Assessing temporal sampling uncertainty in hydrological modelling using multi-sensor observations from GPM and TROPICS

Satellite-based precipitation products are widely used for hydrological modelling; however, temporal sampling uncertainty caused by infrequent satellite overpasses can significantly affect runoff estimation, particularly in mountainous and flood-prone watersheds. This study evaluates the influence of temporal sampling uncertainty in precipitation on runoff simulation using multi-sensor combination data from two different satellite constellations (Global Precipitation Measuring Mission-GPM and Time Resolved Observations of Precipitation Structure and Storm Intensity with a Constellation of Smallsats - TROPICS) constellations over the Ranikhola watershed in the eastern Himalayas, India. Five GPM sensors (GMI, SSMIS, MHS, AMSR2, and ATMS) and three TROPICS CubeSats (T3, T5, and T6) were analysed individually and in different combinations. Sensor overpass times were collocated with IMERG precipitation data, and the derived rainfall was used to simulate runoff in the HYSIM (lumped model) hydrological model during 2016–2020. Results show that increasing the number of sensors improved sampling frequency, reduced revisit gaps, and enhanced runoff simulation accuracy. The combined TROPICS configuration (T3–T5–T6) achieved the best performance, with NSE = 0.6983, R 2 = 0.9568, RMSE = 7.75 m 3 s −1, and MAE = 5.62 m s −1, whereas the GMI–SSMIS combination showed poor performance with NSE = −1.7385 and RMSE = 23.36 m 3 s −1. The TROPICS constellation provided more than 12 overpasses per day, compared to ∼1–2.5 overpasses per day for individual GPM sensors. The findings demonstrate that high-frequency CubeSat constellations substantially reduce temporal sampling uncertainty and improve runoff estimation, highlighting their potential for flood forecasting and hydrological applications in data-scarce mountainous regions.

Robust satellite constellations have a unique ability to increase temporal sampling deemed crucial in estimating precipitation which is an essential climate variable. As future climate projections indicate increasing rainfall extremes and flood risk (Malik et al., 2020a), accurate temporal sampling of precipitation will become increasingly important for reliable runoff estimation and effective flood management planning. Global precipitation estimation largely depends on constellations of passive microwave sensors along with satellite-based radars and infrared sensors in geosynchronous earth orbits (Dogra et al., 2023). Like the Global Precipitation Measuring Mission (GPM) constellation that offers excellent precipitation information every 3 h or less at 90% of time at any location within 60-degree latitude North-South (Kidd et al., 2021). With satellite constellations like GPM, a pressing issue continues to be the sampling considerations especially for passive microwave sensors causing sampling uncertainty. In this study, the focus is on temporal sampling uncertainty, which arises from the space time intermittency between successive passive microwave (PMW) sensor overpasses, causing systematic under sampling of sub-hourly to sub-daily precipitation variability (lida et al., 2006). Previous studies have shown that sampling errors can be the dominant source of uncertainty in satellite precipitation products (Maggioni et al., 2022; Nijssen and Lettenmaier, 2004; Bell et al., 1990; Shin and North, 1988). The existing satellite constellations may have very well redefined the scale of precipitation measurement, but our ability to collect satellite data far outpaces our ability to understand how their sampling uncertainties manifest in hydrological models.

Existing studies on sampling uncertainty of satellite-based precipitation rely on precipitation climatology potential effects of climate change on flood occurrence and intensity either from individual platforms or from constellations (Bell et al., 1990; Steiner, 1996; Bell and Kundu, 2000; Indu and Kumar, 2015; Roy et al., 2020; Malik et al., 2020b). Over recent years, numerous studies (Gebremichael and Krajewski, 2004; Nijssen and Lettenmaier, 2004; Gebregiorgis et al., 2012; Ganesh et al., 2019; Pradhan et al., 2021) have investigated the impact of both retrieval and sampling errors on hydrological modelling. A key hydrological challenge is understanding how temporal gaps in satellite precipitation observations influence hydrological simulation (Aerts et al., 2023). With the emergence of high-frequency constellations like TROPICS, there is a need to assess whether improved revisit frequency can enhance streamflow estimation compared to established systems such as GPM. Owing to their high revisit frequency, small-satellite constellations have emerged as a promising approach for improving precipitation retrieval and supporting low-latency precipitation monitoring (Afzali Gorooh et al., 2023). Recent study like Cerbelaud et al., 2025; Junqueira et al., 2021 explain effect temporal revisit on understanding river dynamics. Among various hydrological variables, river discharge is especially critical, as it reflects the integrated response of a catchment to precipitation over time and space (Garzon et al., 2023). Reliable discharge estimates are essential for flood forecasting, drought monitoring, and effective water resource management (Sun et al., 2026). A consistent increase in discharge leads to higher vulnerability to flooding (Malik and Pal, 2021). When inaccurate or incomplete records of precipitation data are used in hydrological models, it may misrepresent the timing and magnitude of river discharge, which can compromise early warning systems and limit our ability to prepare for extreme hydrological events (McMillan et al., 2011).

