评估堆叠和混合合奏学习方法，以估计日常参考蒸发

A R T I C L E I N F O

Handling Editor - Dr. B.E. Clothier

Keywords:

Reference evapotranspiration Spatio-temporal change Climate change

Rice area expansion

A B S T R A C T

It is of great significance to study the variation in reference evapotranspiration (ETo) and the factors that in- fluence it to ensure the sustainable and efficient utilization of water resources. In this study, the spatio-temporal changes in ETo and rice areas as well as the relationship between them in Heilongjiang Province, China, from 1951 to 2018 were analysed by using trend, correlation and contribution analyses. The results showed that rice area expansion is an important factor affecting ETo changes. ETo in Heilongjiang Province decreased during the study period. The areas with high ETo in the Songnen and Sanjiang Plain showed a gradually shrinking trend. Climate change and rice area expansion both contributed to the decline in ETo. The decreases in sunshine duration and average wind speed in Heilongjiang Province were the main reasons for the decrease in ETo. Rice

area in Heilongjiang province continued to increase at a rate of 0.54 × 105 ha yr—1. The expansion of rice area in

Sanjiang Plain was the largest, followed by Songnen Plain. In response to the rice area expansion, the rise in average temperature slowed down, the average relative humidity was increased, while wind speed was reduced, which directly lead to the decrease in ETo. The effects of the transformation of different land types into rice fields on ETo were different. The cumulative effects of wetlands and dry fields conversion to rice fields on ETo were

7.67 mm yr—1 and — 20.15 mm yr—1 in Sanjiang Plain respectively.

1. Introduction

Evapotranspiration is an important process in the water and energy cycles, and it is also one of the key factors affecting crop growth, dry biomass production and yield (Qiu et al., 2019). A detailed under- standing of the changes in reference evapotranspiration (ETo) and the factors that influence it is essential for the scientific management of water resources and sustainable protection of agricultural production (Nam et al., 2015; Lv et al., 2018). Climate and land use changes have a strong impact on ETo (Chen et al., 2011, 2019; Dias et al., 2015; Li et al., 2017; Odongo et al., 2019). However, the impact of land use changes caused by agricultural practices, such as irrigation, on local hydrological conditions is still unclear (Ozdogan and Salvucci, 2004). Furthermore, scientific irrigation management is key to ensuring stable high yields and high water use efficiency of food crops (Tang et al.,2018). Analysing changes in ETo with the synergy of climate and irrigation is very

important for planning regional irrigation, adjusting agricultural planting structures, and protecting the ecological environment (Han et al., 2010; Jia et al., 2016).

In the context of climate change, changes in all meteorological var- iables, such as temperature, wind speed, rainfall, and solar radiation, have led to changes in ETo (Scheff and Frierson, 2014; Nam et al., 2016; Luo et al., 2018). For example, a decrease in sunshine duration and wind speed have led to a decrease in evapotranspiration, but the opposite is true for an increase in temperature (Roderick and Farquhar, 2002; Gao et al., 2006; Xie et al., 2008; Nan et al., 2012; Pour et al., 2020), and ETo has also been significantly affected by relative humidity (Li et al., 2015; Kong et al., 2018). Chen and Huo (2016) used a method of removing meteorological factors to assess the impact of climate change on ETo. Temperature and wind speed were believed to be the main factors that influence ETo in the Heihe River Basin (Jia et al.,2016). Yassen et al. (2020) observed that ETo in Egypt was greatly affected by climate

* Corresponding author.

E-mail address: yfluo@whu.edu.cn (Y. Luo).

https://doi.org/10.1016/j.agwat.2021.106912

Received 18 December 2020; Received in revised form 5 April 2021; Accepted 7 April 2021

Available online 29 April 2021

0378-3774/© 2021 Elsevier B.V. All rights reserved.

Fig. 1. Map of Heilongjiang Province.

change in most regions, with more profound impact in summer than winter. The relationship between climate conditions and terrestrial surface processes in different geographic regions varies (Koster, 2004). Although similar studies have been conducted in Northeast China, cli- matic impacts on ETo in cold rice-growing areas, especially in rice-concentrated areas in Heilongjiang Province, are still unclear.

