Quantifying the impact of climate variables on reference evapotranspiration in Pearl River Basin, China

ABSTRACT

The impact of climate variables on monthly reference evapotranspiration (ET_o) is a critical issue in water resources management and irrigation planning. The spatio-temporal contribution of climate variables to ET_o in the Pearl River Basin (PRB), China, from 1960 to 2016 were calculated based on sensitivity and relative change of each climatic variable. The results show that annual ET_o total decreased by 1.64% and diminished in magnitude from the southeast to the northwest. Sunshine duration, wind speed and relative humidity decreased by 15.5%, 7.4%, and 4.0%, respectively, while average temperature increased by 4.25%. The ET_o showed a positive sensitivity to all variables except relative humidity, which showed a negative sensitivity. Sunshine duration had the highest contribution of −4.26%, and the overall decrease in ET_o was mainly caused by the declines in sunshine duration and wind speed, which offset the positive impact of rises in average temperature and reduction in relative humidity.

KEYWORDS:

• evapotranspiration
• spatio-temporal
• climate variable
• sunshine duration
• Pearl River Basin

Editor A. Castellarin Guest editor M. Borga

1 Introduction

Global climate change is projected to result in major impacts on the water cycle and energy balance at regional scales (Prudhomme et al. 2013, Guo et al. 2017). Evapotranspiration (ET) is a major component of both the hydrological cycle and the land surface energy balance, and is a critical link between the two.

Evapotranspiration from a hypothetical grass reference crop, with an assumed crop height of 0.12 m that closely resembles an extensive surface of green, well-watered grass of uniform height, actively growing and completely shading the ground, is called the reference evapotranspiration and is denoted as ET_o (Allen et al. 1998). The ET_o in China as been reported to show a decreasing trend as a result of global change between 1954 and 1993 (Thomas 2000). Hobbins et al. (2004) reported pan ET data for the USA showed a decrease from 1950 to 2002 in 64% of the stations examined. Although it is conjectured that ET_o should have an increasing trend with the rising global temperatures (Tabari et al. 2011), some researchers, as cited earlier, report that ET_o has shown a decreasing trend. Some have called this apparent contradiction the “evaporation paradox” (Brutsaert and Parlange 1998). However, as ET_o is controlled by several other atmospheric variables besides temperature, climatically linked changes in other variables have been cited as possible drivers for observed regional ET_o trends. The reduction in ET_o in some regions (e.g. Canada and the Tibetan Plateau) may be linked to decreases in wind speed (Burn and Hesch 2007, Zhang et al. 2009b), or affected by the reduction of radiation, as reported for the Yangtze River (Xu et al. 2006, Cong et al. 2009) and the Northern Hemisphere (Roderick and Farquhar 2002), or an increase in relative humidity (Bandyopadhyay et al. 2009). Most of these studies are based on conventional trend and correlation analysis methods (Herath et al. 2017).

Recently, Yin et al. (2010) combined sensitivity analysis with a relative change methodology to quantify the impact of different climate variables on ET_o. Sensitivity analysis attributed relative humidity as the main variable controlling ET_o in the Songnen Plain (Ma et al. 2017), while it was sunshine duration for the Jialing River (Herath et al. 2017), radiation for Jiangsu (Chu et al. 2017), and average temperature for Australia (Guo et al. 2017). A similar analysis found different variables controlled ET_o in different regions: net radiation in the Yangtze River (Xu et al. 2006), average temperature in southwest China (Liu et al. 2016), relative humidity in the Loess Plateau (Ning et al. 2016), and wind speed in the Yellow River region (She et al. 2017).

Because many studies use multi-monthly seasonal or annual datasets for their analysis, they may miss significant relationships shown with finer time scales such as seen in monthly climatic data (Sun et al. 2017). In addition, monthly time scales may be more important to water resource managers and irrigation planners than coarser temporal scales. Additional factors include spatial heterogeneity of climatic variables and ET_o that may necessitate interpolation methods; different studies have used different methods (Xu et al. 2006).

The Pearl River has the second largest annual average flow in China, with the Pearl River Basin (PRB) spanning tropical and sub-tropical geoclimatic regimes (Zhang et al. 2018). The PRB is critical to the water security and agriculture in the basin (Li et al. 2013). For example, the Pearl River Delta (PRD) is the centre of Guangdong-Hong Kong-Macao Greater Bay Area region and accounts for more than 10% of the Chinese Gross Domestic Product (GDP) (Zhang and Chen 2017). The Pearl River’s tributary, the Dongjiang River, is the main drinking water source for Shenzhen City and Hong Kong (Zhang et al. 2009a). Previously, there have been no studies on the impact of climate change on ET_o in the PRB.

