Rainfall Measurements Research Articles (Page 1)

HighlightsWhat are the main findings?Wind affects rainfall measurements taken by the OTT Parsivel2 disdrometer, with significant under- or overestimation depending on wind speed and direction.Numerical simulation enables the wind-induced bias of disdrometer measurements to be quantified in terms of site-independent catch ratios.What is the implication of the main finding?Raw measurements should be adjusted using the derived catch ratios, based on the measured drop size distribution, wind speed and direction.Positioning of the instrument to align with the prevailing wind direction is crucial to achieve acceptable accuracy in rainfall measurement.The impact of wind on precipitation measurements from the OTT Parsivel2 optical transmission disdrometer is quantified using computational fluid dynamics simulations. The numerical velocity field around the instrument body and above the instrument sensing area (the laser beam) shows significant disturbance that depends heavily on the wind direction. By computing the trajectories of raindrops approaching the instrument, the wind-induced bias is quantified for a wide range of environmental conditions. Adjustments are derived in terms of site-independent catch ratios, which can be used to correct measurements in post-processing. The impact on two integral rainfall variables, the rainfall intensity and radar reflectivity, is calculated in terms of collection and radar retrieval efficiency assuming a sample drop size distribution. For rainfall intensity measurements, the OTT Parsivel2 shows significant bias, even much higher than the wind-induced bias typical of catching-type rain gauges. Large underestimation is shown for wind parallel to the laser beam, while limited bias occurs for wind perpendicular to it. The intermediate case, with wind at 45°, presents non negligible overestimation. Proper alignment of the instrument with the laser beam perpendicular to the prevailing wind direction at the installation site and the use of windshields may significantly reduce the overall wind-induced bias.

Increasing air temperatures are expected to change streamflow patterns in Arctic watersheds. However, the hydro-meteorological datasets necessary to evaluate these changes remain limited in the Arctic. We revised and updated a unique hydrologic dataset from Arctic Alaska with field measurements of continuous streamflow, precipitation, and evapotranspiration to identify changes in the hydrologic cycle. We analyzed water balance components and peak streamflow in Imnavait Creek, Upper Kuparuk River, and Kuparuk River during snowmelt and summer periods. Field measurements show that summer streamflow (mid-June to late-September) has increased in volume and magnitude. While snowmelt runoff still constitutes a sizable portion of annual runoff (28–58%), the relative proportion of summer runoff is notably increasing. High interannual and spatial variability is observed for both rainfall and summer runoff, whereas comparatively lower interannual variability is observed for snowmelt runoff. This paper compares streamflow changes occurring at three watersheds of varying sizes and identifies reasons for observed changes using snow, rainfall, and evapotranspiration measurements. We also provide hydrologic analysis for design considerations at Arctic Alaska communities, the Dalton Highway, and the oil pipeline; as this infrastructure may be at risk of increased erosional damage due to more frequent and higher magnitude summer flow events.

Water is fundamental to daily life and plays critical roles across numerous sectors. To address increasing water-related challenges, innovative solutions are being developed, such as adaptive management and advanced remote sensing, along with concepts like water security and global information integration. Dam safety, in particular, is becoming more critical as aging infrastructure, seismic activity, and extreme weather events increase the risk of dam damage or failure. In response, dam safety has become a top priority for national disaster management strategies. Governments are enacting regulatory measures to ensure dam safety, and various organizations are implementing both institutional and technical safeguards. one of the primary gaps in dam safety is the lack of standardized protocols for water release, especially in emergency situations. This project proposes an AI-driven dam management system that leverages a Stacked Dense LSTM (Long Short-Term Memory) model to monitor and stabilize dam conditions. The LSTM model processes real-time water level data, along with temperature, humidity, and rainfall measurements, which are stored in the cloud. This machine learning framework helps predict and manage water flow, guiding the automated release of water through dam gates to prevent overflow and flooding. floods, often triggered by rising rivers, lakes, or heavy rainfall, can occur unexpectedly at any time and pose significant risks to lives and property. These events can force families from their homes, devastate farms, and create lasting hardships for affected communities. The proposed system’s predictive capabilities offer a proactive approach to dam management, helping to mitigate flood risks and enhance community resilience

