Evaluation of satellite rainfall estimates using PERSIANN-CDR and TRMM over three critical cells in Jordan

Precipitation data are a crucial element in the study of water resource management, hydrological forecasting, and climate change. Current data collection methods, such as ground-based point stations, suffer from limitations in spatial and temporal coverage. To overcome these restrictions, several agencies have launched satellites that estimate rainfall over a gridded area using infrared and microwaves, converted into areal rainfall data through pretested algorithms. Despite their potential, the use of satellite-derived precipitation products is limited due to the lack of thorough quality assessment (Kidd & Levizzani 2011).

The main measuring methods for precipitation data collection are ground-based measurements: radar and point rain gauge stations. These classic recording methods are the most popular source of current hydro-meteorological rainfall data and are generally trusted in hydrological studies (Bernard & Gregoretti 2021). The measuring technology is well established and has been used for decades (Girons Lopez et al. 2015). However, the data are collected at specific sites and thus requires interpolation to obtain the areal rainfall (Camera et al. 2014).

On the contrary, satellite sensors use different techniques such as visible/infrared (VIS/IR) methods, active and passive microwave (MW) techniques, and combined VIS/IR and MW approaches to provide global and consistent measurements of precipitation (Qin et al. 2014). These methods employ unique algorithms, including PERSIANN-CDR and Tropical Rainfall Measuring Mission (TRMM), which offer long-term historical records of rainfall (Maggioni et al. 2016). Each of these used distinct approaches for estimating rainfall through satellite data, each characterized by unique data sources and methodologies (Tapiador et al. 2012; Sun et al. 2018).

TRMM 3B43 relies on both passive MW and radar data from the TRMM satellite, utilizing a combination of physical models and statistical techniques to estimate precipitation (Kofidou et al. 2023). In contrast, PERSIANN-CDR exclusively utilizes geostationary infrared satellite data and employs artificial neural networks to establish statistical relationships between cloud properties and rainfall (Gavahi et al. 2023).

In preceding research, rain gauge data exhibited spatial-temporal limitations. Evaluation of five satellite rainfall estimates (SREs) in Iran showed diverse performance, e.g., TRMM and PERSIANN-CDR, across climatic zones and time frames (Alijanian et al. 2017). In the Upper Huaihe River Basin, PERSIANN-CDR displayed stability but overestimated annual precipitation (Wei et al. 2023). Furthermore, an analysis of three high-resolution SREs – TRMM and PERSIANN-CDR among them – across varied Chinese basins highlighted their different accuracies in estimating precipitation and hydrological events. TRMM demonstrated superior performance in humid regions, while PERSIANN-CDR exhibited potential in arid areas. Nonetheless, both SREs require enhancement in hydrological (Zhang et al. 2022). Kesarwani et al. (2023) evaluated five SREs for drought monitoring capabilities in India, finding TRMM to have a higher correlation and lower errors compared to PERSIANN products. Multiple studies worldwide have examined SREs' potential to influence hydrology, water resource management, and flood forecasting and prevention (Beaufort et al. 2018; Ren et al. 2018; Eini et al. 2022; Dayal et al. 2023; Kumar et al. 2023b).

In Jordan, ground-based rainfall data covering the whole country are lacking and the accuracy of many of them is questionable due to reasons of station place (schools and police stations), collector's skills (reporting persons), timely reporting, and the measuring device, itself. Nevertheless, the stations report point values and are sparsely distributed. This research aimed to evaluate the performance of satellite-based rainfall estimation in comparison with ground-based measurements in Jordan in the long term from 2000 until 2013. In particular, the study utilized two of the most widely used satellite rainfall products, TRMM and PERSIANN-CDR, and compared them with ground-based data collected from rain gauge stations in three distinct regions: the north, middle, and south. The selection of these areas was based on their different climatic and geographical conditions and daily data availability. The north region, with its mountainous terrain and high annual precipitation, presented a favorable location to test the accuracy of satellite-based rainfall estimation. In contrast, the south region, characterized by low rainfall and sparse distribution, represented a challenging case to assess the performance of satellite-based estimation algorithms.

The study's findings have important implications for water resources management, hydrological forecasting, and climate change research in Jordan. Accurate rainfall estimates are essential for managing water resources effectively and forecasting floods, droughts, and other extreme weather events (Alsilibe et al. 2023; Wei et al. 2023). With its unique challenges in obtaining reliable rainfall data, Jordan can benefit significantly from satellite-based rainfall estimation. We contribute valuable insights that address the specific data challenges faced by this region, thereby expanding the applicability and relevance of satellite-based rainfall estimates in a distinct setting.

