**1. Introduction**

Satellite observations have been widely used during recent decades for several meteorological, hydrological and climatological applications incorporating precipitation data worldwide [1–6]. In order to fill in where ground observations are absent or sparse, satellite estimations have been evolving using sophisticated algorithms that can identify rainfall, snow and/or other hydrometeors [7–11]. However, although Satellite Precipitation (SP) products are able to, overall, capture the variability and magnitude of rainfall, still they cannot accurately estimate the localized rainfall variations. Thus, validation of satellite precipitation products is often needed against ground-based measurements.

The Tropical Rainfall Measuring Mission (TRMM) platform placed in orbit during the 1997–2015 period, provided reliable data of high spatial (≈25 km) and temporal resolution (3 h), at a geographical coverage between 50◦ N and 50◦ S [12–17]. TRMM's successor, namely, the Global Precipitation Mission (GPM) has been in orbit since 2014, giving estimates at even higher resolutions (≈10 km; 30 min) and geographical coverage from 60◦ S to 60◦ N, making it available for a variety of applications, including the assimilation of GPM data in numerical weather prediction models to improve model forecasting skill [15,16], the monitoring of severe weather events [13,15–17], hydrological hazards [18,19], etc.

Several studies attempted to demonstrate the accuracy of TRMM and GPM IMERG (Integrated Multi-satellitE Retrievals) estimates in various geographical areas. In their study over mainland China, Wu et al. [20] found that both SP products overestimate light rainfall. This is attributed to the fact that hydrometeors detected by infrared and microwave sensors as well as precipitation radars may partially or even totally evaporate before they are registered by the rain gauges. Furthermore, these authors found an underestimation of moderate and heavy rainfall by both products. A slightly better performance by GPM-IMERG, according to these authors, is attributed to the satellite overpasses and sensor capabilities.

In a similar study over Pakistan, Anjum et al. [21] found a slight dominance of IMERG; however, both products correlate well with the in situ measurements at a monthly scale, adequately following the temporal pattern. Again, underestimation of moderate and heavy rainfall and overestimation of light events was reported.

In their study regarding the area of Singapore, Tan and Duan [22] presented similar results, showing good correlation on a monthly scale, with rain gauges for both products and moderate correlation for daily values. The authors underlined that the better performance of IMERG was not that notable and that the main advantage of the new product was mostly its finer resolution.

In a study over the Tibetan plateau (Hexi region), Wang et al. [23] found a better correlation for IMERG, ascribed mostly to the ability to detect better moderate and heavy rainfall; however, they concluded that the improvement was not significant.

In their study (China, 2015–2017), Chen et al. [24] evaluated the performance of IMERG (v5) and TRMM 3B42 (v7) and found that, at monthly and annual scale, both datasets were highly correlated with rain gauge observations. Considering daily values, satellite estimates overestimate precipitation for intensities within the range 0 to 25 mm/day and underestimate precipitation for light and heavy intensities. Considering various statistical scores, they found that IMERG, in general, performed better in detecting the observed precipitation.

In a similar study (China, March 2014 to February 2017), Wei et al. [25] found severe underestimation with high negative relative biases for both IMERG (v5) and TRMM products. However, IMERG product performed better than TRMM 3B42 in the detection of precipitation events in terms of specific statistical scores (i.e., probability of detection), over China and across most of the sub-regions.

Sunilkumar et al. [26] evaluated the GPM-IMERG (v5) final precipitation product against a ground-based gridded data set over Japan, Nepal and the Philippines for two years (2014–2015). Their results showed generally good performance (in terms of statistical scores, like correlation, mean bias, root mean square error) of GPM-IMERG over three regions, although an underestimation was noticed during heavy rainfall events. They also noticed that GPM-IMERG estimates improved its capability in terms of detecting light and heavy precipitation events, although their performance was found to be seasonally dependent.

