The reliability of satellite precipitation estimates during tropical cyclone Shaheen

Abstract

Climate change has increased the frequency and intensity of extreme weather events worldwide, amplifying the global need for reliable weather prediction. Cyclones and flash floods pose serious threats to human life and infrastructure, with advance forecasts and flash flood guidance provided by numerical weather models for disaster mitigation. These models depend on data from satellites and/or ground instruments. In the greater part of the world, less industrialized countries rely widely on the free ubiquitous satellite data, amid limited availability of rain gauges and radars. Assessing the reliability of satellite data during an extreme event in diverse geographic regions is therefore vital. Here we present the assessment of satellite rainfall estimates during tropical cyclone Shaheen, which hit the Arabian Sea and the Arabian Peninsula, in particular, Oman, where flash flood prediction is highly dependent on the MWGHE satellite estimator. We show that satellite data overestimate gauge cumulative precipitation, with daily totals showing a consistent positive bias. Six-hourly analysis reveals fluctuations, with overestimation varying by intensity and timing. Comparisons with UAE radar observations further highlight these discrepancies, particularly in the spatial distribution and intensity of precipitation. Our study underscores the need for a full range of products and/or enhancements of current satellite estimators.

Introduction

Global climate change has caused unusual and severe weather conditions, including hotter, drier climates, increased flooding, and unprecedented heatwaves in many areas around the world¹. These changes contribute to more frequent and intense extreme weather events. Global warming, salinity, and rising sea levels are examples of these changes, with increasing evidence indicating that they are primarily caused by human activity. This emphasizes the urgent need to improve adaptation capabilities and mitigate the risks associated with climate change, including reliable data, models, and thus forecasts.

Tropical cyclones and flash floods are directly affected by climate change, causing alterations in their frequency and intensity¹ and an increase in extreme weather conditions, including severe flooding in various regions, which disrupts social, technical and environmental systems. In addition to immediate health consequences, such as injury, these events also pose a threat to vulnerable groups, including elderly individuals, children, pregnant women, and first responders². The impact of these events must be considered both in terms of their immediate effects as well as their indirect effects on individuals’ health in the years to come. Globally, extreme weather events such as droughts, epidemics, floods, natural-technological events, and earthquakes have historically caused significant deaths, although there has been a decline in deaths since the 1930s².

In recent decades, tropical cyclones have become less frequent but more intense in some regions worldwide, including the southern Indian Ocean, the north Australian Sea, and the northwest Pacific³. Although cyclones are rare in the Arabian Sea, which is part of the North Indian Ocean, the frequency and duration of these events have increased over time³. The frequency and intensity of heavy precipitation and flash floods also increased significantly, with new record levels registered in April 2024 in Oman and the United Arab Emirates, leading to more than a dozen death and other severe damage (see New York Times (17 April 2024), https://nyti.ms/3EiEjrR). Climate change, particularly global warming, has been linked to increased cyclone intensity and significance in the Arabian Sea and Oman Sea. The effects of climate change have been observed in factors such as mesoscale eddies (ME) and changes in salinity during the pre- and post-monsoon seasons. With the rise in cyclones, there is a risk to the economy, especially as it pertains to oil extraction, processing, and export in the Arabian Gulf, as well as maritime transportation and coastal infrastructure³. Gonu (2007), Phet (2010), and Shaheen (2021) were the last three intense tropical cyclones that hit the region. They caused approximately 6.07 billion dollars damage³. The history of tropical cyclones in the region and their tracks from 1980 to 2023 are shown in Fig. 1a, which we produced using data from a combination of satellite observations and reanalysis datasets provided by the International Best Track Archive for Climate Stewardship (IBTrACS)^4,5. Compared to the earlier periods of 1980–1994 and 1995–2009, a higher frequency of cyclones with longer duration occurred during 2010–2023. Indeed, the number of cyclones per year has roughly doubled during 2010–2023 compared to 1980–1994 as shown in Fig. 1b.

Fig. 1

(a) Tropical cyclone tracks in the North-West Indian Ocean during 1980–2023 with the Shaheen cyclone track highlighted. (b) Number of tropical cyclones per year during the same period. Data are from the IBTrACS database^6,7. The map was generated by the authors using Python version 3.9.16.

Tropical cyclone Shaheen

Tropical cyclone Shaheen was one of the unusual cyclones (Its track is highlighted in Fig. 1a ). It formed in the Arabian Sea on September 30, 2021, from the remains of the Gulab tropical cyclone in the Bay of Bengal⁸. It took a long path and crossed India before reaching Oman coastal area with wind speed of 102–116 km/h⁹ and wind gust up to 167 km/h⁸. The cyclone made landfall, on October 3, 2021, near $23.9^\circ$N, $57.3^\circ$E in northern Oman. It is the first such event in over 130 years to hit the southeast corner of the Arabian Peninsula and it was extended inland through Oman to near the borders with the United Arab Emirates (UAE) and Saudi Arabia. The cyclone dissipated on October 4, 2021 before entering Saudi Arabia’s borders⁸. The cyclone underwent four distinct stages as it progressed: initially intensifying upon reaching the coastal areas, transitioning into a weakening phase over land, and finally dissipating on October 4, 2021, before entering Saudi Arabia’s borders Fig. 2.

Fig. 2

Satellite imagery showing the progression of Cyclone Shaheen across Oman in four stages: from its intensification upon reaching the coastal areas to its dissipation inland near the borders of Saudi Arabia on October 4, 2021. GSMaP satellite data was used as the source of this image, which was edited using QGIS software. Adopted with permission from⁹.

Significant high waves were noted along Oman’s coast and the maximum amount of precipitation recorded by rain gauges between 2 to 4 of October 2021 was 465 mm, in Al Khaburah, Oman, as recorded by the stations of Oman’s Ministry of Agriculture, Fisheries and Water Resources⁸. The cyclone was associated with high-speed wind that reached a maximum speed of 130 km/h (80 mph), with a minimum mean sea level pressure recorded at 976 hPa in its core¹⁰. As the cyclone streaked along the coast, citizens suffered storm surges, torrential rain, and flash flooding. This event caused several landslides⁸. As a result of flash flooding, a large amount of sediment had been deposited in the coastal areas and valleys. Some areas of Iran experienced severe dust storms during the passage of the cyclone across the Oman Sea, which impacted people’s health⁸. Many infrastructures were destroyed and power was disrupted. About 14 people died and many others were lost¹¹. The sea surface temperature (SST) decreased by $5^\circ$ C¹² after the cyclone, mainly due to the significant influence of high wind stress and cyclonic eddy circulation. Economically, Shaheen caused approximately USD 100 million of damage in Oman¹³.