Several studies have utilized satellite-based precipitation products for runoff and discharge estimation across different spatial and temporal scales. Kumar et al. (2022) examined the performance of CMORPH and IMERG for monthly runoff estimation in the Himalayas. A comparative evaluation by Pakoksung and Takagi (2016) reported that GPM and CMORPH performed best for runoff simulation, with GPM providing estimates closest to observations. At the global scale, Hong et al. (2007) demonstrated the potential of satellite precipitation products for runoff estimation from basin to continental levels, despite methodological uncertainties. In semi-arid catchments, Najmaddin et al. (2017) showed that satellite-derived precipitation inputs to hydrological models can substantially improve runoff assessment in data-scarce environments. The impact of precipitation uncertainty on runoff modelling has also been highlighted by Fekete et al. (2004). Moreover, satellite sampling errors can significantly affect discharge estimation, as irregular and infrequent sensor revisits often fail to capture the full spatial and temporal variability of precipitation events (Behrangi and Wen, 2017). Other studies, including Nourani et al. (2021); Adane et al. (2021), Mei et al. (2017), and Vergara et al. (2014), further investigated the influence of satellite-derived precipitation on discharge estimation across different hydrological settings. Previous studies have examined the effects of satellite precipitation uncertainty and GPM temporal sampling on hydrological simulations (e.g., Nijssen and Lettenmaier, 2004; Gebremichael and Krajewski, 2004; Pradhan et al., 2021). However, these studies mainly focused on sampling uncertainty of GPM core observatory instruments (2, namely, GMI and DPR) and their resulting precipitation products. To the best of author’s knowledge, the sampling uncertainty of new missions like TROPICS CubeSat constellation (4 cubesat constellation) has not yet been investigated. One of the direct manifestations of sampling uncertainty is accurate estimation of runoff for hydrological applications. This manuscript seeks to address the research hypothesis on whether a 4 cubesat constellation from TROPICS delivers acceptable runoff estimates when used in a hydrological model in comparison to the instruments on board GPM. To address the same, this study systematically evaluates the influence of TROPICS and GPM sampling configurations on precipitation representation and runoff simulation using IMERG-collocated precipitation and the HYSIM hydrological model.

This study focuses on quantifying sampling uncertainty of precipitation using a multi-sensor combination from two different constellations, namely, a) the GPM constellation and b). Time Resolved Observations of Precipitation Structure and Storm Intensity with a Constellation of Smallsats (TROPICs). Increasing satellite orbit density and combining platforms with different orbital inclinations can significantly reduce sampling inaccuracies and improve observation frequency (Nourani et al., 2025; Agyemang et al., 2026). In this study, temporal sampling uncertainty specifically refers to the timing and frequency of satellite overpasses (i.e., revisit gaps between successive observations at a given location), rather than differences in sensor footprint, spatial resolution, or retrieval physics. Traditional satellite missions often provide only one or two observations per day, limiting their ability to capture rapidly evolving convective systems. In contrast, small-satellite constellations offer higher temporal sampling frequencies, improving the monitoring and estimation of intense precipitation events (Seto and Kanae, 2025). The major aim is to identify the manifestation of precipitation temporal sampling uncertainty in the estimation of runoff on a daily scale. In particular, the sensors utilized are TROPICS-03, TROPICS-05 and TROPICS-06 from TROPICS constellation and five sensors, namely, GMI, SSMIS, MHS, AMSR2, and ATMS from the GPM constellation. Many previous GPM sampling studies primarily evaluate rainfall uncertainty using either individual sensors or the full constellation. This study furthers existing research by examining sampling uncertainty contributed by sensors on board GPM and TROPICS by considering them individually as well as in combination. For each sensors and their combinations, precipitation is sampled from IMERG at the corresponding sensor overpass times, ensuring a consistent reference framework for precipitation while varying only the temporal sampling frequency by the sensors. This framework allows a direct assessment of how increased sampling frequency and improved capture of short-lived precipitation events affect in runoff simulation by HYSIM hydrological model. The novelty of work lies in demonstrating the potential of the new TROPICS constellation in comparison to the established GPM sensors, with particular emphasis on their capability to capture frequent precipitation events, represent the majority of precipitation occurrences with a minimum number of satellites in constellation, and their implications of these uncertainties when applied in the HYSIM hydrological model for runoff estimation.

Study area

The present study focuses on the Ranikhola watershed (Figure 1), located in the Gangtok District of Sikkim, India, covering an area of approximately 254 km 2 between latitudes 27°14′15″N—28°23′49″N and longitudes 88°29′25″E−88°43′19″E. The watershed exhibits an elevation range from 292 to 4061 m above mean sea level, with steep slopes and uneven terrain characteristic of the mid-Himalayan region. The watershed receives a mean annual rainfall of about 3085 mm (1991–2020, IMD), of which nearly 72% occurs during the southwest monsoon season (June–September). The mean annual temperature is between 11 °C and 14 °C, with considerable seasonal variation. The watershed is drained by the Ranikhola River, which flows around 36 km in a southwest direction before joining the Teesta River at Singtam. The study region is vulnerable to flash floods and landslides (Roy et al., 2024a). On 4 October 2023, a glacial lake outburst flood (GLOF) from South Lhonak Lake severely impacted central to southwestern part of the watershed, including major cities like Singtam, Ranipool, and Gangtok (Gaikwad et al., 2025; Singha et al., 2025). The watershed sustains local livelihoods by supplying water for domestic use, irrigation, and hydropower, while also supporting biodiversity, slope stability, and remaining highly sensitive to floods and soil erosion due to its steep terrain and intense monsoon rainfall. The basin is characterized predominantly by loamy soils, with grain-size distribution and drainage properties indicating moderate to high infiltration capacity, consistent with previous hydrological assessments of the region (Roy et al., 2024b). Land-use analysis indicates that forest cover declined from approximately 84% in 2008 to 72% in 2020, while built-up area increased from about 7% to 14% over the same period, reflecting increasing anthropogenic influence within the watershed.

