Spatial and temporal variation patterns of reference evapotranspiration across the Qinghai-Tibetan Plateau during 1971–2004

Authors


Abstract

[1] Reference evapotranspiration (RET), an indicator of atmospheric evaporating capability over a hypothetical reference surface, was calculated using the Penman-Monteith method for 75 stations across the Qinghai-Tibetan Plateau between 1971 and 2004. Generally, both annual and seasonal RET decreased for most part of the plateau during the study period. Multivariate linear models were used to determine the contributions of climate factors to RET change, including air temperature, air humidity, solar radiation, and wind speed. Spatial differences in the causes of RET change were detected by K-means clustering analysis. It indicates that wind speed predominated the changes of RET almost throughout the year, especially in the north of the study region, whereas radiation was the leading factor in the southeast, especially during the summertime. Although the recent warming trend over the plateau would have increased RET, the combined effect of the reduced wind speed and shortened sunshine duration negated the effect of rising temperature and caused RET to decrease in general. The significant decrease in surface wind speed corresponded to the decreasing trends of upper-air zonal wind and the decline of pressure gradient, possibly as a result of the recent warming.

1. Introduction

[2] Warming-induced hydroclimatic changes, if fully realized, would likely vary greatly across any one continent [Ziegler et al., 2005], and climatic changes affect all elements of hydrologic cycle, such as precipitation, runoff, infiltration, groundwater flow, and evaportranspiration in a variety of ways [Kundzewicz and Somlyody, 1997]. Recent studies on regional-scale hydroclimatic changes pay much attention to the spatial and temporal patterns of evapotranspiration, involved in the maintenance of forest and oasis ecosystems, management of farmland and pasture irrigation, sustainable water supply for industrial and domestic demands et al. [e.g., Yu et al., 2002; Jin et al., 2004; Francisco, 2005; Xu et al., 2006; Gao et al., 2006; Brunel et al., 2006; Calanca et al., 2006; Burns et al., 2007].

[3] The Qinghai-Tibetan Plateau, famous as “the roof of the world,” is rich in lakes, glaciers, and wetlands and is the main source area of several major rivers in Asia [Shen and Chen, 1996; Luosang, 2005]. The management and utilization of affluent water resources and hydropower on the plateau are not only important issues concerned with China, but also the neighboring countries located in the lower reaches of the international rivers originated from the plateau [Liu and Qimeiduoji, 1999; Liu et al., 2006a]. Significant warming trends were detected on the plateau during the last several decades [e.g., Liu et al., 2006b]. The possible shift of water balance caused by climate change might have impacted the water resources on the plateau.

[4] As a result of monsoon climate, most precipitation and evapotranspiration on the plateau occurs from June to September. Evapotranspiration not only is a basic component of water balance, but also plays an important role in the energy budget in the earth-atmospheric system. In the central and southeastern plateau, the latent heat flux dominates the surface energy balance and is a major energy source of the atmosphere during the summer monsoon [Li, 2002; Xu et al., 2005]. Therefore the spatial and temporal patterns of evapotranspiration may have a significant impact on the transport of water vapor and latent heat and further local hydroclimatic courses.

[5] Actual evapotranspiration (AET) is influenced by various factors of both local climate and land surface condition. Potential evapotranspiration (PET) was first defined as “the amount of water transpired in a given time by a short green crop, completely shading the ground, of uniform height and with adequate water status in the soil profile” [Penman, 1948, 1956]. Similarly, the term of maximum possible evaporation or evaporation capacity usually refers to the amount of water evaporated from surfaces not limited by the supply of water [Gao et al., 1978]. FAO (Food and Agriculture Organization of the United Nations) accepted the definition of reference surface and recommended the concept of reference evapotranspiration (RET) instead of other ambiguous expressions such as PET [Allen et al., 1998]. The FAO Penman-Monteith method is recommended as the single standard method for determining RET, which represents the evaporating power of the atmosphere at a specific location and time and does not consider the crop characteristics and soil factors [Allen et al., 1998].

[6] As an observational and physical representation of PET, pan evaporation has declined in many parts of the world during the last several decades [e.g., Peterson et al., 1995; Chattopadhyay and Hulme, 1997; Golubev et al., 2001; Liu et al., 2004; Roderick and Farquhar, 2004; Tebakari et al., 2005]. Recent studies suggested that the decline of pan evaporation, PET, or RET over the Qinghai-Tibetan Plateau was mainly attributable to the decrease of surface wind speed, regardless of the rising temperatures [Chen et al., 2006; Zhang et al., 2007]. Similar attributions have been reported for some other parts of China [Chen et al., 2005; Xu et al., 2006], the northeast of India [Jhajharia et al., 2009], the south of Canada [Burn and Hesch, 2007], and the most parts of Australia [Roderick et al., 2007]. Chen et al. [2006] indicated that changes in relative humidity played the supporting role affecting PET trends of the plateau during 1961–2000; while changes in sunshine duration played an insignificant role. Still, Zhang et al. [2007] suggested that the decrease of net total radiation was the secondary cause of the decreasing trend in RET of the plateau during 1966–2003. Since a climate factor may have a different impact on the RET process, its relative contribution to RET variation may also be different spatially. The relative significance of the climate factors of RET variation needs to be clearly quantified. Additionally, Chen et al. [2006] reckoned that the forcing factors of RET trends across the plateau were generally the same in all months, i.e., the decreasing wind speeds. However, there might be still some seasonal variation of the relative significance of the climate factors. In our study, we focused on the spatial and seasonal differences in the contributing factors to the RET trends over the plateau, which has not been specifically revealed in previous studies. Furthermore, we investigated the causes why the surface wind speed had decreased over the plateau, since it was identified as the main cause of the RET reduction.

