Annual and Seasonal Scale

The annual averages of precipitation, as estimated from reference and gridded data products (GPPs) over Punjab region, are shown in Figure 2. The temporal trends indicated that the annual precipitation amounts that were underestimated by the GPPs with the exception of TRMM and PERSIANN-CDR products, which overestimated the magnitude as compared with the reference data. The temporal magnitude in all the precipitation products was different and deviate up to 100–200 mm as compared with in-situ measurements. However, the most identical temporal fluctuation pattern of precipitation was observed in GPCC and TRMM products with the relative mean error of −16.59% and 2.88%, respectively. Figure 3 illustrates the spatial distribution of GPPs and observed annual average precipitation. The results indicated that, the spatial configuration of GPPs exhibited the similar pattern of precipitation as compared with observed estimates. Across all of the GPPs, the northern part of study area received more amount of precipitation than southern part and exhibited the precipitation pattern towards south to north gradient with the range of 15–1500 mm. In all GPPs, the spatial consistency was observed with a major deviation in northern Punjab. However, the most accurate and consistent spatial pattern was observed in GPCC and TRMM product and the lowest spatial accuracy was observed in CPC following by UDel and PERSIANN-CCS products particularly in northern Punjab. These products were relatively less accurate to capture the higher amount of precipitation distribution over the northern part of the study region. In contrast, the APHRODITE product showed less precipitation magnitude in the central and southern Punjab as compared with observed precipitation. The spatial distribution of the APHRODITE product showed better results at high catchments of the study region as compared with plain areas. These results are consistent with the findings of [27], who reported the better performance of APHRODITE product as compared with reference data at high-altitude catchments in Pakistan.

**Figure 2.** Mean annual variation in GPPs and reference data.

**Figure 3.** Spatial distribution of annual average of (**A**) GB and (**B**) SB precipitation products during the whole study periods.

The descriptive statistical measures divulge the characteristics of GPPs when compared with the reference data. Figures 4–6 indicates the plots as A (GB) and B (SB) precipitation products against the reference data during annual, winter, and summer monsoon, respectively. The result showed that, all the selected GPPs underestimated the annual precipitation amount, except TRMM and PERSIANN-CDR product, which indicates the overestimation of precipitation over the study region. The highest accuracy was observed in the GPCC and TRMM products with lowest mean error and higher R-squared value with the magnitude of −76.35 (0.90) and 14.26 mm (0.90) respectively. These products are comparatively better to capture the annual precipitation variability over the study region. However, the lowest accuracy was observed in CPC and PERSIANN-CCS products with the magnitude of mean error −163.57 (−34.71%) and−16.08 mm (−3.23%), respectively. These datasets are relatively less accurate for capturing the annual variability amount of precipitation during the whole study period. Moreover, the GSMap and PERSIANN-CCS products overestimated the precipitation magnitude ≤ 500 mm and underestimated the precipitation amounts, which are ≥ 500 mm. The inter-comparison of gridded products indicated a similar consistency, except TRMM and PERSIANN-CDR products, which showed positive bias. The underestimation was more conspicuous in GB products as compared with SB products. The SB products showed lower bias and higher accuracy as compared with GB products, which could be due to higher pixel resolution of SB products over the study region [28]. Overall, TRMM product indicated better quantitative performance and showed reasonable consistency against reference precipitation over the study region.

**Figure 4.** Statistical indicators for the assessment of (**A**) (GB) and (**B**) (SB) precipitation products against reference data (Annual timescale).

The GB and SB precipitation products were further evaluated during winter monsoon season. All of the GB products underestimated the winter precipitation with the highest and lowest accuracy being observed in CRU and CPC product with the estimated mean error of −25.08 (−20.49%) and 37.99 mm (−31.04%), respectively. In contrast, the SB products overestimated the precipitation magnitude except CHIRPS product, which indicates the underestimation of precipitation amount over the study region. The highest and lowest accuracy was observed in TRMM and GSMap products with the estimated mean error of 24.06 (19.39%) and 80.66 mm (65.03%) respectively. The rest of the precipitation products also showed better quantitative agreement with a higher range of deviation from the mean observed precipitation. Overall, the performance of CRU and TRMM were consistent and show relatively less error during the winter monsoon season.

**Figure 5.** Statistical indicators for the assessment of (**A**) (GB) and (**B**) (SB) precipitation products against reference data (winter monsoon).

Furthermore, the evaluation of precipitation products during the summer monsoon season indicated the similar pattern of statistical metrics as assessed during annual time scale. The summer monsoon precipitation dominated the regional water balance and plays a significant role in annual precipitation variability. All of the GB and SB products underestimated the observed precipitation. The highest and lowest accuracy in GB products was observed in GPCC and CPC with the estimated mean error of −50.03 (−16.17%) and −114.50 (−36.48%), respectively. Moreover, the TRMM and GSMap showed the highest and lowest accuracy as compared with in-situ with the estimated mean error of −15.18 (−4.48%) and −136.06 mm (−40.16%), respectively. The products underestimation was more

conspicuous during winter monsoon period as compared with annual and summer monsoon season. Moreover, all of the SB products showed less range of mean error during annual scale following by summer and winter monsoon seasons. However, the GB products exhibited lower mean error in winter following by annual and summer scale. Overall, the performance of GPCC and TRMM products showed the best agreement with in-situ in terms of higher R<sup>2</sup> values of <sup>≥</sup> 0.90 and lower values of RME ≤ 25% during annual, winter, and summer monsoon seasons.

