In this study, CA extending from 35° N to 55° N and from 45° E to 87° E, and covers five countries: Kazakhstan, Kyrgyzstan, Tajikistan, Uzbekistan, and Turkmenistan (Figure 1). The geomorphological features include high mountains and flat basins, with the elevations ranging from −132 m to about 7,500 m. Overall, CA is geographically high in the southeast and low in the northwest (Guo et al., 2017; Yang et al., 2020). The climate of CA is experiencing sharp differences in temperature, heterogeneous precipitation, and intense evaporation. Thus, it is sensitive and vulnerable to variations in climate change (Lioubimtseva and Henebry, 2009; De Beurs et al., 2018; Zhu et al., 2020). The annual precipitation can reach on 1,000 mm in the moutains areas in the southeastern CA, and the arid and desert climates with annual precipitation generally below 300 mm in the low areas in the northwestern CA (Lai et al., 2020; Lu et al., 2021).

According to the Köppen climate classification method (Chen and Chen, 2013. http://hanschen.org/koppen), we classified CA into three sub-regions (Figure 1B): temperate continental climate (Dfa) in the north (NCA), dry arid desert climate (BWk) in the southwest (SWCA), and Mediterranean continental like climate (Dsb) in the southeast (SECA). The total annual precipitation of CA is concentrated in cold seasons (Dai and Wang, 2017). Generally speaking, the CA is a typical region with complex rainfall conditions that deserves systematical evaluation the gridded daily precipitation products.

The spatial features of mean annual and seasonal precipitation from the 49 rain gauge measurements over CA during 1985–2005 have been presented in Figure 2. Mean annual precipitation exhibit the trend of a gradual precipitation decreased from southeastern to northwestern CA regions with precipitation hotspots located in SECA (more than 360 mm) and NCA (range of 200–300 mm).The area with the lowest average annual precipitation (less than 200 mm) expanded over the SWCA, which has an arid climate pattern and consists of desert areas (Figure 2A).

Figures 2B–E illustrate the spatial pattern of mean OBS seasonal precipitation over CA. In spring and winter, Areas with abundant precipitation are mainly distributed in the SECA (Figures 2B,E).The area-averaged precipitation was 99.6 and 86.4 mm, respectively. In contrast, less precipitation was observed in NCA during the cold seasons. The highest precipitation in the past 21 years over NCA concentrate in summer. In general, the annual and seasonal evolution of the OBS precipitation across CA were characterized by two major features: (1) mean annual precipitation decreased from the southeast to the north and southwest parts of the CA; and (2) there was more precipitation in the cold seasons (spring and winter) than in the warm seasons (summer and autumn). These two features of precipitation based on data from 49 rain gauges are consistent with those documented in the literature (Chen et al., 2009).

Figure 3 shows the spatial distribution of OBS, APHRODITE, GPCC, and CPC mean daily precipitation during 1985–2005 over CA. The OBS and three gridded products interpolated into each site displayed in (Figure 3A,B1–B3), and the gridded precipitation presented in (Figure 3C1–C3). All these three products accurately reproduced the spatial distribution of mean daily precipitation.

The spatial pattern of mean daily OBS precipitation (Figure 3A) was similar to the mean annual precipitation, presented an apparent decrease trend of precipitation from the southeast to north of CA. With the large mean daily precipitation in SECA (whithin 1–2 mm/ day), followed by NCA (whithin 0.6–0.8 mm/ day), and was the lowest in SWCA (less than 0.6 mm/ day). All three products captured the spatial distribution of mean daily precipitation characyeristics well, while CPC underestimated daily precipitation in the SECA area (Figure 3B3). The precipitation of five sites both in NCA and SWCA, and fifteen sites in SECA from CPC was lower than that of OBS. CPC had more sites where precipitation was underestimated as compared to APHRODITE and GPCC. Figures 3C1–C3 show the spatial distribution of gridded daily precipitation derived from APHRODITE, GPCC and CPC reflected the same features. However, GPCC showed more rainy in the NCA than APHRODITE and CPC. CPC showed that precipitation is scarce in SECA as compared to the other two products.