2. Data and Methods

2.1. Meteorological Data

[7] Meteorological data are required for the FAO Penman-Monteith method. The discontinuities in the meteorological data over the plateau are mostly caused by weather station relocations and missing observations. In our study, observations with poor integrality before 1970s were discarded and weather stations that moved more than 30' in longitude or latitude or 100 m in altitude were filtered out to reduce any potential impact of observation discontinuities on the analyses. We abandoned five stations located in Qinghai Province, i.e., Mangya (no. 51866), Menyuan (no. 52765), Xinghai (no. 52943), Henan (no. 56065), and Banma (no. 56151), which were used in the studies by Chen et al. [2006] and Zhang et al. [2007].

[8] The data from 75 stations in China's National Meteorological Observatory (NMO) network were used for the RET calculation (Figure 1, Table 1) in our study. Daily maximum, minimum and mean air temperatures, mean relative humidity, wind speed and sunshine duration were provided by the National Meteorological Information Centre (NMIC) of China Meteorological Administration (CMA). The length of records for 74 stations is 1971–2004, but 1971–2000 for Chaka (no. 52842). As an average for the 74 stations, about 1.1% of the days were considered missing, if any of the weather observations was missing. Table 2 presents the number of months with missing data, i.e., month with all days missing, for all the 75 stations used for RET calculation during the period of 1971–2004. Besides, radiation observations from 16 stations (Figure 1, Table 1) were used to calibrate the Angstrom coefficients for estimating the solar radiation in the FAO Penman-Monteith method. Differing from the study by Zhang et al. [2007] which used 11 stations to calibrate the Angstrom coefficients, 5 more stations located in Xinjiang Uygur Autonomous Region and Gansu Province were used in our study to improve the estimates of solar radiation over the north of the plateau.

Figure 1.

Locations of meteorological stations on and around the Tibetan Plateau. Black dots indicate stations used for calculating reference evapotranspiration, and open circles stand for stations with radiation observations. All the stations are labeled by their WMO numbers.

Table 1. Information for the National Meteorological Observatory Stations Used in This Study
WMO NumberNameLatitude (°N)Longitude (°E)Altitude (m a.s.l.)
  • a

    Stations with radiation observation.

51709aKashi39.4775.981291.3
51777aRuoqiang39.0388.17890
51828aHetian37.1379.931374.7
51855Qiemo38.1585.551248.4
51931Yutian36.8581.651423.3
52418aDunhuang40.1594.681139
52533aJiuquan39.7798.481477.2
52602Lenghu38.7593.332770
52633Tuole38.898.423367
52645Yeniugou38.4299.583320
2652Zhangye38.93100.431482.7
52657Qilian38.18100.252787.4
52679Wuwei37.92102.671530.9
52707Xiaozaohuo36.893.682767
52713Dachaidan37.8595.373173.2
52737Delingha37.3797.372981.5
52754aGangcha37.33100.133301.5
52787Wuqiaoling37.2102.873045.1
52818aGeermu36.4294.92807.6
52825Nuomuhong36.4396.422790.4
52836Dulan36.398.13191.1
52842Chaka36.7899.083087.6
52856Qiapuqia36.27100.622835
52866aXining36.72101.752295.2
52868Guide36.03101.432237.1
52876Minhe36.32102.851813.9
52908Wudaoliang35.2293.084612.2
52996Huajialing35.381052450.6
55228aShiquanhe32.580.084278
55279Bange31.3890.024700
55299aNaqu31.4892.074507
55472Shenzha30.9588.634672
55578Rikaze29.2588.883836
55591aLhasa29.6791.133648.7
55598Zedang29.2591.773551.7
55664Dingri28.6387.084300
55680Jiangzi28.9289.64040
55696Longzi28.4292.473860
55773Pali27.7389.084300
56004Tuotuohe34.2292.434533.1
56018Zaduo32.995.34066.4
56021Qumalai34.1395.784175
56029aYushu33.0297.023681.2
56033Maduo34.9298.224272.3
56034Qingshuihe33.897.134415.4
56038Shiqu32.9898.14200
56046Dari33.7599.653967.5
56067Jiuzhi33.43101.483628.5
56079Ruoergai33.58102.973439.6
56080Hezuo35102.92910
56093Minxian34.43104.022315
56106Suoxian31.8893.784022.8
56116Dingqing31.4295.63873.1
56125Nangqian32.296.483643.7
56137aChangdu31.1597.173306
56144Dege31.898.583201.2
56146aGanzi31.621003393.5
56152Seda32.28100.333893.9
56167Daofu30.98101.122957.2
56172Maerkang31.9102.232664.4
56173aHongyuan32.8102.553491.6
56178Xiaojin31102.352369.2
56178Xiaojin31102.352369.2
56182Songpan32.65103.572850.7
56202Jiali30.6793.284488.8
56227Bomi29.8795.772750
56247Batang3099.12589.2
56257Litang30100.273948.9
56287Yaan29.98103627.6
56312Linzhi29.6794.332991.8
56357Daocheng29.05100.33727.7
56374Kangding30.05101.972615.7
56444Deqin28.4898.923319
56462Jiulong29101.52987.3
56543Zhongdian27.8399.73276.1
56548Weixi27.1799.282325.6
56043aGuoluo34.47100.253719
Table 2. Months With Missing Data for the 75 Stations Used for RET Calculation During 1971–2004
WMO NumberDaily Maximum, Minimum, or Mean TemperatureDaily Mean Relative HumiditySunshine DurationWind Speed
52633 February 1980  
52707April–December 1974April–December 1974April–December 1974April–December 1974
52818December 1993December 1993December 1993December 1993
528422001–20042001–20042001–20042001–2004
52876February 1992February 1992February 1992February 1992
55299October–December 1994October–December 1994October–December 1994October–December 1994
56038   June 1974
56257January, June–August, November 1971; November, December 1972   

