ABSTRACT
Satellite precipitation estimations (SPEs) have become important to estimate rainfall in remote and inaccessible areas. The study evaluates two high-resolution SPEs (IMERG and CHIRPS) in Peninsular Malaysia from 2011 to 2020. In situ rain gauge observation data were used as reference data, and a series of statistic indices were used to evaluate the performance of SPEs. In order to identify the source of error in the SPEs, an error decomposition technique was proposed whereby the bias is segregated into four different independent components. The study found that IMERG outperformed CHIRPS, with both satellites performing well in the east coast region but poor in the central region. A superior correlation between the SPEs and rain gauge observations was found during the northeast monsoon. The false bias has shown the widest range compared to other error components, indicating that it is the main contributor to the total bias of both SPEs in Peninsular Malaysia.
HIGHLIGHTS
Regional evaluation of IMERG and CHIRPS satellite estimations across Peninsular Malaysia.
Seasonal evaluation of IMERG and CHIRPS satellite estimations across Peninsular Malaysia.
Identification of the source of errors in satellite rainfall estimations using the error decomposition technique.
INTRODUCTION
An accurate estimation of precipitation is essential in predicting the changing trends of monsoon rainfall to manage the floods that affect millions of people and the destruction of ecology and farmlands (Tan et al. 2016). However, the rain gauge networks distributed sparsely in developing countries face severe challenges in obtaining reliable spatiotemporal precipitation information. In the past decades, satellite technology has begun to be adopted to improve flood visualization and reduce flood modelling uncertainties. In particular, in real-time mode, the assimilation of satellite technology may serve to keep forecasts obtained from flood simulations on track; while in hindcast mode, it assists in obtaining better estimates of the dynamic footprints of past flood events. Due to extensive spatial coverage and finer space and time resolutions, satellite precipitation estimations (SPEs) offer distinct advantages over ground sensors. These estimates are useful in data-sparse and ungauged basins especially oceanic and mountainous regions (Tian et al. 2009; Moazami et al. 2013; Gado et al. 2017), where rainfall data cannot be obtained from any resources. Nowadays, several high-resolution SPEs have been operationally available, including the National Oceanic and Atmospheric Administration Climate Prediction Center morphing technique product (CMORPH) (Joyce et al. 2004), the Precipitation Estimation from Remotely Sensed Information using Artificial Neural Networks (PERSIANN) (Sorooshian et al. 2000), the Tropical Rainfall Measuring Mission (TRMM) Multi-satellite Precipitation Analysis products (TMPA) (Huffman et al. 2007), the Global Satellite Mapping of Precipitation (GSMaP) (Kubota et al. 2007), Climate Hazards Group InfraRed Precipitation with Station data (CHIRPS) (Funk et al. 2015), etc.
Numerous studies evaluating the performance of SPEs have been done varying with seasons, locations, topography, climatology factors, etc. The conventional descriptive and continuous statistic metrics are widely utilized to measure the errors of SPEs, regardless of point-to-pixel scale or basin scale (Moazami et al. 2013; Dembélé & Zwart 2016; Soo et al. 2018; Tan & Santo 2018). However, the application of these statistical metrics is still limited because of the lack of effective quality assurance. Apart from that, the performance of SPEs can vary over time and other environmental and geographical factors and can be prone to bias and stochastic errors, which highly rely on the region's hydro-climatic features (Wang et al. 2018).
Given the aforementioned issues, to understand the error characteristics of SPEs and their sources of errors, many researchers further investigate these errors computed. For instance, Vergara and authors performed separate satellite error propagation investigations over mild-slope-terrain basins with data showing that the rainfall–runoff cycle buffers the satellite error variance and that this damping effect exhibits basin-scale dependence (Vergara et al. 2014). Some researchers applied the error decomposition schemes to further decompose these indicators to pinpoint the source of errors. Tian et al. (2009) designed a scheme to decompose the error of SPPs into three categories which are missed precipitation, hit bias (HB), and false precipitation to evaluate the performance of Integrated Multi-satellitE Retrievals for GPM (IMERG) and TRMM estimations, respectively, in Mainland China. This method is efficient at detecting the primary source of error in the total bias (TB) and can overcome the possibility of total underestimation due to the cancellation of individual error components. Thus, such a method has been applied by many researchers to evaluate and assess different SPEs in different parts of the world (Xu et al. 2016; Lekula et al. 2018; Su et al. 2018), but no study has been conducted for Malaysia.
