Evaluation of Satellite Rainfall Products over the Mahaweli River Basin in Sri Lanka

The availability of accurate spatiotemporal rainfall data is of utmost importance for reliable predictions from hydroclimatological studies. Challenges and limitations faced due to the absence of dense rain gauge (RG) networks are seen especially in the developing countries. Therefore, alternative rainfall measurements such as satellite rainfall products (SRPs) are used when RG networks are scarce or completely do not exist. Noteworthy, rainfall data retrieved from satellites also possess several uncertainties. Hence, these SRPs should essentially be validated beforehand. The Mahaweli River Basin (MRB), the largest river basin in Sri Lanka, is the heart of the country’s water resources contributing to a significant share of the hydropower production and agricultural sector. Given the importance of the MRB, this study explored the suitability of SRPs as an alternative for RG data for the basin. Daily rainfall data of six types of SRPs were extracted at 14 locations within the MRB. Thereafter, statistical analysis was carried out using continuous and categorical evaluation indices to evaluate the accuracy of SRPs. Nonparametric tests, including the Mann-Kendall and Sen’s slope estimator tests, were used to detect the possibility of trends and the magnitude, respectively. Integrated MultisatellitE Retrievals for Global Precipitation Measurement (IMERG) outperformed among all SRPs, while Precipitation Estimation from Remotely Sensed Information using Artificial Neural Networks (PERSIANN) products showed dire performances. However, IMERG also demonstrated underestimations when compared to RG data. Trend analysis results showcased that the IMERG product agreed more with RG data on monthly and annual time scales while Tropical Rainfall Measurement Mission Multisatellite Precipitation Analysis–3B42 (TRMM-3B42) agreed more on the seasonal scale. Overall, IMERG turned out to be the best alternative among the SRPs analyzed for MRB. However, it was clear that these products possess significant errors which cannot be ignored when using them in hydrological applications. The results of the study will be valuable for many parties including river basin authorities, agriculturists, meteorologists, hydrologists, and many other stakeholders.


