracy, they have generally performed well on flat terrain but not in complex landscapes and mountainous regions, such as the Chashma Barrage area. In our study, we used Misfit (> 0.3) and pulse peakiness (< 1.5) factors, which yielded statistically significant results even in the complex terrain which makes this study novel because it shows reliable results even in complex terrain like Chashma Barrage. The correlation coefficient (R) and the root mean square error (RMSE) values slightly reduced their performance when analyzing the data with other varying values of filter parameters. The RMSE significantly declined to 0.27–0.28 m from 3.42 m when filter parameters (Mf > 0.3 & pp < 1.5) were applied, which may be considered a substantial improvement in data quality. The correlation between the two datasets also improved from 0.88 to 0.935–0.954 after analyzing altimetry data using filter parameters.

It is also important to note that data filters may improve the quality, but the quantity is expected to be reduced. Figure 8 indicates that the number of satellite data points visible at any given time drastically decreases when filter criteria are applied. Although the reduction in data quantity is not immediately apparent in the map due to data point overlap, it is evident that applying filter criteria greatly reduces the amount of visible data. Figure 8 clearly shows that altimetry satellites do not always follow identical tracks, with a slight shift occurring at each consecutive pass. This dynamic behavior causes changes in satellite pass tracks, sometimes remaining the same across different days (with similar spatial coverage but at varying times) or showing different spatial coverage at different times. The improved outcomes can be attributed to applying the Misfit (Mf) filter and Pulse Peakineess (PP), which effectively limit the amount of noise in the satellite pulses.

By retaining pulses with minimal noise, these filters help greatly reduce errors in water level estimates. However, employing these filters results in fewer observations when noisy pulses are eliminated, which lowers the total number of pulses detected on the water’s surface. We did not even get a single ‘good’ waveform for some dates, resulting in a drastic decline in observations. It depicts that many waveforms were contaminated due to land surface or other environmental impacts. The topography and surroundings close to the barrage may affect the quality of the radar altimetry return signal since the radar instrument frequently loses lock or cannot provide useful data across complicated terrain and steep slopes because of its larger footprint as also indicated in Rehman et al., (2022). Due to the uneven terrain in the lower-middle Indus River at the Chashma Barrage, the Sentinel 3 sample size, containing ‘good’ observations, was reduced. Applying Mf and PP filters allowed only ‘good’ waveforms without or with a low level of land contamination.

The data loss occurred due to the application of filter criteria, which significantly impacted the consistency of river monitoring processes, including flood forecasting and water supply management. Regular and reliable water level data are essential to ensure the main canal gates are operated correctly and other hydrological operations are effectively managed. Figure 9 illustrates the filter criteria on satellite pulses.

Fig. 9
figure 9

Map illustrates the effect of different filter criteria on the number of observations (satellite pulses) within the study area. Applying specific criteria for pulse peakiness (PP) and Misfit (Mf) results in reductions of observed data

Full size image

It is also observed that some errors are persistent throughout all scenarios, likely due to uncertainties in in-situ data. WAPDA records in-situ water level readings once a day, at 8 AM, while satellite overpasses occur at varying times throughout the day. Events such as rainfall, barrage gate openings, or flooding can occur between the in-situ readings and satellite overpasses, leading to changes in pond levels. These time discrepancies may introduce errors in the validation process.

Despite having some limitations, the findings of this study significantly contribute to the field of hydrology and water resource management by demonstrating the efficacy of satellite radar altimetry for accurate water level estimates, especially in a complex terrain. Accurate water level information is crucial for hydrologic studies where precise water level data are essential for modeling and analysis. Additionally, traditional ground-based gauge stations are sparsely distributed, and it is even difficult to install and maintain these stations in mountainous regions. The water level data obtained from satellite radar altimetry, improved through the filtering techniques discussed, can bridge these data gaps by delivering consistent and reliable water level measurements across extensive and often inaccessible regions. Reliable estimates of water levels are also very crucial for flood hazard modeling, early warning systems, and disaster preparedness.

