Comparison of precipitation products to observations in Tibet during the rainy season

Zhuo Ga , Za Dui , Duodian Luozhu , Jun Du

1. Lhasa Branch of Chengdu Institute of Plateau Meteorology, China Meteorological Administration, Lhasa, Tibet 850000,China

2. Tibet Climate Center, Tibet Meteorological Bureau, Lhasa, Tibet 850000, China

3. Lhasa Meteorological Bureau, Lhasa, Tibet 850000, China

ABSTRACT Precipitation is an important component of global water and energy transport and a major aspect of climate change. Due to the scarcity of meteorological observations, the precipitation climate over Tibet has been insufficiently documented. In this study, the distribution of precipitation during the rainy season over Tibet from 1980 to 2013 is described on monthly to annual time scales with meteorological observations. Furthermore, four precipitation products are compared to observations over Tibet. These datasets include products derived from the Asian Precipitation-Highly-Resolved Observational Data(APHRO), the Global Precipitation Climatology Centre (GPCC), the University of Delaware (UDel), and the China Meteorological Administration (CMA). The error, relative error, standard deviation, root-mean-square error, correlations and trends between these products for the same period are analyzed with in situ precipitation during the rainy season from May to September. The results indicate that these datasets can broadly capture the temporal and spatial precipitation distribution over Tibet. The precipitation gradually increases from northwest to southeast. The spatial precipitation in GPCC and CMA are similar and positively correlated to observations. Areas with the largest deviations are located in southwestern Tibet along the Himalayas. The APHRO product underestimates, while the UDel, GPCC, and CMA datasets overestimates precipitation on the basis of monthly and inter-annual variation. The biases in GPCC and CMA are smaller than those in APHRO and UDel with a mean relative error lower than 10% during the same periods. The linear trend of precipitation indicates that the increase in precipitation has accelerated extensively during the last 30 years in most regions of Tibet. The CMA generally achieves the best performance of these four precipitation products. Data uncertainty in Tibet might be caused by the low density of stations, complex topography between the grid points and stations, and the interpolation methods, which can also produce an obvious difference between the gridded data and observations.

Keywords: APHRO; GPCC; UDel; CMA; Tibet; precipitation

1 Introduction

The Tibetan Plateau (TP) is of considerable importance to the Asian monsoon and global general circulation via mechanical and thermal forcing (Ye and Gao, 1979; Yanaiet al., 1992; Duan and Wu, 2005)due to its unique altitude and horizontal extent. The Tibet Autonomous Region (Figure 1) is located on the southwest border of China, southwest of the Qinghai–Tibet Plateau. It is difficult to carry out a meticulous study of the precipitation over Tibet in the bulk of the TP due to limited measurements.However, precipitation is the most important atmospheric input to the terrestrial hydrologic system and varies greatly in both time and space, particularly at fine-spatial and temporal scales. Reanalysis products and gridded observational datasets offer a convenient way to analyze weather forecasts and global climate change. It is therefore essential to evaluate their quality and credibility in regions with sparse meteorological stations to understand what type of data is more feasible and efficient in Tibet.

Figure 1 Elevation above sea level in Tibet with the locations of meteorological stations,red dots refer to the locations in the paper (unit: m)

Gridded datasets could provide greater spatial representation of precipitation compared with observations, which is useful for many types of climate research including analysis of climatic change and variability (Daiet al., 1997; ?en and Habib, 2000), precipitation patterns (Klein Tanket al., 2002), and interannual and decadal variations (Ensor and Robeson,2008; Juárezet al., 2009). Datasets obtained from satellite products have been used to assess mean annual and seasonal diurnal rainfall cycles (Kiddet al.,2013; Liu, 2015).

A large number of studies have evaluated precipitation estimates over the TP with different observational datasets and satellite products. Specifically,precipitation trends (Zhanget al., 2015), diurnal variation (Baoet al., 2011), and annual and seasonal precipitation (Tonget al., 2014; Youet al., 2015) have been compared with gridded data and with gauge observations in the TP. The spatial pattern and seasonality (Maussionet al., 2014) of precipitation have also been compared with satellite-based precipitation data from the Tropical Rainfall Measurement Mission.Satellite precipitation estimate products have also been used to model the three-dimensional structure of precipitation (Longet al., 2016) and to evaluate variations in precipitation (Liet al., 2010; Wu and Zhai,2012; Maet al., 2016).