Introduction
Water is being regarded as a remarkable substance due to its critical role related to human life. Reduced precipitation amounts can lead to prolonged droughts while excess precipitation rates can result in ooding. However, both extreme conditions disrupt the socioeconomic activities of people and harm the environment adversely. Water plays an important role in a country's economy in terms of irrigated agriculture, hydropower generation, provision of drinking water, and industrial use [1]. is marks the importance of proper water management for both social and economic bene ts for a country. For proper management of water, accurate predictions from hydrological modeling are required. Hence, the role of spatially distributed precipitation data is crucial in this context [2].
Precipitation and evapotranspiration are the major components of the hydrologic cycle, and both these components carry signi cant shares of the hydrologic cycle [3]. Until the early 1980s, the only method used to obtain these precipitation data was rain gauges (RGs) located across the world [4]. Although RG data are the most accurate method of measuring rainfall [5], there are significant challenges faced when maintaining rain gauge networks. ese challenges are due to complex geographical features, harsh climatological environments, and high financial burdens [6]. Some of the reported shortcomings of RG data are missing rainfall days, data transmission dropouts, calibration errors of the RGs, and local wind effects affecting rainfall level estimations [7,8]. Incidences during the period of war in countries also lead to collapsing of these rain gauges which ultimately resulted in no rainfall records in certain regions of the world [9].
With the continuous evolvement of science and technology, currently, several other methods are available to obtain rainfall data. ey are radars, satellites, and reanalysis products. However, high costs for installment and maintenance of radar networks are unbearable for the developing world to adopt radar networks [6]. On the other hand, reanalysis products have significant biases [10]. erefore, the satellite rainfall product (SRP) data have grabbed the attention of many researchers due to its ability to provide spatiotemporal rainfall data addressing most of the shortcomings faced when using RGs. e SRPs are attractive since they are available at no cost. In addition, the SRPs are available from fine spatial and temporal scales such as 0.10°a nd 30 minutes. For instance, both the near-real-time product of GSMaP and IMERG provides 30-minute resolution while the IMERG product provides rainfall in 0.10°r esolution [11]. ere are many SRPs available and most of them have been used in similar research studies as well. Some of them are Tropical Rainfall Measurement Mission (TRMM), Multisatellite Precipitation Analysis (TMPA) [12], Climate Prediction Center Morphing (CMORPH) [13], Precipitation Estimation from Remotely Sensed Information using Artificial Neural Networks (PERSIANN) [14], and Multisource Weighted Ensemble Precipitation (MSWEP) [15]. But these products have been found to possess inherent systematic uncertainties and sampling errors which need to be clearly studied and identified before using them in hydrological applications [16,17]. However, for flood and drought predictions, understanding global climate change, and determining the availability of water resources for agriculture and industrial purposes, the climatological data needs to be accurate. erefore, comparing SRPs by benchmarking with RGs is essential to obtain a sound conclusion on their levels of uncertainty and thereafter to determine their suitability for many practical applications.
Following this requirement, many research studies have been conducted around the world to assess the accuracy and reliability of SRPs with respect to RG data [6,[18][19][20]. Many studies have demonstrated that TRMM (TMPA) is comparatively better than other SRPs since it used available gauge measurements in its calibration process [21,22]. Higher efficiency [23] and higher accuracy [24] shown by TRMM 3B42 in research studies had proven its accuracy compared to its near-real-time product (TRMM B42RT). Research studies incorporating CMORPH had observed that it shows dependencies on rainfall estimates [25]. GPM (IMERG) when compared with TMPA in a study conducted in India showed better performance in capturing heavy rainfall events [26]. A study in Ethiopia that focused on CMORPH and PERSIANN-CCS discovered that almost 50% of underestimations can be expected from these SRPs and is cautious to be directly used in flood prediction models [16]. GSMaP-MVK data was proved to produce overestimates when compared with gauge data in a study conducted in Indonesia [27]. Although in the developed world, many studies have been carried out to examine the efficiency of SRPs as an alternative for RGs, the developing region has been least focused in this regard. In the context of Sri Lanka, only a few studies have been carried out to examine the efficiency of SRPs. A previous study was carried out for the Mundeni Aru Basin, Sri Lanka, to evaluate the applicability of SRPs in flood hazard mapping [28]. Two other studies attempted to determine the accuracy of SRPs in streamflow simulation for the Seethawaka watershed, Sri Lanka, using a hydrological model [29,30].
e Mahaweli River Basin (MRB) in Sri Lanka covers approximately one-sixth of the landmass in the country ( Figure 1). e MRB is imperative due to its major role played in hydropower generation and irrigation water supply for many parts of the country. e MRB is the main water supplier for agricultural activities in the eastern dry zone of the country irrigating more than 1,000 km 2 of land. e hydroelectricity produced from six dams of the MRB ( Figure 1) supplies more than 40% of the country's electricity. More importantly, all three climatic zones of Sri Lanka fall within the MRB [31,32]. Despite the vital importance of the MRB to Sri Lanka, the amount of research carried out in this study region is handful. Previously, many studies have been carried out in the subbasins of the MRB and only a very few have focused on the entire MRB. Some of the disciplines which investigated the entire MRB were hydrologic modeling [33], climate change impacts on paddy cultivation in dry zones in the MRB [34], assessing the streamflow variability and rainfall trend of climatic zones [35], El-Nino and La-Nina impact on streamflow [31], drought occurrence and atmospheric circulation [36], and projection of future climate [37]. Other research studies which focused on the subbasins of the MRB investigate the prediction of streamflow in ungauged catchments [38] and flood modeling [39]. Additionally, the Uma-Oya subbasin of the MRB has been an area of great interest for soil erosion. Some of the studies carried out in the Uma-Oya watershed focused on the assessment of the impact of siltation on hydropower generation [40], soil erosion studies [41,42], land-use change impact on landslides [43], and land-use change impacts on river health [44,45]. Several field-scale studies have also been conducted to investigate water pollution [46,47] and Chronic Kidney Disease (CKD) [32]. e present study investigates the suitability of SRPs by comparing them with RGs and examines trends of SRPs and RG in the MRB of Sri Lanka. e accuracy of SRPs is evaluated against RGs through continuous evaluation and categorical indices. In addition, the trend analysis of precipitation products was carried out using the Mann-Kendall test and Sen's slope methods. e current study is the first of its kind which examines the efficiency of SRPs over the MRB. Hence, the present study carries a significant novelty to the research community, opening doors for the use of SRPs in the absence of RG data in the MRB. Stakeholders of the MRB including river basin authorities, agriculturists, meteorologists, hydrologists, and stakeholders (including government authorities and farmers) will be benefited from the results of the study.