[9] In order to examine the changes in circulation patterns and intensity as possible causes of climatic factors over the plateau, we obtained the monthly 500 hPa zonal wind speed and geopotential height data from the NCEP/NCAR Reanalysis data set [Kalnay et al., 1996] provided by the NOAA/OAR/ESRL PSD (Boulder, Colorado, USA, http://www.cdc.noaa.gov/).

2.2. FAO Penman-Monteith Method

[10] The reference surface is defined as “a hypothetical reference crop with an assumed crop height of 0.12 m, a fixed surface resistance of 70 s m−1 and an albedo of 0.23,” and the FAO Penman-Monteith method [Allen et al., 1998] to estimate RET (mm/d) is given as:

equation image

where:

Rn

net radiation at the crop surface [MJ m−2d−1]

G

soil heat flux density [MJ m−2 d−1]

T

mean daily air temperature at 2 m [°C]

u2

wind speed at 2 m [m s−1]

es

saturation vapor pressure [kPa]

ea

actual vapor pressure [kPa]

Δ

slope of the saturation vapor pressure curve at air temperature T [kPa °C−1]

γ

psychrometric constant [kPa °C−1].

[11] To ensure the integrity of computation, the weather measurements should be made at 2 m (or converted to that height) above an extensive surface of green grass, shading the ground and not short of water [Allen et al., 1998]. In China's NMO network, air temperature and humidity are measured at 1.5 m and wind speed at 10–12 m above ground. The measured daily minimum temperature, daily maximum temperature, and daily mean specific humidity at 1.5 m are respectively about 99%, 101%, and 101% of the measurements at 2 m, estimated according to the field observations during the Second Qinghai-Tibetan Plateau Experiment (TIPEX) from May to July 1998 [Liu et al., 2002]. In our study, the normal measurements of air temperature and humidity at 1.5 m were used instead of the required measurements at 2 m, since the differences are rather small and could be ignored. Wind speed at 2 m height was converted from the normal measurement at 10–12 m based on the logarithmic wind speed profile equation given by the FAO Penman-Monteith method [Allen et al., 1998]:

equation image

where:

z

height of measurement above ground surface [m]

uz

measured wind speed at z m above ground surface [m s−1].

2.2.1. Calibration of the Angstrom Coefficients

[12] In the FAO Penman-Monteith method, the solar radiation, if not measured, can be calculated using the Angstrom formula [Allen et al., 1998]:

equation image

where:

Rs

solar or shortwave radiation [MJ m−2d−1]

n

actual duration of sunshine [hour]

N

maximum possible duration of sunshine or daylight hours [hour]

n/N

relative sunshine duration [-]

Ra

extraterrestrial radiation [MJ m−2 d−1]

As

regression constant, representing the fraction of extraterrestrial radiation reaching the earth on overcast days (n = 0)

as + bs

the fraction of extraterrestrial radiation reaching the earth on clear days (n = N).

[13] In our study, the observations of n and Rs from 16 stations (Figure 1, Table 1) were used to calibrate the Angstrom coefficients as and bs. Ra and N were computed based on date and latitude according to the equations given by the FAO Penman-Monteith method [Allen et al., 1998]. To ensure the continuity of Rs and n, the years with over 10 missing days in any month were excluded (Table 3). Rs/Ra and n/N were then computed and as and bs were derived form the least squares fit of Rs/Ra on n/N by solving the linear model.

Table 3. Calibration of the Angstrom Coefficients for the 16 Stations With Radiation Observationa
WMO NumberNameLength of RecorddasbsR2
  • a

    Column d is the total number of days with available data. All values of as and bs are statistically significant (p < 0.001).

51709Kashi1958–2003147290.2480.4850.698
51777Ruoqiang1958–2003158790.2340.5140.742
51828Hetian1958–2003152890.2760.4750.744
52418Dunhuan1958–2003152360.230.5370.759
52533Jiuquan1993–200338360.2290.5270.811
52754Gangcha1993–200339090.1930.6690.831
52818Geermu1958–2003161220.2620.5630.81
52866Xining1959–2003152060.2050.570.717
55228Shiquanhe1972–2003104280.1340.6620.386
55299Naqu1961–2003129420.2080.5610.516
55591Lhasa1961–2003138080.2840.5430.616
56029Yushu1961–2003133490.2030.6220.715
56043Guoluo1993–200338310.2560.5790.819
56137Changdu1961–2003143530.2030.6350.667
56146Ganzi1994–200335050.3010.5180.825
56173Hongyuan1994–200334160.1970.6530.828

[14] The Angstrom coefficients of the 16 stations (Table 3) are rather close to the values reported by Chen et al. [2004]. Generally, the magnitudes of as + bs, i.e., the fraction of extraterrestrial radiation reaching the earth on clear days, are above 0.8 in higher altitude areas, but below 0.8 in lower altitude areas. Overall, the Angstrom model is suitable for daily global radiation estimation on the plateau as indicated by high R2 values (Table 3). For stations with no observation of solar radiation but sunshine duration, as and bs were estimated by Kriging interpolation using Surfer 8.0 (Golden Software, Golden, Colorado, USA).