Variations in rainfall patterns across different regions reduce the accuracy of existing satellite channel models. As satellite services and 5G applications continue to advance, the development of accurate rain-impairment-aware channel models has become essential. This paper presents a prediction model for rain-induced impairments in High Throughput Satellite (HTS) and 5G satellite-to-land communication channels. The proposed model integrates three novel algorithms designed to characterize and analyze rain-induced attenuation and channel quality. Specifically, these algorithms calculate rain-specific attenuation, effective slant path lengths through rainfall, overall rain-induced attenuation, signal carrier-to-noise ratios, and symbol error rates across three conventional modulation schemes. Additionally, the study introduces a new database detailing rain-induced attenuation on HTS channels, considering various frequencies and rainfall intensities. Results indicate substantial fluctuations in HTS-to-land fade levels and signal quality during rainfall events, which could lead to communication link outages, particularly at higher-order modulation schemes. This study provides practical methods to analyze channel characteristics using actual rainfall measurements, thereby facilitating the effective design and deployment of future HTS and 5G system.

Abstract East Africa relies heavily on satellite-based rainfall estimates due to the lack of in situ data. However, satellite rainfall products often perform poorly in this region. In this study, data from the Trans-African Hydrometeorological Observatory (TAHMO) were used to build a regional rainfall product in East Africa based on the Soil Moisture to Rain (SM2Rain) algorithm. Subsequently, this regional product was merged with a reanalysis product (ERA5) and two microwave (MW)/infrared (IR)-based rainfall products (IMERG and CHIRPS) based on the Statistical Uncertainty Analysis-Based Precipitation Merging (SUPER) framework. Within this framework, merging weights are derived from the error variances of the rainfall products determined from quadruple collocation on a pixel-to-pixel basis. The merged and individual products are evaluated using data from individual TAHMO stations. We tested SUPER with various interproduct dependency assumptions and found that, in the best-performing configuration, IMERG contributed the most to the merged product, followed by CHIRPS, ERA5, and SM2Rain. SM2Rain showed performance comparable to other rainfall products but is more useful for detecting the offset of the rainy season in drier climates and less reliable under wet conditions. The findings indicated that the merged product outperforms the individual products in most performance metrics. Additionally, we demonstrated the importance of comparing satellite and ground-measured precipitation time series, alongside evaluating performance metrics. The ultimate goal of this study is to develop a workflow to enhance the accuracy of rainfall measurements in East Africa by leveraging information from TAHMO data and different existing products, contributing to the improvement of satellite-based rainfall estimates in East Africa.

Soil erosion is a major environmental challenge that necessitates meticulous investigation and the implementation of sustainable management practices. The objective of this study is to provide a thorough assessment of soil erosion in the Bay region from 2020 to 2023, utilizing the Revised Universal Soil Loss Equation (RUSLE) and advanced geospatial technologies, particularly Google Earth Engine, to guide sustainable land management strategies. The study integrates multiple datasets, including CHIRPS for rainfall measurement, MODIS for land use analysis, and a digital elevation model for slope calculation, to offer a comprehensive understanding of the factors contributing to soil erosion. The rainfall erosivity (R) factor is calculated using CHIRPS data, while the soil erodibility (K-factor) is derived from the soil dataset. The topographic (LS-factor) is computed using the digital elevation model, and the cover-management (C) and support practice (P) factors are determined from the NDVI and land use data, respectively. The findings reveal considerable spatial variation in soil erosion across the Hirshabelle regions. The results are categorized into five levels based on the severity of soil loss: Slight (<10), Moderate (10-20), High (20-30), very high (30-40), and Severe (>40). While areas classified under “Slight” soil loss are dominant, indicating relatively stable soils, regions under “Severe” soil loss signal potential land degradation and the need for immediate intervention. Furthermore, the study revealed the intricate interplay of slope, vegetation, and land use in influencing soil erosion. Areas with steeper slopes and less vegetation were more susceptible to soil loss, emphasizing the need for targeted soil conservation measures in these regions. The land use factor played a crucial role, with certain land uses contributing more to soil erosion than others.