**Figure 6.** Statistical indicators for the assessment of (**A**) (GB) and (**B**) (SB) precipitation products against reference data (summer monsoon).

In order to evaluate the temporal pattern variability of different GPPs against observation data, Taylor diagrams were plotted [62] to quantify the precise agreement between the observation and GPPs in terms of correlation coefficients (CC), standard deviation (SD), and root mean square deviation (RMSD), which are shown in Figure 7 as A (GB) and B (SB) precipitation products during annual, winter, and summer seasons. In the diagram, correlation coefficient (CC) is denoted by blue lines adjoining perpendicular to the parabolic scale, standard deviation (SD) is denoted by radii of the black cycles and root mean square deviation (RMSD) donated by the radii of the green cycles. In the diagram, if the value of the GPPs is closer to the observation data, then it is considered to be a better product. The diagram statistics provide the evaluation of the temporal pattern of the GPPs against observed data.

**Figure 7.** Statistical evaluation of (**A**) (GB) and (**B**) (SB) precipitation products against reference data by using Taylor diagram.

The results indicated that the GPCC and TRMM precipitation products outperformed the other products during the annual scale in terms of higher CC of around 95% and less value of the RMSD. The GPCC and TRMM points marked closer to the reference point indicated that the products relatively performed better and they are suitable for the study region. The rest of the precipitation products also depicted good agreement with observed data over the study region, with notable mean error all the way. These results are in agreement with the findings of [27,72] who reported the better performances of TRMM and GPCC products. The range of agreement is different, which could be due to a difference in region, study period and statistical metrics. Moreover, the Taylor analysis during winter precipitation indicated the better performance of CRU and TRMM products as marked closer to the reference data with relatively higher CC of around 97% and lower values of RMSD. The GPCC, APHRODITE, and CHIRPS products also showed better efficiency in terms of high CC and lower RMSD values. On the other hand, the evaluation of precipitation products during summer monsoon revealed the better performance of GPCC and TRMM products, as indicated by a higher CC of around 98% and lower RMSD values. These results are consistent with the outcomes of products evaluation during the annual time scale over the study region. The efficiency of GPCC and TRMM precipitation products during annual and summer season was found to be similar and consistent. The major influence of annual precipitation variability depends upon the summer monsoon season, as it receives the major amount of precipitation and main driver of annual hydrological cycle [73]. Overall, the performance of GPCC and TRMM products were best during the annual and summer time scale over the study region. However, CRU product shows relatively better performance during the winter monsoon period.

Figure 8 shows the spatial distribution of GB and SB precipitation products against reference data in terms of spatial pattern of statistical indices (ME, RMSD and CC) over the Punjab region during the whole study periods. The results indicated that all of the GB products underestimated the precipitation amounts as compared with observed precipitation. The range of deviation was more conspicuous in northern Punjab with the dominated pattern of underestimation from north to south gradient. In the entire study period, the GPCC product exhibited best agreement, as indicated by the distribution of higher CC and lower values of ME and RMSD. However, the accuracy of GPCC product was more precise over northeast and southern part of Punjab. The inter-comparison of GB products indicated the similar pattern of underestimation, except the UDel and CPC product, which indicated the higher range of underestimation over the whole study region. Both of the products showed the lowest agreement following by APHRODITE product as compared with the observed pattern of precipitation. On the other hand, the SB precipitation products showed the visible pattern of overestimation over the whole study region. However, the pattern was more conspicuous over the northern and eastern part of the study region. Moreover, all of the products showed a similar pattern of underestimation over south-west part of the study region. However, the pattern of underestimation was more noticeable in PERSIANN products over south to south west gradient as compared with observed precipitation. The statistical pattern of all the SB products pointed out the better performance of TRMM product in terms of higher CC and lower ME and RMSD values. The TRMM product outperformed the rest of the product by showing the lowest error and highest accuracy. However, the pattern of overestimation was more noticeable over the northern region of the study area. The overestimation in these areas could be related to the higher rate of evaporation. These results are consistent with the findings of [27], who reported the higher overestimation of satellite products in the foothills of northern mountain range of Hindukush, Karakoram and Himalaya (HKH in Pakistan. Generally, the GPCC and TRMM products showed the better accuracy and consistency in terms of spatial distribution of statistical metrics with higher values of CC and lower values of ME and RMSD.

**Figure 8.** Spatial statistical indicators for the assessment of (**A**) (GB) and (**B**) (SB) precipitation products against reference data.

Overall, the extent of under/over estimation of precipitation by GB and SB products suggests the significance of data correction before their reliable utilization in climate and hydrological studies. For the correction of precipitation products, we analyzed the estimated mean error in different data products and introduce a correction factor of each GB and SB product for annual, winter, and summer monsoon season over the Punjab region. The regional correction factor of each product for annual, winter, and summer monsoon during the period of 1979–2017 and 2003–2017 for GB and SB precipitation products are summarized in Table 2. The correction factor needs to be multiplied with the respective data products in order to minimize the percentage of error over the target region. The corrected product data could be directly used in climate modelling and other relevant studies [27].


**Table 2.** Annual and seasonal correction factors for each GPPs over Punjab region.