The extent to which tropical cyclone Shaheen impacted Oman and the surrounding countries, the extent of damage occurred, and the role of ocean-atmospheric interaction were reported by several studies^10,14,15. It is important to note that some of these studies relied on weather station measurements only, whereas others relied on satellites to estimate the precipitation levels like those found in¹¹. To date, no study has investigated the accuracy of the satellite estimates during the Shaheen cyclone.

Satellite, radar, and rain gauge precipitation estimates

Satellite-based precipitation estimation has been the most widely used data in addition to radar measurements and rain gauge observations for a number of reasons¹⁶. Although rain gauges measure rainfall directly, they can be inaccurate due to factors such as wind under-catch, limited distribution, or damage incurred during extreme weather. Wind-induced under-catch can reduce the accuracy of rain gauge measurements by 2% to 10%¹⁷ and the error in rain gauge readings can increase by approximately 1% for every 1.6 km/hr increase in wind speed¹⁸. For this reason, meteorological agencies set specific conditions for the placement and shape of rain gauges. However, even with these safeguards, extreme events such as Tropical Cyclone Shaheen, which reached the coast with wind speeds of 102–116 km/h⁹, can still affect the accuracy of rain gauge measurements. These are in addition to other typical challenges associated with the collection of accurate rainfall data from rain gauges, such as systemic errors of the instruments and occasional downtime caused by extreme weather phenomena. As a result of the gaps in data collection as well as the inherent limitations of the instruments, major impacts may result, especially if there are no other rain gauges in the affected area.

Radars, on the other hand, cannot measure rainfall directly but instead, they analyze the signals reflected back from raindrops or other atmospheric particles. A radar’s ability to send and receive those signals can be affected as well by beam blockages and obstacles. Inaccuracies in the data provided by radar systems can also be caused by gaps between radar installations. Nevertheless, the accuracy of radar readings is dependent upon factors such as the location and terrain surrounding the radar¹⁹. However, the advantage of radars lies in their ability to provide high temporal and spatial resolution.

Microwave-enhanced geostationary hydro-estimator (MWGHE)

Satellites can advantageously offer frequent and spatially continuous estimations of precipitation, although they too can suffer from issues such as indirect measurements of rainfall and resolution limitations of the satellite product. Globally, high-resolution geostationary satellites and polar orbital satellites are used to estimate precipitation by adjusting the output rain rate from the infrared imagery channel using various algorithms. Satellite-based estimates of rainfall rates were introduced many years ago. In the 1970s, rainfall rates were derived using enhanced imaging satellites with visible and infrared windows²⁰. Infrared and visible channels were used in many satellite products. Microwave channels are also used for estimating or adjusting the output rain rate from the infrared imagery channel^20,21. Some techniques are only capable of calculating convective precipitation. However, other methods can estimate both convective and stratiform precipitation. One example of such a method is the Hong algorithm²². In fact, a variety of satellite-based precipitation estimation products are available using different algorithms and with different coverage and spatio-temporal resolutions. Each product offers its own advantages and limitations and can be used to monitor different aspects of precipitation. A list of some of the main operational satellite estimators with varying platforms available is provided by Sadeghi et al.²³.

Microwave satellite sensors, deployed on polar orbital satellites by the National Oceanic and Atmospheric Administration (NOAA) and the European Organisation for the Exploitation of Meteorological Satellites (EUMETSAT), are highly effective tools for estimating precipitation. These sensors operate at various frequencies between 23 GHz and 190 GHz, allowing them to detect precipitation particles directly from clouds. By utilizing multi-channel instruments, microwave sensors can capture both low frequencies absorbed by raindrops and high frequencies scattered by hydrometeors like raindrops, hail, and snowflakes²⁴.

Despite their high performance, microwave sensors have certain limitations. Since the polar orbital satellites follow an orbital sequence, they cannot continuously detect clouds at a single location. Additionally, incomplete beam-filling errors introduce uncertainty in detecting convective clouds²⁴.

With the Climate Prediction Center’s morphing method (CMORPH) infrared (IR)-based cloud top temperatures are used to derive propagation vectors for cloud tops that are interpolated with microwave-based precipitation estimates to produce 8 km resolution precipitation estimates every half-hour with an approximate 18-hour latency. CMORPH’s MW-based satellite precipitation estimates are used to enhance Global Hydro Estimator’s (GHE) IR-based accumulated precipitation estimates. The adjusted product is called MWGHE, provided by the Hydrologic Research Center (HRC). As a result of this adjustment, biases associated with the MW data are addressed and estimation are improved²⁵. MWGHE tackles challenges in estimating different types of precipitation, like light rainfall, solid precipitation, and orographic rainfall, through advanced algorithms and processing techniques. It also addresses the characterization of errors associated with single and multiple sensors.

Using a grided scale, MWGHE provides 1-hour, 3-hour, 6-hour, and 24-hour accumulations of satellite-based rainfall estimates. The updates are made every hour, with a latency of approximately 45 minutes.

How reliable are the satellite precipitation estimates?

Satellite precipitation estimation plays an important role in predicting and mitigating the impact of extreme weather events, especially in flood-prone areas. Recent studies have highlighted the importance of accurately predicting precipitation, especially in some vulnerable regions²⁶. For example, sea level rise puts coastal communities at serious risk because it can cause flooding and worsen the effects of heavy rainstorms, especially in areas with low-lying topography. Tropical Cyclone Shaheen is an example of such a case¹¹. For effective disaster management and resource allocation, it is essential to understand the reliability of satellite precipitation estimates compared to ground-based observations.

Previous studies focused on the evaluation of various satellite products, such as GPM-IMERG, GSMaP, and PERSIANN-CCS, across a variety of climatic conditions and geographies. For example, Chen et al.²⁶ and Pradhan et al.²⁷ demonstrated the importance of accurately capturing precipitation intensity, particularly where discrepancies between satellite estimates and ground observations persist. Other studies have found that the IMERG data tends to underestimate rainfall, particularly during high-intensity, short-duration events²⁸. Similarly, Li et al.²⁹ found that satellite-based products tend to overestimate light precipitation and underestimate heavy precipitation, suggesting that further refinement is necessary. The satellite products had difficulties estimating precipitation in regions with sparse rain gauge data and in highest topography³⁰. This finding underscores the challenges of the satellite estimate in our area due to such a limitation.