Study Area.
e MRB is the longest river in Sri Lanka extending to 335 km. e Mahaweli River starts from the SriPada Mountains in the central hills. e drainage area of the MRB is 10,448 km 2 , covering one-sixth of the landmass in the country. Figure 1(a) illustrates the elevation map of MRB obtained by extracting Digital Elevation Models (DEMs) from the Shuttle Radar Topography Mission (SRTM) with a resolution of 30 m × 30 m. Depending on the climatic and geographical distribution of annual rainfall amount, three climatic regions (wet, intermediate, and dry zones) fall in the MRB (Figure 1(b)) [36]. e annual rainfall in the country is influenced by four monsoon seasons of Southwest Monsoon (SWM) from May to September, Northeast Monsoon (NEM) from December to February, first intermonsoon (FIM) during March and April, and second intermonsoon (SIM) during October and November. Having the wet zone of the river basin in the southwest part of the country and the dry zone in the northeast part, the annual rainfall on the MRB is dependent on all these four seasons with the main ones being SWM and NEM [48]. e river basin in three climatic zones can be clearly seen in Figure 2(a). e annual rainfall in this basin is estimated to be 2500 mm of which 900 mm is discharged back to the sea. is produces an annual average discharge of 8.4 billion m 3 in the basin [49]. Tea estates and forest areas can abundantly be seen upstream of the MRB, whereas cultivated areas are frequent downstream of MRB. e landuse types of MRB obtained from the Department of Surveys, Sri Lanka, with a resolution of 1 km × 1 km are given in Figure 1(c). erefore, as illustrated in Figure 1(c), the upper catchment of the MRB is mainly occupied by agricultural lands (mostly tea cultivation) while the lower catchment is by forest cover.