No glacier or glacial lake is present within the Ranikhola watershed; therefore, glacier melt or glacial lake outburst flood (GLOF) processes do not contribute directly to runoff generation in the study area.

Methodology and datasets

The study examines temporal sampling uncertainty by analysing the contributions of five key passive microwave sensors from the GPM constellation (data period 2016–2020) and three key sensors from TROPICS constellation (data period of 2024) in runoff simulation using HYSIM model (Figure 2). These sensors are the GPM Core Observatory Microwave Imager (GMI), the DMSP F18 Special Sensor Microwave Imager/Sounder (SSMI/S), the Metop-C Microwave Humidity Sounder (MHS), the NOAA-20 Advanced Technology Microwave Sounder (ATMS), and the GCOM-W1 Advanced Microwave Scanning Radiometer 2 (AMSR2). These sensors were selected as they are the key contributors to the GPM constellation, providing essential observations for global precipitation monitoring (Kidd et al., 2021). The SAPHIR sensor (Sounder for Probing Vertical Profiles of Humidity) was excluded due to its intermittent operation (Kidd et al., 2021). For the TROPICS mission, only the three CubeSats that were operational during the study period were included in the analysis. For each sensor, individual overpasses were identified and collocated in space and time with IMERG to extract the corresponding precipitation estimates. The sensors were then systematically combined in incremental configurations — (i) GMI and SSMI/S, (ii) GMI, SSMI/S, and MHS, (iii) GMI, SSMI/S, MHS, and AMSR2, and (iv) GMI, SSMI/S, MHS, AMSR2, and ATMS—to individually assess the impact of sensor in affecting the precipitation temporal sampling. A similar approach was applied to the new TROPICS CubeSat constellation. TROPICS consists of four operational CubeSats (initially six, with two lost during launch) operating at frequencies between 90 and 205 GHz to monitor atmospheric moisture, precipitation, and temperature profiles, particularly for tropical cyclones (Blackwell et al., 2025; Blackwell et al., 2018a; Kidd et al., 2022). All datasets were pre-processed prior to analysis. First, the swath geometry of each sensor was extracted and clipped to the area of interest for every overpass. The corresponding overpass times were then identified and matched with the nearest IMERG time step to ensure temporal collocation. Finally, the precipitation values extracted from IMERG for each overpass were combined and aggregated to obtain daily accumulated precipitation. In this study, TROPICS orbital overpass times and swath geometry from 1 January 2024 to 31 December 2024 were used. Since the TROPICS constellation was not operational during 2016–2020, no actual TROPICS precipitation observations exist for that period. Therefore, a sampling-based approach was adopted in which the full set of 2024 TROPICS swath geometries and overpass times was sequentially applied to the IMERG precipitation fields for each year from 2016 to 2020. Specifically, the 2024 TROPICS swath pattern was first overlaid on IMERG data for 2016 to extract rainfall values at the corresponding overpass locations and times; the same 2024 swath pattern was then applied independently to IMERG 2017, 2018, 2019, and 2020 in an identical manner. In this way, IMERG provides the precipitation estimates for each year, while TROPICS contributes only the temporal and spatial sampling pattern. This framework isolates the impact of TROPICS temporal sampling on rainfall and runoff estimation without introducing uncertainties associated with TROPICS retrieval algorithms. The approach assumes that the 2024 TROPICS orbital configuration is representative of the constellation’s nominal sampling behaviour. This assumption does not account for potential orbital drift, constellation phasing, or seasonal changes in local overpass times. We acknowledge that applying the 2024 TROPICS overpass patterns to the 2016–2020 IMERG baseline assumes a degree of temporal representativeness that cannot be directly verified without multi-year TROPICS orbital records. As the TROPICS CubeSats operate in non-sun-synchronous, low-inclination orbits (∼30°) without active station-keeping, they experience continuous drift in their Local Time of Day (LTOD) of equatorial crossing, cycling through the full 24-h diurnal cycle over a period of several months (Blackwell et al., 2018b). The primary objective of applying the 2024 orbital geometry is to simulate the statistical sampling capability of the fully deployed constellation specifically the geometric revisit intervals and spatial coverage gaps, low-inclination CubeSat configuration rather than to reconstruct precise historical overpass times. We acknowledge this as a structural limitation arising from the recent deployment of the TROPICS constellation and the consequent absence of multi-year orbital archives. Variations in constellation phasing, orbital drift, and local overpass timing may alter the probability of observing short-duration precipitation events, particularly in regions exhibiting strong diurnal rainfall cycles. Future work will incorporate a dedicated sensitivity analysis using fully simulated synthetic orbits across multiple years to formally quantify the localised diurnal bias introduced by this extrapolation. To avoid the retrieval uncertainty from various sensors used, IMERG was retained as reference datasets. IMERG demonstrate high accuracy and high temporal latency with global validation of precipitation product (Huffman et al., 2015; Tan and Santo, 2018; Tang et al., 2020; Jiang and Bauer-Gottwein, 2019) Details of all the datasets used for this study are tabulated in Table 1.