While the Microwave MWGHE multi-spectral product offers a global perspective on precipitation estimation, there are still challenges associated with calibrating satellite data to account for regional and seasonal biases and variability. Other findings from the analysis of the MWGHE product by Georgakakos et al.³¹ and Konstantine et al.³² showed notable biases in estimating regional precipitation, encompassing both underestimation and overestimation across diverse seasons and geographical regions.

Multiple factors can affect the satellite estimates of the precipitation. Some satellite estimates use approaches sensitive to deep convection processes and misestimate another type of precipitation³⁰. In addition, deserts and snow areas can reflect signals to the satellite, which are read as precipitation signatures³³. The lack of ground-based data used for bias calibration is another cause for the misestimation of precipitation³⁰. This overestimation, as observed during Tropical Cyclone Shaheen, is likely influenced by errors in the interpretation of cloud structures and by limitations in the distinction of convective from stratiform precipitation.

The MWGHE satellite products overstated the high intensity precipitation in Tropical Cyclone Shaheen. This was mainly due to the difficulty in distinguishing between types of precipitation. In addition, the incorrect interpretations of cloud structure by satellite algorithms can be another reason why MWGHE is overestimated. Non-precipitating cloud anvils can be mistakenly detected as heavy rainfall by satellite retrievals, which often rely on cloud-top brightness temperatures and microwave signals. At the same time, the cyclone’s movement inland led to underestimations due to reduced convection over complex terrain, the influence of dry air, and dust aerosols, which suppressed rainfall. The limited availability of ground-based data for calibrating the satellite estimates made the issue worse. In addition, the region’s topography and the dominance of warm rain processes, which don’t produce strong ice-phase signals detectable by satellites, added to the difficulty in accurately estimating precipitation during this extreme event.

In this work we evaluate the accuracy and reliability of the precipitation estimation using MWGHE-corrected data products during the Shaheen Tropical Cyclone by comparing them with (a) rain gauge measurements in Oman, where the cyclone made landfall, and (b) Radar products in the UAE to validate their accuracy and improve satellite products reliability for Oman. The performance evaluation specifically focuses on the period from October 2–4, 2021, that is the day of the landfall, the day before, and the day after.

The paper is structured into five main sections following the introduction. In the Results, we present the findings of the spatial and statistical analyses. In the Discussion we elaborate on these findings and provide insight and interpretation, as well as the implications of the results and some recommendations. In Methods, we provide details of the study domain and spell out the methodology applied to the data, focusing on the case of tropical cyclone Shaheen and its reference data. In $Data\; Availability$, we provide information on the availability of the data used in the study, while the $Code \; Availability$ accounts for the availability of tools used in the data analysis.

Results

Rain gauge comparison

In the following analysis, precipitation amounts below 0.1 mm were excluded to minimize the influence of sensor noise, and values above 5 mm were emphasized in the Oman region to highlight high precipitation areas. Positive values in spatial analysis indicate an overestimation of MWGHE relative to surface measurements, whereas negative values represent an underestimation. This convention is maintained consistently across all spatial analysis figures.

Figure 3 shows daily precipitation from the rain gauge network (Fig. 3a–c) and the interpolated (polygon) approach (Fig. 3d–f) over Oman on October 2, 3, and 4, 2021, respectively. (Fig. 3g–i) is the daily satellite base precipitation estimate over the same period for direct comparison of the three products. The point-based maps (Fig. 3a–c) highlight the sparse distribution of gauges across the region. This leaves large areas without direct observations. In contrast, polygon maps (Fig. 3d–f) produce a continuous surface that assigns precipitation values to every location, even those without direct measurements.

Fig. 3

Spatial daily precipitation of rain gauges point distribution (a–c), rain gauges interpolated to basin (d–f), and MWGHE satellite data (g–i) for October 2–4, 2021, over Oman. The map was generated by the authors using Python version 3.11.10.

However, interpolation can reduce the accuracy of the data, especially in areas far from any gauge stations. Table 4 illustrates how maximum rainfall estimates may differ substantially between actual gauge data and interpolated values. For example, on October 2, the point-based gauges recorded a maximum of 4.8 mm, whereas the interpolation yielded 11.0 mm. Similar trends appear on October 3 and 4, underscoring that the interpolation method can both overestimate and underestimate localized extremes when gauge coverage is limited. A larger data set is used for cumulative precipitation compared to the daily and 6-hour data sets. Because of this, we performed a direct point-to-point comparison between gauge observations and satellite estimates. This approach selects the nearest satellite grid point to each gauge station, allowing a straightforward assessment of agreement. However, for daily and 6-hour analysis, we use interpolated data sets at the basin scale for more reliable spatial consistency. In addition, both the box plot and the scatter plot in cumulative analysis emphasized values of 5 mm and above. The cumulative difference between the precipitation of the gauge and the satellite is illustrated in Fig. 4a, where the individual gauge stations are plotted with their respective differences. The geographic distribution highlights regions with significant overestimation, particularly in coastal and northern areas. Some inland stations show better agreement. The gridded difference map Fig. 4b extends this comparison by interpolating point-based differences in the study region. The areas with larger overestimation align with regions experiencing higher cumulative precipitation, emphasizing the satellite product’s tendency to overestimate in these regions (Fig. 4).

Fig. 4

Comparison of cumulative differences between rain gauge measurements and MWGHE satellite estimates over Oman for the full period: (a) point-to-point and (b) interpolated. Positive values indicate MWGHE overestimation; negative values indicate underestimation. The map was generated by the authors using Python version 3.11.10.

The box plot Fig. 5a summarizes the cumulative difference between the gauge and the MWGHE estimates at all locations. The median difference is approximately 10.79 mm, indicating a tendency of the satellite to overestimate cumulative precipitation. The spread of the data suggests variability in satellite performance across different stations, with a few significant outliers highlighting locations where satellite data have largely deviated from gauge observations.