Our study uses the daily averages of relative humidity, sunshine duration average temperature and wind speed gathered at 54 stations from 1960 to 2016, using these values to calculate monthly ET_o based on the Penman-Monteith formula. Spatial interpolation is implemented to create a regional dataset of ET_o. These data are compared with pan evaporation data from 38 stations. The main objectives of this study are: (1) to quantify the spatio-temporal change of ET_o and climatic variables in the PRB, (2) to establish ET_o sensitivity to climate variables, and (3) to quantify the contribution of climate variables to monthly ET_o.

2 Materials and methods

2.1 Study area

The Pearl River Basin (PRB) is drained from the three tributaries, the Xijiang, Beijiang and Dongjiang rivers, and the Pearl River Delta (PRD) around Guangzhou-Taishan-Shenzhen (Fig. 1). The PRB is located in tropical and subtropical monsoon climate zones of southern China between 102°14′–115°53′E and 21°31′–27°0′N, with an area of 47.27 × 10⁴ km². The elevation gradually increases from east to west, where the Yunan-Guizhou Plateau lies. Within this transect, the highest altitude is approximately 2933 m a.s.l. To the southeast is the coastal area, and to the north, mountains (Zhang and Chen 2017). The annual average temperature is approximately 19.4°C, and the annual average precipitation is 1496 mm, mainly concentrated between April and September (Sun et al. 2016).

Figure 1. Location of the Pearl River Basin study area.

2.2 Data

There are more than 100 meteorological stations in and around the basin, with data available from the China National Meteorological Information Centre archive (http://data.cma.cn/). Stations were excluded if they had missing data for more than 1% of time, or if they lacked data on 60 or more continuous days. Finally, we chose 54 stations that had high quality daily relative humidity (RH), sunshine duration (SD), average temperature (TAV) and wind speed (WS) for the period from 1960 to 2016. Spikes in data were corrected by replacing them with the mean value of their adjacent days; this process did not influence the long-term trend of meteorological data.

2.3 Analysis method

2.3.1 Penman-Monteith equation

The ET_o was calculated with the Penman-Monteith equation using coefficients appropriate for well-watered grass (Goyal 2004):

(1) $\mathrm{E}{\mathrm{T}}_o=\frac{0.408\mathrm{\Delta }\left(R_n-G\right)+\gamma \frac{900}{T+273}{u}_2\left(e_s-e_a\right)}{\mathrm{\Delta }+\gamma \left(1+0.34u_2\right)}$(1)

where ET_o is the reference evapotranspiration (mm/d); R_n is the net radiation (MJ m⁻² d⁻¹); G is the soil heat flux density (MJ m⁻² d⁻¹); T is the mean daily air temperature at 2 m height (ºC); u₂ is the wind speed at 2 m height (m s⁻¹); e_s is the saturation vapour pressure (kPa); e_a is the actual vapour pressure (kPa); Δ is the slope of the vapour pressure curve (kPa ºC⁻¹); and γ is the psychrometric constant (kPa ºC⁻¹). Variables needed for Equation (1) that were not directly measured were calculated using the FAO protocols (Allen et al. 1998), such as obtaining net radiation R_n from sunshine duration data.

2.3.2 Data comparison with pan evaporation

Pan evaporation (ET_pan) is a simplicity and efficient method to measure daily ET, widely used over the word. The long-term measured ET_pan data over different regions can be used to verify the spatio-temporal characteristics of ET_o. The ET_pan data from 38 stations with E601 between 2000 and 2014 were compared with ET_o calculated using Equation (1). The correlation coefficient between the annual ET_o of the 38 stations and annual ET_pan was 0.86, with a slope of 0.5716 (Fig. 2), and is consistent with the relationship of ET_pan to ET_o under humid conditions (Grismer et al. 2002). This agreement suggests that the quality of the temporal and spatial ET_o data is reliable.

Figure 2. Relationship between the observed annual pan evaporation, ET_pan, and calculated annual reference evapotranspiration, ET_o.

2.3.3 Interpolation

The ET_o for each of the 54 stations (Section 2.2) was used to create a spatial dataset for the PRB region using spatial interpolation of the meteorological variables and ET_o. The interpolation was assessed by using 40 stations for developing the spatial interpolation, and 14 stations used to compare with the interpolation. Ordinary kriging with spherical variogram (Ks), ordinary kriging with exponential variograms (Ke), universal kriging with linear drift variograms (Kl) and inverse distance weighting (IDW) were used with ArcGIS software for the interpolation (Zhang and Chen 2017).