To summarize, a large number of studies with different products have been used to model the precipitation over the TP and have obtained multiple significant scientific results. However, due to the scarcity of meteorological observations, the precipitation distribution over Tibet has been insufficiently documented.A couple of studies (Liet al., 2010; Baoet al., 2011;Wu and Zhai, 2012; Maussionet al., 2014; Tonget al., 2014; Youet al., 2015; Longet al., 2016; Maet al., 2016) have specialized in the variability of precipitation over the TP, demonstrating the temporal and spatial variability of precipitation, the ability of gridded products and satellite precipitation datasets to detect precipitation amount based on meteorological observations. Obviously, gridded data cover a wide range of spatial and temporal scales with higher resolution and are potentially helpful in data poor regions.

The purposes of this study is twofold: (i) to describe the characteristics of precipitation (the amount and variability) at monthly to annual time scales from observations which include meteorological measurements in Tibet and four gridded products from the Asian Precipitation-Highly-Resolved Observational Data (APHRO), the Global Precipitation Climatology Centre (GPCC), the University of Delaware (UDel),and the China Meteorological Administration (CMA)and (ii) to better understand the application and credibility of precipitation products in climate change research over Tibet in terms of the mean bias and rootmean-square error (RMSE) verified against rain gauges, even though these products might have different interpolation methods and stations involved in the production of their datasets. Therefore, our results should be viewed as conservative estimates of the differences between the various data sources. This study lays a foundation for understanding and improving the performance of precipitation products, thereby enhancing the utility of precipitation data in research and operational applications over mountainous regions with sparse meteorological stations.

2 Data and methods

2.1 Meteorological measurements

The overall topography of Tibet is high in the northwest and low in the southeast, and the Plateau edge is higher than the central. Therefore, most meteorological stations are situated along the Yarlungzangbo River Valley and in the inhabited eastern region of Tibet, lower altitude and longer sunshine duration, as shown in Figure 1. Rain gauge data were collected from a total of 38 artificial meteorological stations (Song and Wang, 2013) in Tibet. Only a few stations are located in the vastly uninhabited regions of western Tibet. Even though a large number of automatic weather stations (automatic observed and transmitted by instrument) were set up in the past few years providing significant amounts of information for weather forecasts and disaster prevention, many uncertainties still exist between artificial and automatic observations because of different instrument and observation methods. Precipitation obtained from artificial stations is currently one of the most efficient sources of long-term precipitation estimates over the study area. In addition, most stations in Tibet were set up around the 1960s and the 1970s, thus meteorological measurements were used as true values to analyze the spatial distribution and variation tendency in Tibet during 1980–2013, and the rain gauge precipitation was also selected to compare with gridded products from different sources.

Since precipitation from May to September accounts for 80%–95% of annual precipitation amount in Tibet (Song and Wang, 2013), rain gauge precipitation and gridded products during the rainy season(rainy season refer to the period from May to September) will be used to quantitatively analyze the difference of rain gauge precipitation from various gridded products.

2.2 Gridded products

Table 1 presents the precipitation products used in this paper from different sources. These datasets include APHRO from Japan, GPCC and UDel from the U.S.A., CMA from China with different coverage and time periods respectively. The same resolution and time period except APHRO have been used in the comparison with rain gauge measurements in Tibet.All these precipitation products have different number of stations used to estimate the monthly total precipitation along with interpolation methods while some stations are same as rain gauge stations in Tibet.

Table 1 Description of the precipitation products used in this study

APHRO: The Asian Precipitation-Highly-Resolved Observational Data Integration Towards Evaluation of the Water Resources (APHRODITE's water resources, or APHRO for short) product (Yatagaiet al., 2009, 2012) is a precipitation dataset based on rain gauge data developed by a consortium of the Research Institute for Humanity and Nature and the Meteorological Research Institute of the Japan Meteorological Agency. The data used in APHRODITE analysis were (1) GTS-based data (global summary of the day), (2) data precompiled by other projects or organizations, and (3) APHRODITE's own collection. The products are gridded precipitation data and rain/snow discrimination information over the monsoon Asia.Precipitation distribution from May to September was studied using meteorological observations and gridded products. Therefore, the data here only consists of liquid precipitation for each grid. Monthly data was calculated on the basis of daily precipitation during the period 1980–2007.