Flooding is one of the natural disasters that can occur in various parts of the world and may arise suddenly. However, flood events can be predicted or anticipated through relevant scientific approaches. One such method is by estimating the flood discharge in a given area. Rainfall data is one of the essential inputs required to determine flood discharge. In practice, however, ground-based rainfall measurements often have limitations. To overcome these shortcomings, satellite-based rainfall data can be utilized. There are notable differences between directly measured rainfall data and satellite-derived rainfall data; therefore, satellite data must be calibrated or validated prior to conducting further analysis. One of the most widely used satellite rainfall datasets is the GPM (Global Precipitation Measurement) satellite data, which has a spatial resolution of 0.1° x 0.1°. This study employs a combination of two statistical methods—validation and calibration—to evaluate rainfall data. Prior to evaluation, the RMSE and NSE values did not meet acceptable standards, and the correlation value was low. However, after the evaluation using both methods, improvements were observed: RMSE and NSE values became acceptable, and the correlation increased. These results indicate that the applied methods are effective for evaluating rainfall data. For future research, monthly or annual rainfall data can be utilized to further explore the relationship between different temporal scales of rainfall observations.

Climate change has increased flood risks in downstream Nigeria, driven by altered hydrology, dam operations, and land-use changes threatening infrastructure, livelihoods, and ecosystem stability with growing frequency and severity. This study analyzes flood patterns, identifies key environmental drivers, and predicts flood-prone areas through an integrated machine learning and geospatial analysis approach. Data sources included Synthetic Aperture Radar (SAR) imagery from Sentinel-1, rainfall measurements, Shuttle Radar Topography Mission (SRTM) elevation data, and surface water level records. Machine learning models Random Forest (RF), Support Vector Machine (SVM), and Artificial Neural Network (ANN) were applied using geospatial tools such as Google Earth Engine and ArcGIS 10.5 to assess flood dynamics from 2018 to 2024. Downstream regions (elevation: 78–235.1 m) exhibited greater flood susceptibility than upstream areas (up to 1399.43 m). Flood extents rose by 10.9% in August (from 2441.91 km2 in 2018 to 2707.75 km2 in 2024) and by 39.8% in October (from 3083.44 km2 to 4311.55 km2). The RF model achieved the highest accuracy (92%), outperforming SVM (88%) and ANN (85%). Inundated areas increased from 20 to 35% of downstream zones. Rainfall intensity rose by 15–20%, with annual totals exceeding 4311 mm in some areas. Forest cover declined by 15–20%, further exacerbating flood risks. The findings demonstrate that climate change, land-use alteration, and dam operations are major contributors to flooding. Mitigation strategies include 10–15% reforestation, embankment construction, and machine learning–driven early warning systems, which can reduce flood damage by up to 30%. These approaches support sustainable flood risk management in Nigeria.