A few studies with evaluation of satellite precipitation products over Cyprus are reported in the literature. Retalis et al. [27] performed an analysis of precipitation data from satellite data TRMM 3B43 (versions 7 and 7A) over Cyprus and compared them with the corresponding gauge observations and E-OBS gridded data (i.e., a European daily high-resolution gridded dataset of surface temperature and precipitation to be used for validation of Regional Climate Models and for climate change studies) for a 15-year period (1998–2012). They concluded that correlation between TRMM and E-OBS was higher in summertime (≈0.97), but significantly lower in the winter period (≈0.60–0.74). It was noticed that the annual correlation tends to decrease considerably with time. They also found that the coefficient of determination between TRMM, E-OBS estimates and gauge data were relatively high (0.929 and 0.932, respectively); however, the variations noticed were attributed to the elevation differences.

A study for a 30-year period (1981–2010) for the precipitation database Climate Hazards Group Infrared Precipitation with Station data (CHIRPS) in Cyprus was presented by Katsanos et al. [28]. The CHIRPS database was evaluated against gauge stations data. Results showed good correlation between monthly CHIRPS values and recorded precipitation with the correlation coefficients found

to be around 0.85 and January the month with the highest correlation. The corresponding values for the annual mean ranged between 0.70 and 0.74, with the mountainous stations showing a slightly higher correlation.

In a later study, Katsanos et al. [29] examined the performance of several climatic indices for the CHIRPS precipitation dataset and rain gauges records on high spatial (0.05◦) and temporal (daily) resolution for a period of 30 years (1981–2010). Results indicates quite a promising performance regarding indices related to daily precipitation thresholds, resulting in high correlation scores. However, for indices referring to number of days, results showed medium or no correlation, probably due to the criteria used for the identification of a wet (rainy) day on the CHIRPS dataset.

Furthermore, Retalis et al. [30], in their study on the accuracy of the GPM IMERG estimates over Cyprus (April 2014 to February 2017), concluded that, overall, a very good agreement (based on the statistical analysis) between monthly IMERG estimates and gauge data was established (coefficient of determination r2 value <sup>≈</sup> 0.93), presenting a tendency of IMERG for underestimation when higher elevation (>1000 m) was considered. They also examined the daily dependency of IMERG estimates and gauge data, considering a series of extreme precipitation events, and they concluded that this is case dependency, while elevation does not have an apparent effect.

The objective of this study is to evaluate statistically the performance and improvement of the GPM IMERG product compared to TRMM 3B43V7 estimates, thus exploring, the continuity and uniformity between IMERG and TRMM-era data sets over Cyprus so that they can be used in climate studies as a combined and consistent dataset. The present research is a continuation and extension of previous studies by the same authors.

The current research aims at comparing the two products, namely, GPM IMERG and TRMM 3B43, in order to determine and highlight possible differences, advantages and disadvantages of each one of them, based on the performance of several statistical skill scores, along with cross-evaluation against the dense rain gauge dataset over Cyprus during the period from April 2014 to June 2018.

### **2. Study Area**

Located in the north-eastern corner of the Mediterranean Sea, the island of Cyprus has a typical eastern-Mediterranean climate. The major characteristic of this type of climate can be concisely described by a bimodal seasonality with alternating relatively short wet winters and prolonged dry summers. As can be seen from the geomorphological map in Figure 1, the island is transversed by two mountain ranges: the high Troodos massif in the southwest with the highest peak, Olympus at 1951 m, and the elongated east-west oriented narrow Pentadaktylos range, rising to 900 m which borders the northern coast from east to west. Between the two mountain ranges, lies the central Mesaoria plain. Narrow, relatively flat strips of land surround the island along its coast.

**Figure 1.** Geomorphology map of Cyprus highlighting the distribution of the dense gauge station network grouped according to elevation.

Most of the winter dynamic systems which affect Cyprus originate from the southwest to west [31,32]; hence, the highest average annual precipitation values are recorded on the southern side of the highest peaks of the Troodos mountain and the lowest over the rain-shadowed areas north of Troodos and at coastal stations on the east part of the island [33].