GPCC: The Global Precipitation Climatology Centre (GPCC) product (Schneideret al., 2011) is primarily comprised of precipitation data on a monthly basis from a variety of sources. The GPCC used the additional acquisition and processing of daily precipitation data to fulfill the needs of near real-time weather analyses, prediction, and climate monitoring.The GPCC has acquired precipitation data from approximately 190 countries based on quality-controlled data from 67,200 stations worldwide. Monthly precipitation data are routinely obtained from synoptic weather reports (SYNOP) at Deutscher Wetterdient (DWD), National Oceanic and Atmospheric Administration (NOAA), and Japan Meteorological Agency (JMA). This product contains monthly totals on regular grids with different spatial resolutions of 0.5°×0.5°, 1.0°×1.0° and 2.5°×2.5°, and data with 0.5°×0.5° resolution have been used here.

UDel: Monthly total rain gauge-measured precipitation from the University of Delaware (UDel) (Legates and Willmott, 1990) was compiled from several updated sources including the global historical climate network, the Atmospheric Environment Service/Environment of Canada, and other archives.Station values of the monthly total rain gauge-measured precipitation (P) were interpolated to a 0.5°×0.5°latitude/longitude grid, where the grid nodes are centered on 0.25°. The resulting number of stations used to estimate the monthly total precipitation ranges from approximately 4,100 to 22,000 globally.

CMA: Gridded precipitation for this product was derived from the National Meteorological Information Center of the Chinese Meteorological Administration (CMA) based on interpolations from 2,472 daily observation stations on the basis of the Thin Plate Smoothing Splines interpolation method(Hutchinson, 1998a, 1998b; Xuet al., 2009) in China.

Because most of the meteorological stations in Tibet were set up in the 1970s, observations and products from 1980 to 2013 were used to indicate the distribution of regional precipitation. Furthermore,APHRO during 1980 to 2007 was used in the study which is different from other products due to the gridded daily precipitation of the monsoon Asia is only available for 1961–2007.

2.3 Interpolation method

Because gridded precipitation products possess different spatial resolutions from conventional observations, to compare gridded precipitation with rain gauges, all datasets have been interpolated to locations of the meteorological stations in Tibet to better understand the quality of each product. There are a large number of interpolation methods, such as Inverse Distance Weighting, Cubic Spline, and Kriging,to interpolate data from a grid to a point and vice versa. In mathematics, bilinear interpolation (Masty?o,2013) is an extension of linear interpolation to interpolate functions of two variables (e.g.,xandy) on a rectilinear 2D grid by performing linear interpolation first in one direction and then again in the other direction. Even though each step is linear in the sampled values and the position, the interpolation as a whole is not linear but rather quadratic in the sample location.The bilinear interpolation method can obtain good interpolation results with high quality and continuity;therefore, it is widely used in research and operational work (Xiao, 2011; Chenet al., 2016).

Tibet accounts for nearly one-eighth of the total land area of China; however, meteorological, observational and operational stations are scarce over this vast area. Values are much closer to real values when using simple interpolations such as the bilinear interpolation method with fewer calculations within this region.

2.4 Statistical analysis

Linear regression has the advantage of being commonly accepted, and trend lines are often used to argue that a particular action or event caused the observed changes at a given point in time. This method does not require a control group experimental design or a sophisticated analysis technique. Therefore, linear analysis was used to analyze trends in annual precipitation for each meteorological station from 1980 to 2013, even though this method suffers from a lack of scientific validity in cases where other potential changes can affect the data. In this study, linear regression was used to understand the estimations of the long-term variation tendencies in different grid products.

Rain gauge measurements were used to indicate the true value in Tibet and are also one of the most reliable products to analyze climatology and the trend of precipitation in the rainy season during the period of 1980–2013. Linear regression was used in the variation tendency of the precipitation. Variables such as Error (E), Relative Error (RE), Standard Deviation(SD), and Root-Mean-Square Error (RMSE) were calculated to examine the differences between observations and gridded precipitation. Correlation coefficients are also used to analyze the credibility of the four products in climate studies of precipitation over Tibet. Because there is a lack of observations prior to 1980 in most regions of Tibet, the study period of the precipitation starts in 1980.