Figure 4 shows the density-colored scatterplots of 49 OBS mean daily precipitation versus three interpolated gridded products over the entire CA and subareas from 1985 to 2005 at daily scale. The correlation coefficients (R) derived from the three gridded products and OBS over the entire area and subareas were above 0.83, GPCC and APHRODITE more agreement with the OBS. The regression coefficient between GPCC and OBS was the highest, and almost coincided with the 1:1 line. In addition, the RMSE of GPCC displayed a lower value (0.28 mm/ day) and positive bias. APHRODITE also fitted well with OBS daily precipitation, while CPC showed worse performance with significantly underestimated values (RBias = −24.39%). Both APHRODITE and GPCC are performance better than CPC, as these has high Rs value (0.978 and 0.969) and low RMSEs value (0.24 mm/day and 0.28 mm/ day). CPC performs poorest as it exhibits the lowest R (0.84) and highest RMSE and RBias of 0.49 mm/ day and –24.39%, respectively. In the semi-arid climatic region (NCA, Figure 4B), all three products underestimated OBS daily precipitation, and the RBias were −12.39%, −3.02%, and −15.89% for APHRODITE, GPCC and CPC, respectively. RMSE values for APHRODITE, GPCC, and CPC were 0.57 mm/ day, 0.49 mm/ day, and 0.56 mm/ day, respectively. GPCC was obviously superior than other two precipitation products for higher R and lower RMSE. Based on the statistical index, APHRODITE performs best among the three products in SWCA region. GPCC ranks second with smaller Bias (Figure 4C). Figure 4D shows that the APHRODITE products had high accuracy with largest R (0.986) and lowest RMSE (0.42 mm/ day) at the daily scale in SECA region. The performance of GPCC was comparable than that of CPC, with the slightly better R and RMSE value. In general, CPC underestimated the daily precipitation over the whole CA. Based on RMSE that the error of CPC is larger than that of both APHRODITE and GPCC, the accuracy of CPC in daily scale is significantly lower than other two products.

Figure 5 shows the spatial distribution of the four statistical metrics for APHRODITE, GPCC, and CPC products against with OBS over the CA from 1985 to 2005. Figures 5A–C show the spatial distribution of R derived from three interpolated gauge-based gridded precipitation datasets in each individual station. All three gridded products displayed a strong correlation, with R values higher than 0.8 accounting for 83.7, 83.7, and 65.3% of all stations for APHRODITE, GPCC and CPC, respectively. Figure 5A shows the higher R values of APHRODITE located in the SWCA and SECA. Figures 5B,C indicates the R values of GPCC (CPC) in NCA and SECA were marginally higher than those in the SWCA. Figures 5D–F show the spatial distribution of RMSE in each individual sites. The mean RMSE values of APHRODITE, GPCC, and CPC over the entire CA were 1.46 mm/ day, 1.5 mm/ day, and 1.73 mm/ day, respectively. The RMSE values of APHRODITE were higher than those of GPCC and CPC in the NCA. In the arid desert climatic zone (SWCA), the RMSEs were the lowest, the number of stations marked by blue (higher than 1.5 mm/ day) of CPC (Figure 5F) over SECA were more than that of APHRODITE and GPCC, which indicates that RMSE values of CPC were higher than those of APHRODITE and GPCC. The spatial distribution of Bias for each pair of products were depict in Figures 5G–I. Approximately 59.2% of the stations showed positive bias for APHRODITE and GPCC against OBS CPC underestimated daily precipitation over almost all the regions, 95.9% of the stations showed negative bias of CPC across the entire CA. The last column in Figure 5 shows the spatial distribution of RBias between gridded products versus OBS. Better performance in terms of RBias values at the range of −10 to 10% that was mainly distributed in SECA (SWCA) for the APHRODITE (GPCC) products. In general, both APHRODITE and GPCC products shows positive deviation in most regions, while CPC present large negative deiation over almost all the CA.

The monthly change of regional precipitation for OBS and three interpolated products from 1985 to 2005 over CA and each sub-regions were given in Figure 6. It is clearly indicate that there are seasonal differences in each climatic zones. Figure 6A shows the precipitation in the entire CA is concentrated in spring and winter from March to May and from November to December, respectively. In SWCA (Figure 6C) and SECA (Figure 6D), a similar monthly precipitation pattern was illustrated as in CA, corresponding to the double peak precipitation type. On the contrary, monthly precipitation is concentrated in summer from May to August in NCA (Figure 6B), corresponding to the single peak type, where the maximum value occurred in July (43 mm/ month).

Monthly precipitation obtained from all three gridded precipitation products was consistent with the variation of OBS in each region. In NCA (Figure 6B), the monthly precipitation estimated by the three interpolated gridded products reasonably described the actual observations while GPCC was the closest. In general, variations of monthly precipitation from three products were similar to the OBS, and the estimates of monthly precipitation from APHRODITE and GPCC were closer to OBS has better performance. while the CPC product underestimated the precipitation obtained from gauge measurements and showed poor performance.