This project aimed to evaluate the use of meteorological satellite-derived rainfall data to estimate water inflows to dams. In this study, the Lam Pao Dam in the Chi Basin, Thailand, was used as a case study. Rainfall data were obtained using the PERSIANN technique. To improve accuracy, satellite-derived rainfall estimates were adjusted using ground-based rainfall measurements from stations located near and within the catchment area, applying the 1-DVAR method. The Kriging method was employed to estimate the spatial distribution of rainfall over the catchment area. This approach resulted in a Probability of Detection (POD) of 0.92 and a Threat Score (TS) of 0.72 for rainfall estimates in the Chi Basin. Rainfall data from the Weather Research and Forecasting (WRF) numerical models were used as inputs for the HEC-HMS model to simulate water inflows into the dam. To refine rainfall estimates, various microphysics schemes were tested, including WSM3, WSM5, WSM6, Thompson, and Thompson Aerosol-Aware. Among these, the Thomson Aerosol-Aware scheme demonstrated the highest accuracy, achieving an average POD of 0.96, indicating highly reliable rainfall predictions for the Lam Pao Dam catchment. The findings underscore the potential benefits of using satellite-derived meteorological data for rainfall estimation, particularly where installing and maintaining ground-based measurement stations is difficult, e.g., forests/mountainous areas. This research contributes to a better understanding of satellite-derived rainfall patterns and their influence on catchment hydrology for enhanced water resource analysis.

Abstract In road construction groundwater especially surface and flows play very important roles, because road paths must cross over narrow or wide drainage basins through culverts, bridges or aqueducts. Under the bridges and aqueducts water problems do not pose very dangerous situations, but culverts as comparatively small-scale engineering structures are subject to most dangerous surface water (runoff) problems and thus need more attention. Culverts under the highways and railways must be designed and constructed in such a way that they pass the upstream runoff without critical problem to downstream area. In the literature there are many reports about the water passage and dimensional design formulations for culvert works that have empirical bases and validity for the region they are planned. In most of the culvert drainage basins, there are no measurements of rainfall and therefore empirical, logical rational approaches are needed. One of such method is the bivalent logic (two-value, crisp) rule-based approach is the Talbot culvert design approach, which does not depend on complex mathematical equations and cannot consider various uncertainties (vague, imprecise, incomplete) effectively. This paper proposes fuzzy logic inference model (FLIM) approach for transformation of Talbot bivalent logic method steps into more meaningful fuzzy logic method. The application of the proposed method is made in a local area where a serious train accident occurred due to a poorly maintained culvert in Tekirdağ province near Istanbul on the European side of Turkey.

This study investigates the impact of desert dust on precipitation patterns using multi-model simulations. Dust-based processes of formation/removal of ice nuclei (IN) and cloud condensation nuclei (CCN) are investigated by using both the online access model WRF-CHIMERE and the online integrated model WRF-Chem. Comparisons of model predictions with rainfall measurements (GRISO: Spatial Interpolation Generator from Rainfall Observations) over the Italian peninsula show the models’ ability to reproduce heavy orographic precipitation in alpine regions. To quantify the impact of the mineral dust transport concomitant to the atmospheric river (AR) on cloud formation, a sensitivity study is performed by using the WRF-CHIMERE model (i) by setting dust concentrations to zero and (ii) by modifying the settings of the Thompson Aerosol-Aware microphysics scheme. Statistical comparisons revealed that WRF-CHIMERE outperformed WRF-Chem. It achieved a correlation coefficient of up to 0.77, mean bias (MB) between +3.56 and +5.01 mm/day, and lower RMSE and MAE values (~32 mm and ~22 mm, respectively). Conversely, WRF-Chem displayed a substantial underestimation, with an MB of −25.22 mm/day and higher RMSE and MAE values. Our findings show that, despite general agreement in spatial precipitation patterns, both models significantly underestimated the peak daily rainfall in pre-alpine regions (e.g., 216 mm observed at Malga Valine vs. 130–140 mm simulated, corresponding to a 35–40% underestimation). Although important instantaneous changes in precipitation and temperature were modeled at a local scale, no significant total changes in precipitation or air temperature averaged over the entire domain were observed. These results underline the complexity of aerosol–cloud interactions and the need for improved parameterizations in coupled meteorological models.