To test the interpolation, SPSS software (IBM; ibm.com) was applied to by calculate correlation coefficients. We calculated the correlation between interpolated and observed values of ET_o, and the meteorological variables at the 14 test stations for each year in the period 1960–2016. There have many papers that compared ordinary kriging and IDW interpolation (e.g. Tegos et al. 2015). For this study, the results (Table 1) indicate that kriging and IDW have similar results, but Ks (0.857) is much better than IDW (0.318) for TAV interpolation, so we chose Ks to interpolate ET_o, and meteorological variables into 1 km × 1 km spatial resolution gridded data.

Table 1. Correlation coefficients between interpolation result and observation data. IDW: inverse distance weighting; Ke: ordinary kriging with exponential variograms; Kl: universal kriging with linear drift variograms; Ks: ordinary kriging with spherical variogram. ET_o: reference ET; RH: relative humidity; SD: sunshine duration; TAV: average temperature; WS: wind speed.

	Method
Factor	IDW	Ke	Kl	Ks
ETo	0.827**	0.824**	0.809**	0.817**
RH	0.755**	0.466**	0.752**	0.749**
SD	0.947**	0.951**	0.938**	0.950**
TAV	0.318*	0.850**	0.857**	0.851**
WS	0.343**	0.366**	0.283*	0.349**

**indicates significance at the 0.01 probability level; *indicates significance at the 0.05 probability level.

2.3.4 Trend

Sen’s slope was used to get a less biased slope estimate (Sen 1968), while the nonparametric Mann-Kendall (MK) trend test (Mann 1945, Kendall 1948) was employed to detect the significance of temporal ET_o changes of each grid (Tabari et al. 2011).

2.3.5 Sensitivity

The relative sensitivity of ET_o to the changes in each particular meteorological variable was quantified with a partial derivative method (McCuen 1974, Saxton 1975). Practically, the partial derivatives were estimated by changing each variable by 10%, while the other variables were held constant (Chu et al. 2017):

(2) $S_{v_i}=\underset{\mathrm{\Delta }{v}_i/v_i}{lim}\left(\frac{\mathrm{\Delta }\mathrm{E}{\mathrm{T}}_{\mathrm{o}}/\mathrm{E}{\mathrm{T}}_{\mathrm{o}}}{\mathrm{\Delta }{v}_i/v_i}\right)\approx \frac{\mathrm{\partial }\mathrm{E}{\mathrm{T}}_{\mathrm{o}}}{\mathrm{\partial }{v}_i}\times \frac{v_i}{\mathrm{E}{\mathrm{T}}_{\mathrm{o}}}$(2)

where S_vi is a dimensionless sensitivity coefficient, ∆ is the difference operator for each variable (not to be confused with the vapoir pressure slope symbol used in Equation (1)), and v_i is each climate variable indexed by i. If S_vi > 0, this indicates that ET_o increases with increases in the particular climate variable, while S_vi < 0 indicates that ET_o decreases with increases in a particular climate variable.

2.3.6 Contribution

Although Equation (2) reveals the ET_o sensitivity to relative changes, it does not account for the relevance of any particular variable for a given region or season where its relative change may be negligible or substantial, compared to the other variables. For any particular set of conditions, the contributions of a target variable to ET_o can be defined by (Yin et al. 2010, Li et al. 2017a):

(3) $\mathrm{C}\mathrm{o}{\mathrm{n}}_{v_i}=S_{v_i}\times \mathrm{R}{\mathrm{C}}_{v_i}$(3)

(4a) $T_{v_i,j}=\frac{\mathrm{\partial }{v}_i}{\mathrm{\partial }t}\delta t_j$(4a)

(4b) $\mathrm{R}{\mathrm{C}}_{v_i}=\frac{n\times T_{v_i,j}}{\left|\mathrm{a}{\mathrm{v}}_i\right|}\times 100\mathrm{\%}$(4b)

where Con_vi is the time-trend based contribution of climate variable v_i to ET_o; S_vi is the climate variable sensitivity coefficient defined in Equation (2); RC_vi is the relative change over time frame n – defined in Equation (4b); n is the length of the study period (57 years in this study); T_vi,j is the yearly or monthly change (designated by subscript j) of each climate variable v_i; and av_i is the mean variable averaged over multiple years or months, depending on the desired time frame. In this study j was set to a monthly time interval. If Con_vi > 0, the climate variable has a positive contribution to the variation of ET_o over the 57 months, while if Con_vi < 0, this indicates a negative contribution to ET_o.

3 Results

3.1 Spatio-temporal change of ET_o

The annual ET_o in the PRB showed a slightly decreasing trend from 1960 to 2016 (Fig. 3), varying from a minimum of 998.9 mm to a maximum of 1180.0 mm (annual average: 1080.6 mm). A greater decreasing trend is seen from 1960 to 1997, but then there is an increasing trend from until 2016. Monthly ET_o showed significant seasonal change, with a high of 131.7 mm and a low of 51.1 mm in January.