The filtered altimetry data can also function as a virtual gauge station in data-scarce regions, eliminating the need for physical infrastructure. This is particularly useful for developing countries, where installing and maintaining ground-based stations is challenging. The methodology discussed can also support integrated water resources management (IWRM) by providing comprehensive and reliable water level data, which are very important for planning and managing water resources at a basin-wide scale. This can help in making informed decisions for water allocation, irrigation, and conservation. Lastly, it can also help resolve international transboundary water distribution issues by providing objective information on water levels. This way, it can promote cooperation and sustainable management of shared water resources.

Furthermore, this study emphasizes how crucial it is to consider certain topographical features when using filter criteria for processing satellite altimetry data. Any area of interest may have a complicated surrounding environment. Therefore, specific filtering techniques are needed to effectively handle noise while minimizing the loss of important data points. This methodology constitutes a noteworthy progression in the refinement of altimetry data analysis methodologies for inland water monitoring, providing invaluable perspectives for upcoming investigations and practical implementations in water resource management.

6 Limitations

The major limitation of utilizing filter parameters on satellite pulses is the loss of data points, which is very crucial for any scientific findings. Although applying filter parameters enhances water level estimation, it results in loss of data points as well. Loss of data points is very concerning because even a single data point was not found for some dates, ultimately impacting the time-frequency of water level measurement. The smaller number of data points eventually impacts the long-term hydrological modeling, climate change impact assessment, flood risk assessment, and transboundary water management strategies. Therefore, this limitation must be addressed by further developing advanced algorithms that produce reliable satellite data, thereby preventing gaps in the time series of water level data. In this study, we focus on one type of satellite data; however, the frequency of water level estimation can be improved by integrating data from multiple satellites or by coupling satellite data with available ground observations. Furthermore, although the Mf and PP have shown good results in sorting out the pulses based on minimum noise, including specular reflection and land contamination, further clarification is required to comprehend the characterization of these noise sources. Future research may benefit from a more thorough investigation of how various filtering parameters target noise sources in detail, ensuring broader applicability across varying environments. Finally, the values for Mf and PP were chosen from past studies and broadly by trial-and-error method. Future research could include a detailed sensitivity analysis by carefully accessing a range of Mf and PP filter thresholds. Researchers can find values that can enhance data accuracy across various water bodies and climatic situations by gradually modifying thresholds and comparing the outcomes with ground-truth data. Additionally, by examining the trends in big datasets, machine learning approaches might enhance threshold settings. Insights into possible modifications required for particular applications would also be obtained by testing thresholds over time and in a variety of settings, including rivers, lakes, and areas with seasonal variations. This approach would enhance the filter’s adaptability, improving satellite altimetry data reliability in hydrology.

7 Conclusions

This study aimed to assess a viable filtering method for estimating water levels in complex terrains with and without land-contaminated waveforms. This approach ensures that the benefits of improved accuracy are realized without compromising the availability and reliability of water level data for critical applications.

The Indus River’s topography contributes to waveform noise at high altitudes with narrow reaches. Evaluations compared altimetry-derived water levels under different Misfit and Pulse Peakiness filter settings with observed data at upstream Chashma Barrage. Among the scenarios tested, the combination of PP > 0.3 and Mf < 1.5 proved most effective, showing improved statistical outcomes. However, this improvement came with some data loss, highlighting the influence of various factors on data quality. Therefore, the findings of this study offer an improved approach for estimating water levels in complex terrains while minimizing the potential for error.

Implementing Mf and PP filters showed promising results, suggesting filtered altimetry datasets could serve as virtual gauge stations in data-scarce regions. However, the improved data quality, achieved at the expense of data quantity or by introducing gaps, may affect various applications where data continuity is essential. To mitigate these effects, it is essential to integrate multiple data sources, use predictive algorithms, and develop comprehensive monitoring systems that combine satellite and ground-based observations.

Future research may explore alternative filtering methods and advanced re-tracking techniques to optimize accuracy and data availability in barrage water level measurements. Integrating different altimetric missions could enhance temporal and spatial resolutions. Beyond the Chashma Barrage, additional accuracy evaluations of other inland water bodies are necessary because the finding of this at Chashma Barrage is promising and may possibly be site-specific. Future research and validation in different inland water bodies and diverse environments are required to confirm the broader applicability of the method.