IMERG was used as a common reference framework to avoid ambiguity and systematic differences among precipitation retrievals from multiple sensors, allowing the analysis to focus solely on temporal sampling effects. IMERG pixels corresponding to the overpass time of each sensor were extracted through collocation. For spatiotemporal collocation, each individual overpass of the sensors was evaluated to determine whether the Ranikhola watershed was fully covered within the swath. Although several sensors considered in this study (e.g., SSMI/S and ATMS) possess native footprint dimensions substantially larger than the 254 km 2 Ranikhola watershed, their precipitation retrievals were not used directly. Instead, only the sensor overpass timing and swath geometry were employed to sample precipitation from the common IMERG reference dataset. Consequently, all runoff simulations were driven by precipitation fields derived from the same IMERG spatial resolution, ensuring that differences among sensor configurations primarily reflect temporal sampling characteristics rather than footprint-size or retrieval-related effects. Only those overpasses providing complete spatial coverage of the watershed were retained for further analysis. The date and time of each qualifying overpass were recorded and rounded up to the next 30-min interval to align with the IMERG temporal resolution. IMERG precipitation estimates corresponding to these time intervals were then extracted for the Ranikhola watershed. Consequently, the procedure may introduce a temporal mismatch of up to 30 min. This timing uncertainty is particularly important for short-duration convective precipitation events, especially in mountainous regions where rainfall intensity and spatial organization can evolve rapidly within sub-hourly timescales. In such cases, the selected IMERG interval may not fully represent the exact precipitation conditions at the actual satellite overpass time, potentially resulting in localized underestimation or overestimation of rainfall intensity. Nevertheless, the influence of native sensor footprint size on precipitation representation over small mountainous watersheds remains an important source of uncertainty and warrants further investigation.

Although IMERG was adopted as a common precipitation reference framework to isolate temporal sampling effects, it is important to acknowledge that IMERG itself contains uncertainties, particularly in complex mountainous environments. Satellite precipitation retrievals may be affected by topographic influences, mixed-phase precipitation, beam-filling effects, and limitations in gauge-based bias correction over regions with sparse observations. Consequently, any systematic bias present in IMERG may propagate into all runoff simulations performed in this study because the precipitation values used for each sensor configuration originate from the same reference dataset. Such biases may influence the magnitude and timing of simulated discharge and therefore affect the absolute values of NSE, RMSE, MAE, and PBIAS. However, because all sensor combinations were evaluated using the identical IMERG precipitation framework, the relative differences among sensor configurations primarily reflect temporal sampling characteristics rather than retrieval-related differences. Therefore, the results should be interpreted as a comparative assessment of temporal sampling uncertainty conditioned on the accuracy of the IMERG reference product.

The precipitation information derived from IMERG for each sensor combination and overpass configuration was then used as input to the HYSIM hydrological model to simulate river discharge. From 2016 to 2020, a total of 17,494 satellite overpasses were analyzed in this study, comprising 2,730 from AMSR2, 3,480 from ATMS, 1,929 from GMI, 3,563 from MHS, and 2,986 from SSMIS. For the TROPICS CubeSat constellation, we assumed that the number of annual overpasses remained approximately constant throughout the study period. Using 2024 as a reference year, the annual overpasses were estimated as 1,295 for T3, 1,605 for T5, and 1,751 for T6. Multiplying these values by the five-year study period yielded totals of 6,475 for T3, 8,025 for T5, and 8,755 for T6, respectively.

HYSIM is well-suited for runoff simulation in high-altitude regions due to its flexible structure and ability to perform effectively with limited observed data (i.e., precipitation and evapotranspiration), making it ideal for remote, data-scarce mountainous areas (Manley, 1978; Remesan et al., 2019; Dey and Remesan, 2022; Roy et al., 2024c). Observed rainfall data from the India Meteorological Department (IMD) (Pai et al., 2014), along with five static hydraulic parameters (Table 2), were used as the primary inputs for the calibration period (2016–2018). Calibration of HYSIM was performed using an iterative Monte Carlo–based parameter sampling approach rather than a deterministic optimisation algorithm. A total of 13 calibration iterations were conducted, with 200 simulations in each iteration. In each iteration, hydrological parameters were randomly sampled within physically realistic bounds, and model performance was evaluated using NSE, R 2, and PBIAS. Poorly performing parameter sets were progressively discarded, and the feasible parameter ranges were refined in subsequent iterations. The number of iterations was selected based on convergence behaviour, as performance metrics showed negligible improvement beyond the 13th iteration, indicating stabilization of the optimal parameter space while maintaining computational efficiency. Similar iterative calibration strategies have been adopted in previous HYSIM applications in data-scarce and high-altitude watersheds.