S-band radar beams are easily obstructed by terrain during propagation. After the beam is blocked, the radar cannot receive the echo signal of the target area, forming a data blind spot. Traditional methods cannot obtain a complete precipitation` dataset, which increases the difficulty of precipitation estimation and leads to errors in precipitation estimation results, resulting in lower scores. This study conducted quantitative precipitation estimation using S-band dual-polarization radar under a convective scale ensemble simulation. Firstly, a certain region in Hebei Province was selected as the research object to conduct convective scale ensemble simulation to obtain more precipitation datasets. Then, the precipitation data was smoothed and used to invert the radar precipitation intensity every 6 min. The estimated hourly rainfall of the radar was matched with the hourly rainfall measurement of a single-point rainfall station. Finally, based on deep learning theory, a quantitative precipitation estimation model for S-band dual-polarization radar was constructed. The experimental results show that using the proposed method, the root mean square error (RMSE) value is less than 0.372, the mean absolute error (MAE) is less than 0.247, the correlation coefficient (CC) value is higher than 94.7%, the TS score is higher than 95.1%, and the quantitative precipitation estimation effect is good.

Abstract Lightning and rainfall are associated with severe weather conditions. These two parameters, occurring either simultaneously or with a lag, have an intrinsic relationship. This study focuses on characterizing this inherent lightning–rainfall relation, which in turn is used to perform corrections of rainfall rate estimation based on satellite measurements. The approach is to analyze the lightning–rainfall relation at varying spatial and temporal scales and to select the best resolution at which the lightning flash rate can best model rainfall volume. Then, the parameters of a power-law model are optimized to assimilate lightning flashes into the Global Satellite Mapping of Precipitation (GSMaP) rainfall product. The results revealed that the lightning and rainfall relationship varies depending on the weather system. The lightning–rainfall relation, which can be empirically described by a power law, is most apparent in thunderstorms followed by frontal systems and least in typhoons. The differences in relationships are due to the variations in lightning and rainfall collocation per weather system. Rainfall volumes are also found to vary heavily on low lightning flash rates. Considering the best-case resolution of 100 km and 60 min for thunderstorms, the lightning-corrected GSMaP rainfall volume shows a reduction in root-mean-square error (RMSE) by about 59% and an increase in the correlation from 0.6 to 0.7 concerning radar rainfall volume. The correction scheme is also effective from resolutions of 40 to 250 km in thunderstorms, achieving moderate positive correlations with RMSE about 5 mm h−1 and below. Significance Statement Accurate and timely information on precipitation is necessary to mitigate the impacts of rainfall-related hazards. While satellites provide consistent rainfall information at wide coverage, satellite-based measurements are still prone to error. Considering the association between rainfall and lightning occurrences, we came up with a technique to improve the performance of the GSMaP rainfall measurements through the introduction of lightning data. This study is a step forward to leveraging the future global coverage of lightning measurements from satellite-based lightning mappers that can enable more accurate global rainfall estimates.

High-resolution, satellite-retrieved precipitation products are useful input data for hydrological predictions and water resources management, especially in developing countries where the availability of ground-based rainfall measurements with high spatial coverage is very limited. This research explores the temporal variability of rainfall, crucial for understanding hydrology and water resource management, particularly in vulnerable regions like Bihar, India. satellite-based precipitation estimates from the Tropical Rainfall Measuring Mission (TRMM) was evaluated using observed rainfall at meteorological station using key statistical parameters and also carried long-term trend analysis during 2000–2023 by applying Mann–Kendall Test and estimating Standardized Anomaly Index (SAI). The results reveal that, Rainfall was underestimated by satellite product and Before bias correction, TRMM data exhibited significant discrepancies in rainfall estimates, with varying biases and mean errors across grid points. After bias correction, the agreement between TRMM and observed rainfall significantly improved, with Pearson correlation coefficients stabilizing between 0.8 and 1.0, bias reduced to − 0.1 to 0.2, and mean errors minimized to − 0.1 to 0.1. Additionally, root mean square error (RMSE) improved, and R2 values, indicating enhanced reliability of the corrected data. The analysis of annual rainfall and the Standardized Anomaly Index (SAI) indicates significant variability without a clear trend. This variability exposes the region to extreme weather, with flooding in wet years like 2007, 2008, 2011 and 2021 and droughts in dry years such as 2005 and 2019. Thus, there is an urgent need for adaptive water management and agricultural strategies.