Irrigation is considered an important anthropogenic process affecting the climate and hydrology (Lobell et al., 2009). In agricultural areas, diversion and irrigation changes affect the regional microclimate. For example, the expansion of irrigation areas has led to, among other changes, a decrease in local temperature (Bonfils and Lobell, 2007; Han and Yang, 2013; Han et al., 2018), a decrease in wind speed (Alpert and Mandel, 1986; Jia et al., 2016), an increase in humidity (Zhao et al., 2016), and an increase in precipitation (Jo´dar et al., 2010; Mao et al., 2011), which affects potential evapotranspiration (Ozdogan and Sal- vucci, 2004). Therefore, crop and net irrigation water requirements will be significantly affected by these changes. Understanding how reference evapotranspiration changes with irrigation may have important impli- cations for future agricultural water management. Han et al. (2014) found that ETo decreased with irrigation expansion. In general, there are few studies on the impacts of irrigation and other agricultural activities on evapotranspiration, especially for the large-scale expansion of rice areas. It is necessary to systematically explore the impact of increased rice areas on ETo.

Rice is one of the three major food crops in China, and its production has been increasing continuously (China National Bureau of Statistics, 2019). With agricultural development and the powerful support of the government, rice areas in Heilongjiang Province expanded rapidly, ac- counting for 13.05% of the total rice area of China (Liu et al., 2013; Gao et al., 2020). The expansion of rice areas will inevitably cause a sub- stantial increase in irrigation water use, which will affect ETo. It is very important to explore the impact of climate change under natural con- ditions and the rice area expansion caused by human activities on regional ETo changes. The objectives of this work are therefore (1) to analyse the temporal and spatial changes in ETo and rice areas in Hei- longjiang Province; (2) to analyse the effects of meteorological variables change on ETo in Heilongjiang Province; and (3) to explore the impact of increased rice areas on ETo under the background of climate change in Heilongjiang Province.

2.
Materials and methods

2.1. Study area and data

The study area, Heilongjiang Province, is located in the northern- most part of Northeast China, between 43◦26′ N~53◦33′ N and 121◦11′ E~135◦05′ E (Fig. 1). Its total area is approximately 4.73 105 km2 which includes 12 prefecture-level cities and the Daxinganling.

Heilongjiang Province has higher elevations in the northwest and southeast and lower plains in the northeast and southwest. It is in the cold temperate zone and has a temperate continental monsoon climate. It is the province with the lowest temperature in China, and the tem- perature difference between the northern and southern areas is large (Gao and Liu, 2011). Summers are short and rainy here, and winters are cold and long. Average temperature in Heilongjiang is approximately 1–5 ◦C, and annual precipitation is 400–600 mm. Precipitation from

June to September accounts for more than 65% of the annual precipi- tation. With the development of agriculture, Heilongjiang Province has become an important commercial grain production base in China (Su et al., 2021). The western Songnen Plain and the eastern Sanjiang Plain are the two main grain production areas in Heilongjiang Province, both of which mainly produce rice (Lu et al., 2017).

The statistical data of rice areas from 1951 to 2018 were obtained from the Heilongjiang Statistical Yearbook (Heilongjiang Bureau of Statistics, 1987) and the Statistical Yearbook of Heilongjiang State Farms (Bureau of Statistics of Heilongjiang State Bureau of Agriculture Reclamation, 1993). These yearbooks are annual statistical publications reflecting economic and social development and are published by the China Statistics Press. The meteorological data used in this study, including the daily maximum temperature, daily minimum tempera- ture, average temperature, average relative humidity, sunshine dura- tions, average wind speed and daily precipitation at 33 National Meteorological Observatory (NMO) stations from 1951 to 2018, were provided by the China Meteorological Data Network. A total of 5.72 million data were collected, with an error rate of almost 0 and a missing rate of 0.06%. The datasets are available and have been quality controlled by the National Climate Information Center with a data loss rate of less than 0.1% (Zhao et al., 2020). The few missing data were interpolated based on the data from adjacent stations (Lin et al., 2018).

Table 1

Information for the National Meteorological Observatory Stations used in this study.

Name

These climate variables were used to calculate ETo in Heilongjiang Province according to the FAO-56 Penman-Monteith method. The location and distribution of meteorological stations within Heilongjiang Province are shown in Fig. 1. Detailed information on these stations, including the station code, name, location, latitude, longitude, and altitude, is presented in Table 1. The land use information in 1980 and 2018 was obtained from the Date Center for Resources and Environ- mental Sciences, Chinese Academy of Sciences (http://www.resdc.cn). These data were generated by visual interpretation from Landsat TM/ETM satellite images with a horizontal resolution of 1000 m. The increase of construction land in Heilongjiang Province from 1960 to 2010 was concentrated in the South (Zhang et al., 2017). In the analysis

of the effects of rice area expansion, the data of Harbin (HRB) and other stations with an increase of construction more than 30 km2 were deleted to eliminate the interference of urbanization and industrialization.