The correlation coefficient is a measure of the strength and direction of the linear relationship between two variables and is defined as the (sample)covariance of the variables divided by the product of their (sample) SDs. Here, the temporal coefficient indicates the relationship between observations and gridded products at each meteorological station along the time series during the period 1980–2013 and the spatial coefficient indicates the annual relationship between observations and gridded products based on the spatial series.

These statistical variables can be expressed as follows.

where,Oiis the observations,Siis the gridded precipitation,is the average value ofOi,is the average value ofSi,irefer to different station, andnis the number of years in the time series.

3 Results

3.1 Spatial distribution of precipitation and error between observations and gridded products

There are considerable spatial variations in rainfall in the Tibetan region. The mean total precipitation amount (Figure 2) during the rainy season ranges from 58.6 mm to 607.4 mm and decreases from southeast to northwest of Tibet with increasing latitude.The spatial distribution of precipitation shows that areas with large rainfall are primarily located in the middle and eastern regions of Nagqu, Lyingchi, and precipitation in most areas of Qamdo even exceeds 400 mm. The precipitation ranges from 300 mm to 400 mm in northern Tibet such as western Nagqu, the middle areas along the Yarlungzangbo River (the upper reaches of Brahmaputra), and the southern slopes of Tibet. There is very little precipitation in western Tibet, which is considered to consist of desert and rock areas. One station (Shiquanhe) located in western Tibet (32.50°N, 80.08°E) observed only 58.6 mm of rainfall.

In general, gridded precipitation can generate similar spatial distributions to observations. To further understand the applicability of gridded precipitation products, the bias of spatial distribution was calculated between the four products and observations. Figure 3 demonstrates that precipitation in APHRO (Figure 3a) underestimates observations for most regions of Tibet, especially in the middle areas along the Yarlungzangbo River. However, it overestimates precipitation in southern Tibet,e.g., in Nielamu, Pali, and B-axu, and its deviation ranges from 80 mm to 200 mm.The deviation distributions between GPCC (Figure 3b) and UDel (Figure 3c) and observations are quite similar to that of APHRO. The difference is that GPCC precipitation describes the precipitation along the Yarlungzangbo River quite well, while the UDel product greatly overestimates the rainfall on the southern edge of Tibet along the Himalayas, especially at the Nielamu, Pali, Dingri, and Lazhi stations.The differences between the CMA (Figure 3d)product and observations are quite small. The CMA product only overestimates rainfall in northern Qamdo and slightly underestimates rainfall in some areas along the Yarlungzangbo River. Therefore, CMA precipitation is the closest to the spatial distribution of observations among four gridded products. All in all,gridded products could present the distribution of precipitation as rain gauge data in Tibet, but they all exhibit obvious deviation especially along the south and southeastern edge of Tibet. The reason might be caused by the number of rain gauge station in Tibet which is used to interpolate the gridded product and terrain complexity to some extent.

3.2 Spatial distribution of SD

Figures 4a–4e show the corresponding SDs of these products. Broadly speaking, the spatial distribution of the SD of gridded products closely follows the averaged distribution during the rainy season, indicating that all datasets basically capture the spatial distribution of the concentration degree of precipitation.Regions with high precipitation in observations (Figure 4a) are primarily located in southeastern Tibet and along the Yarlungzangbo River. The largest deviation regions in the GPCC (Figure 4c) and UDel (Figure 4d) precipitation are situated in southern Tibet. The distribution of the concentration degree in APHRO(Figure 4b) is not obvious compared to observations.The CMA (Figure 4e) product displays heavy precipitation located in southeastern Tibet as the rain gauge observation; however, it fails to capture precipitation in the eastern areas along the Yarlungzangbo River.