As a tropical country, Malaysia suffers from different kinds of climate hazards, such as flooding and landslides, every year. The impacts of climate change and degradation are increasingly felt. Therefore, to improve and enrich the existing findings, a more extensive and in-depth analysis of various precipitation measuring tools in Malaysia is still required. The present study evaluates the performance of two high-resolution satellite estimations at regional and seasonal scales during the last decade (2011–2020) for Peninsular Malaysia. The two satellites to be focused on are IMERG and CHIRPS. The bias in the satellite estimations is analysed and disintegrated into independent components by using the error decomposition technique proposed by Tian et al. (2009), whereby the TB is segregated into three categories, namely the HB, miss bias (MB), and false bias (FB). However, according to Chaudhary & Dhanya (2021), inherent deficiencies may exist in the error decomposition technique, whereby there is a possibility of cancelling out of individual positive and negative hit rainfall values while computing the HB, and this may lead to an underestimation of HB. Thus, the HB in this study is further divided into two components which are the over-hit bias (OHB) and under-hit bias (UHB), to provide an improved representation of errors.
METHODS
Data collection
For the present study, daily rainfall data from 550 rain gauge stations, starting from 1 January 2011 to 31 December 2020 were collected from the Department of Irrigation and Drainage Malaysia (DID), as shown in Figure 1(b). The study period was selected because during this period both SPEs and rain gauges are able to provide complete data at daily intervals with minimal missing values.
Next, two high-resolution Satellite Precipitation Products (SPPs), which are the Integrated Multi-satellitE Retrievals for GPM (IMERG) and CHIRPS were employed in this study. Both satellite datasets were extracted from the Climate Engine Application for Peninsular Malaysia starting from the year 2011 to 2020 (10 years). IMERG is an international satellite mission to provide next-generation observations of rain and snow worldwide every 3 h. It is the unified algorithm that provides rainfall estimates combining data from all passive-microwave instruments in the GPM Constellation. The primary enhancement is the mixture of high spatial-temporal resolutions, near-global coverage, and high quality by using active instruments to support the use of passive instruments. IMERG was obtained with 0.1° × 0.1° (approximately 10 km × 10 km) spatial resolution. There are three main IMERGs, namely Early, Late, and Final Runs. The IMERG Final Run daily version 6 products at a spatial resolution of 0.1° were employed in the present study as it is more accurate and bias-corrected using the Global Precipitation Climatology Centre (GPCC) precipitation gauges (Huffman et al. 2020). CHIRPS is an over-35-years quasi-global satellite rainfall dataset, spanning 50°S–50°N (and all longitudes) and ranging from 1981 to the present. The CHIRPS dataset provides various temporal rainfall resolutions (daily, monthly, pentad, and decadal), and registers with a spatial resolution of 0.05° (approximately 5.3 km × 5.3 km). In the present study, the daily CHIRPS version 2 products at a spatial resolution of 0.05° were adopted.
Evaluation indexes
Statistic metrics . | Equations . | Optimal value . |
---|---|---|
CC | 1 | |
MAE | 0 | |
RMSE | 0 | |
Percent bias/total bias | 0 | |
Under-hit bias | 0 | |
Over-hit bias | 0 | |
Miss bias | 0 | |
False bias | 0 |
Statistic metrics . | Equations . | Optimal value . |
---|---|---|
CC | 1 | |
MAE | 0 | |
RMSE | 0 | |
Percent bias/total bias | 0 | |
Under-hit bias | 0 | |
Over-hit bias | 0 | |
Miss bias | 0 | |
False bias | 0 |
Note: S, satellite measured precipitation; G, rain gauge measured precipitation; , mean of rain gauge observation; , mean of satellite rainfall estimations; i, index of data; n, the total number of measurements; , Under-hit rainfall events measured by satellite; , Over-hit rainfall events measured by satellite; , Miss rainfall events measured by the satellite; , False rainfall events measured by the satellite; .