Furthermore, hydraulic parameters, were treated as static throughout the simulation period. This assumption is consistent with the lumped conceptual structure of HYSIM, which does not explicitly represent dynamic channel adjustment processes such as bed deformation or floodplain morphology changes during high-flow events. In the absence of event-scale hydraulic observations for the study watershed, static parameterization provides a pragmatic and widely adopted approach for runoff simulation in mountainous, data-limited watershed (Roy et al., 2024a; Roy and Indu, 2026). Calibration of HYSIM was performed using an iterative Monte Carlo parameter-sampling approach rather than a deterministic optimization algorithm. Calibration focused solely on hydrological parameters, conducted over 13 iterations with 200 simulations each. In each iteration, hydrological parameters were randomly sampled within physically realistic bounds, and model performance was evaluated using statistical indicators. Poorly performing parameter sets were progressively discarded, and the feasible parameter ranges were refined in subsequent iterations. The calibration was terminated at 13 iterations because further iterations produced no major change in parameterization or model performance, indicating stabilization of the feasible parameter space (Roy et al., 2024b). The Singtam, the only gauging station (Figure 1), located at the watershed outlet, was used to compare observed and simulated daily runoff for the calibration period (2016–2018) and validation period (2019–2020). To evaluate HYSIM’s performance in the Ranikhola watershed, standard statistical indicators were used, including Nash–Sutcliffe Efficiency (NSE) (Equation 1), Coefficient of Determination (R 2) (Equation 2), Percentage Bias (PBIAS) (Equation 3), Root Mean Square Error (RMSE) (Equation 4), Mean Absolute Error (MAE) (Equation 5), and RMSE–Observations Standard Deviation Ratio (RSR) (Equation 6) (Ayele et al., 2024). These metrics collectively provide a comprehensive assessment of the model’s accuracy and reliability in streamflow simulation.

The HYSIM hydrological model requires reliable observed datasets, particularly precipitation (sourced from the India Meteorological Department, IMD) and discharge (obtained from the Central Water Commission, CWC), to facilitate the optimization and calibration of its hydrological parameters (Dey and Remesan, 2022; Roy et al., 2024c). Accurate calibration of the model is essential for ensuring that simulated hydrological processes closely represent the observed behavior of the watershed. However, for high-altitude regions, the availability of station-based meteorological data is often limited or entirely absent due to sparse monitoring networks and challenging terrain conditions. To address this limitation, gridded meteorological data provided by IMD were employed as the primary observed input for this study (Pai et al., 2014; Singh et al., 2019). The study area, the Ranikhola watershed, is entirely encompassed within a single IMD grid, thereby allowing the gridded dataset to be directly utilized without the need for additional spatial interpolation or merging of multiple grid points.

0 i is the i th observed value.

P i is the i th predicted value.

O ¯ is the mean of the observed values.

n is the total number of observations

y ¯ i = Predicted or a point value

y m e a n = Mean of all the values

y i = Actual values where

O i = Observed data at i th

S i = Simulated data at i th

n = total number of observation where

x i = Predicted value at i th

y i = Observed value at i th

N = Total number of values where

N = Total number of values

x i = Predicated value at i th

x ¯ i = Actual value at i th where

y o b s, i = Observed value at i th

y s i m, i = Simulated value at i th

μ = Mean value

The well-calibrated and optimized HYSIM model was used to simulate runoff using IMERG rainfall and rainfall derived from its collocated overpasses with different sensor combinations. This approach allows for evaluating the impact of satellite constellation configurations and temporal sampling on precipitation estimates and the accuracy of daily runoff simulation.

This section presents the estimation of runoff over the Ranikhola watershed using multiple sensor combinations from both the GPM and TROPICS constellations. The objective is to evaluate how temporal sampling uncertainty influences river discharge patterns. For this purpose, a lumped hydrological model, HYSIM, was optimized based on observed meteorological and hydrological inputs.

In this study, four different multi-sensor combinations from the GPM constellation were considered: (i) GMI and SSMIS, (ii) GMI, SSMIS, and MHS, (iii) GMI, SSMIS, MHS, and AMSR2, and (iv) GMI, SSMIS, MHS, AMSR2, and ATMS. Additionally, three CubeSat sensor combinations (T3 and T5; T3 and T6; and T3, T5 and T6) and three individual CubeSats (T3, T5, and T6) were evaluated, as summarized in the accompanying table.

Sampling consideration

Relationship between individual sensors and their combinations in terms of sampling frequency (passes per day) and sampling interval (minutes) (Figure 3). Individual sensors such as AMSR2, ATMS, GMI, SSMIS, and MHS exhibit relatively low overpass frequencies, typically ranging from ∼1 to ∼2.5 passes per day, which corresponds to longer sampling intervals of approximately 600–1500 min. The shaded regions around each data point represent the variability in sampling frequency and sampling interval, calculated as the standard deviation of the number of overpasses for each individual sensor and their combinations across the study period, reflecting the natural day-to-day inconsistency in satellite revisit timing. Systematically combining additional sensors, the overall sampling frequency increases exponentially, while the mean sampling interval decreases, showing the benefit of multi-sensor integration for reducing temporal gaps in satellite observations (Winkelried et al., 2023).

TROPICS CubeSats (T3, T5, T6) individually achieve higher sampling frequencies (∼3–5 passes/day) with shorter sampling intervals compared to traditional sensors. The combination of two TROPICS CubeSats provides sampling performance equivalent to that of all five GPM constellation sensors combined. Demonstrating the significant potential of small satellite constellations to enhance temporal coverage. TROPICS constellation (T3, T5, T6) provides the highest overpass frequency (>12 passes/day) and the shortest sampling intervals (∼140 min), providing the best temporal resolution for monitoring rapidly evolving precipitation systems. This finding is consistent with Blackwell et al. (2018a), who reported that a fully deployed TROPICS four-CubeSat constellation could reduce the median revisit time to approximately ∼1 h over tropical regions.