Figure 2 Spatial distribution of precipitation at meteorological stations in Tibet (unit: mm)

Figure 3 Error of the gridded products with respect to meteorological observations (unit: mm): (a) APHRO minus observations, (b)GPCC minus observations, (c) UDel minus observations, and (d) CMA minus observations. The pink triangles indicate meteorological stations where the deviation in precipitation was larger than 300 mm

Figure 4 Distribution of SD from different data sources (unit: mm). (a) observations,(b) APHRO, (c) GPCC, (d) UDel, and (e) CMA

3.3 Spatial distribution of RMSE

Figure 5 shows the distribution of the RMSE in different gridded products. The largest deviation regions are located in Lyingchi and southern Tibet with values of 100–200 mm in APHRO (Figure 5a). The UDel (Figure 5c) product has a similar distribution with a larger extent and deviation than that of APHRO, especially in southern Tibet. The RMSEs in the GPCC (Figure 5b) and CMA (Figure 5d) products are relatively smaller than those in the APHRO and UDel products on average. Regions with larger deviations in GPCC appear on the southern edge of the Himalayas and in the southeast areas such as Lyingchi and Qamdo. The CMA product better displays the amount of precipitation with deviations of less than 100 mm in most regions of Tibet. In summary, nearly all products overestimate the actual precipitation, especially APHRO and UDel, and the main errors are located on the southern edge of the Himalayas and in southeastern Tibet. This may be caused by the complex topography in those two areas, which contain the Everest Mountain Qomolangma and the Yarlung Zangbo Grand Canyon, respectively. The CMA product includes more information from meteorological stations than other products, which it uses to interpolate the gridded precipitation; consequently, it has fewer constraints along the border of Tibet. It is important to be careful when using gridded products in areas with complicated terrain. However, gridded products still have a considerable potential to provide high-resolution precipitation information over areas without measurements.

Figure 5 Distribution of RMSE between gridded products and observations (unit: mm):(a) APHRO, (b) GPCC, (c) UDel, and (d) CMA

3.4 Monthly precipitation

In this section, we study the monthly variation of precipitation with meteorological observations and values in these four gridded products presented in Figure 6. Tibet can be roughly divided into four subareas according to local climatic and agricultural characteristics,i.e., arid, semi-arid, humid, and semi-humid areas. In general, monthly precipitation (Figure 6a) has two types of variations: single peak and double peak. Precipitation in most regions of Tibet(arid, semi-arid, and semi-humid areas) belongs to the single type in that maximum rainfall appears in July or August and minimum in January or December. Precipitation in southeastern Tibet and on the southern ridge of Tibet (humid areas) has two high values that appear in March–April and in August–September.The humid areas have large amounts of precipitation.

Furthermore, we divided Tibet into western and eastern regions based on their longitude (90°E) and their climatic characteristics to compare monthly precipitation from different sources. In general, the precipitation obtained from different gridded products is basically consistent with the variation in observations in western (Figure 6b) or eastern (Figure 6c) Tibet and a large deviation between the four products and observations appears in summer. In particular, UDel clearly overestimated precipitation during the rainy season, especially in July and August. The amount of precipitation estimated by APRHO, GPCC, and CMA are similar to observations; however, they also slightly overestimate precipitation in July and August in western Tibet. APHRO slightly underestimated precipitation during the entire rainy season; UDel and CMA overestimated precipitation while GPCC was nearly consistent with observations in eastern Tibet.In general, all the products can detect the basic variation tendency of monthly precipitation. UDel has the biggest deviation, with an amount nearly twice as big as observations in western Tibet; however, all four products well-exhibited the observed variation tendency in eastern Tibet.

Figure 6 Monthly precipitation in (a) arid and humid sub-areas, in (b) western and (c) eastern Tibet from gridded products and observations during 1980–2013 (1980–2007 for APHRO) (unit: mm)

3.5 Inter-annual and inter-decadal variation of precipitation

From the regional mean precipitation during the rainy season obtained from the meteorological stations, we noted the climatology and long-term trends in precipitation over Tibet from 1980 to 2013. The multi-year mean precipitation in Tibet is 384.15 mm,with a maximum of 472.35 mm in 2000 and minimum of 277.52 mm in 1983. The linear trend of precipitation was 8.55 mm per decade (hereinafter simplified as mm/10a) in the rainy season during 1980–2013, indicating that generally the precipitation has weak increase tendency over Tibet. The decadal mean precipitation shows that precipitation has increased steadily in the last decades, especially from the 1980s to the 1990s with values of 363.23 mm and 392.85 mm, respectively, and then the increase in precipitation slowed in the 2000s with a value of 396.69 mm.