RESULTS
Spatiotemporal analyses of average precipitation
Regional assessment
Regions . | Performance indicators . | IMERG . | CHIRPS . | ||
---|---|---|---|---|---|
Lowest . | Highest . | Lowest . | Highest . | ||
Northern | CC | 0.20 | 0.56 | 0.12 | 0.41 |
PBIAS (%) | −32.50 | 48.99 | −6.19 | 68.21 | |
MAE (mm/day) | 5.25 | 11.31 | 6.52 | 14.04 | |
RMSE (mm/day) | 10.84 | 20.44 | 11.46 | 21.23 | |
East Coast | CC | 0.07 | 0.74 | 0.02 | 0.60 |
PBIAS (%) | −41.88 | 73.17 | −96.53 | 82.44 | |
MAE (mm/day) | 5.80 | 12.38 | 6.65 | 14.35 | |
RMSE (mm/day) | 10.71 | 37.05 | 11.93 | 42.55 | |
Central | CC | 0.19 | 0.56 | 0.16 | 0.30 |
PBIAS (%) | −15.48 | 83.34 | −18.22 | 93.86 | |
MAE (mm/day) | 5.14 | 8.80 | 6.15 | 9.69 | |
RMSE (mm/day) | 10.94 | 17.17 | 11.30 | 16.88 | |
Southern | CC | 0.16 | 0.64 | 0.14 | 0.48 |
PBIAS (%) | −32.3 | 86.27 | −29.37 | 76.01 | |
MAE (mm/day) | 5.44 | 9.37 | 6.14 | 10.21 | |
RMSE (mm/day) | 11.22 | 22.89 | 11.04 | 23.20 |
Regions . | Performance indicators . | IMERG . | CHIRPS . | ||
---|---|---|---|---|---|
Lowest . | Highest . | Lowest . | Highest . | ||
Northern | CC | 0.20 | 0.56 | 0.12 | 0.41 |
PBIAS (%) | −32.50 | 48.99 | −6.19 | 68.21 | |
MAE (mm/day) | 5.25 | 11.31 | 6.52 | 14.04 | |
RMSE (mm/day) | 10.84 | 20.44 | 11.46 | 21.23 | |
East Coast | CC | 0.07 | 0.74 | 0.02 | 0.60 |
PBIAS (%) | −41.88 | 73.17 | −96.53 | 82.44 | |
MAE (mm/day) | 5.80 | 12.38 | 6.65 | 14.35 | |
RMSE (mm/day) | 10.71 | 37.05 | 11.93 | 42.55 | |
Central | CC | 0.19 | 0.56 | 0.16 | 0.30 |
PBIAS (%) | −15.48 | 83.34 | −18.22 | 93.86 | |
MAE (mm/day) | 5.14 | 8.80 | 6.15 | 9.69 | |
RMSE (mm/day) | 10.94 | 17.17 | 11.30 | 16.88 | |
Southern | CC | 0.16 | 0.64 | 0.14 | 0.48 |
PBIAS (%) | −32.3 | 86.27 | −29.37 | 76.01 | |
MAE (mm/day) | 5.44 | 9.37 | 6.14 | 10.21 | |
RMSE (mm/day) | 11.22 | 22.89 | 11.04 | 23.20 |
Further analysis was conducted to compare statistical analysis results of IMERG and CHIRPS daily precipitation in the current study with past studies conducted at different regions globally in terms of CC and RMSE as shown in Table 3. Table 3 shows that IMERG recorded the strongest correlation with rain gauge observations in Colombia with a CC value of 0.88 (Rodríguez-Sandoval et al. 2024). Generally, past research has shown a moderate correlation between IMERG and rain gauge observations in China, Northern Vietnam, Malaysia, Korea, Japan and Singapore with CC values falling in the range between 0.50 and 0.68 (Kim et al. 2017; Tan & Santo 2018; Nguyen & Do 2021; Liu et al. 2022). Whereas, a weak correlation between IMERG and rain gauge observations is found in Yarlung Zangbo River Basin, Mexico, Chindwin River Basin and Southern Tibetan Plateau with CC values ranging from 0.22 to 0.46 (Xu et al. 2017; Yuan et al. 2017; Ji et al. 2022; Rincón-Avalos et al. 2022). Besides, the RMSE of IMERG from past studies ranges from 4.74 to 23.41 mm/day. Statistical analysis results in this study show that IMERG obtained a larger range for both CC (0.07–0.74) and RMSE (10.71–37.05 mm/day). On the other hand, past studies show that CHIRPS have a low correlation with rain gauge observations in Mexico and Antioquia, Northwestern Colombia with CC values recorded at 0.31 and 0.29, respectively. Meanwhile, the RMSE of CHIRPS for the past studies ranges from 1.63 to 22.19 mm/day. Similar to IMERG, CHIRPS in this study also exhibited a larger range for both CC (0.02–0.60) and RMSE (11.04–42.55 mm/day). As discussed, both IMERG and CHIRPS demonstrated a diverse range of CC and RMSE values across different geographical regions and climates. Therefore, this indicates that the performance of IMERG and CHIRPS is solitary in Peninsular Malaysia due to the study area's unique hydro-climatic features. This statement is further backed by Wang et al. (2018), Nikolopoulos et al. (2013), and Mei et al. (2016) who recognized that the performance of SPEs is heavily influenced by geographical conditions, basin conditions, and regional rainfall patterns.