Jordan is situated within the latitudinal range of 29° 11′ N to 33° 22′ N and the longitudinal range of 34° 19′ E to 39° 18′ E. Covering an expanse exceeding 89,000 km², the country's landscape is predominantly arid, accounting for over 80% of its total area, and receiving an annual rainfall of under 100 mm (Al-Bakri et al. 2011). The distribution of rainfall is influenced by latitude, longitude, and elevation, creating variations across the territory (Wang et al. 2022).

Precipitation diminishes progressively from the northern to the southern regions, from west to east, and from higher altitudes to lower elevations. On average, the north experiences around 600 mm of rainfall annually, while the south and east receive less than 50 mm (Hadadin et al. 2010). Jordan's rainy period spans from October to May, with a substantial 80% of the yearly precipitation concentrated between December and March (Oweis & Hachum 2006; Renzi et al. 2023). Throughout this rainy season, a significant portion of the rainfall results from orographic effects, triggered by the passage of frontal depressions near Cyprus across the Mediterranean region (Al-Bakri & Suleiman 2004). Additionally, notably, Jordan is unaffected by the monsoon season, further delineating its unique precipitation patterns (Mehta & Yadav 2021; Singh et al. 2022). Because Jordan's rainfall varies appreciably from north to south in view of the orographic effect and distance from the center of the frontal system (Ziv et al. 2015; Marra et al. 2022), each cell was studied separately and was not lumped with other cells.

North cell: This particular cell depends on 36 monitoring stations and occupies geographical coordinates ranging from 35.75° to 36° E longitude and 32.25° to 32.50° N latitude. Notably, this specific cell stands out due to its distinctively higher values in terms of recorded rainfall measurements when compared to the other groups of rainfall stations. Furthermore, an elevation gradient is clearly observable within this cell, gradually increasing from its eastern boundary to its western limit.

Middle cell: Serving as the central cell for a cluster of 24 stations, this cell finds its geographic position nestled between the coordinates of 35.75° and 36° E and 31.75° and 32° N. In contrast to other groups of monitoring stations, this cell showcases moderately measured values of rainfall gauges. Adding to its distinct characteristics, a noticeable elevation pattern unfolds within this cell, ascending progressively from its western extremity to its eastern border.

South cell spans across geographic extents that extend from 35.25° to 35.50° E and 30° to 30.25° N. It sets itself apart with the lowest recorded values of rainfall gauges among the three cells. Within this cell, a total of 11 rainfall stations are actively engaged in data collection efforts. Moreover, an evident elevation trend is observed within this cell, where the elevation rises incrementally from its western boundary to its eastern edge.

The ground-based rainfall data are collected from point stations within the zone. The data consist of daily rainfalls from 2000 to 2013 obtained from the Jordan Ministry of Water and Irrigation (MWI). The station data were first tested for consistency and correctness with other stations in proximity. About 36, 24, and 11 rainfall stations were used in the north, middle, and south cells, respectively (Figure 1).

In 1997, TRMM was jointly launched by the National Aeronautics and Space Administration (NASA) and the Japan Aerospace Exploration Agency (JAXA) to estimate precipitation for latitudes ranging from 50° S to 50° N. The TRMM satellite was equipped with various instruments including the Visible Infrared Radiometer (VIRS), TRMM Microwave Imager (TMI), Cloud and Earth Radiant Energy Sensor (CERES), Lightning Imaging Sensor (LIS), and the first spaceborne precipitation radar. These instruments provided indirect measurements of precipitation, and the data gathered were subsequently used to develop several precipitation retrieval techniques. The global precipitation measurement (GPM) satellite is one such technique that has replaced TRMM in recent years.

The precipitation retrieval algorithms use data obtained from MW and infrared sensors on board the satellite, as well as rainfall measurements obtained from ground-based gauges. These algorithms create 3-hourly rainfall estimates at a spatial resolution of 0.25°. Multiple studies, including those conducted by Zhang et al. (2022) and Brasil Neto et al. (2022), have employed these algorithms to yield valuable insights into precipitation patterns.