2.2.2. Calculation of the Soil Heat Flux

[15] The soil heat flux (G) is the energy utilized in soil heat exchange. Although G is small compared to Rn and may often be ignored, the amount of energy gained or lost by the soil should be theoretically subtracted or added to Rn when estimating evapotranspiration [Allen et al., 1998]. In the FAO Penman-Monteith Method, the monthly value of G is given as:

equation image

or, if Tmonth, i+1 is unknown:

equation image

where:

Gmonth,i

soil heat flux of month i [MJ m−2 day−1]

Tmonth,i

air temperature of month i [°C]

Tmonth, i−1

air temperature of the previous month [°C]

Tmonth, i+1

air temperature of the next month [°C].

2.3. Statistical Methods

2.3.1. Long-term Means and Intraannual Variation Patterns of RET and Related Variables

[16] The monthly means of air temperature (T, °C), relative sunshine duration (S, %), relative humidity (RH, %), 2 m wind speed (U, m s−1) and sums of RET (mm) were calculated for spring (MAM), summer (JJA), fall (SON), and winter (DJF) for all 75 stations from 1971 to 2004. The missing data for all 75 stations but Chaka (no. 52842) were filled using the means of the values from the other years.

2.3.2. Regression Analysis

[17] Linear regression was applied to examine the trends of RET during the period of 1971–2004. Spatial distribution pattern of RET was examined using stepwise regressions, where mean annual RET was regressed against latitude ϕ, longitude λ, and altitude z. To determine the relative contributions of the climatic variables, we also regressed annual and seasonal RET against T, S, U, and RH for each station. The significance level for a predictor to be added into the model was set as 0.05, and the significance level for a predictor to be removed from the model was 0.1. The regression analysis was performed using Matlab 7.0 (MathWorks, Natick, MA, USA).

2.3.3. K-Means Clustering Analysis

[18] K-means clustering analysis [Seber, 1984; Spath, 1985] was used to identify the difference in the annual RET series of 74 stations from 1971 to 2004. K-means clustering can best be described as a partitioning method that divides the series into K mutually exclusive clusters. K-means treats each series as an object having a location in space. It finds a partition in which objects within each cluster are as close to each other as possible, and as far from objects in other clusters as possible. In our study, the “correlation distance” which equals to one minus the sample correlation between series was selected from the five different distance measures provided by Matlab 7.0. Each cluster in the partition is defined by its member objects and by its centroid, or center, to which the sum of distances from all objects in that cluster is minimized. K-means uses an iterative algorithm that minimizes the sum of distances from each object to its cluster centroid over all clusters. To avoid local minima as much as possible, we partition the 74 series into K clusters (varying from 2 to 74) with ten replicates performed for each partitioning solution. Each of the ten replicates begins from a different randomly selected set of initial centroids. For each clustering solution, there is a corresponding series of silhouette index, which ranges from +1.0, indicating points that are very distant from neighboring clusters, through 0, indicating points that are not distinctly in one cluster or another, to −1.0, indicating points that are probably assigned to the wrong cluster. Practically, the average silhouette value sharply increased with the number of clusters when the 74 series were partitioned into several decade clusters, but of little significance. Therefore in our study, the correct number of clusters should be smaller than ten while with a high average silhouette value and none (or few) minus silhouette values occurring.

3. Results and Discussion

3.1. Spatial Pattern of Mean Annual RET

[19] Spatial distribution of mean annual RET is shown in Figure 2. The regression equation was RET = 2342.028 − 12.055λ − 0.062z, with R2 = 0.332. This indicates that the mean annual RET generally decreases by 6 mm with rising 100 m of altitude and by 12 mm moving one degree of longitude eastward. Solar radiation increases with altitude because of thinner aerosphere, less air density, water vapor and aerosols; wind speed generally increases with altitude as an observed fact across the plateau; relative humidity is not clearly related to altitude because of the complex impacts of topography and atmospheric circulation [Zhang et al., 1982; Guan et al., 1984; Dai, 1990]. The decrease of RET with altitude is in accordance with the decrease of air temperature with altitude. The decrease eastward of RET, which does not agree with the spatial distribution of temperature, is determined by the spatial patterns of solar radiation, relative humidity, and wind speed, which generally decrease from the northwest to the southeast of the plateau throughout the year [Dai, 1990]. The predictor ϕ was excluded from the regression model, implying that the influence of latitude on the distribution of RET is smaller than and being replaced by that of altitude or longitude. Additionally, the low R2 value suggests that the spatial pattern of RET over the plateau is much determined by the complex influence of diverse terrain on the meteorological factors across the Qinghai-Tibetan Plateau [Dai, 1990].

Figure 2.

Mean annual reference evapotranspiration for the period of 1971–2000.