Rain Gauge Data.
For the analysis of this study, based on the availability of rainfall data, 14 rain gauge stations in the wet, the intermediate, and the dry zones of the MRB were obtained from the Department of Meteorology, Sri Lanka. e selected 14 rain gauge stations with their location and period are listed in Table 1. e missing data percentage in the rain gauge stations was below 5%; therefore, rainfall data from neighboring stations were used to fill the missing values.
e average rain gauge density in the Mahaweli catchment is 1.34 gauges per 1000 km 2 . is is a significantly high value when compared with several other studies. Mu et al. [50] had an average rain gauge density of 0.53 gauges per 1000 km 2 , Paca et al. [51] Figure 3.
TRMM-3B42 and 3B42RT: TRMM 3B42/3B42RT satellite products were produced through a combined mission between the National Aeronautics and Space Association (NASA) and the Japan Aerospace Exploration Agency (JAXA) with the finest temporal resolution of 3 hours and a spatial resolution of 0.25°× 0.25°. TRMM 3B42 version 7 product is a research-grade satellite product and is one out of the two products of TRMM 3B42 retrievals. TRMM Combined Instrument (TCI) estimate employing TRMM Microwave Imager (TMI) and the Precipitation Radar (PR)  data are the sources used to estimate precipitation through the TRMM satellite [12]. e spatial coverage of TRMM is between 50°N and 50°S [53]. is rainfall data is produced after two months from the observation time and is called a post-real-time product. e combined use of microwaves, infrared waves, and ground observed rainfall measurements makes it a better product in estimating rainfalls overall [19]. TRMM 3B42RT version 7 is a near-real-time product which is the remaining product of TRMM 3B42 retrievals. IMERG: Integrated MultisatellitE Retrievals for Global Precipitation Measurements (IMERG) is another satellite product developed by NASA. e recently released version (V06 B) of IMERG Final Run was used in this study. is product has a spatial resolution of 0.10°× 0.10°and spatial coverage of 50°N-50°S. Moreover, the finest temporal resolution is 30 minutes. IMERG product is based on low-earth orbit satellites and geostationary satellites together with high spatiotemporal observed rainfall data. e algorithm used to produce IMERG is intended to intercalibrate, merge, and interpolate "all" satellite microwave precipitation estimates, together with microwave-calibrated infrared satellite estimates, precipitation gauge analyses, and potentially other precipitation estimators. e V06B product is considered a huge improvement over the superseded versions of IMERG since it was updated further in data processing, algorithms, and verification which effectively enhanced its detection accuracy [54,55].
PERSIANN: Artificial Neural Network (ANN) algorithm is used in Precipitation Estimation from Remotely Sensed Information using Artificial Neural Networks (PERSIANN) to estimate rainfall. In order to estimate precipitation through PERSIANN IR brightness temperature data from global geostationary satellites is obtained from Climate Prediction Center (CPC) and National Oceanic and Atmospheric Administration (NOAA). is product is available from 2000 to the present and the finest temporal coverage of 1 hour is available. e spatial resolution of the product is 0.25°with a spatial coverage of 60°N-60°S. In order to account for the high uncertainties in associating the relationships between precipitation and cloud-top brightness temperature due to cloud properties and atmospheric conditions, the ANN parameters are updated with rainfall estimates from low-orbital satellites whenever independent estimates of rainfall are available [14,56].
PERSIANN-CCS: PERSIANN-Cloud Classification System (PERSIANN-CCS) is a real-time global high-resolution SRP. e ANN algorithm used in PERSIANN-CCS extracts local and regional cloud features from infrared geostationary satellite imagery. e information is extracted from the whole cloud patch and provides multiple infrared brightness temperature versus rainfall rate relationships for different cloud classification types enabling this product to generate variable rain rates at a given brightness temperature and variable rain/no-rain IR thresholds for different cloud types   [57,58]. is product has a spatial resolution of 0.04°× 0.04°a nd the finest temporal frequency of 1 hour is available. e PERSIANN-CCS data is available from 2003 to the present. PERSIANN-CDR: PERSIANN-Climate Data Record (PERSIANN-CDR) developed using the PERSIANN algorithm was developed from the National Climatic Data (NCDC) CDR program of the National Oceanic and Atmospheric Administration (NOAA) [59]. e finest temporal resolution of this product is 1 day with a 60°N-60°S spatial coverage and a 0.25°× 0.25°spatial resolution. Infrared brightness temperature CDR from Gridded Satellite (GridSat)-B1 of the International Satellite Cloud Climatology Project (ISCCP) is used as the input used to train the neural network of the PERSIANN model [19]. e PER-SIANN-CDR data obtained were then bias-corrected using 2.5°monthly Global Precipitation Climatology Project (GPCP) data. is SRP was mainly developed with the aim of providing a long-term, high-resolution, and global precipitation dataset to be used in studies to determine changes and trends in daily precipitation, especially extreme precipitation events, due to climate change and natural variability [60].

Data Extraction.
e six satellite products used in this study were extracted in different methods. PERSIANN group of products were directly obtained from the CHRS data portal as CSV files. IMERG and TRMM products were obtained as NetCDF (Network Common Data Form) files from the NASA GESDISC portal. Afterward, IMERG was extracted through the process of merging the files in Climate Data Operator (CDO) followed by the extraction using R coding in RStudio. e TRMM products were merged using a similar approach as in IMERG, but the extraction of the point rainfall data was done using MATLAB 9.6.

Overall Methodology.
Statistical indices were used to determine the accuracy of the satellite data sets with respect  to RG data. Nonparametric tests were used to analyze the trend and to find the magnitude of the trends observed.