Figure 3. Variation of the annual ET_o in the Pearl River Basin in the period 1960–2016.

Spatially, annual ET_o showed an increasing trend from the northwest to the southeast (Fig. 4). The highest ET_o (1286.3 mm) was in the Pearl River Delta region, near Shenzhen. The lowest annual ET_o (907.9 mm) was within an arc running from Shuicheng in the west to Rongjiang in an easterly direction. Areally, 13.1%, 44.7% and 39.7% region of the basin had annual ET_o between 900 and 1000 mm, 1000 and 1100 mm, and 1100 and 1200 mm, respectively. Only 2.5% of the basin area had ET_o greater than 1200 mm.

Figure 4. Spatial distribution of the annual ET_o in the Pearl River Basin.

The highest monthly ET_o, near the Pearl River Delta and Mengzi-Yuxi, occurred during January to April and east of Guilin-Nanning in July (Fig. 5), on a spatial (not seasonal) basis. After August, the western region first showed a decreasing monthly ET_o trend, with the entire basin showing a decreasing trend reaching its lowest values in November–December.

Figure 5. Monthly ET_o in the Pearl River Basin.

The annual ET_o change expressed per year, calculated by the Sen slope method, shows that annual ET_o varied from – 1.98 to 2.29 mm year⁻¹ (Fig. 6). Most of the basin (72.9%) showed a decreasing ET_o trend (Fig. 6 and Table 2). In the central basin around Nanning-Laibin-Guiping, ET_o had the highest decreasing trend of 10.2%. Overall, ET_o averaged over the entire the Pearl River Basin showed a decrease of about 1.6% from 1960 to 2016.

Table 2. Percentage of positive (PS) and negative (NE) trend region of basin and the change trend of ET_o and climate variables in the PRB in the period 1960–2016. See Table 1 for climate variable abbreviations./dec: per decade.

		Jan.	Feb.	Mar.	Apr.	May	Jun.	Jul.	Aug.	Sep	Oct.	Nov.	Dec.	Annual
ETo	PS	10.23	76.61	8.00	58.91	9.85	44.26	7.92	28.80	26.11	85.87	68.83	55.14	27.15
	NE	89.77	23.39	92.00	41.09	90.15	55.74	92.08	71.20	73.89	14.13	31.17	44.86	72.85
	mm/dec	−0.81	0.51	−0.94	0.19	−0.72	−0.18	−1.00	−0.25	−0.83	0.58	0.31	0.14	−2.96
	mm	51.08	56.49	79.04	96.69	114.33	113.91	131.66	125.60	107.16	87.36	64.11	53.12	1080.55
RH	PS	67.27	4.48	15.37	5.30	0.45	12.47	0.00	0.00	0.01	0.00	1.68	3.22	0.00
	NE	32.73	95.52	84.63	94.70	99.55	87.53	100.00	100.00	99.99	100.00	98.32	96.78	100.00
	/dec	0.08	−0.48	−0.33	−0.57	−0.52	−0.32	−0.57	−0.86	−0.69	−0.87	−0.58	−0.82	−0.55***
SD	PS	2.16	31.29	0.00	19.51	5.65	10.16	0.00	0.06	0.00	29.04	29.62	6.67	1.06
	NE	97.84	68.71	100.00	80.49	94.35	89.84	100.00	99.94	100.00	70.96	70.38	93.33	98.94
	h/dec	−0.20	−0.02	−0.15	−0.06	−0.13	−0.11	−0.19	−0.15	−0.20	−0.03	−0.06	−0.09	−0.12***
TAV	PS	79.45	100.00	93.43	99.19	94.67	99.65	94.58	96.11	93.03	99.23	100.00	98.50	99.55
	NE	20.55	0.00	6.57	0.81	5.33	0.35	5.42	3.89	6.97	0.77	0.00	1.50	0.45
	℃/de	0.07	0.33	0.06	0.21	0.07	0.15	0.08	0.12	0.10	0.23	0.24	0.10	0.14***
WS	PS	15.91	18.05	16.64	17.99	28.40	53.89	41.50	90.46	64.58	40.11	28.89	32.00	35.23
	NE	84.09	81.95	83.36	82.01	71.60	46.11	58.50	9.54	35.42	59.89	71.11	68.00	64.77
	m/dec	−0.05	−0.07	−0.08	−0.06	−0.04	0.01	−0.00	0.03	0.01	−0.02	−0.03	−0.03	−0.03**

***indicates significance at the 0.001 probability level; **indicates significance at the 0.01 probability level.

Figure 6. Spatial change trend of the ET_o in the Pearl River Basin.