3.2. RET Trends During 1971–2004

[20] The trend slopes of annual and seasonal RET are shown in Figures 3 and 4. Significant decreasing trends were detected in annual RET for 39 stations (i.e., 52% of all stations), while increasing trends were found for only 6 stations (i.e., 8% of all stations). The remaining 40% of the stations presented no significant trends. Annual RET for stations with significant negative trends decreased by 20–60 mm/decade, except for Qiemo (no. 51855), which had a trend slope of −141.9 mm/decade. For most of these stations, significant negative trends were found in at least two seasons. However, annual RET for stations with significant positive trends increased by only about 20 mm/decade, and the increasing trends mainly occurred in autumn and summer.

Figure 3.

Annual trend slopes of reference evapotranspiration during 1971–2004.

Figure 4.

Seasonal trend slopes of reference evapotranspiration during 1971–2004.

3.3. Climate Factors for RET Change

[21] According to the K-means analysis on the annual RET series of the 74 stations from 1971 to 2004, two clusters can be detected. The stations of Cluster 1 are mainly to the south of 33°N and stations of Cluster 2 to the north (Figure 5). This pattern implies the possible differences in the relative contribution of climate factors for RET change, since RET is governed by solar radiation, air temperature, air humidity, and wind speed.

Figure 5.

Partition of stations according to the two-cluster solution by K-means analysis on the annual series of reference evaportanspiration from 1971 to 2004. The size of the proportional symbol stands for the inversed distance to the cluster centroid, which equals to one minus the correlation coefficient between the member series and the centroid.

[22] Stepwise regression was completed for each of the 75 stations, with the annual means of T, S, U and RH as the predictors and the annual RET as the dependent variable from 1971 to 2004 (2000 for Chaka, Table 4). The regression models are characterized by high R2 values, with an average of 0.915. For 58 stations, all four climate variables were included in the regression models, which means that air temperature, radiation, wind speed and air humidity all had significant influences on RET variation at these stations. As an exception, however, U was excluded for Pali (no. 55773) since its regression coefficient is negative, which is against the physical process of evaporation in which wind speed is positively correlated to ET, even though it is statistically significant (p < 0.1). RH did not enter the models for three stations (no. 56227, 56093, and 56257), suggesting that humidity played a less important role than the other factors in humid areas. T entered all models but for three stations (no. 51855, 55299, and 56543), and S was excluded for only two stations (no. 51777 and 55299). U did not enter the models for twelve stations, mostly found in the central plateau. Naqu (no. 55299) is distinguished from all other stations, for only RH was included in its regression model.

Table 4. Stepwise Regression Analysis With Annual Means of T (°C), S (%), U (m s−1), and RH (%) as Predictors and Annual RET (mm) as Dependent Variable, Based on the Period of 1971–2004a
WMO NumberIntercept (mm)Regression Coefficient forR2
TSURH
  • a

    All the models are statistically significant (p < 0.01).

51709705.0415.843.79231.49−6.190.95
51777796.7134.680287.72−9.310.91
51828497.9721.154.92246.33−4.870.95
5185551307.54223.75−3.380.99
51931340.3219.795.4265.31−2.90.97
52418504.9428.144.73215.04−6.240.94
52533510.6827.024.8164.9−5.50.94
52602508.5432.955.55142.64−5.440.95
52633591.8622.953.9276.81−2.690.85
52645698.1129.864.8439.77−4.120.9
52652674.2431.35.24135.01−7.690.96
52657615.1631.395.9566.89−4.970.86
52679427.3719.796.69156.87−4.540.97
52707832.0441.322.66144.04−8.380.96
52713593.4827.664.94117.65−4.940.84
52737436.0729.26.96116.23−4.120.96
52754792.5237.314.845.67−5.990.91
52787813.08285.4134.4−8.230.91
52818554.9834.975.76163.53−7.360.97
52825622.2842.944.12132.06−5.430.94
52836743.9233.124.14106.22−6.310.95
52842806.1234.795.0170.17−7.620.96
528561094.324.590103.22−6.720.93
52866539.6616.34.997.41−2.780.95
52868567.8936.555.1592.75−5.230.88
52876503.2329.934.6122.38−3.60.95
52908973.3424.494.250−6.610.92
52996906.0826.716.170−7.590.82
55228450.9823.386.5885.04−2.660.85
55279842.0534.687.140−7.210.94
552991439.8000−11.120.86
55472733.1735.845.7449.13−5.780.96
55578675.630.925.6106.07−5.090.96
55591644.826.577.52114.7−6.980.93
55598805.6435.675.6118.12−9.670.99
55664313.0929.5510.8968.97−5.060.96
55680522.2234.897.3103.37−4.940.97
55696786.5935.026.291.43−7.980.93
55773912.5919.075.78−10.75−6.060.9
56004741.3315.966.1422.81−6.30.82
56018709.8121.875.1345.86−3.970.87
56021613.2126.627.3636.43−4.790.84
56029381.9224.948.2284.88−1.880.92
56033804.1918.495.180−4.760.88
56034698.0222.475.250−2.890.87
56038662.5223.526.740−3.420.9
56046746.0424.584.9224.32−3.650.93
56067766.7716.676.090−4.580.65
56079994.8218.884.70−7.080.76
56080507.0322.986.0750.66−2.680.85
56093247.2423.717.2763.2200.9
56106684.6425.315.4763.54−4.610.93
56116597.7725.277.4757.34−4.740.93
56125584.5624.497.1176.52−4.450.93
56137453.3722.558.07122.55−3.060.97
56144606.0329.96.7292.84−4.680.98
56146578.5131.726.956.8−3.660.94
56152752.1627.075.6626.55−4.220.87
56167523.231.326.76100.11−3.670.94
56172468.2821.569.6277.16−3.540.93
56173637.9127.727.670−4.150.86
56178432.1633.118.85176.47−6.320.98
56182499.2717.588.4450.51−2.810.8
56202836.221.075.0224.17−5.570.86
56227213.4921.619.9380.0900.92
56247375.8425.756.96226.87−1.950.97
56257319.9222.38.1853.6700.9
56287858.2113.3610.2161.48−5.990.93
56312780.8517.467.2378.33−6.370.91
56357697.9619.958.1976.17−6.730.95
56374763.8633.325.9729.69−5.960.94
56444959.1728.347.350−8.040.94
56462822.4334.888.3170.49−9.190.96
56543959.6808.330−6.60.87
56548639.7215.79.3753.2−4.250.93