The TRMM satellite analysis algorithm generated two primary products: the TRMM 3B42 and TRMM 3B43. Daily comparisons were made using TRMM 3B42 Version 7, while monthly comparisons were made using TRMM 3B43 Version 7. The data are readily available online via https://giovanni.gsfc.nasa.gov/giovanni/. The seasonal and annual estimates were calculated based on the monthly values derived from the TRMM algorithms. Due to the long-term nature of this study, spanning from 2000 to 2013, TRMM was chosen over GPM as GPM was not yet available during this time period. The use of TRMM highlights the importance of this study in enhancing our understanding of precipitation patterns in Jordan, especially within the critical cells analyzed.

The Climate Data Record (CDR) program, a collaboration between the National Oceanic and Atmospheric Association (NOAA) and the University of California, Irvine, produced the PERSIANN-CDR product, a retrospective satellite-based precipitation dataset that spans from 1983 to the present day. This dataset offers daily precipitation estimates with a spatial resolution of 0.25° × 0.25°, making it a valuable source of information for long-term studies (Ashouri et al. 2015).

The PERSIANN-CDR product is a multi-satellite, high-resolution precipitation dataset that utilizes infrared satellite data from global geosynchronous satellites to retrieve precipitation data (Sun et al. 2019). The model is calibrated using National Centers for Environmental Prediction stage IV hourly precipitation data and subsequently run on the entire GridSat-B1 IR data archive (Nguyen et al. 2018). The resulting estimates are further refined using the Global Precipitation Climatology Project (GPCP) monthly 2.5° precipitation products to minimize any predicted precipitation biases (Ashouri et al. 2015).

This product is a crucial resource for studying precipitation patterns, and it is available for download at https://chrsdata.eng.uci.edu/. The product's high spatial and temporal resolutions, coupled with its multi-satellite approach, make it an excellent tool for analyzing precipitation patterns on a global scale.

The correlation between the ground-based data and satellite data on a daily basis is presented in Table 2. Because satellites measure rainfall by infrared and MW reflections from clouds and those measurements are sensitive to clouds' content of rain, it was decided to divide the rainfall depth into three, light (less than 5 mm/day), moderate (5–20 mm/day), and heavy (≥20 mm/day). On this basis, the ground-based north cell number of light, moderate, and heavy rainy days was 873, 304, and 58 days, respectively, of the total 1,235 rainy days. Similarly, in the middle cell, the number was 858, 204, and 60 of the total 1,122 rainy days. And, in the south cell, it was 446, 75, and 8 of the total 529 rainy days. For TRMM 3B42 satellite data, the north cell had 1,087, 89, and 59 light, moderate, and heavy rainy days, respectively, of the total 1,235 rainy days. The middle cell had 1,008, 84, and 30 of them, respectively, of the total 1,122 rainy days; and in the south cell, there were 502, 22, and 5 light, moderate, and heavy rainy days of the total 529 rainy days. For PERSIANN-CDR satellite data, the north cell light, moderate, and heavy rainy days were 1,098, 125, and 10 days, respectively, of the total of 1,233 rainy days; in the middle cell, they were 1,025, 89, and 7, respectively, of the total 1,121 rainy days and in the south cell, they were 501, 26, and 2 of the total 529 rainy days.

Along with the above, Table 2 presents the comparison in percent relative deviation between the number of light, moderate, or heavy rainy days in satellite data and the ground-based data. On light rainy days, the deviation in the three cells varies between 13 and 25% in TRMM 3B42 data, and 13 and 26% in PERSIANN-CDR data; on moderate rainy days, it varies between −71 and −59% in TRMM 3B42 data and −65 and −56% in PERSIANN-CDR data; and in heavy rainy days, it varies between −1.7 and −50% in TRMM 3B42 data and −75 and −88% in PERSIANN-CDR data. Based on these numbers, it seems that the satellite's detection of light rainy days is more successful than moderate or heavy rainy days.

To compare the satellite rainfall depths with the ground-based ones, various correlation measures were calculated and presented in Table 2. In summary, the RMSE ranges between 5.39 and 9.08 mm/day for TRMM data and 5.15 and 8.02 mm/day for PERSIANN-CDR data. RE ranges between −62.3 and −68.2% for TRMM data and −44.97 and −69.35% for PERSIANN-CDR data. The CC ranges between 0.17 and 0.32 for TRMM data and 0.12 and 0.22 for PERSIANN-CDR data. MBE ranges between −1.78 and −2.07 mm/day for TRMM data and −1.67 and −2.87 mm/day for PERSIANN-CDR data. MAE ranges between 2.76 and 4.96 mm/day for TRMM data and 2.74 and 4.47 mm/day for PERSIANN-CDR data. Irrespective of other measures, the RE high value and the CC low value for both datasets indicate a low correlation with the ground-based data and a weak fit to a straight line between the two (Figure 4). Multiple studies have assured that the accuracy of the TRMM 3B42 and PERSIANN-CDR daily data is poor relative to other scales (Tan et al. 2015; Hamza et al. 2020).