On a global scale, the impacts of climate change on the hydrological cycle are clearly visible and documented. Some damages to ecosystems and humanity tend to be irreversible, particularly extreme events such as floodings and droughts, which are expected to occur more frequently (IPCC 2023). Quantifying the processes at work in the hydrological cycle at the catchment scale is very complex due to landscape heterogeneity and non-linear interactions. However, it can be addressed by constraining the problem using different sources of data and modelling tools. In this study, conducted at the ORE AgrHyS agro-hydrological observatory (Brittany, NW France), we compared ground-based measurements of actual evapotranspiration (AET) and soil water content (SWC) with satellite-based AET estimates and the outputs of a semi-distributed hydrological model. AET fluxes were measured by eddy-covariance at the FR-Nzn flux tower site (part of the EUROFLUX network), located on grazed grassland at the catchment head. SWC measurements were performed with TDR (Time-Domain Reflectometry) sensors at different locations and measurement depths in the watershed (5cm depth at most sites, but also down to 50 cm at the flux tower). Satellite-based AET estimates were computed using Landsat 8 OLI-TIRS products as well as local and meteorological model data (AROME model, by Météo France). The hydrological model used multi-site daily rainfall measurements and potential evapotranspiration to simulate daily specific discharge at the outlet and further constrain soil water balance, water table depth, flow rates and AET fluxes at the catchment scale. Additionally, the SWC TDR measurements performed over the catchment were compared to estimates obtained from the Copernicus Sentinel-1 radar product. Modelled and measured AET fluxes were compared over the period 2016-2024 to explore their inter-annual variability, extreme weather events (heat waves and droughts), and to discuss discrepancies between models and measurements. The potential of spatialized AET combining local measurements and modelled outputs is also studied. Such a combination of data sources is expected to improve modelling tools and to reduce uncertainties in AET and water balance at the catchment-scale.

Toward accurate and scalable rainfall estimation using surveillance camera data and a hybrid deep-learning framework.

Developing flood forecasting techniques at short timescales improve early warning systems to mitigate severe flood risk and facilitate effective emergency response strategies at vulnerable sites. In this study, we develop a hybrid deep learning algorithm, C-GRU, by integrating Convolutional Neural Networks (CNN) with Gated Recurrent Unit (GRU) model and evaluate its effectiveness in forecasting an hourly flood index () in five flood-prone, specific study sites in Fiji. The model incorporates statistically significant lagged with real-time hourly rainfall measurements obtained from rainfall stations, and comparative analysis is performed against benchmark models: CNN, GRU, Long Short-Term Memory and Random Forest Regression. The proposed model’s outputs comprise the predicted at each specific site at a lead time of 1-h. The results demonstrate that the proposed hybrid C-GRU model outperforms all the other models in accurately forecasting over a 1-hourly forecast horizon. Across all of the study sites, the proposed model consistently generates the highest r (0.996–0.999) and the lowest RMSE (0.007–0.014) and MAE (0.003–0.004) in the testing phase. The proposed hybrid C-GRU model also achieves the highest Global Performance Index (GPI) values and the largest percentage of forecast errors (FE) ( 98.9–99.9%) within smaller error brackets (i.e., ) across all study sites. Using the methodologies developed, we show the practical application of the proposed framework as a decision support system for early flood warning, demonstrating its potential to enhance real-time monitoring and early warning systems with broader application to flood-prone regions.