[23] The absolute values of the standardized regression coefficients for T, S, U, and RH are mapped as bar graphs in Figure 6 to reveal the spatial pattern of relative importance of the climate factors on RET variation. Apparently, the influence of the same climate factor can vary spatially. T generally played a more important role in the northeast of the plateau. S had a greater impact in the southeast than in the north. The influence of U was much stronger to the north of 33°N than that to the south. RH mainly affected the variation of RET at several stations in the central plateau.

Figure 6.

The absolute value of the standardized regression coefficients of T (annual mean temperature), S (annual mean relative sunshine duration), U (annual mean 2-m height wind speed), and RH (annual mean relative humidity), with annual reference evapotranspiration as the dependent variable for the period of 1971–2004. Stepwise regression is used to derive the linear model.

[24] K-means clustering was also performed on the absolute values of the standardized regression coefficients of T,S,U, and RH, as a method to further determine the leading factor of RET variation in different areas. The 75 stations were partitioned into three clusters (Figure 7), with the leading factors of 1) wind speed, 2) humidity, and 3) radiation. The smaller increasing rates of daytime temperatures than that of daily mean temperatures on the Qinghai-Tibetan Plateau had limited the importance of mean temperature trends for changes of PET rates [Chen et al., 2006]. The increase in mean temperature showed little correlation with the declining trends in RET and pan evaporation [Zhang et al., 2007]. Therefore there is no cluster with temperature as the leading factor. Generally, there is a contiguous area in the north of the study region dominated by Cluster 1 stations, while the southeastern part has mostly Cluster 3 stations. The Cluster 2 stations are intermingled with the other two clusters. This suggests that the changes in RET for most stations, especially those north of 33°N, were mainly attributable to the changes in surface wind speed; while changes in sunshine duration were generally responsible for RET changes in the southeast of the plateau.

Figure 7.

Partition of stations according to the three-cluster solution, based on the K-means analysis on the absolute value of the standardized regression coefficients of T (annual mean temperature), S (annual mean relative sunshine duration), U (annual mean 2-m height wind speed), and RH (annual mean relative humidity), with annual reference evapotranspiration as the dependent variable during 1971–2004. The size of the proportional symbol stands for the inversed distance to the cluster centroid, which equals to one minus the correlation coefficient between the member series and the centroid.

[25] The relative importance of the climate factors on RET may vary with seasons. Again, high R2 values indicated strong robustness of the models of the 75 stations (R2 = 0.97, 0.98, 0.91 and 0.95 for MAM, JJA, SON and DJF, respectively). Figure 8 indicates the seasonal variation of the relative significance of the climate factors for changes in RET across the plateau. In summer, radiation was the dominant factor for most stations to the south of 33°N, while wind speed for those to the north, similar to that of the annual RET (Figure 6). In spring and fall, the number of stations strongly influenced by radiation was much smaller than in summer. In fall, different from spring, humidity became the dominant factor at several stations mainly located in the central plateau, e.g., Bange (no. 55279), Tuotuohe (no. 56004), and Maduo (no. 56033). In winter, the impact of radiation was much weaker than the other factors for most part of the plateau. Therefore wind speed dominated almost throughout the year for most stations to the north of 33°N.

Figure 8.

Same as in Figure 6, but for the seasonal scales.

3.4. Main Reasons for RET Decrease

[26] The annual and seasonal trends of the climate factors, i.e., T, S, RH, and U, were examined for each station during 1971–2004 to determine the reason for the detected RET trends. In general, significant increases in T and significant decreases in U were detected almost across the plateau throughout the year; S significantly decreased to the south of 33°N in JJA, SON, and MAM; RH significantly increased to the south of 33°N, but most of the increasing trends were detected in DJF.

[27] Five stations with significant RET trends were selected to show the contributing factors and causes for the RET trends (Table 5). Zedang is located in the middle reaches of the Yalungzangbo River to the north of the Himalayas, and Maduo in the source area of the Yellow River in central Qinghai Province. Dege and Hezuo are closest to the centroids of Cluster 1 and Cluster 2, respectively, in the K-means analysis on annual RET change (Figure 5). Qiemo was selected because of its extreme decreasing trend of RET.