Categorical Evaluation Indices.
To examine the detection accuracy of SRPs, there are four categorical indices, namely, the False Alarm Ratio (FAR), the Critical Success Index (CSI), the Probability of Detection (POD), and the Proportion Correct (PC). FAR gives the fraction of times the satellite detects rainfall events that did not actually occur. POD provides a quantitative measure of the number of times the satellite will detect rainfall events accurately. CSI provides the fraction of times the rain gauge and/or SRP data were correctly predicted and PC provides a quantitative measure of the accuracy of the detected rainfall [61]. e equations (1)-(4) associated with each of the indices are as given below. False Alarm Ratio (FAR) Probability of Detection (POD) Critical Success Index (CSI) Proportion Correct (PC) where H represents the rainfall accurately detected (correct hits), M represents the rainfall not detected (missed data), F represents the precipitation that was falsely detected (false alarms), C represents the correct negatives, and T F , T H , T M , and T F are the number of times each of the cases occurred, respectively. e analysis using categorical indices was done for 1 mm/day and 10 mm/day thresholds representing light and heavy rainfall, respectively. e threshold to represent light and heavy rainfall was decided based on Table 3 [62] and the contingency table used for the analysis is shown in Table 4.
If the rainfall threshold � x mm, the categorical indices were calculated based on the following conditions.

Continuous Evaluation Indices.
Continuous evaluation indices were used to evaluate the performance of SRP with respect to RG data. To find out the extent to which these two datasets agree with each other (correlation), Pearson's correlation coefficient (CC) was used. To determine the absolute average magnitude of the error between the two datasets, Mean Absolute Error (MAE) was used. Root Mean Square Error (RMSE) which gives more relevance to the larger errors when compared with MAE was used again to determine the absolute average error magnitude. To determine the degree of overall underestimation or overestimation, Relative Bias (RB) was used [61,64,65]. Finally, to measure the goodness of fit between the observed and SREs which was initially developed by Gupta et al. [66] and further modified by Kling et al. [67], Kling Gupta Efficiency (KGE) was used. KGE is less sensitive to extreme rainfall and therefore can interpret the overall fitness of rainfall having different intensities [68]. Equations (5)-(9) associated with each of the indices are as given below.
where S i , S, and S d (for SRP data) are the i th station, mean values of SRP data, and standard deviation of SRP data, respectively, G i , G, and G d (for Rain gauge data) are the i th station, mean values of gauge data, and standard deviation of gauge data, respectively, and n represents the total number of data considered. e strength range of correlation coefficient (CC) was defined as interpreted by e Political Sciences Department at Quinnipiac University [69] and is given in Table 5. Intensity class of rainfall Rainfall thresholds per day No/tiny rainfall P < 1 mm Light rainfall 1 mm ≤ P < 2 mm Low moderate rainfall 2 mm ≤ P < 5 mm High moderate rainfall 5 mm ≤ P < 10 mm Heavy rainfall P ≥ 10 mm where sgn( An increasing trend will be the outcome if S is having a very high positive value and decreasing trend if S is having a very low negative value. To compute the probability associated with the calculated S and the sample size to obtain a significance of the trend, [71] describes a normal approximation test incorporating the Mann-Kendall Statistic, S. A normalized test statistic Z is computed along with the probability associated with the Z value, f (z).
where Var(S) � (n(n − 1)(2n + 5) − t i�1 t i (i)(i − 1)(2i + 5))/18, where n is the number of datasets, t is the number of tied groups, and t i is the number of datasets in the i th group.
e probability density function f (z) for a normal distribution with a mean of 0 and a standard deviation of 1 is given by Taking a 95% significance level, the trend will be determined to be decreasing if Z is negative and the probability is greater than 0.95. Similarly, the trend will be determined to be increasing if Z is positive and the probability is greater than 0.95. If the probability is lesser than 0.95, then it was concluded that no trend is present in the datasets [72].

eil-Sens Slope Estimator Test.
To quantify and obtain a magnitude of the trends observed from the MK test, Sens Slope Estimator was used [73]. Since the data sets used correspond to the same time intervals and upon arranging the data sets in ascending order with time, the slope of each time series data pair was calculated using the following equation: where j > k, X corresponds to a data value at a j/k time and k � 1, 2, . . . N Upon arranging the N values in ascending order, the median value of the Sens Slope Q i will be calculated using the equation given below:

Detection of Light Rainfall Events.
e mean values calculated for each satellite product after the calculation of categorical indices with the 14 rainfall gauge stations yielded the following results. Figure 4 shows the graphical representation of the obtained results. e threshold used for the light rainfall representation was 1 mm/day. e best performance in terms of FAR was shown by TRMM-3B42 and TRMM-3B42RT products with a mean value of 0.40 and a stationwise variation of 0.01 to 0.54 while the worst performance was shown by PERSIANN-CDR with a mean of 0.54 and a stationwise variation of 0.01 and 0.75. With respect to POD and CSI, the best performance was shown by the IMERG rainfall with a mean of 0.79 and 0.47, respectively, as further proved by Moazami et al. [74]. Stationwise distribution of these indices was 0.47 to 0.88 in POD and 0.37 to 0.57 in CSI. e worst performance in POD was represented by TRMM-3B42 with a mean of 0.66, and in CSI, PERSIANN-CCS and CDR showed the worst performances giving a mean of 0.39. TRMM-3B42 performed best in PC with a mean of 0.73 and stationwise variation of 0.41 to 0.84 while PERSIANN-CDR showed the worst performance with a mean value of 0.62 and stationwise variation of 0.55 and 0.68. From the stationwise mean values, IMERG turned out to be the best, having high performances in both the number of times of accurate rainfall detection (POD) and correct rainfall prediction (CSI). PERSIANN-CDR proved to be the worst by performing the worst rainfall detection (PC), the number of times of correct rainfall prediction (CSI), and the highest fraction of false rainfall ratio (FAR). When comparing the categorical indices results with the climatic zone distribution of stations, IMERG stood out in

Detection of Heavy Rainfall Events.
e results for categorical indices for each satellite product with a 10 mm/ day threshold to represent heavy rainfall are summarized below. Figure 5 shows the graphical representation of the obtained results.
Best performance in FAR, POD, and CSI was shown by IMERG product with a mean value of 0.47, 0.5, and 0.35, respectively, which were in consistent with the range of values obtained in a study done in China incorporating different IMERG products [75]. e stationwise variation   CCS also proved to be the worst by performing relatively badly in both accurate rainfall detection (PC) and the number of times of correct rainfall prediction (CSI). When comparing the climatic zone distribution of stations with the categorical indices, IMERG stood out in all four indices having higher mean values in dry zones. PERSIANN-CCS performed worst in both POD and CSI in wet zones and in FAR in dry zones. Likewise, PERSIANN-CDR also poorly performed in FAR and POD in dry and wet zones, respectively.
After comparing all the results obtained from light and heavy rainfall detection, it was clear that IMERG outstands the highest with a value of 0.9 in PC which is the highest out of all mean values recorded. It showed this higher accuracy mainly in the dry zones of the river basin. A study carried out on the performance evaluation of SRPs over varying climates and complex topographies has also proved this finding with the IMERG product [76]. PERSIANN-CCS and CDR were found to outstand the least having mean values in the range of 0 to 0.4 in both POD and CSI. is was further proved by Gadouali et al. [60], who performed a study on SRPs in Morocco. His results showed that PERSIANN-CDR was worst at detecting rainfalls. Another study which was performed on the evaluation and comparison of SRPs in Burkina Faso, West Africa, also proved the same with PERSIANN products [77].