2.2. Methods

2.2.1. Calculation of reference evapotranspiration

The Penman-Monteith method recommended by the Food and Agriculture Organization (FAO) of the United Nations has a complete theoretical basis and a small calculation error. It is the most widely used

surface (m s—1), es is the saturation vapor pressure (kPa), ea is the actual

vapor pressure (kPa) and Δ is the slope of the saturation vapor pressure curve (kPa ◦C—1).

2.2.2. Linear trend analysis and Mann-Kendall test

A linear regression equation was established to analyse the changing trends of ETo and meteorological variables in Heilongjiang Province. Time t was taken as the independent variable and ETo or meteorological variables x as the dependent variable. The regression equation is as follows:

x = at + b (2)

where x is the value of ETo or meteorological variables, a is the slope of the equation, which is used to represent the linear trend of the time series, t is the year, and b is the regression constant. The series has a positive trend when a is greater than 0, and it has a negative trend when a is less than 0. The larger the absolute value of a is, the more significant the positive or negative trend of time series is.

The reliability test of the linear trend results adopts the Mann- Kendall test. This method is suitable for the non-parametric test of time series trends (Yavuz, 2018). The statistic S of time series x is calculated as follows:

ETo calculation method. This study uses it to calculate ETo, and the

n—1 n

formula is as follows (Allen et al., 1998):

900

S = sgn xj — xi (3)

i=1 j=i+1

ETo

0.408Δ(Rn — G) + γ u2(es — ea)

= Δ + γ(1 + 0.34u2)

(1)

where xj is the value of jth data, n is the length of the data series and sgn is the sign function defined as follows:

where ETo is the daily reference evapotranspiration (mm day—1), Rn is the net radiation at the crop surface (MJ m—2 d—1), G is the soil heat flux (MJ m—2 d—1), γ is the psychometric constant (kPa ◦C—1), T is the mean air temperature (◦C), u2 is the mean wind speed at 2 m above the ground

sgn.xj — xi) =

1 if xj > xi 0 if x x

⎩ —1 if xj < xi ⎭

(4)

When n ≥ 8, the statistic S approximately obeys the normal distri-

[1.5h

(h)3 ]

E(S) = 0 (5)

n(n — 1)(2n + 5) — ∑ t (i — 1)(2i + 5)

where h is the distance between the samples; C0 is the nugget; C1 is the part of sill; C0 +C1 is the sill; a is the variation distance.

Var(S) =

i

i=1

18

(6)

2.2.4. Correlation analysis

Pearson correlation coefficient indicates the degree of correlation

where E S is the mean of the statistic S; Var S is the variance of the statistic S; m is the number of tied groups; ti denotes the number of data points in the ith group. The test statistic ZMK is:

between two sets of data (Sun and Qin, 2016). It was used to analyse the

relationship between ETo and meteorological variables. The following formula adapted from (Revadekar et al., 2012; Kang et al., 2017) was

used for this purpose:

⎧⎪ S — 1 ⎫⎪

N

⎪⎪⎨ √̅V̅̅̅a̅̅r̅̅(̅̅S̅̅)̅̅

S > 0

⎪⎬

∑i=1[(xi —

X) ×

(yi —

Y) ]

(11)

ZMK =

0 S 0

⎪

(7)

rxy = √̅∑̅N̅̅̅̅(̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅)̅̅̅2̅̅̅̅̅̅̅̅∑̅̅N̅̅̅̅̅(̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅)̅̅̅2̅

S 1

⎩ √̅V̅̅̅a̅̅r̅̅(̅̅S̅̅)̅̅

S < 0 ⎪⎪⎭

i=1

xi — X

×

i=1

yi — Y

The null hypothesis (H0) of no trends is accepted if —

Z1—α ≤ ZMK ≤ Z1—α , otherwise H0 is rejected and the existence of a trend

where rxy is the correlation coefficient between x and y, while xi is the

ETo value in i year, y is the meteorological variable value in i year. X and

is accepted. The significance of the trends was tested at the significance

level α. In this study, α is set at 0.05 and 0.01 significant level. Besides,

the Sequential version of MK test statistic was employed to identify the

approximate starting point (here year) of change in time series of rice

area.