The correlation measures of the two monthly satellite datasets were calculated and presented in Table 3. The negative values mean that the satellite data are less than the ground-based ones. As noticed in the table, the RRMSE, RMBE, and RMAE measures of the TRMM data are lower than those of the PERSIANN-CDR data, and the CC is higher, which means the TRMM data fit the ground-based data better than PERSIANN-CDR data. The fitness of both datasets is good as reflected in the values of all correlation measures, particularly CC, which range between 0.83 and 0.95 for TRMM data and 0.69 and 0.88 for PERSIAN-CDR data. In comparison to the daily error measures in Table 2, the monthly data fitness is superior and can be considered multiple times better. To secondly confirm this superiority, the relative RMSE, MBE, and MAE (scaled to the mean) for monthly data were calculated too (Table 3) and they were found multiple times lower than those in the daily data in Table 2. As noted by CC values in Table 3, the fit to ground-based is better for north and south cells (0.95, 0.94, 0.87, and 0.88) than the south cell (0.83 and 0.69). This means that a better fit of satellite data to ground-based data is proportional to rainfall depth.

For the north cell, recognized for its significant rainfall, TRMM exhibits superior performance with a higher CC value of 0.89 compared to PERSIANN-CDR's 0.81. TRMM's lower RMSE, MBE, MAE, and RE% values demonstrate its enhanced accuracy in estimating heavy rainfall, possibly attributed to its advanced technology tailored for higher precipitation levels.

Conversely, within the middle cell characterized by moderate rainfall, TRMM displays less favorable performance compared to PERSIANN-CDR, indicated by its lower CC value of 0.69 in contrast to PERSIANN-CDR's 0.73. Higher RMSE and MBE values and slightly less favorable MAE and RE% further underscore TRMM's relatively reduced accuracy in estimating moderate rainfall. This performance discrepancy might be due to technological constraints affecting TRMM's precision in this specific precipitation range. In the south cell with the lowest rainfall, TRMM showcases more effective performance compared to PERSIANN-CDR. TRMM's higher CC value of 0.73 in contrast to PERSIANN-CDR's 0.22, along with lower RMSE, MBE, MAE, and RE% values, suggests TRMM's technological adeptness at detecting lower rainfall levels. This proficiency reflects its more precise estimations in lighter rain, emphasizing the technology's optimization for such conditions.

The research analyzed the precision of satellite data in detecting and measuring rainfall by comparing two products, TRMM and PERSIANN-CDR, with ground-based data across three distinct geographic areas. The results revealed that daily, both satellite products exhibited less accuracy and higher discrepancies with ground data, with a better ability to detect light rainfall compared to moderate or heavy rainfall. However, at the monthly level, both products showed good results, with TRMM 3B43 performing better than PERSIANN-CDR. Additionally, TRMM 3B43 was more effective in the heavy rainy season, while PERSIANN-CDR had a higher CC value in the middle cell.

The study concludes that satellite data may not be as dependable in detecting and measuring rainfall daily but can still be useful on other scales, with variations in accuracy depending on the specific satellite product and geographic area.

The study was limited by the lack of freely available dense ground observations. Including additional gauge observations would have enhanced the evaluation of rainfall in TRMM and PERSIANN-CDR sub-daily products within other cells. Moreover, future studies could evaluate the performance of other SREs, such as the GPM satellite-based rainfall dataset. Further research is needed to examine the ability of satellite-based rainfall products to simulate hydro-meteorological phenomena such as floods and droughts in Jordan.

These findings contribute significantly to the understanding of the accuracy and limitations of satellite-based rainfall estimation products in Jordan, particularly in critical cells. These findings can be used as a basis for future research to develop improved techniques for estimating rainfall using satellite data, which could provide valuable information for water resources management and disaster risk reduction in the region.

The authors of this paper are grateful to Jordan University of Science and Technology for their support fund.

Data cannot be made publicly available; readers should contact the corresponding author for details.

The authors declare there is no conflict.