To determine the inter-annual variation obtained from the various datasets, we further calculated the difference between the products and observations.Figure 7 shows the relative precipitation errors of the products with respect to observations during the rainy season. UDel has the largest difference compared with the observation in most years during the period 1980–2013, with a mean RE of 27.75%, and its biggest RE of 77.60% appears in 2009. The relative error in APHRO is smaller than that in UDel, with a mean relative bias of 16.52%, and its biggest RE of 40.82% occurs in 1983. The mean relative biases in the GPCC and CMA precipitation are 5.79% and 6.34%, respectively, with REs lower than 10%, except in a few years. In general, the APHRO precipitation underestimates while the UDel product overestimates observations. The GPCC and CMA products perform much better than the APHRO and UDel products with respect to in situ observations in Tibet.

Figure 7 Temporal variation in the RE between gridded products and observations during the period 1980–2013 (unit: %)

Table 2 indicates the inter-decadal variation of mean RE. UDel shows the largest bias of these products, especially in 1980s with a value of 33.94%.The biases in GPCC and CMA in 1990s are 4.71%and 5.55%, respectively. Their mean REs have values lower than 10% during other periods.

Table 2 Inter-decadal variation of mean RE between gridded products and observations (unit: %)

3.6 Variation tendency

To accurately characterize the reliability of precipitation products, the spatial (Figure 8) and temporal variation (Table 3) tendencies were investigated during the rainy season in Tibet. The mean precipitation trend (Figure 8a) is distributed heterogeneously over Tibet. Precipitation exhibits an increasing tendency in most regions of Tibet from 1980 to 2013. The maximum increasing trend is identified at Mangkang(32.87 mm/10a) while the minimum is identified at Shiquanhe (0.42 mm/10a). The area with most significant increased precipitation is located in a river valley in the middle of Tibet. Regions with drying trends are located in most regions of Lyingchi and in the southern regions of Ngari and Qamdo. The maximum drying trend is located in Bomi with a value of 29 mm/10a and the minimum drying trend is located in Pulan (2.04 mm/10a) County of Ngari region.Therefore, the regions with increasing precipitation are located in most regions of Tibet with different magnitudes and those with an opposite variation trend are primarily situated in southeastern and southern Tibet along the Himalayas.

Figure 8 Spatial distribution of mean precipitation trend (unit: mm/10a) in the rainy season from observations and gridded products during the period 1980–2013 (unit: mm/10a). (a) observations, (b) APHRO, (c) GPCC, (d) UDel, and (e) CMA

Figure 8a depicts the results from in situ observations; it can be seen that precipitation demonstrates obvious increasing tendencies in most regions of Tibet with a variation range of 0–30 mm/10a during the period 1980–2013, except in Lyingchi, Pulan, and Nielamu. Even though the APHRO (Figure 8b), GPCC (Figure 8c), and UDel (Figure 8d) datasets show increasing trends in most regions of Tibet as well, the range and intensity of the increasing/decreasing trends have notable discrepancies, especially in UDel. The CMA precipitation (Figure 8e) exhibits nearly the same variation distribution. Precipitation in APHRO enlarges the decreasing extent in southeastern Tibet while the UDel data mostly narrows the increasing extent. In addition, most areas with increasing trends exceed a significance level of 90%.

Table 3 presents the temporal variation of the linear trend coefficient in the four precipitation products and in situ observations during the period 1980–2013.All datasets exhibit an increasing trend in precipitation during the rainy season with different amplitudes.The coefficients of the UDel and APHRO datasets are smaller than that of observations with values of 2.90 mm/10a and 7.08 mm/10a, respectively. The GPCC precipitation is slightly larger with a value of 9.89 mm/10a, which is the closest to the value of the observations, while the CMA precipitation overestimates (12.70 mm/10a) the temporal variation of the observations.

Table 3 Temporal variation of the linear trend for gridded products during the period 1980–2013 (unit: mm/10a)

3.7 Correlation coefficient

With the aim of obtaining a reliable picture of the similarities between different products and observations, we calculated the spatial and temporal correlation coefficients (Figure 9) of the datasets and in situ measurements. The spatial-wise correlation coefficient (Figure 9a) demonstrates that precipitation in GPCC and CMA is positively correlated to observations with high coefficients; the coefficients between GPCC and CMA and observations reach 0.796 and 0.881, respectively. The difference in the values of the coefficients of GPCC and CMA gradually becomes large, especially after the year 2000. This might be closely associated with the automatic meteorological stations that have been set up in Tibet since 2000,CMA used these new observations in its interpolation while GPCC did not, and the number of observation stations and the degree of data richness will definitely affect the accuracy of the interpolation data to a certain extent. The coefficient between APHRO and observations is smaller than the former two products with values of 0.666. The correlation coefficient of UDel to observations is the smallest among the four precipitation products with a value of 0.319. All coefficients have a significance level greater than 95%(0.312), except that of the UDel product.