2.2.3. Spatial data processing

The kriging interpolation method in ArcGIS 10.2 was used to inter-

polate ETo and meteorological variables to a raster surface, and WGS-

1984 coordinates were used as the map coordinate system. Kriging

interpolation based on the semivariogram model performs unbiased

optimal estimation of regional variables in a limited region. It extends

discrete point information to continuous spatial information and obtains

ideal prediction results (Lv et al., 2019; Oliver and Webster, 1990). In

this study, ordinary kriging (OK), which is the most popular type of

kriging, was used to interpolate ETo and meteorological variables at

different timescales (Moghbeli et al., 2020). Suppose there are alto-

gether n measured points in the neighborhood of the estimated point x0,

i.e., x1, x2., xn. The interpolation formula of OK is as follows:

n

Z∗(x0) = λiZ(xi) (8)

i=1

where Z xi are the observations selected in the x0 neighborhood to perform the estimation of Z∗ x0 ; Z∗ x0 is the kriging estimate at loca- tion x0; λi denotes the weight of the ith observation. Under the condition of unbiased and minimum variance, the equation system of the weight coefficient λi is:

are positively related when r > 0. X and Y have a negative correlation

when r < 0. The greater the absolute value of the number, the greater

the strength of the correlation between the two variables. SPSS software

was used for statistical analysis. The significance of correlation coeffi-

cient was tested by t-test (Zhang et al., 2011). The statistical significance

in this study corresponded to a significance level of p < 0.05(*) and

p < 0.01(**).

2.2.5. Contribution assessment

Han and Hu (2012) and Jia et al. (2016) considered that climate

change and irrigation were important causes of regional evapotranspi-

ration change. Rice area expansion in Heilongjiang Province mainly

occurred in the Sanjiang and Songnen Plain (Lu et al., 2017). That is, the

irrigation water consumption of the Sanjiang and Songnen Plain

increased significantly. Rice area expansion was the main land use

change type in these two plains during the period of 1980–2015 (Chen

et al., 2018). Thus, it has been assumed that the changes in ETo in the

Sanjiang and Songnen Plain were only affected by climate change and

rice area expansion.

First order differential method was used to remove the trends of

meteorological variables to effectively separate the trend component

and the non-trend component. Trend components include the trend

caused by climate change and the trend caused by rice area expansion.

The first order difference is the difference between two consecutively

adjacent items in the sequence. After the original sequence Xt is first

order differentiated, a new sequence Yt without trend can be obtained

(Zhang et al., 2016):

⎧⎪ ∑n

λ C.x x ) μ

C.x

x ) j

1 2 … n

Yt = Xt+1 — Xt (12)

⎨ i=1 i

i, j + =

0 , j ( = , , , )

(9)

T X Y

(13)

⎩⎪ ∑n

λi = 1

t = t — t

i=1

where C(xi, xj) is the covariance function of A and B. The most

commonly used semivariogram models are the spherical, exponential,

and Gaussian models (Dalezios et al., 2002; Moghbeli et al., 2020).

Spherical model is the most commonly used and has good accuracy. In

this study, the Kriging interpolation method with spherical model as

semivariograms is used for spatial interpolation analysis. The spherical

model is expressed as follows:

where Xt is the original sequence, Yt is the detrended sequence, Tt is the

trend component, t is the sample number, and t 1, 2, …, n, where n is

the sequence length.

To quantify the contribution of climate change and rice area

expansion to ETo, the following steps were carried out on the Sanjiang

and Songnen Plain: First, we assumed that if rice area in a region was

unchanged during the period, then any change in ETo in that region was

caused by climate change. It means ∆ETo ∆EToclimate and ∆ETorice 0.

Second, we assumed that in case of change in the rice area, the changes

800

750

700

650

600

550

Fig. 2. Average ETo during (A) 1951–1967, (B) 1968–1984, (C) 1985–2001 and (D) 2002–2018.

EToclimate and ∆ETorice are ETo changes due to climate change and rice area expansion, respectively.

3. Results and discussion

3.1. Spatio-temporal changes in ETo

The spatial distribution of ETo in each period is shown in Fig. 2. It is different in each period. The areas with high ETo were mainly in the Songnen and Sanjiang Plain. In 1951–1967 and 1968–1984, ETo was higher in most areas of the Songnen Plain and the central part of the Sanjiang Plain (Fig. 2A and B). The areas with high ETo in 1985–2001 and 2002–2018 were located in the southwestern Songnen Plain and the southern Sanjiang Plain (Fig. 2C and D).

The comparison between Fig. 2A and B indicated an expansion in the

1951 1961 1971 1981 1991 2001 2011

Year

Table 2

Interannual variation in ETo in each city in Heilongjiang Province.

Fig. 3. Trend in ETo in Heilongjiang Province during the 1951–2018 period.