Continuous Evaluation Indices.
e correlation between the observed and satellite rainfall data was observed for both daily and monthly time scales. In both time scales, all CCs were in the range of 0 < CC < 1, showing an overall positive correlation in all products. Out of the six satellite products, the highest median of 0.6 was observed in the IMERG product which interpreted a strong correlation. All the other products showed relatively poor correlation within the range of 0 < CC < 0.65. In all 6 products, higher CC was observed mainly in dry zones while the lowest was identified in wet zones of the river basin. Studies conducted by Amjad et al. [76] in Turkey and Alijanian et al. [56] in Iran also showed that lower CC values are obtained for wet regions. In the monthly time scale analysis, the highest median of 0.84 was shown by IMERG and TRMM-3B42 products. e highest positive values in both time scales were recorded from the dry zones of the river basin and the lower values from the wet zones. e order from strong positive to strong negative correlations was from IMERG and TRMM products to PERSIANN products in both time scales.
KGE shows a perfect fit when the value is close to 1 [68]. In both the daily and monthly analysis, IMERG showed the highest median value of 0.47 and 0.69 for daily and monthly scales, respectively. e worst fit of data was exhibited by PERSIANN-CDR on each time scale. All products showed relatively poor performance for daily data with all values less than 0.75. Most of the higher KGE values were observed in the dry zone and lower values were seen in the wet zones on the daily time scale. e goodness of fit within datasets was best with IMERG followed by TRMM products and PERSIANN products where the worst fit was observed. However, a better performance was observed by monthly values when compared with daily values.
Overall, both good correlation (CC) and fit between the datasets (KGE) were shown by the IMERG product further being proved by Adane et al. [78] from a similar study done in Northeastern Ethiopia. Poor correlation was mainly showcased by PERSIANN-CDR product.
is poor performance can be a result of cloud-top IR observations as this product is mainly based on [15]. e graphical representation of the zonewise distribution of the results for the daily time scale from CC and KGE is shown in Figure 6 and the results of the same in monthly time scale are provided in Figure 7.
Both daily and monthly analyses with MAE, RMSE, and RB showed similar behavior (Figures 8-10  the intermediate zone and lower errors in the dry zone of the river basin. is was further confirmed by a study done in Pakistan which stated that higher RMSE is prominent in regions having higher average annual rainfall [79]. e highest underestimation was shown by PERSIANN-CDR while the highest overestimation was from PERSIANN-CCS with medians of −7.55% and 7.79%, respectively. However, in monthly analysis, the highest underestimations were observed in PERSIANN-CDR and the highest overestimations were in PERSIANN product. All products showed higher underestimations in the intermediate zone and higher overestimations in the dry zone of the basin in both analyses. Overall, it was noted that all SRPs are better at capturing rainfalls accurately with very less errors in the dry zone of the basin while the errors increased when moving from wet to the intermediate zone. IMERG was the best with very less errors in capturing accurate rainfalls when intercompared with the RG data. However, IMERG causes underestimations of rainfall data (median RB of −4.83% in both daily and monthly time scales). Comparatively, given that the IMERG product is set aside, TRMM products also showed better performance and PERSIANN products showed relatively the worst performances in all indices. e better performances of IMERG and TRMM products were further proved by Anjum et al. [80], who found that these products perform well with reference to the RG data at a monthly time scale. Also, overestimations were prominent only in PERISANN products while all other products produced underestimations.
From the performance shown in all the continuous evaluation indices, all products cannot efficiently reproduce the temporal variability in the observed rainfall gauge data on both daily and monthly time scales. Among the six evaluated products, IMERG followed by TRMM products performed well and PERSIANN products performed the worst in each index. Yang et al. [81] had yielded results similar to the present study. eir findings also showed that IMERG V06 and TRMM-3B42 V07 were in the best overall agreement with RG data in all temporal scales. Likewise, the categorical evaluation of these products is similar to what was observed from continuous evaluation indices. IMERG stood out here as well. e final run IMERG product, which is the product used in this study, has proved its higher performance in the Middle East as well [82]. It was also clear that these products possess significant errors which cannot be ignored when using them in hydrological applications. An approach for error correction would be required beforehand. Figures 8-10 show the box plot representation of the results for MAE, RMSE, and RB in both daily and monthly analyses, respectively.