Similarly, the temporal correlation coefficient(Figures 9b–9e) illustrates that GPCC and CMA are closely correlated to observations with mean coefficients larger than 0.8 in most regions of Tibet, while coefficients of APHRO and UDel to observations are smaller than 0.40. The correlation coefficients are even negatively related to observations for the APHRO and UDel precipitation in most regions of Tibet. Nearly all meteorological stations do not exceed a significance level of 95%.

4 Conclusions

The spatial and temporal variation of the precipitation amounts during the rainy season derived from the APHRO, GPCC, UDel, and CMA products were compared to rain gauge measurements at meteorological stations in Tibet from 1980 to 2013. The results indicate that these datasets can detect precipitation distribution which decreases from southeast to northwest over Tibet. GPCC and CMA are quite similar to observations in their spatial distributions. The APHRO product underestimates while the UDel, GPCC, and CMA datasets overestimate precipitation on the basis of monthly and inter-annual variations. The biases are smaller in GPCC and CMA than in other products with a mean RE lower than 10% during the same periods. The linear trend indicates that precipitation in recent decades has shown extensive increases in section 3.5 over most regions of Tibet as well as the variation of precipitation over the Tibetan Plateau(Youet al., 2012). CMA product generally achieves the best performance of the four products.

It should be emphasized that gridded products can broadly capture the distribution of precipitation but that not all datasets are equally biased and the choice of dataset is important. Areas with positive bias are primarily located in southwestern Tibet and those with negative bias are primarily located in northwestern Tibet. Local factors such as topography partly account for these differences, with northern Tibet covered with broad grasslands, and southwestern Tibet surrounded by mountains and canyons with forest. However, the heterogeneity in the precipitation records, which results from changes in instrumentation, recording practices, and station locations/environment, may cause changes in the precipitation records (Groismanet al., 1991; Karlet al.,1993). Better datasets have better sources of in situ data to interpolate onto grids; this might contribute to the final higher quality of gridded datasets. CMA product used all observations from meteorological stations in Tibet during the interpolation process.Therefore, the gridded product obtained from CMA has the highest quality compared with other three products.

Figure 9 Correlation coefficients of the (a) spatial and (b–e) temporal variations between products and observations at meteorological stations in Tibet during the period 1980–2013. (b) APHRO, (c) GPCC, (d) UDel, and (e) CMA. The value 0.312/0.329 represents time/area that exceeds a significance level of 95% respectively

In addition, interpolation algorithms can contribute to improvements in the interpolation accuracy in different manners. Data uncertainty over Tibet caused by the low density of stations, complex topography between grid points and stations, and the interpolation methods can produce obvious biases between gridded data and observations (Zhao and Fu, 2006;Youet al., 2010). In this paper, only precipitation during the rainy season of limited stations was analyzed with the gridded datasets due to the scarcity of meteorological measurements in Tibet. Inter-annual precipitation variability with high-resolution precipitation products and satellite-based precipitation estimates will be the subject of a future study.

Acknowledgments:

This study was supported by the National Natural Science Foundation of China (Grant No. 41130960)and Key Science and Technology Plan of Tibet Autonomous Region (Grant No. XZ201703-GA-01).The authors would like to acknowledge the data providers: the Asian Precipitation – Highly-Resolved Observational Data Integration (APHRO), Global Precipitation Climatology Centre (GPCC), University of Delaware (UDel), Chinese Meteorological Administration (CMA) and Tibet Meteorological Information Center (TMIC) for providing the precipitation products that made this study feasible.We would like to express thanks to Colleague ZhiQiang Lin for his beneficial suggestions. The authors would like to thank Enago (www.enago.cn)for the English language review. We are also grateful to anonymous reviewers as well as the editor, whose insightful comments and constructive criticism have improved the paper substantially.