In the seven months of January, March and May–September, ET_o showed a decreasing trend (Table 2), with more than 70% of the basin showing this decrease. Within this time range, July had the most substantial decreasing trend of approx. – 10.0 mm year⁻¹, with more than 92.1% region of the basin showing a decrease. On the other hand, October showed the most rapidly increasing trend of 58 mm year⁻¹, with 85.9% of the basin showing an increasing trend.

3.2 Spatio-temporal change of climate variables

The sunshine duration, wind speed and relative humidity showed decreasing trends, as indicated by the Sen slope analysis, with decreases of 0.12 h decade⁻¹, 0.03 m s⁻¹ decade⁻¹, and 0.55% decade⁻¹, respectively (Table 2; see also Section 4.1), and relative change (RC) decreases of 15.5%, 7.4% and 4.0%, respectively (Fig. 7). Mean temperature had an increasing trend, of 0.14ºC decade⁻¹, which is consistent with the global climatic temperature change, representing an increase in RC of 4.25%. The Mann-Kendall significance tests showed all variables passed the significance test at p = 0.001, except for wind speed that passed at p = 0.01.

Figure 7. Relative change, RC, of climate variables in the Pearl River Basin. RH: relative humidity; SD: sunshine duration; TAV: average air temperature; WS: wind speed.

Spatially, RH, WS and SH showed a downward trend in 100%, 98.9% and 64.8% of the basin area, respectively (Table 2). Average temperature (TAV) trends increased everywhere in the basin, with the highest changes around the Pearl River Delta and the western high-altitude region (Fig. 7).

Monthly RH showed a decreasing trend in 80% of the basin for all months except January, with the greatest decreases of around 0.87% decade⁻¹ in October, and the RC decreasing at 6.6% (Table 2, Fig. 8). Sunshine duration showed decreases during all months for more than 70% of the basin area. Average temperature showed an increasing trend in more than 95% of the basin for all months, except for 20% of the region, which showed an increasing trend in January. February had the most rapid increase of about 0.33ºC decade⁻¹, with RC of 16.2%. Wind speed showed an increasing trend in June, August and September, but a decreasing trend in all other months. Wind speed in March had the most rapid decreasing trend of 0.08 m s⁻¹ decade⁻¹, and the largest RC decreases of 21.6%.

Figure 8. Monthly relative change, RC, sensitivity coefficient, Sen, and contribution, Con, of the climate variables in the Pearl River Basin.

3.3 Spatio-temporal change in the sensitivity S_v

The sensitivity coefficients, S_v, averaged – 0.45, 0.53, 0.27 and 0.11 for RH, TAV, SD and WS, respectively (Fig. 9). Relative humidity had a negative sensitivity throughout the entire basin, with the lowest magnitudes in the central basin and the highest magnitudes in the western basin and the Pearl River Delta, and a maximum magnitude of – 0.71. The S_v for air temperature decreased from east to west and from south to north, with the highest sensitivity in the Pearl River Delta region reaching 0.65. The spatial distribution of SD sensitivity was similar to that of TAV, but the highest value was mainly in the southern basin, with values reaching 0.32. The sensitivity of ET_o to wind speed was generally low, with the highest value of 0.17 in the northern and western basin.

Figure 9. Sensitivity coefficient (absolute) between ET_o and climate variables in the Pearl River Basin. RH: relative humidity; SD: sunshine duration; TAV: average air temperature; WS: wind speed.

Sensitivity varied greatly in different months (Fig. 8). From April to October, TAV had the highest sensitivity coefficient, while RH exhibited the highest magnitude sensitivity in the other months. Relative humidity had negative sensitivities throughout the year, with January having the highest magnitude sensitivity coefficient of about – 0.66, and August having the lowest magnitude of – 0.26 in August. Sensitivity of TAV and SD increased from 0.35 and 0.15, respectively, in January to 0.66 and 0.43, respectively, in August. Wind speed had a lower sensitivity coefficient compared to the other variables, ranging from 0.17 in December to 0.06 in July.

3.4 Spatio-temporal change in the contribution of Con_vi

The contribution of different variables (Con_vi) to the change in ET_o was examined with the sensitivity (S_vi) and relative change (RC_vi) of each climate variable (Equations (3) and (4)). Sunshine duration and wind speed showed a negative Con of −4.3% and −1.0%, respectively (Fig. 10). Air temperature showed a Con of 2.3% and that for relative humidity was 1.8%. All four of the climate variables combined had a Con of −1.1%. The relative change (RC) of ET_o had a spatio-temporal average about – 1.6%, varying between −10.2% and 11.4%. The correlation coefficient between the sum contribution of all four climate variables and the RC of ET_o was 0.91 over the basin. Sunshine duration had the largest contribution, with a spatial value varying from −8.1% to 1.8%, with 98.9% of the basin exhibiting a negative contribution. The contribution from wind speed varied from −8.1% to 4.7%, with 64.8% of the basin showing a negative contribution. Both air temperature and relative humidity exhibited positive contributions throughout essentially the entire basin, varying from – 1.3% to 4.9% and 0.4% to 4.9%, respectively.