The statistical metrics (R, RMSE, Bias, and RBias) are computed from gridded products against with OBS over CA and its sub-regions during 1985–2005. The specific results are shown in Figure 7. In terms of R, all three products showed strong correlations with OBS over CA at monthly scale, with the values ranges from 0.8 to 1 (Figure 7A). The RMSE was high from March to April and October to December, and low from June to September (Figure 7D). This is consistent with the monthly precipitation over CA, though CPC performed worse than APHRODITE and GPCC. As for Bias (Figure 7I), the results show that APHRODITE underestimated the precipitation, except in July. The GPCC overestimated the monthly precipitation from April to July. CPC had a large error and poor performance. Figures 7B,F,J,N illustrates that the agreement between OBS monthly precipitation and the three gridded products was higher over the NCA. The results of RMSE and Bias show that GPCC had a lower error and better performance than APHRODITE and CPC. APHRODITE and CPC underestimated and had the worst performance for precipitation during the rainy season (from June to August). In the mainly arid desert SWCA area, with respect to RMSE (Figure 7G), the three gridded products showed low errors in scarce precipitation months (from June to September), and APHRODITE had lower error than GPCC and CPC. APHRODITE and GPCC showed better performance than CPC, as reflected by the results of Bias (Figure 7K) and RBias (Figure 7K), and CPC had the highest negative deviation with the ranges from −22.9% to −13.7%. Based on the R values over the SECA region (Figure 7C), the three gridded products had relatively higher correlations in most months (except July and August). APHRODITE had lower RMSE values than GPCC and CPC at the monthly scale, APHRODITE and GPCC underestimated the precipitation during the rainy months and overestimated precipitation in the scarce precipitation months (from June to September), and CPC tended to underestimate precipitation. APHRODITE and GPCC showed better performance than CPC, as reflected by the results of RBias.

For CA and each sub-region, the results for statistic metrics of the three gridded products versus OBS precipitation at seasonal scale, sorted from bad to good performance was supplied in Figure 8. It can be seen that all three gridded products had higher correlations (>0.87, statistically significant at the 99% confidence level), and the R values of APHRODITE and GPCC were larger than that of CPC across CA in all seasons. The RMSE values of APHRODITE were lower than those of GPCC and CPC, Bias and RBias of GPCC were positive and the values were lower than those of APHRODITE and CPC across CA in Spring, Summer, and Winter. In all seasons, the GPCC product has a lower error and good performance than APHRODITE and CPC over NCA. In Spring, GPCC overestimated precipitation, and APHRODITE and CPC underestimations. The RBias values of APHRODITE, GPCC, and CPC were −9.68%, 0.6%, and −15.41%, respectively. The three gridded products underestimated precipitation in Summer, Autumn, and Winter. The RMSE of GPCC was less than that of APHRODITE and CPC in summer over SWCA (0.21 mm/ day for GPCC, 0.23 mm/ day for APHRODITE, and 0.26 mm/ day for CPC). APHRODITE performed better than GPCC and CPC during Spring, Autumn, and Winter seasons. Additionally, GPCC overestimated precipitation, and both APHRODITE and CPC were underestimated precipitation in all seasons over the SWCA area. APHRODITE and GPCC performed better than CPC in all seasons over SECA. Precipitation in the SECA region was overestimated by APHRODITE during Summer (RBias of 13.45%) and underestimated in other seasons. The GPCC product overestimated precipitation in Spring and Summer (RBias for Spring was 2.07% and for summer was 12.73%).

Figure 9 shows the time series of the interannual variability of precipitation derived from OBS and gridded products over CA and three sub-regions period of 1985–2005. The amplitude in terms of the interannual variability for three gridded precipitation products are basically consistent with OBS, but they had apperent differences in the magnitude and trends over whole CA and each sub-regions. Both GPCC and APHRODITE has a similar trend with OBS in each region, while GPCC performs outstand.

The dispersion of boxplots of the statistical metrics for three gridded precipitation products over CA and sub-areas during 1985–2005 were illustrated in Figure 10. APHRODITE showed largest R (within 0.93–0.99) and smallest RMSE (varied from 0.15 to 0.36 mm/ day, with the medians value at 0.23 mm/ day), which indicated that APHRODITE performed best among the three gridded products over the entire CA (Figures 10A,E). GPCC and CPC showed relatively large error based on RMSE, whoes values of GPCC and CPC varied from 0.16 to 0.49 mm/ day and from 0.26 to 0.76 mm/ day, respectively. For the perspective of RBias, APHRODITE and CPC were underestimated the annual precipitation while GPCC overestimated with the 0.74% medians of RBias (Figure 10I) over the entire CA.