City name Max (mm yr—1)

Min (mm yr—1)

Mean (mm yr—1)

Slope (mm yr—1)

in ETo were attributable to the combined effect of climate change and rice area expansion. Therefore, the effects of rice area expansion on ETo in a region can be quantitatively identified by removing the effects of climate change on ETo in that region (refer to Eq. (15)) since the effects of climate change on ETo in a region are the same. The trend of each meteorological variable was removed by first order differential method and ETos was recalculated by using the detrended data series:

∆ETo = ETo — ETos (14)

∆ETo = ∆EToclimate + ∆ETorice (15)

where ∆ETo is the change in ETo; ETos is stationary ETo sequence; ∆

Daxinganling 699.83 534.04 587.19 0.43

Heihe 696.62 492.53 621.42 0.42

Qiqihar 890.39 611.73 776.95 -0.49a

Daqing 994.85 758.02 863.13 -1.27b

Suihua 800.51 563.47 679.05 -0.18

Yichun 674.75 547.66 610.55 -0.29

Harbin 830.34 576.99 696.04 -0.84b

Hegang 767.47 568.81 669.65 -0.13

Jiamusi 772.36 595.25 677.89 -0.16

Shuangyashan 909.88 625.80 732.53 -0.85a

Qitaihe 890.68 668.74 765.19 -0.80a

Jixi 772.60 600.57 680.62 0.14

Mudanjiang 825.84 606.59 688.66 -0.33

a Significance at 0.05;

b Significance at 0.01.

Table 3

Trends of meteorological variables and ETo for cities.

City name Average temperature (℃ yr—1)

Average relative humidity (% yr—1)

Sunshine duration (h d—1 yr—1)

Average wind speed (m s—1 yr—1)

Precipitation (mm yr—1)

ETo (mm yr—1)

Daxinganling 0.040b -0.073b -0.008b -0.015b -0.762 0.427b Heihe 0.048b              -0.091b              -0.006              -0.020b              -0.258                            0.415

Qiqihar 0.034b -0.068b -0.008 -0.025b 0.537 -0.487a Daqing 0.022b              -0.031a              -0.012              -0.025b              0.447              -1.274b Suihua              0.028b              -0.015              -0.010a              -0.012b              0.305              -0.176

Yichun 0.030b -0.036a -0.004 -0.016b 0.289 -0.291 Harbin 0.034b              -0.038a              -0.018a              -0.023b              -0.986              -0.835b Hegang              0.020b              0.010              -0.003              -0.003              -0.569              -0.131

Jiamusi 0.024b -0.015 -0.006 -0.008b -0.318 -0.159

Shuangyashan 0.032b -0.024 -0.005 -0.026b -0.963 -0.848a

Qitaihe 0.029b 0.010 -0.015b -0.008b -0.507 -0.796 Jixi 0.027b              -0.017              -0.007a              -0.010b              0.228              0.138a Mudanjiang              0.031b              -0.040a              -0.019b              -0.012b              0.821              -0.332 Mean              0.031b              -0.033a              -0.009              -0.016              -0.134              -0.335

a Significance at 0.05;

b Significance at 0.01.

Table 4

Correlation coefficients between ETo and meteorological variables for cities.

City name Average temperature Average relative humidity Sunshine duration Average wind speed Precipitation

Daxinganling 0.290a -0.560b 0.319a -0.093 -0.456b Heihe 0.479b              -0.555b              0.493b              -0.146              -0.452b Qiqihar              0.134              -0.589b              0.379b              0.347b              -0.669b Daqing              0.108              -0.535b              0.640b              0.549b              -0.683b Suihua              0.286a              -0.618b              0.620b              0.355b              -0.506b Yichun              0.188              -0.533b              0.699b              0.254a              -0.422b Harbin              0.054              -0.556b 0.730b 0.436b -0.419b Hegang 0.354b -0.675b 0.600b 0.348b -0.522b Jiamusi 0.250a -0.720b 0.638b 0.419b -0.581b

Shuangyashan 0.209 -0.711b 0.802b 0.450b -0.569b Qitaihe 0.228              -0.806b              0.679b              0.419b              -0.538b Jixi              0.381b              -0.687b              0.504b              0.189              -0.517b

Mudanjiang 0.191 -0.611b 0.518b 0.373b -0.539b

a Significance at 0.05;

b Significance at 0.01.

areas with high ETo in the Songnen and Sanjiang Plain. From the second period to the third period, the high ETo area decreased, with more obvious decreasing trend in the Sanjiang Plain (Fig. 2B and C). From the third period to the fourth period, the high ETo area decreased more obviously, and the area with higher ETo in the Sanjiang Plain was only the southernmost area (Fig. 2C and D). The ETo of Heilongjiang Prov- ince from 1951 to 2018 had an overall decreasing trend. At the same time, the high ETo areas gradually decreased.