The scatter plot Fig. 5b further quantifies the performance of MWGHE precipitation estimates compared to gauge data. The moderate correlation (0.63) indicates a reasonable relationship between satellite and gauge measurements, though discrepancies in individual magnitudes persist. The overall positive bias (31.49 mm) suggests an overestimation of cumulative precipitation by the satellite product. The root mean square error (RMSE) of 73.99 mm further emphasizes that, while satellite products capture general precipitation patterns, individual magnitudes may significantly deviate from ground observations.

Fig. 5

Cumulative precipitation: (a) Box plot representation and (b) Scatter plot visualization. The map was generated by the authors using Python version 3.11.10.

The spatial daily precipitation from the rain gauges and satellite data for October 2–4, 2021 is illustrated in Fig. 3. Their differences are given in Fig. 6, where they show a significant overestimation by the satellite in most areas. On October 3, the day of landfall Fig. 3b, the MWGHE satellite estimates missed some high precipitation over northern Oman, with gauge-measured values exceeding 40–80 mm in some regions. However, the average difference map Fig. 3d shows a general overestimation trend, with maximum average errors exceeding 80 mm in certain localized areas and exceeding 40 mm in broader regions.

Fig. 6

Daily difference maps (a–c) and average differences between rain gauges and MWGHE satellite estimates (d) over Oman for October 2–4, 2021. Positive values indicate MWGHE overestimation; negative values indicate underestimation. The map was generated by the authors using Python version 3.11.10.

The scatter plots shown in Fig. 7 further emphasize the discrepancies, with a notable spread in the high intensity precipitation values. For October 2 and 4, Fig. 7a and c, the data show significant scatter and low correlation, reflecting substantial overestimation by MWGHE. For October 3, Fig. 7b illustrates an improved correlation (0.80 Table 1), as MWGHE estimates better aligned with gauge measurements during landfall. The subplot Fig. 7d highlights the domain sum plot, confirming MWGHE’s consistent overestimation of total precipitation by MWGHE, particularly on October 3. Furthermore, it is important to note that the thresholds observed on the different days of the cyclone result from both the natural distribution of the intensity of rainfall during the different phases of the cyclone and the effects of interpolation. October 3 corresponds to the peak precipitation phase, while lower values on October 2 and 4 reflect weaker rainfall. Although both the gauge and the MWGHE precipitation estimates were interpolated to the same polygonal basins, their underlying spatial characteristics differ substantially. The gauge network is sparse and, during spatially localized rainfall events, interpolation can smooth spatial variability, potentially misrepresenting localized extremes within a basin. In contrast, MWGHE originates from a much higher spatial resolution and retains finer-scale precipitation features even after basin-scale averaging. These differences explain the observed patterns in the comparison, including the apparent cutoff at about 10 mm in the gauge data and the vertical alignment of values near 0 mm in several panels.

Fig. 7

Scatter plots comparing daily precipitation from rain gauges and MWGHE satellite estimates over Oman (a–c). The diagonal line representing the perfect agreement. Subplot (d) shows the total precipitation sum over the full domain. The map was generated by the authors using Python version 3.11.10.

The box plots shown in Fig. 8 provide additional insight into the variability and biases in the daily precipitation estimates from the rain gauges compared to MWGHE. The median differences ranged from 1.32 mm on October 3, Fig. 8b, to 2.74 mm on October 4, Fig. 8c, with the average median difference for all days being 3.07 mm, Fig. 8d. On October 3, the day of landfall, the interquartile range remained relatively narrow, suggesting less variability in the bulk of the data. However, numerous extreme outliers were present, with values reaching close to 200 mm, confirming the substantial overestimation by MWGHE.

Fig. 8

Box-plots showing daily precipitation differences between rain gauges and MWGHE satellite data over Oman (a–c). Subplot (d) represents the temporal average differences over the domain. The map was generated by the authors using Python version 3.11.10.

The daily error metrics are summarized in Table 1, with MAE values increasing from 8.89 mm on October 2 to 15.27 mm on October 3, reflecting the peak discrepancies during landfall. RMSE values reached 33.13 mm on October 3, with biases of 11.55 mm, indicating systematic overestimation by MWGHE satellite during intense rainfall (Fig. 4).

Table 1 Error metrics for daily rain gauge and MWGHE satellite comparisons in Oman.

The kernel density estimates indicate differences between satellite precipitation estimates and rain gauge measurements for Oman over the three study days. To ensure a more effective comparison, we applied a Gaussian kernel density estimation (KDE) that is normalized by the number of data points, with logarithmic transformed precipitation data, as described in the methodology. This transformation helps smooth the distribution while maintaining symmetry across datasets.

On October 2, 2021 Fig. 9a, satellite data underestimated precipitation in a very low amount compared to rain gauges, with gauge observations concentrated below 20 mm and satellite densities showing lower values at higher intensities. On October 3 Fig. 9b, the day of Cyclone Shaheen’s landfall, Satellite estimates generally align with gauge data for light precipitation (below 100 mm) but show notable overestimation for moderate to high precipitation intensities (above 150 mm). Satellite estimates precipitation reaches up to 300 mm, while gauge observations recorded less than 150 mm. Despite this overestimation, density distributions indicate improved alignment for lower precipitation values during landfall. By October 4 Fig. 9c, satellite estimates continued to exceed gauge data, recording over 120 mm, whereas rain gauges measured less than 27 mm. Across all three days, satellite data consistently overestimated higher precipitation intensities.

Fig. 9

Kernel density representation of the Satellite estimates and rain gauge measurements of the 24-hour accumulated precipitation for tropical cyclone Shaheen during October 2–4, 2021 over Oman. The map was generated by the authors using Python version 3.9.16.

The temporal and intensity differences are highlighted in the 6 hour analysis, as shown in Fig. 10 to provide a finer temporal resolution, capturing variations in precipitation over shorter time scales. The error metrics shown in Table 2 indicate significant variability in deviation, with values ranging from 32.21 mm, indicating MWGHE overestimation, to − 8.25 mm in a single interval, which indicates MWGHE underestimation. Specifically, on October 3 at 0600 UTC, gauge measurements exceeded the satellite value by an average of 8.25 mm and RMSE of 18.50 mm. The correlation values are moderate during landfall, reaching 0.53, before declining as the cyclone has dissipated (Fig. 10h–l).

Fig. 10

6-hourly difference maps between rain gauges and MWGHE satellite precipitation estimates for Oman during October 2–4, 2021. Positive values indicate MWGHE overestimation; negative values indicate underestimation. The map was generated by the authors using Python version 3.11.10.