Studies . | Study area . | Period . | IMERG . | CHIRPS . | ||
---|---|---|---|---|---|---|
CC . | RMSE (mm/day) . | CC . | RMSE (mm/day) . | |||
This study | Peninsular Malaysia | January 2011–December 2020 | 0.07–0.74 | 10.71–37.05 | 0.02–0.60 | 11.04–42.55 |
Rodríguez-Sandoval et al. (2024) | Colombia | 2001–2019 | 0.88 | – | – | – |
Ji et al. (2022) | Yarlung Zangbo River Basin, China | 2003–2015 | 0.39 | 4.74 | – | – |
Liu et al. (2022) | Mainland, China | 2001–2015 | 0.50 | 5.63 | – | – |
Rincón-Avalos et al. (2022) | Mexico | 2001–2017 | 0.27 | 21.18 | 0.31 | 22.19 |
Nguyen & Do (2021) | Northern Vietnam | 2007–2014 | 0.68 | 11.70 | – | – |
Saddique et al. (2022) | Jhelum River Basin, India | 1998–2007 | – | – | 0.73 | 1.63 |
López-Bermeo et al. (2022) | Antioquia, Northwestern Colombia | 1981–2018 | – | – | 0.29 | 21.53 |
Tan & Santo (2018) | Malaysia | March 2014–February 2016 | 0.50–0.60 | 12.94–14.93 | – | – |
Tan et al. (2018) | Kelantan, Malaysia | March 2014–December 2016 | 0.61 | 13.69 | – | – |
Dinku et al. (2018) | East Africa | 2006–2010 | – | – | 0.87 | – |
Kim et al. (2017) | Korea, Japan | March 2014–August 2014 | 0.53–0.68 | 6.68–23.41 | – | – |
Xu et al. (2017) | Southern Tibetan Plateau | May 2014–October 2014 | 0.46 | 7.16 | – | – |
Tan & Duan (2017) | Singapore | April 2014–January 2016 | 0.53 | 11.83 | – | – |
Yuan et al. (2017) | Chindwin River Basin, Myanmar | April 2014–January 2016 | 0.22–0.32 | 9.1–24.7 | – | – |
Studies . | Study area . | Period . | IMERG . | CHIRPS . | ||
---|---|---|---|---|---|---|
CC . | RMSE (mm/day) . | CC . | RMSE (mm/day) . | |||
This study | Peninsular Malaysia | January 2011–December 2020 | 0.07–0.74 | 10.71–37.05 | 0.02–0.60 | 11.04–42.55 |
Rodríguez-Sandoval et al. (2024) | Colombia | 2001–2019 | 0.88 | – | – | – |
Ji et al. (2022) | Yarlung Zangbo River Basin, China | 2003–2015 | 0.39 | 4.74 | – | – |
Liu et al. (2022) | Mainland, China | 2001–2015 | 0.50 | 5.63 | – | – |
Rincón-Avalos et al. (2022) | Mexico | 2001–2017 | 0.27 | 21.18 | 0.31 | 22.19 |
Nguyen & Do (2021) | Northern Vietnam | 2007–2014 | 0.68 | 11.70 | – | – |
Saddique et al. (2022) | Jhelum River Basin, India | 1998–2007 | – | – | 0.73 | 1.63 |
López-Bermeo et al. (2022) | Antioquia, Northwestern Colombia | 1981–2018 | – | – | 0.29 | 21.53 |
Tan & Santo (2018) | Malaysia | March 2014–February 2016 | 0.50–0.60 | 12.94–14.93 | – | – |
Tan et al. (2018) | Kelantan, Malaysia | March 2014–December 2016 | 0.61 | 13.69 | – | – |
Dinku et al. (2018) | East Africa | 2006–2010 | – | – | 0.87 | – |
Kim et al. (2017) | Korea, Japan | March 2014–August 2014 | 0.53–0.68 | 6.68–23.41 | – | – |
Xu et al. (2017) | Southern Tibetan Plateau | May 2014–October 2014 | 0.46 | 7.16 | – | – |
Tan & Duan (2017) | Singapore | April 2014–January 2016 | 0.53 | 11.83 | – | – |
Yuan et al. (2017) | Chindwin River Basin, Myanmar | April 2014–January 2016 | 0.22–0.32 | 9.1–24.7 | – | – |
Seasonal assessment
. | . | IMERG . | CHIRPS . | ||
---|---|---|---|---|---|
Monsoon . | Performance indicators . | Lowest . | Highest . | Lowest . | Highest . |
NEM | CC | 0.07 | 0.81 | 0.07 | 0.64 |
PBIAS (%) | −52.08 | 116.64 | −47.02 | 145.38 | |
MAE (mm/day) | 3.64 | 15.56 | 3.99 | 17.96 | |
RMSE (mm/day) | 8.71 | 52.68 | 8.58 | 61.58 | |
SWM | CC | 0.06 | 0.60 | 0.02 | 0.35 |
PBIAS (%) | −49.46 | 85.66 | −41.4 | 112.43 | |
MAE (mm/day) | 4.47 | 12.17 | 5.15 | 15.27 | |
RMSE (mm/day) | 9.00 | 23.81 | 9.38 | 25.36 | |
IM1 | CC | 0.02 | 0.70 | 0.03 | 0.61 |
PBIAS (%) | −49.47 | 161.99 | −50.97 | 176.98 | |
MAE (mm/day) | 4.86 | 14.88 | 4.6 | 15.34 | |
RMSE (mm/day) | 9.68 | 34.87 | 7.96 | 33.81 | |
IM2 | CC | 0.03 | 0.67 | 0.02 | 0.54 |
PBIAS (%) | −32.77 | 121.69 | −42.74 | 116.9 | |
MAE (mm/day) | 5.68 | 13.72 | 6.18 | 18.92 | |
RMSE (mm/day) | 10.57 | 36.04 | 9.93 | 35.09 |
. | . | IMERG . | CHIRPS . | ||
---|---|---|---|---|---|
Monsoon . | Performance indicators . | Lowest . | Highest . | Lowest . | Highest . |
NEM | CC | 0.07 | 0.81 | 0.07 | 0.64 |