In the entire study period, the average ETo in Heilongjiang Province had a decreasing trend (Fig. 3). The linear trend was 0.08 mm yr—1,

and the Mann-Kendall test was not significant. The decreasing trend has been confirmed in other studies (Xie et al., 2017; Nie et al., 2019). The

maximum value appeared in 1982 at 786.53 mm yr—1, and the mini-

mum value appeared in 2013 at 614.40 mm yr—1. The multi-year average ETo was 689.51 mm yr—1.

The changes in ETo in 13 cities in Heilongjiang Province from 1951 to 2018 were different (Table 2). There were 3 cities with increasing trends. They are Daxinganling, Heihe and Jixi. Ten cities had decreasing

trends. Qiqihar, Shuangyashan and Qitaihe had significant decreasing trends, while trends in Daqing and Harbin were extremely significant.

ETo in Daqing changed at the fastest speed of 1.27 mm yr—1. The rate

of change in ETo in Hegang was 0.13 mm yr—1, which was the slowest.

ETo in Daqing had the highest average value of 863.13 mm yr—1. The average annual ETo in Daxinganling was the lowest.

3.2.
Changes of meteorological variables and their effects on ETo

The trends in meteorological variables and ETo with time in each city

in Heilongjiang Province from 1951 to 2018 are shown in Table 3. The average temperature in Heilongjiang Province increased at a rate of

0.031 ℃ yr—1. Heihe had the most obvious temperature rise in the

province. The average change rate of the average relative humidity in

Heilongjiang Province was 0.033% yr—1. The trends in Hegang and Qitaihe gradually increased, and the trends in the other 11 cities grad-

ually decreased. The trends of sunshine duration in the 13 cities all

declined, with an average rate of 0.009 h d—1 yr—1. The trends of the average wind speed in the 13 cities also declined. The average rate of the

average wind speed was 0.016 m s—1 yr—1. The trends of precipitation

gradually increased in 6 cities and decreased in 7 cities, with an average

rate of 0.134 mm yr—1. Among the 13 cities in Heilongjiang Province, 10 cities had a decreasing ETo trend, and 3 cities had an increasing

trend, with an average rate of 0.335 mm yr—1. Over the past 68 years,

the average temperature has increased while the average relative hu- midity, sunshine duration, average wind speed, precipitation and ETo showed a decreasing tend in Heilongjiang Province.

The correlations between ETo and meteorological variables in 13 cities in Heilongjiang Province are shown in Table 4. From 1951–2018, ETo in each city was positively correlated with average temperature. The average relative humidity was negatively correlated with ETo in

50

40

30

20

10

0

-10

1951 1961 1971 1981 1991 2001 2011

Year

et al., 2012). In addition, the large increase in average temperature, the decrease in average relative humidity and the decrease in precipitation were the main reasons for the large increase in ETo in Daxinganling and Heihe.

3.3. Changes of rice area and their effects on meteorological variables and ETo

From the time series, rice area in Heilongjiang Province continued to

increase from 1951 to 2018, with a growth rate of 0.54 105 ha yr—1, which passed the Mann-Kendall test with a 0.05 confidence level. More obvious growth in rice area was evident since 1980 (Fig. 4). Since the beginning of 2004, the state has adopted a series of measures to increase grain output and encourage farmers to increase their income (Hou et al., 2015). Rice production in Heilongjiang Province entered a new stage of growth due to the policy of strengthening agriculture and benefiting

farmers. During 2004–2013, rice area expansion was the fastest, with a

Fig. 4. Trend in rice areas in Heilongjiang Province from 1951 to 2018.

Fig. 5. Mann-Kendall test of rice area in Heilongjiang Province from 1951 to 2018.

each city. Sunshine duration was significantly positively correlated with ETo in each city. The correlation coefficients between the average wind speed and ETo in Daxinganling and Heihe were negative, but the cor- relations were positive in other cities. Precipitation and ETo were significantly negatively correlated in each city. In conclusion, ETo in Heilongjiang Province was positively correlated with average tempera- ture, sunshine duration and average wind speed and negatively corre- lated with average relative humidity and precipitation. ETo had a strong correlation with average relative humidity, sunshine duration and pre- cipitation, and it had a weak correlation with average temperature and average wind speed.