The second row shows the boxblots of statistical metrics from three precipitation products at annual scale over NCA (Figures 10B,F,J). Figure 10B shows thes R values of GPCC were higher than those of APHRODITE and CPC. The RMSE values of GPCC were lower than that of APHRODITE and CPC, the medians RMSE of GPCC, APHRODITE, and GPCC were 0.47 mm/ day, 0.54 mm/ day, 0.54 mm/day, respectively. (Figure 10F). The medians RBias values was −12.19% for APHRODITE, −2.79% for GPCC, and −15.88% for CPC. The large deviations are derived from OBS precipitation data and APHRODITE and CPC products. Figures 10C,G,K shows the results for R, RMSE, and RBias in the SWCA sub-area. APHRODITE accurately estimated precipitation, as reflected by a higher R (0.8–0.97 with medians value of 0.92), lower RMSE (0.19–0.44 mm/ day with a medians value of 0.29 mm/ day), and lower RBias (−11.7–2.29% with a medians value of −4.58%). GPCC performed better as the mean RBias values were lower (−3.02–17.63% with a medians value of 3.96%), which overestimated precipitation at the annual scale. In contrast, CPC showed the worst performance with lower R (0.53–0.96 with a medians value of 0.86), higher RMSE (0.21–0.69 mm/ day with a medians value of 0.39 mm/ day), and lower RBias (−0.15–34.42% with a mean value of −18.87%) over the SWCA area. Figures 10D,H,L shows the R, RMSE, and Bias of the three products in the SECA region. As shown in Figure 10D, the mean R values of both APHRODITE and GPCC precipitation gridded products were higher that of CPC (the medians R values were 0.98 for APHRODITE, 0.97 for GPCC, and 0.93 for CPC). Figure 10H shows that the mean RMSE value of APHRODITE was lower than those of GPCC and CPC (medians values were 0.39 mm/ day for APHRODITE, 0.51 mm/ day for GPCC, and 1.04 mm/ day for CPC). Figure 10L shows that the medians RBias value of APHRODITE was lower than those of GPCC and CPC (medians value of APHRODITE was −2.1%, of GPCC was 1.48%, and of CPC was −21.12%). APHRODITE and CPC underestimated the precipitation over SECA region.

To clarify the gauge-based gridded precipitation data in reproduce the spatial patten of the interannual variability, Figure 11 depicts the spatial distribution of the annual precipitation linear trends over CA and subregions during 1985–2005. As shown in Figure 11A, the OBS precipitation displayed widespread significant increasing trends across the entire region of CA with 85.71% of stations having a rate of 20.01 mm/10a (Figure 11A). The annual precipitation for NCA, SWCA, and SECA sub-regions of CA showed rising trends at 83.33, 89.47, and 83.33% of the stations, with rates of 18.7 mm/10a, 5.9 mm/10a, and 36 mm/10 a, respectively.

Figures 11B1,C1,D1 displays three interpolated gridded precipitation products for each site for the period 1985–2005. The annual precipitation trends of APHRODITE were generally lower than those of GPCC and CPC (the trend of APHRODITE was 6.95 mm/10 a, of GPCC was 17.33 mm/10 a, and of CPC was 33.11 mm/10 a) over the whole CA. The GPCC product tendencies were in agreement with the OBS, while the APHRODITE underestimated and CPC significantly overestimated the precipitation trends for rain gauges over CA. In the NCA (Figure 11B1), the annual precipitation displayed increasing trends at 14.5 mm/10 a, 15.7 mm/10 a, and 23.4 mm/10a for APHRODITE, GPCC, and CPC, respectively. This indicates that APHRODITE and GPCC underestimated the precipitation trends, while the CPC product overestimated precipitation trends in the entire NCA. In the SWCA (Figure 11C1), APHRODITE generally matched the precipitation trends over 21 years of OBS (OBS = 5.9 mm/10 a, APHRODITE = 5.6 mm/10 a). GPCC and CPC overestimated the trends of precipitation (OBS = 7.4 mm/10 a, APHRODITE = 23.4 mm/10 a). In the SECA with higher elevation regions (Figure 11D1), APHRODITE underestimated the trends of annual precipitation, while CPC significantly overestimated (annual precipitation trend for APHRODITE was 12.8 mm/10 a and for CPC was 52.7 mm/10 a). The GPCC results were in agreement with OBS at a rate of 37.4 mm/10 a. Figures 11B2,C2,D2 shows the gridded distribution of annual precipitation trends of APHRODITE, GPCC, and CPC. The distribution of APHRODITE and GPCC was similar, though APHRODITE exhibited a decreasing trend in most parts of the SECA than GPCC. CPC showed an increasing trend in most regions of the CA, and overestimated the interannual variability for the underestimation of the climatology.