The scatter plots shown in Fig. 11 further emphasize these findings, showing significant deviations in high-intensity precipitation. For October 3, Fig. 11b illustrates a notable improvement in correlation as MWGHE estimates are better aligned with the gauge data. In contrast, Fig. 11a and c for October 2 and October 4, respectively, demonstrate weaker correlations.

Fig. 11

Scatter plots comparing 6-hourly precipitation estimates from rain gauges and MWGHE satellite data over Oman. They represent individual 6-hour intervals, highlighting temporal variability and correlations. The map was generated by the authors using Python version 3.11.10.

The boxplots in Fig. 12 illustrate how the variability in the precipitation estimates of MWGHE changed over time, with interquartile ranges and whiskers highlighting instances of overestimation. On October 3, while high precipitation variability is evident, Fig. 12g and h reveal that the interquartile range is not the widest. However, the presence of substantial outliers suggest a notable overestimation. These discrepancies likely stem from the cyclone’s intensity, the complex topography of the region, and the sparse distribution of rain gauges, which together contribute to the challenges of accurately capturing extreme precipitation events.

Fig. 12

Box-plots showing 6-hourly precipitation differences between rain gauges and MWGHE satellite data over Oman. The map was generated by the authors using Python version 3.11.10.

Table 2 Error metrics for 6-hourly rain gauge and MWGHE satellite comparisons in Oman.

Table 3 captures the variability in precipitation, wind speeds, and associated uncertainties between different stations during the landfall of Tropical Cyclone Shaheen. The area near where the cyclone struck recorded the highest rainfall at 222.2 mm, accompanied by the highest maximum wind gust of 167.1 km/h. Its wind-induced uncertainty was 12%, reflecting the impact of intense winds on gauge accuracy. In contrast, an area farther away from the inland experienced gusts of 94.1 km/h and exhibited an uncertainty of 8%, emphasizing the role of wind speeds in influencing rainfall measurements.

Table 3 also highlights stations with relatively low uncertainty, such as Muscat city (2%), where the average wind speeds remained moderate at 14.8 km/h. This variability underscores the interplay between meteorological conditions and measurement accuracy. Notably, the significant rainfall reported in stations like Al Amrat (142.4 mm) and Bawshar (81.4 mm) aligns with the cyclone’s trajectory, showcasing localized rainfall intensity influenced by topography and storm dynamics.

These observations provide a critical context for evaluating MWGHE satellite performance, as areas with high wind-induced uncertainties may lead to greater discrepancies between ground and satellite measurements.

Table 3 Combined Data for Rainfall, Average Wind, Maximum Wind Gust, and Estimated Wind Induced Uncertainty (%) on 3rd of October 2021.

Table 4 shows the overestimation tendency of the MWGHE satellite product during Tropical Cyclone Shaheen. On October 3, the day of landfall, MWGHE estimated a maximum of 300.0 mm, exceeding both the actual rain gauge measurement (222.2 mm) and the interpolated polygon value (131.0 mm), This discrepancy demonstrates that the interpolation process, by averaging data over an area, can reduce the precision of point-specific measurements. The significant overestimation of the MWGHE estimate is highlighted in October 2 when the cyclone is still a long way from the coastal area. Similarly, on October 4, as the cyclone dissipated, MWGHE recorded a maximum of 125.8 mm, overestimating the actual gauge maximum of 72.0 mm by nearly 75%.

Table 4 Comparison of the Rain Gauge and Satellite Maximum Precipitation.

Radar comparison

To validate observations from the rain gauge analysis, composite radar data from the UAE region are used to compare precipitation estimates with higher spatial resolution, as no radar data are available for Oman. Here, positive values indicate overestimation by MWGHE satellite compared to the composite radar, while negative values indicate underestimation.

Figure 13a illustrates the spatial differences between cumulative radar and MWGHE precipitation estimates. The map highlights localized variations, with significant overestimation by MWGHE in certain areas, particularly near the northeastern coastline. The scatter plot in Fig. 13b shows a weak relationship between MWGHE and radar-based precipitation estimates, with many points deviating substantially from the 1:1 line. The box plot in Fig. 13c indicates a median difference of 0.78 mm, while outliers reveal instances of both overestimation and underestimation. The computed bias of 3.55 mm suggests a tendency for MWGHE to slightly overestimate precipitation. However, the low correlation (0.04) and the RMSE of 8.98 mm highlight considerable discrepancies, indicating substantial variability in agreement between the two datasets.

Fig. 13

Cumulative radar precipitation analysis: (a) Difference map, (a) Scatter plot, and (c) Box plot. The map was generated by the authors using Python version 3.11.10. https://www.python.org/downloads/release/python-3110.

The spatial daily precipitation for October 2 to 4, 2021, is shown in Fig. 14. On October 2, Fig. 14a, MWGHE underestimated radar-derived precipitation, recording a maximum of only 2.4 mm, compared to the radar’s 29.0 mm, Table 5. This significant underestimate reflects MWGHE’s inability to detect light to moderate precipitation accurately. By October 3, Fig. 14b, MWGHE estimates increased, showing better alignment with radar measurements. However, MWGHE significantly overestimated precipitation on October 4, with a recorded maximum of 49.9 mm against the radar’s 15.2 mm, Table 5. These overestimations can be attributed to MWGHE misinterpreting cloud structures and low-intensity precipitation as higher rainfall.

Fig. 14

Daily precipitation estimates from composite radar (a–c) and MWGHE satellite products (d–f) over the UAE for October 2–4, 2021. The map was generated by the authors using Python version 3.11.10.

The daily differences and average difference maps are presented in Fig. 15a–d. On October 2, Fig. 15a, the radar captured localized precipitation averaging within 20 mm, which was underestimated by MWGHE. However, MWGHE consistently overestimated inland radar precipitation during October 3 and 4, Fig. 15b and c, with discrepancies ranging from 1.86 mm and 3.60 mm, respectively “Table 6”. The average difference map Fig. 15d confirms this overestimation trend, particularly for moderate precipitation intensities. The RMSE peaked at 9.05 mm on October 4, and the correlation remained weak across all days, ranging from 0.01 to 0.15.