The impact of changes in meteorological variables on ETo was ob- tained through trend analysis and correlation analysis in each city. The increase in average temperature and the decreases in average relative humidity and precipitation in Heilongjiang Province could result in the increase in ETo, and the decreases in sunshine duration and average wind speed could result in the decrease in ETo. ETo in Heilongjiang Province had a decreasing trend, so the effect of sunshine duration and average wind speed on ETo was greater than that of average tempera- ture, average relative humidity and precipitation (Gao et al., 2006; Nan

rate of 2.67 × 105 ha yr—1. In 2013, rice area reached a maximum of

40.29 × 105 ha. Increased rice area during the study period accounted for 8.44% of the province’s total area. The expansion rate of rice area also accelerated gradually. Mann-Kendall test of rice area in Hei- longjiang Province from 1951 to 2018 was shown in Fig. 5. There was a significant upward trend from 1980 to 2018 (UF > 1.96).

Rice fields in Heilongjiang Province were mainly distributed in the cities located in the Sanjiang Plain and the Songnen Plain (Fig. 6). The expansion of rice area was most obvious in Sanjiang Plain, followed by Songnen Plain. The proportions of rice area in 13 cities in Heilongjiang Province gradually increased, and the increase varied from city to city. The increases in Jiamusi, Shuangyashan, Jixi, Hegang, Harbin, Suihua and Qiqihar were higher than those in the other cities. The growth of Jiamusi in 2011–2018 was particularly prominent because the munic- ipal party committee and municipal government implemented the Jia- musi "10 Million Mu, 10 Billion Kilo" Rice Development Plan in 2010. Daqing, Yichun, Qitaihe and Mudanjiang had a small increase in rice areas. No increase in rice areas was observed in Daxinganling and Heihe. Rice is often grown in plain areas, such as the Songnen and Sanjiang Plain in Heilongjiang Province (Lu et al., 2017). 93.22% of the rice area expansion occurred in Songnen and Sanjiang Plain. Sanjiang Plain accounted for 61.74% and Songnen Plain for 31.48% (Fig. 7). The sig- nificant stage of rice area expansion was selected to study the response of ETo and meteorological variables to rice area expansion. Counties within the same plain have similar topography and climatic conditions.

The difference in trends of meteorological variables in the plains were attributed to the influence of the change in the rice area proportion.

Average temperatures are increasing in all regions, but at different rates. In areas where the proportion of rice area is large, the temperature increase is small (Fig. 8). This result was found in both plains. The average temperature increase in regions where the proportion of rice area changed more than 20% was greater than that in regions with 15–20% rice area change. More swamps were converted into farmland in these regions, which may lead to higher temperatures. An increase in rice area inevitably leads to an increase in irrigation. Soil moisture in- creases with irrigation and then the surface albedo becomes larger. The ground absorbs less solar radiation as the albedo increases. This leads to a drop in the average temperature (Alpert and Mandel, 1986). From the perspective of energy balance, the irrigation of rice fields increases the evapotranspiration of the farmland and changes the energy distribution on the ground. Less energy is used to heat the ground, which will lower the surface temperature (Bonfils and Lobell, 2007; Sacks et al., 2009). These, in turn, affect the local climate. Rice area expansion has restrained the increase in regional average temperature in most areas.

Fig. 6. Rice area proportions in 13 cities in Heilongjiang Province from 1951 to 2018.

Thus, rice development increases evaporative cooling and lowers the air temperature during the growing season (Bonfils and Lobell, 2007; Han et al., 2018).

The average relative humidity in the Songnen Plain had an increasing trend in the areas where the proportion of rice areas increased more than 15% (Fig. 9A). The average relative humidity in all counties had an increasing trend in the Sanjiang Plain, and the range increased with rice area expansion (Fig. 9B). In areas where rice expanded faster, the average relative humidity had a faster increasing trend. Irrigation of rice fields increases soil moisture and forms water surface, which in turn increases actual evapotranspiration. This leads to an increase in water vapor in the air. Therefore, the expansion of rice area may promote an increase in average relative humidity. Irrigation expansion in cropland regions exerts significant cooling and wetting effects on the near-surface atmosphere (Puma and Cook, 2010).

The sunshine duration in most counties of the Songnen and Sanjiang Plains showed decreasing trends (Fig. 10). The sunshine duration in the two plains varied with the proportion of rice areas, and the effect of rice expansion on sunshine duration is not clear.