PBIAS (%) | −52.08 | 116.64 | −47.02 | 145.38 | |
MAE (mm/day) | 3.64 | 15.56 | 3.99 | 17.96 | |
RMSE (mm/day) | 8.71 | 52.68 | 8.58 | 61.58 | |
SWM | CC | 0.06 | 0.60 | 0.02 | 0.35 |
PBIAS (%) | −49.46 | 85.66 | −41.4 | 112.43 | |
MAE (mm/day) | 4.47 | 12.17 | 5.15 | 15.27 | |
RMSE (mm/day) | 9.00 | 23.81 | 9.38 | 25.36 | |
IM1 | CC | 0.02 | 0.70 | 0.03 | 0.61 |
PBIAS (%) | −49.47 | 161.99 | −50.97 | 176.98 | |
MAE (mm/day) | 4.86 | 14.88 | 4.6 | 15.34 | |
RMSE (mm/day) | 9.68 | 34.87 | 7.96 | 33.81 | |
IM2 | CC | 0.03 | 0.67 | 0.02 | 0.54 |
PBIAS (%) | −32.77 | 121.69 | −42.74 | 116.9 | |
MAE (mm/day) | 5.68 | 13.72 | 6.18 | 18.92 | |
RMSE (mm/day) | 10.57 | 36.04 | 9.93 | 35.09 |
The performance of IMERG and CHIRPS under different seasonal periods in terms of MAE and RMSE was also evaluated. Under IMERG, the maximum MAE and RMSE values were observed during NEM, with values of 15.56 and 52.68 mm/day, respectively. The corresponding values for SWM were 12.17 and 23.81 mm/day, for IM1 were 14.88 and 34.87 mm/day, and for IM2 were 13.72 and 36.04 mm/day, respectively. Meanwhile, under CHIRPS, the maximum MAE and RMSE values were observed during NEM, with values of 17.96 and 61.58 mm/day, respectively. The corresponding values for SWM were 15.27 and 25.36 mm/day, for IM1 were 15.34 and 33.81 mm/day, and for IM2 were 18.92 and 35.09 mm/day, respectively.
Analysis of bias components
Regional assessment
Regions . | Biases . | IMERG . | CHIRPS . | ||
---|---|---|---|---|---|
Lowest . | Highest . | Lowest . | Highest . | ||
Northern | Over-hit (OHB) | 18.40 | 45.90 | 17.87 | 46.05 |
Under-hit (UHB) | −32.86 | −50.69 | −31.22 | −48.88 | |
Miss (MB) | −4.18 | −17.11 | −10.04 | −23.79 | |
False (FB) | 14.56 | 65.86 | 29.45 | 74.44 | |
East Coast | Over-hit (OHB) | 12.26 | 50.64 | 14.04 | 55.52 |
Under-hit (UHB) | −31.02 | −55.9 | −27.07 | −51.69 | |
Miss (MB) | −3.60 | −31.42 | −10.96 | −29.70 | |
False (FB) | 7.96 | 47.08 | 14.76 | 72.01 | |
Central | Over-hit (OHB) | 22.43 | 51.65 | 16.25 | 38.93 |
Under-hit (UHB) | −32.58 | −48.73 | −31.58 | −47.88 | |
Miss (MB) | −6.59 | −20.06 | −11.40 | −27.68 | |
False (FB) | 18.59 | 75.94 | 29.29 | 115.28 | |
Southern | Over-hit (OHB) | 17.95 | 45.64 | 13.66 | 40.64 |
Under-hit (UHB) | −33.64 | −48.81 | −24.79 | −47.72 | |
Miss (MB) | −5.92 | −18.04 | −12.60 | −32.76 | |
False (FB) | 17.13 | 65.26 | 25.54 | 78.92 |
Regions . | Biases . | IMERG . | CHIRPS . | ||
---|---|---|---|---|---|
Lowest . | Highest . | Lowest . | Highest . | ||
Northern | Over-hit (OHB) | 18.40 | 45.90 | 17.87 | 46.05 |
Under-hit (UHB) | −32.86 | −50.69 | −31.22 | −48.88 | |
Miss (MB) | −4.18 | −17.11 | −10.04 | −23.79 | |
False (FB) | 14.56 | 65.86 | 29.45 | 74.44 | |
East Coast | Over-hit (OHB) | 12.26 | 50.64 | 14.04 | 55.52 |
Under-hit (UHB) | −31.02 | −55.9 | −27.07 | −51.69 | |
Miss (MB) | −3.60 | −31.42 | −10.96 | −29.70 | |
False (FB) | 7.96 | 47.08 | 14.76 | 72.01 | |
Central | Over-hit (OHB) | 22.43 | 51.65 | 16.25 | 38.93 |
Under-hit (UHB) | −32.58 | −48.73 | −31.58 | −47.88 | |
Miss (MB) | −6.59 | −20.06 | −11.40 | −27.68 | |
False (FB) | 18.59 | 75.94 | 29.29 | 115.28 | |
Southern | Over-hit (OHB) | 17.95 | 45.64 | 13.66 | 40.64 |
Under-hit (UHB) | −33.64 | −48.81 | −24.79 | −47.72 | |
Miss (MB) | −5.92 | −18.04 | −12.60 | −32.76 | |
False (FB) | 17.13 | 65.26 | 25.54 | 78.92 |
According to Table 5, CHIRPS has the highest OHB in the central region, while IMERG has the highest UHB in the east coast region. IMERG has higher OHB in the central and southern regions, whereas CHIRPS shows higher OHB in the northern and east coast regions. The east coast region has the highest magnitude of UHB in both SPEs. IMERG performs better in MB and FB for all regions except for the MB in the east coast region. In general, IMERG shows higher accuracy in detecting precipitation events, with fewer false events and better consistency with rain gauge observations. Both SPEs exhibit the highest FB in the central region, with CHIRPS having a higher FB than IMERG.
