**4. Discussion**

The performance of SM2RAIN-CCI (2000–2015) and SM2RAIN-ASCAT (2007–2015) precipitation products against the RGs observations (obtained from PMD and WAPDA departments of Pakistan) was assessed using six statistical and four categorical metrics. Both precipitation products were evaluated over the complex topography and diverse climate of Pakistan, and their performance was also compared with TRMM TMPA 3B42v7 (TMPA) during the period from 2007 to 2015. SM2RAIN-CCI precipitation observations were not available over snow cover, frozen soil, high topographical regions, and rainforests because of the mask used to remove the areas characterized by issues in SM retrieval [45] (shown in Figure 3). Inputs for SM2RAIN-CCI are based on integrating the active and passive ESA CCI SM datasets with bias-corrected and calibrated observations from GPCC-FDD [37]. However, SM2RAIN-ASCAT observations are obtained from application of the SM2RAIN algorithm to ASCAT SM data without consideration of any filter [32]. The observations are calibrated and bias-corrected based on the ERA5 reanalysis data [43].

Soil moisture-based precipitation products, i.e., SM2RAIN-CCI and SM2RAIN-ASCAT, captured the precipitation both spatially and temporally relatively well. However, both of them tended to fail to estimate the precipitation in the humid region characterized with high precipitation intensity and heavy magnitude (>700 mm), and more specifically in the hydraulically developed region with dams, barrages, headworks, and an extensively developed irrigation canal system (Figures 3 and 4). The performance of SM2RAIN-based products is strongly influenced by spatial and temporal variation of precipitation. Comparatively better performance is observed in the arid and hyper-arid region, which is characterized by low precipitation magnitude (<100 mm). Furthermore, the temporal (seasonal) evaluation also revealed poor performance in monsoon and pre-monsoon seasons, having high precipitation intensity and magnitude. References [9,30] confirmed the poor performance of SM2RAIN algorithm that is unable to adequately estimate precipitation when the soil is close to saturation because the algorithm is unable to derive SM variation when the SM is constant. Similarly, references [41,42] evaluated the performance of SM2RAIN-CCI over north-eastern Brazil and concluded its poor performance in wetter precipitation regimes.

SM2RAIN-CCI and SM2RAIN-ASCAT highly overestimated and underestimated the precipitation in the coastal hyper-arid and mountainous regions of Pakistan, respectively (Figures 3a and 4a). This can be considered as SM retrieval uncertainties, which significantly deteriorate the accuracy of precipitation estimates from the SM2RAIN algorithm (i.e., due to error propagation) [43]. Underestimation of precipitation is associated to poor infiltration capacity and soil moisture storage capacities of rocks (hills), permanent snow and glacier cover, hydraulic developmental activities, and intense and heavy precipitation during monsoon and pre-monsoon seasons, while overestimation is associated to low precipitation intensity and magnitude.

Though SM2RAIN-ASCAT underestimated the precipitation significantly in humid and glacial regions, the locations of heavy rainfall were relatively accurately detected as compared to SM2RAIN-CCI (Figures 3 and 4). Larger biases and uncertainties at those locations are due to the impact of complex topography and diverse climate, which have been well documented in previous literature [47,59,67,68]. Higher positive or negative biases for SM2RAIN-based products are expected over densely vegetated and forest regions (glacial and hilly areas of the humid region) where signals of satellite sensors do not breach into dense vegetation [69], or in the region where the soil remains saturated for longer period of time such as flooded regions (the humid region, especially the hydraulically developed areas) [9], respectively. Considering other statistical metrics, all the metrics showed poor performance in precipitation dominated regions and seasons (Figures 3 and 4, Tables 4 and 5). References [9,41] confirm the findings of the current study.

Concerning the error components, the analysis revealed that SM2RAIN-CCI and SM2RAIN-ASCAT were similar in the spatial distribution of systematic and random errors (Figure 3d,e and Figure 4d,e). The results revealed that systematic error contributed significantly in comparison with random errors, supporting the findings of previous studies conducted by Prakash [6] and Paredes-Trejo, Barbosa and dos Santos [41] in India and Brazil, respectively. Therefore, these products require refinement and corrections before their integration into a particular hydrological application [70].

Seasonal evaluation of SM2RAIN-CCI and SM2RAIN-ASCAT presented a better performance in moderate to low precipitation seasons, i.e., post-monsoon and winter seasons (Tables 4 and 5). During these two seasons, precipitation has had enough time to infiltrate into the soil and then saturate; therefore, the precipitation estimation capabilities of SM2RAIN-based precipitation products are high and hence produce relatively accurate results. However, precipitation in summer (monsoon season), which is distinguished by intense and short duration precipitation events, is the influential factor that affected the performance of both SM2RAIN-based products in terms of biases. Therefore, the possibility

exists that the sensors might have not accurately estimated (or missed) the precipitation during the period [41].

The TMPA precipitation product performed reasonably well across different climate regions when compared with SM2RAIN-CCI and SM2RAIN-ASCAT (Figure 5). A different trend is observed in the humid and arid regions where SM2RAIN-based products overestimate and underestimates the precipitation; TMPA overestimates the precipitation. In contrast, in the hyper-arid region, TMPA underestimates while SM2RAIN-based product overestimates the precipitation. SM2RAIN-CCI depicted poor performance, implying high uncertainties when compared to SM2RAIN-ASCAT and TMPA across all three climate regions. TMPA showed comparatively good agreement with gauged based observations in different climate regions and different seasons relative to the SM2RAIN-based products.

The performance of SM2RAIN-based products across glacial, humid, arid and hyper-arid regions during the common period (2007-2015) are summarized in Tables 7 and 8.


**Table 7.** Summary of the performance of SM2RAIN-CCI datasets across the four climate regions of Pakistan during 2007–2015.


**Table 8.** Summary of the performance of SM2RAIN-ASCAT dataset across four climate regions of Pakistan during 2007–2015.

Based on spatial and temporal evaluation of SM2RAIN-CCI/ASCAT precipitation products across Pakistan along different climate regions, and the obtained results, the following recommendations are suggested for further studies and applications: (1) Agricultural water management and irrigation scheduling in the arid region (the Punjab province, which is agricultural hub of the country), where SM2RAIN-based PPs performed comparatively better, (2) An early flood warning system (EFWS) and flood simulation, where soil moisture (besides the precipitation intensity) plays important role, and bias-corrected SM2RAIN-based products will be helpful for reducing the impact of flood on society, (3) Drought monitoring in drought-prone arid and hyper-arid regions, including hydrological drought (shortage in water storage as well as net precipitation at same time) and meteorological drought (shortage in the catchment's water fluxes such as precipitation), (4) Vegetation and crop growth monitoring, and (5) Groundwater modeling and rainwater harvesting studies.