Table 5. Trend Slope and Absolute Value of the Standardized Regression Coefficient of the Climatic Factors for Five Selected Stations During 1971–2004a
NumberNameRET (mm/10a)Trend Slope/Absolute Value of the Standardized Regression Coefficient
TSURH
  • a

    A star (*) indicates significance at p < 0.05 and two stars (**) at p < 0.01 through the t test. Trend slope of T, S, U, and RH are expressed in °C/10a, %/10a, m s−1/10a, and %/10a, respectively.

55598Zedang−57.6**0.365**/19.03−1.5**/11.92−0.464**/62.860.8/23.26
56033Maduo18.9**0.425**/15.341.2**/12.89−0.107**/0−1.9**/18.67
56144Dege−58.1**0.097/12.46−5.1**/38.75−0.189**/21.522.1**/14.52
56080Hezuo12.3**0.407**/12.961*/16.38−0.112**/7.7−0.8*/5.49
51855Qiemo−141.9**0.407**/0−2.9**/35.73−0.512**/116.441.6**/10.43

[28] At Zedang, the impact of wind speed was the strongest among the four climatic factors and, therefore the significant decrease of wind speed should be cause of the decrease of annual RET (Table 5). At the seasonal scale, the impact of wind speed was essentially the same as that of radiation in summer, but predominated in the other seasons. At Dege, the leading factor for annual RET change was radiation (sunshine duration), followed by wind speed. Summer and fall were dominated by radiation; while wind speed had a slightly greater impact in spring and predominated in winter. At Qiemo, the impact of wind speed was much greater than that of the other factors throughout the year. Therefore the observed significant decrease of RET was mostly attributed to the significant decline of wind speed at both annual and seasonal scales, although the decreasing sunshine duration and increasing relative humidity also played a role. At Maduo, the impact of humidity was slightly stronger than that of temperature and radiation for annual RET change, while wind speed had no significant impact. Similar patterns were found at the seasonal scale. At Hezuo, radiation played the most important role in annual RET change. Summer and fall were predominated by radiation, while temperature and radiation contributed similarly in spring, and winter was predominated by wind speed.

[29] The above analyses indicate that the decrease of surface wind speed was mainly responsible for the RET reduction in the north of the plateau. To determine the cause of the decrease in surface wind speed over the plateau, we first examined the trends of the zonal wind speed at the 500 hPa geopotential height surface for the warm months of March through October when most RET occurs. The Spearman's rank correlation coefficients between the 500 hPa zonal wind speed and the time variable YEAR were used as the indicator for temporal trends during 1971–2004 [Mitchell et al., 1966]. For most part of the plateau, the March–October 500 hPa zonal wind speed decreased during the study period, with significant decreasing trends found in the northeast of the region (Figure 9). This finding points to the changes in circulation patterns and intensity as the possible cause of the decline of surface wind speed. To confirm this, we analyzed the 500 hPa geopotential heights at both 27.5°N and 40°N, with each averaged for the longitudes 80°–102°E. Although the geopotential heights at these two latitudes both increased during 1971–2004, the South-North pressure gradient declined with statistical significance (Figure 10), which led to the decreases in the upper-air zonal wind speed and the reduced strength of circulation over the plateau.

Figure 9.

The Spearman's rank correlation coefficients between March and October mean 500 hPa zonal wind speed and the time variable YEAR during 1971–2004. The thin solid contours indicate positive values; while the dotted contours and the shaded area indicate negative values. The thick solid contours stand for statistical significance levels.

Figure 10.

Change of the South-North gradient of geopotential heights for March–October on the 500 hPa surface over the Tibetan Plateau. The gradient is indicated by ΔGH, i.e., the difference in geopotential heights between the latitudes of 27.5°N and 40°N, averaged for 80°–102.5°E. The linear trend line of ΔGH is statistically significant at p < 0.01.

[30] Our analyses also suggest that the reduced radiation (sunshine duration) was the leading factor for RET decrease at most stations in the southeastern plateau during summertime. This conforms to the general decrease in pan evaporation in the Northern Hemisphere associated with observed large and widespread decreases in sunlight due to increasing cloud coverage and aerosol concentration during the past 50 years [Roderick and Farquhar, 2002]. Visibility in the clear sky reduced by the presence of aerosols has resulted in net global dimming over land from 1973 to 2007 [Wang et al., 2009]. Liu et al. [2004] suggested that the aerosol-caused decrease in solar irradiance (sunshine duration) was most likely the driving force for the reduced pan evaporation in China, which coincided with the study by Ren and Guo [2006]. Sunshine duration was also identified as the most influencing variable for pan evaporation changes in the northeast of India in winter, monsoon and pre monsoon seasons [Jhajharia et al., 2009]. Du et al. [2007] indicated that the decrease of annual and summer sunshine duration in Tibet during 1971–2005 was mainly related to the increase of atmospheric water vapor pressure and precipitation. Up to now it is difficult to confirm the impact of atmospheric aerosol on sunshine duration because of the absence of observations on atmospheric aerosol in Tibet. However, a recent study revealed that a significant amount of brown clouds that consist of a mixture of light-absorbing and light-scattering aerosols is being generated in South Asia by fossil fuel and biomass burning [Ramanathan et al., 2007], which may be transported over the plateau by the summer monsoonal circulation pattern and cause the decreases in sunshine duration in recent years.