Trend Analysis.
Mann-Kendall trend test was used to identify any significant trends in the observed data and SRPs. en, to quantify the trends obtained, Sen's slope estimator was used on the datasets.  e MK test that was performed for monthly, seasonal, and annual time scales of the observed rainfall gauge data showed significant increasing trends mostly in the intermediate zone of the river basin. Increasing trends in the seasonal analysis were mostly in the second intermonsoon season. In the annual and monthly analysis, the wet zone also showed significant increasing trends similar to the finding of Pawar and Rathnayake [83]. IMERG product agreed more with the trends observed in the rainfall gauge data in the monthly and annual time scales. TRMM-3B42 showed significant increasing trends during the second intermonsoon in the intermediate zone. In TRMM-3B42 and PERSIANN-CDR, from the monthly analysis, negative significant trends (dry zone) were mostly observed. In all three time scales, PERSIANN-CCS showed increasing trends in the wet zone. However, PERSIANN-CDR and TRMM-3B42RT showed mixed results while PERSIANN showed no significant trends in all three time scales.

Sen's Slope Estimator Test.
Sen's slope for the trends obtained from the MK test indicated a slope >1.5 mm/month in the monthly analysis for the observed data. For the annual and seasonal analysis, the slopes were >20 mm/year and >9 mm/season, respectively. In the monthly analysis, IMERG indicated an increasing trend of >6 mm/month. High intensity increasing trends of >30 mm/year in the annual time scale were observed in IMERG and both TRMM   Figure 10 demonstrates the MK and Sen's slope estimator results obtained for the annual time scale. e results of the significant trends in the monthly and seasonal analysis are attached in Tables 6 and 7.
From the nonparametric analysis, the results concluded that the IMERG product agrees more with RG data in monthly and annual time scales while TRMM-3B42 agrees more in the seasonal analysis. ese findings agree with a recent study by Hussein et al. [84]. erefore, depending on these results, a careful choice of products for the different zones in the river basin is required. is showed that even in the same river basin, products behave differently with trend patterns depending on the climatic seasons and zones of the river basin.

Conclusions
In this research study, six SRPs (PERSIANN, PERSINN-CCS, PERSIANN-CDR, IMERG, TRMM-3B42, and TRMM-3B42RT) were evaluated against rainfall gauge data. Observed data at 14 locations spatially distributed in the three climatic zones of the MRB, Sri Lanka, were selected. Four categorical indices, FAR, POD, CSI, and PC, were used to determine the accuracy of rainfall detection and prediction of satellite products during light and heavy rainfall. IMERG product showed better performance while PERSIANN-CDR showed the worst performance in detecting and predicting rainfall during both these rainfall events. e accuracy of the SRPs was also evaluated using five continuous evaluation indices, CC, RMSE, MAE, RB, and KGE. Among the six evaluated products, in general, IMERG showed better performance while PERSIANN products showed poor performance. However, IMERG also caused underestimations of rainfall data. From the performance shown in all the continuous evaluation indices, all products cannot efficiently reproduce the temporal variability of RG data in both daily and monthly time scales. From the nonparametric tests done on the two datasets SRP and observed rainfall data to identify any significant trends, it was concluded that the IMERG product agrees more with observed rainfall data in monthly and annual time scales while TRMM-3B42 agrees more on the seasonal scale. In all three time scales, PERSIANN-CCS showed increasing trends in the wet zone. However, altogether, PER-SIANN-CDR and TRMM-3B42RT showed mixed results while PERSIANN showed no significant trends in all three time scales. It can be concluded that a careful selection of global precipitation products is required prior to using them in any application. is showed that even in the same river basin, products behave differently with trend patterns depending on the climatic seasons and zones of the river basin. It was also clear that these products possess significant errors which cannot be ignored when using them in hydrological applications. However, in places of scarce rainfall data in the MRB, IMERG product proved to be a better choice overall. is study being the first of a kind that incorporated SRPs and observed rainfall data for the Mahaweli River Basin, Sri Lanka, is an immense contribution to many stakeholders and the research community as this basin carries significant importance to the country. is research study was subjected to limitations with the use of six research-based SRPs and 14 rain gauge stations. erefore, further, it is recommended to carry out studies incorporating near-real-time products with more rain gauge stations to avoid point-pixel errors.
Data Availability e climatic data used in this research study are available upon request for research purposes.

Conflicts of Interest
e authors declare no conflicts of interest.