Figure 10. Contribution between ET_o and climate variables in the Pearl River Basin. RH: relative humidity; SD: sunshine duration; TAV: average air temperature; WS: wind speed.

The average Con over the entire basin varied by month (Fig. 8). Monthly sunshine duration had an obvious impact on ET_o in January, March and May–September, with the largest contribution in September reaching −6.9%, and the lowest contribution in February. Wind speed had a negative contribution to ET_o from October to May, with the largest magnitude contribution of −3.0% in March. The contribution of relative humidity was low in the middle of the year, with the lowest value of 0.9% in June, while the largest contribution reached 4.2% in December. The air temperature exhibited a positive contribution over the entire year, reaching Con of 6.5% in February, decreasing to a low of 1.0% in May.

From a spatial perspective, sunshine duration had the greatest annual contribution of all variables, affecting 72.7% of the basin (Table 3). On an annual basis, ET_o was most affected (based on Con) by WS, TAV and RH in 16.1%, 6.6% and 4.6% of the basin, respectively. In the area surrounding Guilin-Duan-Yanshan (Fig. 10), the greatest contribution to ET_o came from WS; TAV was the dominant contributor around the west of Mengzi; and RH only controls a small region around Zhanyi in the western basin. On monthly time scales, SD made the greatest contributions to ET_o in January, March and May–September, with essentially approximately 70% region of the basin being so affected. Air temperature exhibited the widest areal distribution of the greatest contributions in February, April, October and November. Relative humidity accounted for the largest space in December, making ET_o show an increasing trend.

Table 3. Percentage of area influenced by each dominant variable to ET_o (%).

	Jan.	Feb.	Mar.	Apr.	May	Jun.	Jul.	Aug.	Sep.	Oct.	Nov.	Dec.	Annual
RH	1.43	0.00	3.14	7.86	7.59	4.10	0.73	2.23	5.24	17.04	1.24	45.22	4.61
SD	68.66	0.00	66.19	28.40	76.16	65.48	98.61	94.67	92.81	7.28	16.31	23.79	72.70
TAV	10.85	84.14	2.11	42.72	0.60	26.75	0.55	1.52	0.15	63.47	63.96	3.41	6.56
WS	19.05	15.86	28.56	21.02	15.66	3.67	0.11	1.59	1.80	12.21	18.49	27.57	16.13

4 Discussion

4.1 Change in climate variables

Sunshine duration decreased appreciably, by 15.1% from 1960 to 2016 (Fig. 11). In more detail, sunshine radiation had a decreasing trend from the 1950s to the 1980s, but with a slightly increasing trend between the 1990s and 2000s, and then levelling off subsequently. In areas where there has been a decreasing trend and then an increasing trend, this has been called “from dimming to brightening” (Wild et al. 2005, Wild 2009, Liu and Zhang 2013). We speculate that the reduction of sunshine duration in the Pearl River Basin between the 1950s and the 1980s stemmed from two major factors. Firstly, sunshine duration decreased because of increases in aerosol concentrations caused by rapid urbanization and the associated combustion-related aerosol production, as postulated by Zongxing et al. (2012) for southwestern China. It is instructive to remember that sunshine duration is defined as the “the sum of the time, during a given period, for which the direct solar irradiance exceeds 120 W/m² (WMO 2003), so increased turbidity from aerosols would decrease irradiance, thereby decreasing the sunshine duration. A secondary effect of increased aerosols on reduced sunshine duration could involve increased cloudiness associated with increased cloud condensation nuclei and ice nuclei, although the relationship is complex (Seinfeld et al. 2016). Secondly, our results show that wind speeds generally decreased from 1960 and the 1990s. These reduced wind speeds will have exacerbated the air quality problem, as it is well known that low wind speeds are linked to higher planetary boundary layer pollutant concentrations (Satheesh and Krishna Moorthy 2005, Fu et al. 2008). The reduced wind speeds could have been caused by an increase in land surface vegetation, the weakening of atmospheric cycle activities, and the weakening of the East Asian monsoon (Xu et al. 2006, Zhang et al. 2009b, Vautard et al. 2010, Li et al. 2017b). Wind speed in the Pearl River Basin did show an increase after the 2000s (Fig. 11), at the time when the sunshine duration tended to increase and level off, consistent with a potential link to aerosol concentrations. Some researchers have attributed global wind speed decreases as a major factor in decreasing ET_o, but this is attributed to the direct effect of wind speed on the aerodynamic resistance rather than through controls over aerosol concentration (McVicar et al. 2012).

