Decomposition of Trend and Interdecadal Variation of Evaporation over the Tropical Indian Ocean in ERA5

Abstract

Based on ERA5 from 1980 to 2018, we compare and analyze the trend and interdecadal variation of evaporation anomalies over the tropical Indian Ocean by the evaporation decomposition method. This method mainly decomposes the evaporation anomalies into the Newtonian cooling, stability, relative humidity, wind speed, and transfer coefficient terms. The annual mean evaporation anomalies show an increasing trend (0.083 mm/d/decade). The Newtonian cooling term (0.026 mm/d/decade), the relative humidity term (0.032 mm/d/decade), and the wind speed term (0.026 mm/d/decade) play a major role in the increasing trend. The interdecadal variation of evaporation anomalies shows decreases in the 1980s and after the early 2000s, and an increase in the 1990s. The decreased evaporation anomalies in the 1980s are affected by the transfer coefficient term, which is associated with the North Atlantic Oscillation (NAO). The increased evaporation anomalies in the 1990s and the decreased evaporation anomalies since the early 2000s are largely controlled by the wind speed term, which are dominated by the Atlantic Multidecadal Oscillation (AMO). The Pacific Decadal Oscillation (PDO) may have important impacts on the interdecadal increase of evaporation anomalies by affecting the wind speed in the 1990s.

1. Introduction

Ocean evaporation is important for the exchange of heat and moisture between the atmosphere and ocean [1,2,3]. The changes in ocean evaporation can affect the sea surface temperature (SST) by atmospheric feedback and the freshwater flux (evaporation minus precipitation) and atmospheric circulation by energy exchanges [4,5,6]. During recent decades the tropical Indian Ocean SST has experienced rapid warming, with enhanced evaporation [7,8,9]. Therefore, understanding the tropical Indian Ocean evaporation variability is essential for the water and energy budget and the complex interaction between the atmosphere and ocean [10].

Although many studies pointed out that the magnitude of ocean evaporation differs considerably in different evaporation datasets [11,12,13], the ocean evaporation displays a significant increasing trend and has similar interdecadal variation [14,15,16]. Due to the development of modern observations and model simulations, the recently released fifth generation of European Centre for Medium-Range Weather Forecasts (ECMWF) reanalysis (ERA5) shows the most reliable estimate in the surface flux over the tropical Indian Ocean [17].

The ocean evaporation is estimated using similar aerodynamic bulk formulae [18,19], which is significantly affected by different factors, such as SST, surface air temperature, near-surface relative humidity, and near-surface wind speed [20]. Lorenz et al. [21] found that the increased relative humidity mainly contributes to the reduction in the evaporation increase below the Clausius–Clapeyron rate. The increasing trend of evaporation over the tropical Pacific is closely related to the SST warming and the enhanced wind speed [15]. The interdecadal trends of global ocean evaporation are affected by the air-sea humidity difference and the wind speed [22]. The enhanced near-surface wind speed leads to negative SST anomalies by the enhanced evaporation, popularly known as wind-evaporation-SST (WES) feedback [23]. Therefore, it is essential to analyze and understand the contribution of these factors to ocean evaporation.

The sensitivities of ocean evaporation to the factors have been analyzed in previous studies [10,24]. However, these studies were restricted to the influence of the factors on evaporation. In this study, we not only quantify the influence of the factors on the evaporation, but also examine the long-term trend and interdecadal variation of evaporation. Recent studies indicated that the evaporation decomposition method can quantify the contribution of the factors controlling ocean evaporation [25,26]. The increased relative humidity, increased air-sea temperature difference, and decreased wind speed reduce the increase in the global ocean [19] and tropical Indian Ocean evaporation [8] to 2%/K in the model simulations. The wind speed plays a dominant role in the increased evaporation, and moisture difference contributes to the decreased evaporation over the quasi-global oceans on the interdecadal scale [25]. Therefore, this study focuses on the trend and interdecadal variation of the annual and seasonal mean evaporation anomaly for the period 1980–2018 over the tropical Indian Ocean using ERA5. The impacts of the main factors (SST, sea-air temperature difference, relative humidity, wind speed, and transfer coefficient) on the evaporation anomaly are also investigated by the evaporation decomposition method, and the related climate variations are discussed.