Observed data-based runoff simulation

As a baseline for this study, calibration and validation of the hydrological model (HYSIM) were carried out using observed discharge and precipitation data from the CWC and the IMD (Figure 4). IMD observed rainfall and CWC discharge data were used as the primary input for the hydrological model framework for calibration (2016–2018) and validation (2019–2020). Table 2 depicts the calibrated hydrological and static hydraulic parameters during the calibration. CWC discharge observations and IMD daily rainfall data were used for the calibration and validation of the HYSIM model. IMERG precipitation was not used during model calibration because previous studies, particularly Roy and Indu (2026), demonstrated that IMD rainfall data provide superior hydrological performance for HYSIM calibration compared to IMERG precipitation forcing. Accurate calibration is particularly important for HYSIM, as the model relies on calibrated static hydraulic parameters that strongly control runoff simulation. Therefore, the use of IMD rainfall data ensured a more reliable hydrological baseline prior to evaluating the impact of satellite temporal sampling uncertainty using IMERG-based sensor sampling experiments. Figure 4 show how well simulated runoff aligns with the observed discharge information. The statistical evaluation ensures that the model’s static hydraulic and calibrated hydrologic parameters are optimally constrained for reliable runoff simulation (Table 3). The model demonstrates strong performance, with coefficient of determination (R 2) values of 0.885 and 0.889, indicating a high degree of correlation between observed and simulated streamflow. The Nash–Sutcliffe Efficiency (NSE) values of 0.837 and 0.845 further confirm the model’s good predictive capability, while the Percent Bias (PBIAS) remains low at −0.019 for both periods, suggesting minimal systematic underestimation. According to Moriasi et al., 2007, established model evaluation criteria, NSE >0.75 is classified as ‘very good’ performance, confirming that the calibrated HYSIM model reliably reproduces the temporal runoff dynamics of the Ranikhola watershed. Additionally, the low RSR values (0.404 for calibration and 0.383 for validation) indicate that the residual variation is acceptably small relative to the observed data variability. Overall, these performance indicators confirm that the calibrated HYSIM model reliably reproduces the temporal runoff dynamics of the Ranikhola watershed. These performance indicators are consistent with similar lumped hydrological model applications mountainous watersheds region reported in the literature (Kumar et al., 2025; Poméon et al., 2017). Limitation of this study is the assumption of temporally invariant hydraulic parameters throughout the simulation period. HYSIM model was implemented using fixed channel geometry and hydraulic characteristics, whereas mountain streams may undergo morphological adjustments during extreme flood events through bank erosion, channel deformation, sediment transport, and deposition processes. Such changes can alter channel conveyance and flow routing characteristics, potentially influencing simulated peak discharge and runoff timing. Uncertainty may exist in the simulation of extreme flood events, and future investigations should explore the coupling of precipitation sampling analyses with morphodynamic and sediment-transport models capable of representing flood-induced channel adjustments. However, the primary objective of this study is to isolate and quantify the impact of precipitation temporal sampling uncertainty on runoff simulation. By keeping hydraulic parameters constant across all simulations, differences in simulated discharge can be attributed solely to variations in precipitation sampling arising from different satellite sensor configurations. Future studies could explore the integration of time-varying hydraulic parameters or coupled hydrodynamic models to further investigate flood-specific processes.

Runoff -simulation by different sensor combinations using optimized HYSIM

The simulated runoff using different sensor combinations from the GPM constellation is shown in Figure 5 for the configurations: (1) GMI–SSMIS, (2) GMI–SSMIS–MHS, (3) GMI–SSMIS–MHS–AMSR2, and (4) GMI–SSMIS–MHS–AMSR2–ATMS. Whereas, two configurations from the TROPICS constellation (1) T03–T05 and (2) T03–T05–T06. It is observed that the combination of all three TROPICS CubeSats outperforms the other configurations, producing discharge values that closely match the observed discharge. In contrast, the combination of only GMI and SSMIS exhibiting high temporal sampling uncertainty due to its lower overpass frequency, which results in significant missed precipitation information. Missing short-duration extreme rainfall events and potentially result in inaccurate runoff estimation and increased flood risk (Das et al., 2018).