Fig. 15

Daily difference maps (a–c) and average differences (d) between radar and MWGHE satellite precipitation estimates over the UAE for October 2–4, 2021. Positive values indicate MWGHE overestimation and negative values indicate underestimation. The map was generated by the authors using Python version 3.11.10.

The scatter plots shown in Fig. 16 highlight the poor agreement between radar and MWGHE. For October 2, Fig. 16a, the scatter reflects MWGHE’s underestimation of radar measurements, with very low intensities. On October 3, Fig. 16b, the scatter demonstrated slightly improved alignment, although MWGHE frequently exceeded radar measurements. By October 4, Fig. 16c, the scatter became more dispersed, indicating a continued overestimation by MWGHE.

Fig. 16

Scatter plots comparing daily precipitation estimates from radar and MWGHE satellite data (a–c). The diagonal line representing the perfect agreement. Subplot (d) shows the total precipitation sum over the full domain. The map was generated by the authors using Python version 3.11.10.

The boxplots shown in Fig. 17a–c illustrate the variability in daily precipitation differences. On October 2, Fig. 17a, the interquartile range was relatively narrow, with MWGHE underestimating most radar values. On October 3, Fig. 17b, the interquartile range widened slightly, reflecting increased variability during moderate rainfall. Figure 17c (October 4) shows a shift towards positive biases, with whiskers extending beyond 10 mm and notable outliers observed across all days.

Fig. 17

Boxplots comparing daily precipitation estimates from radar and MWGHE satellite data (a–c). Subplot (d) represents the temporal average differences over the domain. The map was generated by the authors using Python version 3.11.10.

The error metrics in Table 6 provide a quantitative summary of MWGHE’s performance against radar data. The MAE varies between − 2.14 mm on October 2 and 4.86 mm on October 4. Furthermore, Table 5 compares the maximum precipitation values, highlighting the overestimation by MWGHE when the cyclone approached the UAE region. For instance, MWGHE recorded 49.9 mm on October 4, overestimating the radar’s maximum precipitation (15.2 mm) by more than 228%.

Comparison of satellite and radar precipitation estimates in the UAE highlights further discrepancies. On October 2, 2021, Fig. 18a, satellite estimates were lower than radar measurements, with radar values concentrated below 25 mm. On October 3, Fig. 18b, during Cyclone Shaheen’s landfall, satellite data overestimated radar precipitation, reaching up to 25 mm compared to radar values below 12 mm. However, both datasets aligned better for lower precipitation intensities, reflecting widespread rainfall patterns during the cyclone. On October 4, Fig. 18c, as the cyclone dissipated, satellite estimates again overestimated precipitation, exceeding 50 mm, while radar observations remained below 20 mm. These differences emphasize the limitations of satellite retrieval algorithms, particularly for medium-to-high precipitation intensities in the radar detection area.

Fig. 18

Kernel density representation of the Satellite estimates and radar products of the 24-hour accumulated precipitation for tropical cyclone Shaheen during October 2–4, 2021 over UAE. The map was generated by the authors using Python version 3.9.16.

The underestimation on October 2 is likely due to MWGHE’s limited ability to detect light to moderate precipitation, as its algorithm focuses more on convective rainfall signals. In contrast, the overestimation on October 4 likely stems from misinterpreting cloud structures or residual moisture as heavier precipitation.

Table 5 Comparison of Radar and Satellite Maximum Precipitation in the UAE Area.

Table 6 Error metrics for daily radar and MWGHE satellite comparisons in UAE.

Sub-daily (6-hourly) radar analyses were not conducted in this study due to data limitations; however, the daily metrics and plots provide comprehensive validation of the MWGHE satellite estimates using radar observations.

Discussion

This study highlights the consistent overestimation of precipitation by MWGHE satellite-based estimates during Tropical Cyclone Shaheen, as demonstrated through comprehensive comparisons with rain gauge and radar observations.

The side-by-side comparison in Fig. 3 reveals how limited gauge coverage affects precipitation estimates. Because most gauges are clustered in a few locations, large portions of the region lack direct measurements. Interpolation fills these data gaps, allowing for broader spatial assessments and facilitating comparisons with satellite. However, this approach can smooth out localized rainfall peaks, especially in areas where data are lacking. The differences in Table 4 reflect the challenges of relying on interpolated data, particularly when gauges are sparse. Still, using the polygon-based method remains essential for capturing region-wide patterns and comparing with grid to grid datasets.

The validation of the rain gauge underscores significant biases in the MWGHE estimates, particularly during high intensity rainfall. The cumulative precipitation analysis provides crucial insights into the performance of satellite estimates compared to gauge observations. The findings indicate that satellite-based precipitation products tend to overestimate cumulative rainfall, particularly in coastal and northern regions. This systematic bias highlights the importance of applying bias correction techniques or blending satellite data with ground observations for more reliable precipitation monitoring.

Spatial difference maps Fig. 3 reveal localized overestimations, with maximum average errors exceeding 80 mm in some areas. Statistical metrics, including biases and RMSE values (Table 1), further emphasize the variability in MWGHE performance, with peak discrepancies observed on October 3, 2021, when the cyclone made landfall. Boxplots Fig. 8 and scatter plots Fig. 7 illustrate systematic biases in satellite-derived precipitation, including substantial outliers during intense rainfall. Notably, the finer temporal resolution of the gauge versus MWGHE analysis shows improved accuracy during the landfall phase (e.g., Fig. 10f), likely due to prominent deep convection, which satellite retrieval algorithms are better equipped to detect. However, as the cyclone weakened and transitioned inland, the complexity of the terrain and reduced convective activity contributed to the declining correlation and increased variability in satellite estimates (Fig. 11).

Radar-based comparisons corroborate these findings, offering higher spatial resolution insights into MWGHE performance over the UAE. While radar data captured localized precipitation effectively, MWGHE consistently overestimated moderate-to-high precipitation intensities (Fig. 15). For example, on October 4, MWGHE estimated a maximum of 49.9 mm, whereas radar observations recorded 15.2 mm (Table 5). Scatter plots Fig. 16 highlight poor agreement between radar and MWGHE estimates. Boxplots Fig. 17 reveal substantial variability, with interquartile ranges and whiskers reflecting MWGHE’s overestimation tendencies, and notable outliers exceeding 20 mm. Error metrics in Table 6 confirm weak correlations and significant biases, with RMSE peaking at 9.05 mm on October 4.