In order to remove the possible impact of increase in buildings sur-

rounding the stations on wind speed, weather stations with an added building area of more than 30 km2 were excluded from data. Average

wind speed in all areas of the Sanjiang Plain had a decreasing trend (Fig. 11B). Moreover, the decreasing range of the average wind speed increased with the increase in the rice area proportion. The average wind speed in the Songnen Plain decreased in all counties. It decreased greatly in locations where rice areas increased greatly, and the decreasing range was the largest when the rice area proportion was more than 20% (Fig. 11A). The expansion of rice areas tends to cause the regional average wind speed to decline, which is consistent with the conclusions drawn by Han et al. (2009) and Jia et al. (2016). On the one hand, rice area expansion in Heilongjiang Province may change the surface roughness and have a significant blocking effect on wind speed. On the other hand, the significant increase in irrigation caused a sig- nificant temperature and humidity difference with the surrounding area, thus affecting the local wind circulation (Alpert and Mandel, 1986; Ozdogan and Salvucci, 2004).

Precipitation in the counties of the Songnen Plain and Sanjiang Plain increased from 1980 to 2018 (Fig. 12). The increasing range was the largest when the rice area proportion was more than 20% in Sanjiang Plain. The impact of irrigation on precipitation is very complex. There is no obvious relationship between the changes of precipitation in different regions and the proportion of rice area. The impact of rice area expan- sion on regional precipitation is not clear.

Fig. 7. Rice areas changes during 1980–2018.

Fig. 8. Changes in average temperature in the Songnen Plain (A) and Sanjiang Plain (B) at different expansion levels.

Fig. 9. Changes in average relative humidity in the Songnen Plain (A) and Sanjiang Plain (B) at different expansion levels.

Fig. 10. Changes in sunshine duration in the Songnen Plain (A) and Sanjiang Plain (B) at different expansion levels.

Fig. 11. Changes in average wind speed in the Songnen Plain (A) and Sanjiang Plain (B) at different expansion levels.

Fig. 12. Changes in precipitation in the Songnen Plain (A) and Sanjiang Plain (B) at different expansion levels.

Fig. 13. Changes in ETo in the Songnen Plain (A) and Sanjiang Plain (B) at different expansion levels.

Fig. 14. Effects of rice area expansion and climate change on ETo in the Songnen Plain (A) and Sanjiang Plain (B).

ETo in the counties of the two plains decreased (Fig. 13). As the proportion of rice areas increased gradually, the decline in ETo in the Songnen Plain increased. Combined with the results of correlation analysis, slowing down the rise in average temperature, increasing the average relative humidity and slowing down the wind speed may lead to the decrease of ETo. The larger the proportion of irrigated area is, the greater the influence on evapotranspiration (Han et al., 2009; Jia et al., 2016). The trend may not be strict because the influence between adjacent areas cannot be ignored.

3.4. Contributions of climate change and rice area expansion to ETo trends

Fig. 14 shows that both rice area expansion and climate change in

Sanjiang and Songnen Plain can reduce ETo. The cumulative impact of rice area expansion and climate change on ETo in Sanjiang Plain reached

12.49 mm yr—1 and 13.89 mm yr—1 respectively; whereas, the cor-

responding values in Songnen Plain were 13.93 and 27.30 mm yr—1. According to the impacts of rice area expansion and climate change on ETo, the contribution of rice area expansion in Songnen Plain to the decline of ETo was 33.79% and that of climate change was 66.21%. Rice

area expansion and climate change in the Sanjiang Plain contributed 47.33% and 52.67% to the decline in ETo, respectively.

The response of ETo in Sanjiang Plain to rice area expansion was even smaller than that in Songnen Plain, which is inconsistent with the hypothesis that larger rice area expansion has greater influence on evapotranspiration (Han et al., 2009). About 25.80% of the rice fields in Sanjiang Plain converted from wetlands and 48.15% from dry fields (Fig. 15). The impacts of these two land use and land cover changes on ETo may be different.

The changes in ETo in dry fields converted to rice fields and wetlands converted to rice fields were extracted and analysed respectively. It was

found that wetlands converted to rice fields can increase ETo, with a cumulative effect of 7.67 mm yr—1. The transformation from dry fields

to rice fields can decrease ETo, with a cumulative reduction effect of

20.15 mm yr—1 (Fig. 16). ETo had a greater response to the trans- formation of dry fields into rice fields. The impact of rice area expansion in Sanjiang Plain was still dominated by decline ETo, and the decline was small. These findings are conducive to discussing the effect of rice area expansion on the ecological environment and strengthening the rational use of land resources. The expansion of rice areas played a role in decreasing regional ETo.

4. Conclusions

This study analysed the spatio-temporal dynamics of ETo and rice areas in Heilongjiang Province from 1951 to 2018 and evaluated the relationship among ETo, rice area and climate change. The conclusions are as follows:

(1) ETo in Heilongjiang Province decreased slightly, and the linear

trend was 0.08 mm yr—1. Moreover, the area of high ETo gradu- ally decreased. Rice areas in Heilongjiang Province continued to