The combination of all three TROPICS CubeSats outperforms the five-sensor configuration from the GPM constellation, indicating that the TROPICS constellation provides higher sampling frequency and captures more precipitation events with minimal gaps. This improved temporal coverage can support more reliable flood risk assessment and management, particularly in vulnerable and flood-prone regions (Das et al., 2019). The improved runoff simulation obtained from the TROPICS constellation is not solely related to the higher number of overpasses, but also to the improved timing of observations relative to rapidly evolving rainfall events. TROPICS CubeSats provide a median revisit time of approximately 1 h, substantially shorter than the revisit intervals of conventional passive microwave sensors in the GPM constellation (Blackwell et al., 2018b). This shorter revisit interval increases the probability of capturing critical storm evolution stages and short-duration precipitation peaks that are often missed by sensors with longer temporal gaps. This is particularly important in mountainous watersheds (like Ranikhola) where convective rainfall evolves rapidly and contributes disproportionately to runoff generation. Therefore, the improved hydrological performance of the TROPICS combinations reflects both enhanced temporal sampling frequency and better synchronization with the timing of intense precipitation events. Notably, even the combination of only two TROPICS CubeSats yields discharge estimates comparable to those produced by the five-sensor GPM configuration. The average diurnal distribution of satellite sensor overpasses over the study area was examined for the period 2016–2020 (Figure 6). A limitation of this study is that the TROPICS overpass timing and swath geometry from 2024 are assumed to represent the typical temporal sampling of the TROPICS constellation in earlier years. As mentioned earlier this assumption ignores orbital drift, constellation phasing, or seasonal sampling biases in local overpass times. However, because the TROPICS CubeSats operate in stable low-Earth orbits with fixed inclination and altitude (Blackwell et la., 2018a). As a result, the 2024 sampling pattern provides a reasonable first-order representation of the constellation’s temporal sampling behaviour. Another limitation is that the passive microwave sensors considered in this study have different native spatial resolutions (e.g., ∼15–25 km for GMI and ∼25–40 km for TROPICS), which may also contribute to uncertainty in precipitation representation. However, the present framework was specifically designed to isolate temporal sampling effects rather than retrieval or spatial-resolution-related uncertainties. Therefore, only the swath geometry and overpass timing of each sensor were used, while precipitation estimates were consistently extracted from the same IMERG reference dataset. By avoiding the direct use of individual sensor retrievals, the analysis primarily reflects differences in observation frequency and revisit timing relevant to hydrological modelling. Nevertheless, spatial-resolution differences may still indirectly influence precipitation sampling characteristics and should be investigated further in future studies. Future studies could further evaluate this assumption by using multiple years of orbital data to assess the sensitivity of the results to seasonal and orbital variability. Maxima in the probability density functions represent periods of high sampling frequency, whereas minima indicate low or no sampling of the region at those times. The horizontal lines corresponding to each sensor in Figure 6 denote the average peak overpass times. The results demonstrate that the TROPICS CubeSats exhibit a higher number of overpasses over the study area compared to sensors from the GPM constellation. While most GPM sensors provide, on average, two overpasses per day (except GMI), the TROPICS CubeSats achieve nearly four overpasses per day, supporting higher temporal sampling frequency.

The statistical evaluation of discharge estimates compared to the observed CWC discharge data is presented in Table 4. The discharges from the individual TROPICS CubeSats (T3, T5, T6) and the two-sensor combination from the GPM constellation (GMI–SSMIS) was found to exhibit relatively poorer performance, indicating that these temporal sampling configurations perform poorly than the mean observed discharge and are therefore unsuitable for reliable operational hydrological applications. The GMI–SSMIS pair was found to result in lowest NSE value (−1.7385), while the individual TROPICS CubeSats have NSE values ranging from −1.4 to −1.6, suggesting that their discharge estimates are less reliable than simply using the mean of the observed values. Similarly, these configurations yield relatively high RMSE values, ranging from 21.9 to 23.4 m 3 s -1, and MAE values between 18.0 m 3 s -1 and 19.2 m 3 s -1. The coefficient of determination (R 2) for individual sensors remains moderate, between 0.31 and 0.52, indicating limited agreement with the reference discharge data. It is evident in the GPM-based combinations, where the inclusion of additional sensors (MHS, AMSR2, ATMS) progressively improves NSE and lowers both RMSE and MAE. The progressive improvement seen with the addition of sensors further confirms the benefits of multi-sensor fusion for precipitation estimation (Liu et al., 2026).

The combination of T3 and T5 resulted in a higher NSE (−0.653) and an R 2 of 0.5934, along with reduced RMSE (18.15 m 3 s -1) and MAE (14.95 m 3 s -1) compared to the individual sensors. A further improvement was observed when T3, T5, and T6 were combined, yielding a positive NSE of 0.6983, a substantially higher R 2 of 0.9568, and a considerable reduction in RMSE (7.75 m 3 s -1) and MAE (5.62 m 3 s -1) with least PBIAS -0.2558 where negative values indicate underestimation of discharge by the model relative to observations and positive values indicate overestimation. The improvement is particularly notable for the T3–T5–T6 combination, which achieves both a positive NSE and a high R 2, indicating strong agreement with reference observations and an effective reduction in estimation errors. The superior performance of the T3–T5–T6 configuration is primarily attributed to its higher temporal sampling frequency and shorter revisit intervals, which increase the probability of observing rapidly evolving precipitation events. Because precipitation for all sensor configurations was extracted from the same IMERG reference dataset and only overpasses providing complete watershed coverage were retained, differences in runoff performance mainly reflect temporal sampling characteristics rather than differences in precipitation retrievals or watershed coverage. However, minor contributions from spatial sampling geometry cannot be entirely excluded, as multiple CubeSats provide overlapping observation opportunities that may reduce the likelihood of missing localized precipitation events. A similar trend is evident in the GMI-based combinations, where the inclusion of additional sensors progressively improves NSE (from −1.7385 for GMI–SSMIS to −0.6751 for GMI–SSMIS–MHS–AMSR2–ATMS) and lowers both RMSE and MAE. The progressive improvement seen with the addition of sensors such as MHS, AMSR2, and ATMS to the GMI–SSMIS combination further confirms the benefits of multi-sensor fusion.