Seasonal assessment
Monsoons . | Biases . | IMERG . | CHIRPS . | ||
---|---|---|---|---|---|
Lowest . | Highest . | Lowest . | Highest . | ||
NEM | Over-hit (OHB) | 8.70 | 72.85 | 14.96 | 87.01 |
Under-hit (UHB) | −31.69 | −61.03 | −19.99 | −61.63 | |
Miss (MB) | −3.27 | −36.35 | −9.40 | −48.67 | |
False (FB) | 3.16 | 95.62 | 8.81 | 143.04 | |
SWM | Over-hit (OHB) | 9.58 | 53.61 | 11.95 | 49.60 |
Under-hit (UHB) | −29.67 | −56.57 | −20.35 | −54.81 | |
Miss (MB) | −2.46 | −33.32 | −6.58 | −51.36 | |
False (FB) | 13.50 | 99.65 | 24.55 | 141.56 | |
IM1 | Over-hit (OHB) | 8.33 | 69.51 | 7.80 | 64.02 |
Under-hit (UHB) | −21.06 | −71.44 | −19.14 | −70.68 | |
Miss (MB) | −0.99 | −42.06 | −1.32 | −51.63 | |
False (FB) | 8.41 | 182.22 | 14.37 | 161.93 | |
IM2 | Over-hit (OHB) | 20.03 | 81.08 | 7.84 | 64.99 |
Under-hit (UHB) | −24.84 | −54.17 | −19.72 | −59.28 | |
Miss (MB) | −1.15 | −24.55 | −3.45 | −35.83 | |
False (FB) | 7.75 | 108.94 | 12.08 | 125.50 |
Monsoons . | Biases . | IMERG . | CHIRPS . | ||
---|---|---|---|---|---|
Lowest . | Highest . | Lowest . | Highest . | ||
NEM | Over-hit (OHB) | 8.70 | 72.85 | 14.96 | 87.01 |
Under-hit (UHB) | −31.69 | −61.03 | −19.99 | −61.63 | |
Miss (MB) | −3.27 | −36.35 | −9.40 | −48.67 | |
False (FB) | 3.16 | 95.62 | 8.81 | 143.04 | |
SWM | Over-hit (OHB) | 9.58 | 53.61 | 11.95 | 49.60 |
Under-hit (UHB) | −29.67 | −56.57 | −20.35 | −54.81 | |
Miss (MB) | −2.46 | −33.32 | −6.58 | −51.36 | |
False (FB) | 13.50 | 99.65 | 24.55 | 141.56 | |
IM1 | Over-hit (OHB) | 8.33 | 69.51 | 7.80 | 64.02 |
Under-hit (UHB) | −21.06 | −71.44 | −19.14 | −70.68 | |
Miss (MB) | −0.99 | −42.06 | −1.32 | −51.63 | |
False (FB) | 8.41 | 182.22 | 14.37 | 161.93 | |
IM2 | Over-hit (OHB) | 20.03 | 81.08 | 7.84 | 64.99 |
Under-hit (UHB) | −24.84 | −54.17 | −19.72 | −59.28 | |
Miss (MB) | −1.15 | −24.55 | −3.45 | −35.83 | |
False (FB) | 7.75 | 108.94 | 12.08 | 125.50 |
Both SPEs also exhibited the highest magnitude of UHB during IM1. The highest magnitude of MB was observed during IM1 for both SPEs, while the largest FB was found during IM1 as well, with extreme values larger than 100%. CHIRPS had a significant amount of MB during IM1 due to the inability of remote sensors to detect light precipitation. In general, the study suggests that IMERG has a better performance compared to CHIRPS as it can detect fewer false precipitation events and is highly consistent with rain gauge observations.