Figure 11. Variation of ET_o and climate variables in the Pearl River Basin. RH: relative humidity; SD: sunshine duration; TAV: average air temperature; WS: wind speed.

The air temperature pattern over the 57-year study period is consistent with the global increase of approximately 0.14°C decade⁻¹ in the report of the Intergovernmental Panel on Climate Change (IPCC 2013). The time course of the air temperature pattern is also similar, except in the Pearl River Basin, the slight decrease or constant temperature pattern in the 1960s does not become an increase until well into the 1980s (Fig. 11), while the global averaged increase becomes more apparent beginning in the late 1970s. The relative humidity shows a gentle overall reduction over time, with a higher frequency pattern of a dip starting in the 2000s, then rising around 2010. Both the general trend of increasing air temperature and reduced relative humidity are expected to be linked the definition of relative humidity; and from Equation (1), RH is a key driver of the evapotranspiration process (Tegos et al. 2017), linked to increasing ET_o (Li et al. 2017a).

4.2 Analysis of ET_o change

The annual ET_o pattern from 1960 to 2016 shows consistency with the effects predicted from Equation (1) and the sensitivity analysis (Fig. 11). In particular, the general decrease in ET_o is consistent with the general decrease in annual average wind speed and sunshine duration. However, the decrease in ET_o is not generally consistent with the annual average air temperature increase and relative humidity decrease, both of which are associated with an increase in annual ET_o from Equation (1) and the sensitivity analysis. This apparent inconsistency can be resolved by thinking of the combination of ET_o sensitivity to and contribution from the key weather variables used in this study.

Increased wind speed decreases the aerodynamic surface resistance and increases ET_o; this is consistent with the pattern in ET_o, which is similar to that of wind speed – an apparent decrease over time until the 1990s, then an increase (Fig. 11). The general trend of a decrease in ET_o and wind speed is also consistent with Equation (1). Similarly, the overall trend in a decrease in sunshine duration would be associated with a decrease in the net radiation term of £quation (1), leading to an expected decrease in ET_o.

Sensitivity analysis and Equation (1) show that an increase in air temperature and the expected related decrease in RH will tend to increase ET_o, The fact that ET_o slightly decreases over time implies that the overall contributions from wind speed and sunshine duration outweigh the air temperature and relative humidity contributions in the Pearl River Basin, as discussed below. Linked to this is the expectation that, had temperature and relative humidity showed no trend over time, but sunshine duration and wind speed changed as observed, then the ET_o would have decreased even more dramatically. A finer scale analysis shows that the minimum ET_o (associated with a negative spike) observed in 1997 can be linked to an increased relative humidity (a positive spike), a minimum in sunshine duration of 3.67 h (a negative spike) and a negative spike in wind speed (Fig. 11). In contrast, the positive spike in ET_o in 1963, resulting in the highest ET_o value in the record, is linked to a positive spike in wind speed, a positive spike in sunshine duration, a negative spike in relative humidity and a positive anomaly in air temperature.

The general decrease in ET_o in the Pearl River Basin matches the same trend reported for Guangdong (He et al. 2015) and some other parts of the globe (Roderick et al. 2009). However, other reports that global and some regional ET_o began to increase in the 1980s, reaching a maximum in 1998 (Jung et al. 2010, Zhang et al. 2016, Dong and Dai 2017), do not match our annual ET_o record, which shows a general decrease from 1960 through to 1997, and then a general increase for around a decade before levelling off (Fig. 11). Some of this apparent discrepancy could arise because the global data of Jung et al. (2010), including the Southern Hemisphere, was based on data commencing in the 1980s, so both the global nature of the data and the limited time span could come into play. Other regional ET_o studies, in Canada (Burn and Hesch 2007) and southwest USA (Senay et al. 2017), also show different trends than the Jung et al. (2010) study. Our results coupled with previous studies reinforce the idea that the ET_o equation is associated with sensitivities that, when examined in terms of contributions of climatic variables and their spatio-temporal distribution, explain different spatio-temporal ET_o patterns for specific regions and the globe in general. These results also show that generalizations at the global scale must be carefully evaluated when considering regional or local scale ET_o changes.

4.3 Analysis of the sensitivity and contribution

As noted in Section 4.2, consideration of the differential spatio-temporal sensitivity of ET_o to climatic variables is a critical endeavour. Our analysis agrees with previous reports that sensitivity varies with region and time scale (Irmak et al. 2012); so while ET_o exhibits the highest sensitivity to relative humidity in some regions (Xu et al. 2006), in other regions such as southwest China, more sensitivity is found with wind speed (Liu et al. 2016). This is not a surprise because sensitivity analysis using partial differentials of Equation (1), as presented in Equation (2), will yield differing values depending on where in the multidimensional curved surface any particular region lies in space and time (that is, what the typical climatic variable values are). Furthermore, because spatio-temporal variation of these climatic variables may differ appreciably between variables, the relative change and contributions to ET_o as assessed using Equations (3) and (4) show different ET_o patterns can be expected in different geoclimatic regions and different time scales (daily, monthly, annual).