Excluding trace precipitation ($<0.1$ mm) and emphasizing spatial analysis values above 5 mm provided a focused analysis of meaningful precipitation during the event, revealing key patterns of overestimation or underestimation by MWGHE. For instance, on October 3, MWGHE estimated a maximum of 300 mm compared to the rain gauge maximum of 222 mm and radar maximum of 29 mm compared to 2.4 mm of the satellite precipitation (Tables 2 and 5). The temporal variability captured in 6-hourly metrics Table 2 and difference maps Fig. 10 further demonstrates the evolving biases in MWGHE performance during the cyclone’s progression.

Satellite precipitation estimation remains critical in regions like Oman and the UAE, where low-lying topography and coastal proximity amplify the risks of flooding during extreme weather events. However, this study reveals persistent challenges in accurately capturing precipitation intensity and variability, particularly under extreme condition. This finding align with previous studies (e.g., Georgakakos et al., 2022³¹; Konstantine et al., 2019³²), which reported similar overestimation patterns and emphasized improved calibration and bias correction methods. Addressing these limitations is essential to enhance the reliability of satellite precipitation products in operational early warning systems.

Many factors that highlight discrepancies in precipitation estimates during the tropical cyclone are influenced by more than just satellite limitations. Satellites detect precipitation in the cloud base, while ground-based instruments only measure what reaches the surface or near the surface after evaporation and other processes^34,35. The precision of satellite gauge comparisons also has a substantial impact on localized precipitation patterns, as heavy rainfall is frequently confined to small bands that may not correspond to satellite sampling grids. Moreover, high winds common in cyclones can lead to undercatch in gauges, which might record less rain than actually falls³⁵. The temporal resolution of the satellite with comparison with ground base measurement can also lead to mismatches.

The practical implications of these results are significant. The overestimation trends observed in MWGHE estimates raise important questions about the potential impact on disaster preparedness and resource allocation. For instance, discrepancies between satellite and ground observations may necessitate more conservative evacuation plans or adjustments to risk management policies. In addition, using satellite-derived cumulative rainfall estimates enhances accuracy and trust in disaster readiness by providing consistent, large-scale precipitation data that supports timely and informed decision-making. Collaborations between scientists, policymakers, and operational agencies are crucial to refining satellite algorithms and integrating reliable precipitation estimates into disaster management frameworks. Moving forward, hybrid approaches that combine satellite data with ground-based observations are essential for improving accuracy and reliability. Enhancing rain gauge networks, particularly in underrepresented regions of Oman, and expanding radar coverage are critical steps. Citizen science initiatives, such as deploying automated rain gauges to local communities, could complement these efforts. By leveraging a combination of satellite, radar, and gauge data, national meteorological agencies like DGMET can develop robust and reliable systems for flood forecasting and early warning. It is noteworthy to point out that this is the first study to our knowledge to evaluate this satellite product in our region within the context of an extreme event.

While the results offer valuable insights, the analysis is limited to a single event. Future research should extend this methodology to multiple events, including the recent flash floods of April 2024, to validate and generalize these findings. Such efforts will enhance the accuracy and trustworthiness of satellite precipitation estimates, ensuring their effective integration into disaster preparedness and response systems. Ultimately, this work emphasizes the importance of continuous improvements in satellite estimation techniques and collaborative efforts to address the growing challenges posed by extreme weather in vulnerable regions.

Methods

Tropical Cyclone Shaheen was chosen for this analysis due to its significant impact on the Arabian Peninsula and its illustration of the challenging conditions under which satellite precipitation estimates must operate. Shaheen’s intense rainfall and dynamic meteorological characteristics provide a pertinent case for assessing the reliability of MWGHE satellite precipitation estimates with the available observational data in the area.

Study domain

The study covers two domains in the southeast of the Arabian Peninsula where the tropical cyclone Shaheen made its landfall and had a measurable impact: Oman and the UAE. Oman area lies between 10 and 25 degrees latitude and 53–60 degrees longitude. In the northern part of the country, where cyclone Shaheen struck the coast, the topography ranges from flat coastal areas and valleys to mountainous areas. From the Oman’s national Directorate General of Meteorology DGMET records, the annual precipitation accumulation ranges from about 5 mm to about 300 mm³⁶. In contrast, the average annual precipitation in most areas did not exceed 76.44 mm in 2022³⁷. The difference arises from the fact that most of the rain falls during a particular time frame and has a significant impact on the regions. Coastal areas suffer from floods during tropical cyclone seasons and similar extreme events. In areas lacking rain gauges and radars, full coverage of precipitation data is crucial. Figure 19a shows the distribution of rain gauges in Oman. As shown, the densest distribution is found in the north, where tropical cyclone Shaheen made landfall.

Fig. 19

Coverage of Oman rain gauges and UAE radar system. The map was generated by the authors using MATLAB version R2022a. MATLAB is available at: https://www.mathworks.com/products/matlab.html.

We also utilized the UAE radars as a reference for satellite validation, specifically in the UAE region, where the radar coverage is comprehensive. Most of the UAE land is flat with some mountains. A small area is directly affected by tropical cyclones. However, some advected clouds produce a good amount of precipitation that can be used for satellite validation purposes. Figure 19b shows the locations of the UAE radars.

Methodology

In October 2021, post-monsoon tropical cyclone Shaheen triggered a flash flood along the coast of Oman sea. The impact of the cyclone is determined through a variety of precipitation measurements and estimation methods. Rain gauges in Oman and radars in the UAE are the two ground-based references for data collection in this study, which we used to assess the accuracy and reliability of the satellite precipitation estimates.

Oman rain gauges

Rain gauges are the most trusted rainfall measurement equipment. They measure the amount of rain in a given volume, in millimeters per hour (mm/h). Rain gauges are extremely accurate and provide a reliable reading of the amount of rainfall. They are also relatively affordable and easy to install and maintain, making them a popular choice for many different applications.