Rainfall is a major changeable atmospheric element. Existing rain gauge networks lack the temporal and geographical coverage required for adequate monitoring. Weather radars are directly sensitive to precipitation components, making them important instruments for observing rainfall. However, their use for accurate precipitation estimate with excellent geographical coverage is limited by existing gaps in radar networks and technological complexity. Satellite measurements offer the benefit of producing spatially homogeneous data across broad regions. The main objective of the present study is to investigate rainfall over Iraq using TRMM Rainfall data where the daily rainfall observations collected by ground stations at different locations in Iraq and compared to rainfall estimates obtained from TRMM data in order to calibrate the acquired rainfall data. The results showed that During the months of January, April and December over Iraq influenced by the Mediterranean low pressure systems where during winter, the prevailing westerly winds may bring moist air from the Mediterranean Sea, leading to increased chances of rainfall. Generally, highest rainfall pattern is observed in the south eastern boundary, also during April, the pattern shifts in around of Sulaymaniyah province, from the south-east border to the near north-east sections. additionally, the precipitation weather system in Iraq influenced by the Red Sea Convergence Zone as a meteorological feature which can contribute to winter rainfall over Iraq where the convergence of air masses from the Red Sea and the Mediterranean can lead to the lifting of moist air and subsequent precipitation. Finally, Daily rainfall measurements for all evaluated stations were compared with TRMM-estimated rainfall values, and the results were very acceptable. These results show that rainfall estimations using TRMM data may be valuable in various applications, including agriculture and water resources.

In the hydrological cycle, rainfall is one of the vital components for the environment. Rainfall data plays an important role in determining the amount of runoff that occurs. Accurate and complete rainfall data estimates are needed in managing water resources; which are still limited, especially for watersheds with less field rainfall measurement stations. This study aims to analyze the use of satellite-based rainfall data, especially GPM (Global Precipitation Measurement) as an alternative rainfall data for watersheds with minimal observational rainfall data. In this study, secondary data collection was carried out by downloading observational rainfall data via BMKG Online and GPM satellite-based rainfall data for the period of 2018-2023. Several statistical methods were used to assess satellite performance and the relationship between BMKG and GPM data in Kemuning Watershed. From the results of the analysis, it can be seen that the correlation value between BMKG and GPM shows sufficient correlation (r = 0.45) and as much as 93% of GPM satellite data can be explained by BMKG data factors. Thus, GPM satellite data can be used as an alternative rainfall data, especially for watersheds with limited measurement stations.

Climate change and rapid urbanization have increased the risk of urban flooding, making timely and accurate flood prediction crucial for disaster response. However, conventional physics-based models are limited in real-time applications due to their high computational costs. Recent advances in deep learning have enabled the development of efficient surrogate models that capture complex nonlinear relationships in hydrological processes. This study presents a deep learning-based surrogate model designed to efficiently reproduce the spatiotemporal evolution of urban pluvial flooding using data from physics-based models. For the Oncheon-cheon catchment in Busan, the spatiotemporal evolution of inundation at a 10 m spatial resolution was simulated using the physics-based model for various synthetic inundation scenarios to train the deep learning model based on a Convolutional Neural Network (CNN). The training dataset was constructed using synthetic rainfall scenarios based on probabilistic rainfall data, while the model was validated using both a synthetic flood event and a historical flood event from July 2020 with observed ground-based rainfall measurements. The model’s performance was evaluated using quantitative metrics, including the Hit Rate (HR), False Alarm Ratio (FAR), and Critical Success Index (CSI), by comparing results against both synthetic and real (historical) flood events. Validation results demonstrated high reproducibility, with a CSI of 0.79 and 0.73 for the synthetic and real experiments, respectively. In terms of computational efficiency, the deep learning model achieved a speedup 16.4 times the parallel version and 82.2 times the sequential version of the physics-based model, demonstrating its applicability for near real-time flood prediction. The findings of this study contribute to the advancement of urban flood prediction and early warning systems by offering a cost-effective, computationally efficient alternative to conventional physics-based flood modeling, enabling faster and more adaptive flood risk management.