A comparison with previous hydrological studies indicates that the runoff simulations obtained from the three-satellite TROPICS configuration are comparable to the performance achieved using fully processed satellite precipitation products in regional hydrological applications (Table 5). Although the present framework utilized only the temporal sampling characteristics and swath geometry of the TROPICS CubeSats over a common IMERG precipitation reference field, the resulting runoff simulations (NSE ≈0.70; R 2 ≈ 0.96) were similar to studies employing direct IMERG precipitation forcing. These findings highlight the importance of high revisit frequency and improved storm-peak sampling for runoff simulation in rapidly responding mountainous watersheds. While Pradhan et al. (2021) reported slightly higher NSE values (∼0.73) using DPR-based precipitation forcing over Mahanadi basin, their framework relied on temporally aggregated synthetic precipitation ensembles generated using SREM2D over a continuous CMORPH reference field. In contrast, the present study directly evaluated the temporal sampling characteristics of individual sensors and CubeSat constellations in a flash-flood-prone Himalayan watershed highly sensitive to short-duration precipitation timing. The comparable NSE achieved in this study therefore demonstrates the strong capability of high-frequency CubeSat constellations to reduce temporal sampling uncertainty under hydrologically challenging conditions.

For the quantitative evaluation, the performance of all the sensor combinations taylor diagram is used (Figure 7). The plot represents the runoff estimation compared to the CWC observed data. Point closer to the reference star (red) with a standard deviation similar to that of the reference, and a correlation coefficient near unity indicates better agreement with the observed data.

It can be observed that the combinations of multiple sensors (e.g., GMI, SSMI/S, MHS, AMSR2, ATMS) tend to cluster closer to the reference, exhibiting higher correlation coefficients between 0.6 and 0.8, standard deviations between 0.1 and 5.0 and RMSD more the 10 m 3 s -1. This indicates that multi-sensor sampling improves the representation of runoff variability and reduces deviation from observed discharge patterns. Individual sensors such as T3, T5, and T6 show relatively lower standard deviations and slightly reduced correlations, indicating an underestimation of variability when used independently. The combination of the three CubeSats (T3, T5, and T6) outperforms all other sensor combinations for runoff estimation. It shows a low RMSE value of ∼7 m 3 s -1 and is notably closest to the reference point, with a standard deviation of 9.2 and a correlation coefficient close to 0.98. Indicating that the TROPICS constellation provides higher sampling frequency and captures more precipitation events with minimal gaps (Blackwell et al., 2018b). This result is consistent with findings from small satellite constellation studies demonstrating that CubejSat constellations can fundamentally transform environmental monitoring capability (Ruf et al., 2018). Overall, the diagram highlights that multi-sensor configurations improve the accuracy and reliability of runoff simulation, with the three-CubeSat combination providing the best agreement with observed discharge.

This study demonstrates the impact of temporal sampling uncertainty for precipitation on runoff simulation using in hydrological extreme-prone watershed using the HYSIM model and multi-sensor observations from the GPM and TROPICS constellations. The model was calibrated and validated with high-performance statistics using observed meteorological and hydrological datasets. These results demonstrate that the model is well-calibrated and suitable as a baseline framework for evaluating the impact of temporal sampling uncertainty in precipitation using multi-sensor data from the GPM and TROPICS constellations.

All the five sensors from GPM (GMI, SSMIS, MHS, AMSR2, and ATMS) constellation is used in different combination is used in this study whereas from TROPICS all three available CubSats (T03, T05 and T06) is used as individual and in a combination. Among all evaluated sensor combinations, the three-satellite TROPICS CubeSat configuration (T3–T5–T6) produced the best hydrological performance, yielding strong agreement with observed discharge (R 2 ≈ 0.96; NSE ≈0.70) and significantly lower RMSE and MAE compared to individual sensors and several GPM sensor combinations. The TROPICS constellation also achieved the highest sampling frequency (>12 passes day -1) and shortest revisit interval (∼1–2 h), increasing the likelihood of capturing short-duration storm peaks that are often missed by conventional passive microwave sensors with longer temporal gaps. These findings indicate that improved runoff simulation is associated not only with a higher number of observations, but also with improved timing of precipitation sampling during critical rainfall events. The progressive improvement observed with increasing sensor combinations further confirms that temporal sampling frequency plays a critical role in reducing uncertainty propagation in rainfall–runoff modelling. The results additionally suggest that, in similar data-scarce flash-flood-prone mountainous regions, a constellation configuration equivalent to at least three high-frequency CubeSats may be necessary to achieve sufficiently short revisit intervals for reliable runoff simulation and improved hydrological performance. These findings suggest that high-frequency CubeSat constellations such as TROPICS may play an important role in improving precipitation monitoring and hydrological applications in similar data-scarce, flash-flood-prone mountainous regions.

However, some limitations should be acknowledged. The TROPICS analysis assumes that the 2024 orbital configuration is representative of earlier years and therefore does not fully account for long-term orbital drift or seasonal variability in revisit characteristics. In addition, IMERG was used as a common precipitation reference framework to isolate temporal sampling effects from retrieval-related uncertainty, while differences in native spatial resolution among sensors were not explicitly evaluated. In addition, any systematic precipitation bias in IMERG may propagate into the simulated runoff and influence the absolute hydrological performance metrics reported in this study. Furthermore, the analysis was conducted over a single watershed, which may limit the broader applicability of the findings.

Future studies should extend this framework to multiple hydroclimatic regions and longer orbital datasets to further evaluate the robustness of high-frequency satellite sampling for hydrological applications. Additional investigation using higher temporal resolution precipitation products, direct sensor retrievals, and coupled flood forecasting frameworks may also improve understanding of how temporal sampling uncertainty propagates into extreme runoff and flash-flood simulations.