Rainfall measurement techniques encompass two primary methods: the tipping bucket and weight-based systems. Oman rain gauges, managed by the National Directorate General of Meteorology (DGMET), are installed in the DGMET stations throughout the country, which uses rain gauge devices called the Vaisala Rain Gauge RG13 Tipping Bucket³⁸. The tipping bucket sensor operates through an electromagnetic switch opening and closing at 0.2 mm intervals. To ensure accuracy, the maintenance department calibrates all rain gauges twice a year, with additional checks if necessary, by comparing and inspecting their readings to manually verify the precision of the 0.2 mm measurements. According to the manufacturer’s specifications, the uncertainty in the precipitation measurement is estimated to be 2% for each 25.4 mm/h³⁹. There are about 75 rain gauge stations distributed throughout Oman as shown in Fig. 19a. A 24-hour accumulation of precipitation data was collected during October 2–4, 2021. Some of the data is available online through the Ogimet ranking weather website (www.ogimet.com).

The data collected from the rain gauge stations were extrapolated into polygons based on the characteristics of the area. Averaging within these polygons takes place during the extrapolation process. However, it is important to note that this averaging can reduce the accuracy of the data. For comparison, the maximum rain gauge measurements per day over the Oman area during the period of Tropical Cyclone Shaheen (October 2–4, 2021) were $4.800 \pm 0.004$ mm, $222.2 \pm 0.2$ mm, and $72.0 \pm 0.1$ mm, respectively. Meanwhile, in the polygons, the maximum measurements were $11.00 \pm 0.01$ mm, $131.0\pm 0.1$ mm, and $26.00 \pm 0.02$ mm, respectively.

Rain gauges, especially in the Arabian region, where deserts cover large areas, have the disadvantage of being spatially limited. As a result, satellite products are considered the most reliable instrument to estimate precipitation rates in Oman due to their more comprehensive coverage.

The United Arab Emirates (UAE) radar

The UAE’s National Center of Meteorology (NCM) operates a weather radar system that operates in the C-band. The radar type/manufacturer is Enterprise Electronics Corporation (EEC) C-band radar. The geographical locations of the seven radars that comprise the system are shown in Fig. 19b. Each radar covers a range of up to 200 km. Radar coverage is divided into smaller units called range gates, each measuring 100 meters in diameter. The radars are capable of scanning different parts of the atmosphere by varying their elevation angle from 0.5 degrees to 32.4 degrees⁴⁰. Further, the radars emit signals with a beam width of 1 degree and perform volume scans every 6 minutes in order to gather data.

Composing rainfall data from the radars involves a number of steps. Radar data is initially de-cluttered and noise-filtered using fault algorithms and correction factors. To further reduce clutter, a fuzzy logic classifier uses features such as radial velocity, differential reflectivity texture, and correlation coefficient. Additionally, the attenuation correction adjusts for signal loss due to atmospheric effects. Following that, the highest reflectivity value technique is used to combine data from several radar overlaps to create the merged plan position indicator. The radar receives regular maintenance and yearly calibrations to keep accuracy within a ±1 dB error tolerance, which is about 0.03 mm/h⁴⁰. The radar system is designed to target mixed-phase cloud processes, which are common in the UAE. They apply a $Z-R$ relationship using a formula provided by the manufacturer, wherein rainfall $R$ and reflectivity $Z$ are connected⁴⁰ by

$Z = 200\, R^{1.455}$

Rainfall amounts are estimated at 6-minute intervals and at 100-meter range gates, with a focus on vertical levels between one and three kilometers. The data is then accumulated daily and re-gridded to a 0.5-kilometer resolution, providing detailed information that is useful to the researcher community⁴⁰.

The radar data utilized in this study were collected from the National Center of Meteorology (NCM). In order to create a resolution grid of 0.03 degrees for the composite UAE radars, we employed nearest-neighbor interpolation techniques that allowed for a more realistic comparison with the satellite products, which are also interpolated to such a fine resolution.

The advantage of the radars in contrast to satellites is their higher spatial and temporal resolution. These range from hundreds of meters to a few kilometers, and from 5 minutes to 15 minutes, respectively⁴¹.

Statistical methods and data analysis techniques

This study evaluates the performance of the satellite-based Microwave-Enhanced Geostationary Hydro-Estimator (MWGHE), which is a key tool in the flash flood guidance system of Oman’s National Directorate General of Meteorology (DGMET). The validation is conducted using ground-based measurements from rain gauges provided by DGMET in Oman. Additionally, composite Doppler radar data from the UAE is used to complement the analysis. The evaluation focuses on the period of October 2–4, 2021, during which Tropical Cyclone Shaheen impacted the region. A comparison is made between the satellite-based total precipitation distribution map and the reference data for the 24-hour accumulation. The spatial statistical differences between the compared datasets over 5 mm in Oman and the UAE were analyzed and evaluated.

The statistical analysis included bias, mean absolute error (MAE), root mean square error (RMSE), and correlation coefficient (CC) to evaluate the satellite performance against rain gauges and radar observations. These metrics were calculated daily and for 6-hour intervals, focusing on key differences between ground-based and satellite precipitation estimates. The statistical calculations provided insights into systematic biases and variability during high-intensity rainfall, particularly on October 3, 2021, when the cyclone made landfall. The scatter plots and boxplots presented in the results sections were used to visualize the distribution of errors and precipitation intensity. Bias values indicate the magnitude and direction of overestimation or underestimation, while RMSE and MAE highlight the overall deviations of satellite estimates from ground observations.

The satellite and radar data were corrected to a resolution of 0.033 degrees. As part of the comparison between the satellite and rain gauge data, the area of Oman was divided into approximately 6700 polygons based on coastal boundaries, soil type, and area elevation. Within these specific areas, rain gauge data was interpolated. Consequently, it is important to recall that the accuracy of the expanded rain gauge data via spatial interpolation is dependent on the density of the rain gauge network. This can increase the uncertainty in regions without rain gauge coverage. We also applied a Gaussian kernel density function (Kernel Density Estimation) with log-transformed to analyze precipitation data, leveraging its smoothness and symmetric shape, which is applied equally to the compared data sets for a simplified comparison. The analysis includes all areas that have precipitation and were extracted from the maps.

To account for the uncertainty of the MWGHE estimator in our analysis, we used the residual standard deviation in precipitation of 13.6 mm, obtained from climatological statistical analysis data from the Indian region, the closest region with available uncertainty. This dataset covers the period from October 2015 to 2019³². We chose the Indian region because it has a large network of nearly 3000 rain gauge stations, providing rich data. Also, our area, which lacks statistical study, shares a similar latitude with the Indian peninsula, making it the closest available for comparison by our study. However, this approach has limitations because of the significant difference in the amount of precipitation the Indian area received due to high monsoon activity and geographical characteristics.