Comparative analysis of error decomposition techniques and results
To expand the insights of this study, error decomposition results of IMERG and CHIRPS in this study were compared with SPEs in other geographical locations globally. Only Chaudhary & Dhanya (2021) and Zhang et al. (2021) studies utilized the same error decomposition technique as in this study, which disintegrated TB into OHB, UHB, MB and FB. As shown in Table 7, Chaudhary & Dhanya (2021) concluded that MB is the main contributor to TB at daily scale analysis while OHB and UHB are the main contributors to TB for TRMM 3B42RT and CMORPH and PERSIANN, respectively. On the other hand, a study conducted by Zhang et al. (2021) indicated that FB is the largest bias in TB for PERSIANN Dynamic Infrared–Rain Rate (PDIR), IMERG, and GSMaP in the Eastern Monsoon region of China at a daily scale, while MB and OHB are the highest biases to TB in the wet and dry seasons, respectively. This study found that FB is the main error component in TB for daily and all seasonal scales. Therefore, the error component that contributes the highest percentage to TB is different for every SPE in different geographical regions. The characteristic of error components depends largely on the magnitude of rainfall, topography, and elevation of the selected study area. However, more research needs to be conducted by using the four error components decomposition technique for a more comprehensive comparative analysis to be carried out. The studies conducted by Ghomlaghi et al. (2022), Tian et al. (2009), and Lei et al. (2022) cannot be directly compared to the error decomposition results in this study as they only utilized three error components decomposition technique which decomposes TB into HB, MB, and FB. HB cannot be directly compared to OHB and UHB as HB is the combination of OHB and UHB whose magnitude is reduced by the adding up of OHB and UHB which render insufficient information on how large the portion of HB is dominated by OHB or UHB.
Studies . | Study area . | Period . | SPEs studied . | Main contributor to total bias . | ||
---|---|---|---|---|---|---|
Daily scale . | Wet season . | Dry season . | ||||
This study | Peninsular Malaysia | January 2011–December 2020 | IMERG, CHIRPS | FB | FB | FB |
Chaudhary & Dhanya (2021) | India | 2001–2016 | TRMM 3B42RT | OHB | – | – |
CMORPH, PERSIANN | UHB | – | – | |||
CHIRPS | MB | – | – | |||
Zhang et al. (2021) | Eastern Monsoon Region, China | January 2003–December 2019 | PDIR, IMERG, GSMaP | FB | MB | OHB |
Ghomlaghi et al. (2022) | Central Iran | 2005–2015 | CHIRPS, CMORPH, ERA5-Land, GPM_3IMERGM, MSWEP V2, PERSIANN, PERSIANN-CCS, PERSAINN-CDR, TErraClimate, TRMM_3B43 | Hit bias (HB) | – | – |
Tian et al. (2009) | Contiguous United States | June 2003–February 2007 | AFWA, (TMPA) 3B42, (TMPA) 3B42RT, CMORPH, PERSIANN, NRL | – | MB | HB |
Lei et al. (2022) | Mainland China | 2000–2017 | GSMaP, IMERG, TMPA, CMORPH, PERSIANN | FB | – | – |
CHIRPS | MB | – | – |
Studies . | Study area . | Period . | SPEs studied . | Main contributor to total bias . | ||
---|---|---|---|---|---|---|
Daily scale . | Wet season . | Dry season . | ||||
This study | Peninsular Malaysia | January 2011–December 2020 | IMERG, CHIRPS | FB | FB | FB |
Chaudhary & Dhanya (2021) | India | 2001–2016 | TRMM 3B42RT | OHB | – | – |
CMORPH, PERSIANN | UHB | – | – | |||
CHIRPS | MB | – | – | |||
Zhang et al. (2021) | Eastern Monsoon Region, China | January 2003–December 2019 | PDIR, IMERG, GSMaP | FB | MB | OHB |
Ghomlaghi et al. (2022) | Central Iran | 2005–2015 | CHIRPS, CMORPH, ERA5-Land, GPM_3IMERGM, MSWEP V2, PERSIANN, PERSIANN-CCS, PERSAINN-CDR, TErraClimate, TRMM_3B43 | Hit bias (HB) | – | – |
Tian et al. (2009) | Contiguous United States | June 2003–February 2007 | AFWA, (TMPA) 3B42, (TMPA) 3B42RT, CMORPH, PERSIANN, NRL | – | MB | HB |
Lei et al. (2022) | Mainland China | 2000–2017 | GSMaP, IMERG, TMPA, CMORPH, PERSIANN | FB | – | – |
CHIRPS | MB | – | – |
Practical implications of the findings
Accurate and reliable precipitation data from SPEs is crucial for water-related applications such as flood management, hydrological modelling, drought monitoring, weather forecasting, etc. Flood management is one of the most important water-related measures nowadays as urbanization has altered the original environment and can cause various types of floods. The findings in this study concluded that IMERG performed better than CHIRPS so that users can implement precipitation estimations from IMERG as input data to train and produce different kinds of hydrological models that mimic river flow and forecast future flooding. IMERG precipitation estimations with high CC values and low PBIAS, MAE, and RMSE values should be used as input data for the flood forecasting modelling as this will render less error in the predicting output. The prediction output from the flood forecasting model acts as an early warning system that provides authorities and communities with timely warnings and enables preventive measures to be taken. Besides, the precipitation estimation data aids in evaluating the flood situation by providing information on regions that are most affected by heavy rainfall and potentially affected by floods so that resource allocation during flood events becomes more effective. Moreover, Flood hazard maps can be created by identifying places that are at risk of flooding through the analysis of historical satellite precipitation data. These data are used to support the planning and construction of drainage systems, flood defences and other essential infrastructure to reduce the risk of floods.