For example, sunshine duration had the highest magnitude negative contribution to ET_o variation in the Pearl River Basin (Figs. 8 and 10), although ET_o exhibited a low sensitivity to it. This was because the relative sunshine duration change was greater than the other variables over the study period. ET_o in northeast India also decreased with decreasing sunshine duration and wind speed (Jhajharia et al. 2012). Average temperature and wind speed had lower overall negative contributions to the ET_o variation numerically, although the patterns shown in Fig. 11 show the pattern for annual averaged wind speed matched that of annual ET_o better than other variables. Relative humidity exhibited a moderate positive contribution.

Monthly timescale analysis showed a finer detail of sub-seasonal effects in the Pearl River Basin. Significant reduction of sunshine duration in January, March and May–September led to a decreasing trend of ET_o, while during the other months, ET_o showed an increasing trend responding to the linked air temperature rise and relative humidity reduction. The monthly ET_o contribution analysis could help improve agricultural water resource management.

Just as the Pearl River Basin ET_o showed differing responses to climatic variables than for other regions or the globe, differential ET_o responses within the basin were seen linked to the spatial heterogeneity of the climatic variables (Fig. 7). Relative humidity and average temperature exhibited a positive contribution to ET_o throughout the entire basin. On the other hand, the contribution of wind speed showed a significant increasing trend in the eastern basin and the west of Duan, and a lower decreasing trend of sunshine duration in the western region. These caused the significant positive contribution of wind speed to ET_o, and offset the smaller negative contribution of sunshine duration (Fig. 10), so ET_o showed an increasing trend in these regions.

Local and regional characteristics may dictate the direction of ET_o changes (Irmak et al. 2012), consistent with our results for the Pearl River Basin. Therefore, it is necessary to study the change in climate variables and ET_o on a global scale and refine the trend at different regional scales. Our analysis indicates that the climatic record of ET_o cannot be reliably used as a proxy for the change in other atmospheric variables; hence, although a decrease in ET_o could be indicative of “global dimming” associated with increased aerosol concentrations in some regions, other variables could be the main cause for decreases in ET_o in other regions.

5 Conclusion

This paper quantitatively analyses the spatio-temporal sensitivity and contribution of climate change to the spatio-temporal change of ET_o in the Pearl River Basin between 1960 and 2016 at the monthly scale, which can provide a high reference for the regional water cycle, water resources management and agricultural irrigation period. In later studies, more and more remote sensing satellite images can be combined to provide support for more accurate ET_o spatial distribution. These may give a comprehensive analysis of climate, carbon dioxide density and vegetation change influence on ET_o with the different intensity of human activities or urbanization in different regions.

The ET_o in the Pearl River Basin shows a decreasing trend at the beginning of the study period (1960–2016), but then this trend reversed towards the end of the study period. Annual average ET_o ranged from 998.9 to 1180.0 mm, and July exhibited the highest monthly ET_o of 131.7 mm. The spatio-temporal analysis shows different regions of the basin respond with different sensitivities to relative humidity, sunshine duration, average air temperature and wind speed.

Spatially, monthly ET_o showed a decreasing trend along a southeast to northwest transect. The spatio-temporal pattern of ET_o was linked to a decreasing trend in sunshine duration, wind speed and relative humidity.

The ET_o was sensitive to different climatic variables, with substantially different magnitudes and directions. ET_o was sensitive to relative humidity in a negative fashion, with a value of up to – 0.45, while it had positive sensitivities to air temperature, sunshine duration and wind speed of 0.53, 0.27 and 0.11, respectively. Monthly sensitivities to air temperature and sunshine duration were greatest in mid-year, while sensitivity to relative humidity and wind speed were lowest during that time.

These results show that any downscaling of global climate change ET_o scenarios requires a high spatio-temporal resolution because of the large variability found in the sensitivity of ET_o to climate variables and the spatial and temporal differences in the climate change variables themselves.

Acknowledgements

We appreciate the assistance of the editors, Dr Tegos and another anonymous reviewer for their critical comments and constructive suggestions on the paper.

Disclosure statement

No potential conflict of interest was reported by the authors.

Additional information

Funding

Portions of Paw U’s work in this study were supported by the US National Institute of Food and Agriculture [CA-D-LAW-4526H and CA-D-LAW-7214-RR].