[31] In our study, using a corrected data set of climatic variables, we confirmed the general pattern of decreasing trends of RET across the Qinghai-Tibetan Plateau as identified by Chen et al. [2006] and Zhang et al. [2007]. Still, we detected the spatial pattern of the RET variations across the plateau, i.e., the difference between the north and south of 33°N (Figure 5), which implies some spatial variations of the controlling factors of the RET changes. We confirmed that the combined effect of the reduced wind speed and shortened sunshine duration negated the effect of rising temperature and caused the RET reduction [Zhang et al., 2007]. However, our analyses suggest that there were both spatial and seasonal differences of the contributing factors to the RET trends across the plateau. It is concluded that wind speed predominated the changes of RET almost throughout the year, especially in the north of the study region; while radiation was the leading factor in the southeast, especially during the summertime. Moreover, we found that the South-North pressure gradient in the upper air decreased significantly over the plateau, which was the main cause of the reduced strength of circulation and the reduced surface wind speed.

3.5. Possible Change in AET Related With Decreasing RET

[32] Ultimately, AET is the hydrologic flux of interest, while pan evaporation, PET, or RET matters insofar as it can offer a useful clue to the direction of the change in AET [Ohmura and Wild, 2002]. However, decreasing pan evaporation, PET, or RET cannot simply indicate decreasing AET. On the contrary, AET and pan evaporation, PET, or RET can be inversely related in certain cases. Actually, the decline in pan evaporation has been reported to be associated with an incline in actual evaporation in some climates [e.g., Lawrimore and Peterson, 2000; Golubev et al., 2001], which supports the proposal by Brutsaert and Parlange [1998] that decreasing pan evaporation actually provides a strong indication of increasing terrestrial evaporation in many situations. Theoretically, PET and AET for large homogeneous surfaces with minimal advection of heat and moisture fall in a complementary relationship, i.e., the Bouchet-Morton complementary relationship [Bouchet, 1963; Morton, 1975]. Ramírez et al. [2005] provided direct observational evidence of the complementary relationship in regional evapotranspiration hypothesized by Bouchet in 1963. This complementary relationship has been examined and adopted to estimate regional AET [e.g., Sugita et al., 2001; Hobbins et al., 2001; Xu and Singh, 2005; Ozdogan et al., 2006].

[33] As for the Qinghai-Tibetan Plateau, regional water budget-derived AET generally presented insignificant increasing trends [Zhang et al., 2007], which coincided with the detected increasing trend in vapor pressure, as well as temperature and precipitation [Wu et al., 2007; Du et al., 2007]. AET and PET exhibited some complementary behavior on the plateau, which, however, did not fall in Bouthet's hypothesis [Zhang et al., 2007]. To explain the trend in water budget-derived AET and the relationship between pan evaporation and AET, Hobbins et al. [2004] emphasized that both two driving components, i.e., the radiative energy budget (Qn) and the vapor transfer budget (EA), must be considered together. These two budgets were separately addressed by Szilagyi et al. [2001], Milly and Dunne [2001], and Roderick and Farquhar [2002]. According to our analyses, the reduced sunshine duration that caused Qn to decrease was mainly responsible for the decreasing trend of RET in the southeast of the plateau, though the decreasing wind speed and increasing relative humidity that caused EA to decrease also contributed to decreasing RET. The possible increasing trend in AET can be explained in the context of the complementary relationship, i.e., AET = λWET − PET, where WET is wet environment evapotranspiration, and λ is a constant greater than one but less than two according to Zhang et al. [2007]. WET decreases with declining Qn, PET decreases with declining Qn and EA, but AET may increase when the decreasing rate of PET exceeds that of λWET.

4. Conclusion

[34] In this study we examined the spatial and temporal patterns of reference evapotranspiration (RET) calculated using the FAO recommended Penman-Monteith method over the plateau and its vicinity during 1971–2004. Generally, both annual and seasonal RET decreased for most part of the plateau during the study period. At the annual and seasonal scales, we identified the relative contribution of wind speed, sunshine duration, temperature, and relative humidity to variations of RET. The combined influence of wind speed, radiation (sunshine duration), and humidity was greater than that of temperature on the change of atmospheric evaporating power, while the impacts of the climate factors of RET varied spatially and seasonally. Wind speed predominated the change of RET almost throughout the year, especially in the north of the study region; while sunshine duration was the leading factor in the southeast, especially during the summertime. Although the recent warming trend over the plateau could have increased RET through changes in temperature, the combined effect of the reduced wind speed and shortened sunshine duration negated the effect of the rising temperature and caused RET to decrease in general across the plateau. The significant decline of surface wind speed over most part of the plateau was in accordance with the detected decreasing trends of zonal wind speed on the 500 hPa surface. We found that the geopotential height generally increased over the plateau for the months March–October, but the 500 hPa pressure gradient decreased statistically over the study region. The reduced pressure gradient between 27.5°N and 40°N, possibly as the result of the recent warming, is identified as the main cause of the reduced surface wind speed.

Acknowledgments

[35] This study was in part supported by the National Basic Research Program of China (2005CB422006), the Chinese Academy of Sciences Knowledge Innovation Program (KZCX2-YW-310), the National Natural Science Foundation of China (40871044, 40625002), NASA (NNG05GB85G/EOS/03-0063-0069), and University of San Diego (FRG 07-08, 08-09). We thank the Climate Data Center of National Meteorological Information Center of China Meteorological Administration for providing surface observations, the NOAA/OAR/ESRL PSD (Boulder, Colorado) for providing the NCEP/NCAR reanalysis data, and the anonymous reviewers for their valuable comments and constructive suggestions.

Ancillary

Advertisement