Other than that, SPEs can substantially improve drought monitoring by offering extensive regional and temporal information on precipitation deficiencies. These data facilitate the early identification of drought situations, enhance conventional drought indicators, and aid in the verification of climate models. Additionally, it strengthens early warning systems, provides information for policy and resource allocation, and enhances public awareness and education. Furthermore, the utilization of extensive satellite records is crucial for examining the effects of climate change on the occurrence and intensity of droughts, facilitating the development of more efficient measures for readiness and adaptability.
CONCLUSIONS
In conclusion, satellite precipitation estimates (SPEs) provide valuable enhancements to hydrological analysis, weather monitoring, and forecasting by improving the spatial coverage and temporal resolution compared to traditional ground-based rainfall data. However, it is important to acknowledge that the accuracy and performance of SPEs can vary across regions and seasons. Thus, identifying the sources of error is crucial for improving the reliability and accuracy of SPEs (Prakash et al. 2015). Based on the study, IMERG showed better performance compared to CHIRPS at both regional and seasonal scales for Peninsular Malaysia. Both SPEs were able to estimate the rainfall at the east coast region of Peninsular Malaysia, especially during NEM, which brings a higher intensity of rainfall to the country annually. Nevertheless, based on the error decomposition analysis, IMERG tends to underestimate the actual rainfall, especially in the east coast region of Peninsular Malaysia. On the other hand, CHIRPS tends to overestimate the actual rainfall, especially in the central region of Peninsular Malaysia. At seasonal scales, both SPEs showed a high OHB during the wet seasons of Peninsular Malaysia (NEM and IM2), and a high UHB during the dry season (IM1). Comparing the four error components, both SPEs also exhibited a high magnitude of FB, indicating many erroneous detections of rainfall at places of an actual non-rainy event. The findings of this study provide insights for researchers to improve the reliability and accuracy of error characterization in SPEs. Besides, as both SPEs are able to estimate rainfall accurately at the east coast region of Peninsular Malaysia, the SPE data at the mentioned region can be utilized in water resource management after being implemented in various hydrological models. SPEs provide substantial advantages for water resource management by offering uninterrupted, real-time information over extensive and isolated regions, thereby enhancing supervision and optimizing water distribution. The estimations assist in the control of floods and droughts, optimize the operations of reservoirs, and provide optimal planning for irrigation as well as assuring the efficient use of water in agriculture. In addition, they enhance the precision of hydrological models, aid in evaluating the effects of climate change, and facilitate the estimation of groundwater recharge rates. Furthermore, SPEs facilitate the sustainable and resilient management of water resources by supporting urban water management, environmental conservation and readiness for disaster and response. Moreover, the findings in this study can provide valuable insights and act as a comparative study to other tropical regions such as Indonesia, Thailand, Singapore, Brunei, Vietnam, and Cambodia as these Southeast Asia countries fall under similar climate conditions as Malaysia. Developing uncertainty modelling and bias-reduction algorithms for SPEs before utilizing them in actual scenarios is an important step towards enhancing their performance. Lastly, enhancement of the SPEs' precipitation predicting capability and bias correction of the precipitation data will be considered in future works by implementing machine learning techniques.
ACKNOWLEDGEMENTS
The authors are grateful to the Department of Irrigation and Drainage (DID) Malaysia for providing the rain gauge data used in this study. Also, special thanks to the developers of all SPEs for providing the downloadable data.
FUNDING
This research was supported by the Ministry of Higher Education (MoHE) Malaysia through the Fundamental Research Grant Scheme project (FRGS/1/2023/WAB02/UTAR/02/1) and was partially supported by Malaysia Toray Science Foundation (4417/0005).
AUTHOR CONTRIBUTIONS
R. J. C. and Y. F. H. conceptualized the whole article; V. H. C. and E. Z. X. S. developed the methodology; V. H. C. rendered support in formal analysis; V. H. C. and R. J. C. wrote the original draft; L. L. and Y. F. H. wrote the review and edited the article; R. J. C., L. L., and Y. F. H. supervised the work; R. J. C. rendered support in funding acquisition. All authors have read and agreed to the published version of the manuscript.
DATA AVAILABILITY STATEMENT
Data cannot be made publicly available; readers should contact the corresponding author for details.
CONFLICT OF INTEREST
The authors declare there is no conflict.