2. Data and Methods

2.1. Data

ERA5 is the latest reanalysis produced by ECMWF from 1979 to present with 0.25° × 0.25° horizontal resolution [27]. The variables, including evaporation, surface air temperature, surface dew point temperature, and surface wind speed, were used in this study. The monthly SST dataset is derived from the National Oceanic and Atmospheric Administration (NOAA) Extended Reconstructed SST version 5 (ERSSTv5) on a horizontal resolution of 2° × 2° [28]. In this study, we focused on the spatial mean of the tropical Indian Ocean (40° E–120° E, 30° S–30° N) during the period 1980–2018 (Figure 1). For a better match of the spatial domain used to calculate area-averaged values, the ERA5 dataset was interpolated into a 2° × 2° grid using bilinear interpolation. In addition, the North Atlantic Oscillation (NAO), Pacific Decadal Oscillation (PDO), and Atlantic Multidecadal Oscillation (AMO) were downloaded from Climate Explorer and used to analyze the related climate variability.

2.2. Evaporation Decomposition Method

We used a diagnostic partitioning of the bulk aerodynamic formulation for evaporation following Richter and Xie [19]:

$${\mathrm{EVP}=\mathrm{C}}_{\mathrm{E}}{\mathsf{\rho }}_{\mathrm{a}}{\mathrm{U}[\mathrm{q}}_{\mathrm{s}}\left(\mathrm{SST}\right)-{\mathrm{RH}\times \mathrm{q}}_{\mathrm{s}}(\mathrm{SST}+\mathrm{S})]$$

(1)

where C_E is the transfer coefficient and ${\mathsf{\rho }}_{\mathrm{a}}$, U, and RH are the surface air density, surface wind speed, and surface relative humidity, respectively. $\mathrm{S}=\mathrm{SAT}-\mathrm{SST}$ is the difference between the surface air temperature (SAT) and SST, representing the surface stability. ${\mathrm{q}}_{\mathrm{s}}\left(\mathrm{SST}\right)$ is the saturated specific humidity. ${\mathrm{RH}\times \mathrm{q}}_{\mathrm{s}}(\mathrm{SST}+\mathrm{S})$ is the surface-specific humidity. To estimate the contributions of each factor to evaporation, the evaporation anomalies (${\mathrm{EVP}}^{\prime }$) were decomposed following previous studies [25,26]:

$${\text{EVP}}^{\prime }=\frac{\partial \text{EVP}}{\partial \text{SST}}{\text{SST}}^{\prime }+\frac{\partial \text{EVP}}{\partial \mathrm{S}}{\mathrm{S}}^{\prime }+\frac{\partial \text{EVP}}{\partial \text{RH}}{\text{RH}}^{\prime }+\frac{\partial \text{EVP}}{\partial \mathrm{U}}{\mathrm{U}}^{\prime }+\frac{\partial \text{EVP}}{\partial {\mathrm{C}}_{\mathrm{E}}}{{\mathrm{C}}_{\mathrm{E}}}^{\prime }+\text{RES}$$

(2)

where the partial derivatives are calculated from monthly climatology, representing the sensitivity of evaporation to each factor. The prime denotes a departure from monthly climatology, which, in our case, is based on the period 1980–2018. The right side of the equation represents the Newtonian cooling term (NC), surface stability term (S*), relative humidity term (RH*), wind speed term (U*), transfer coefficient term (C_E*, which is calculated by dividing the evaporation by ${\mathsf{\rho }}_{\mathrm{a}}{\mathrm{U}[\mathrm{q}}_{\mathrm{s}}{\left(\mathrm{SST}\right)-\mathrm{RH}\times \mathrm{q}}_{\mathrm{s}}(\mathrm{SST}+\mathrm{S})]$), and residual term (RES), respectively. This symbol (*) is used to distinguish surface stability term (S*) from surface stability (S).

2.3. Statistical Methods

The trend of evaporation and its main factors for the period 1980–2018 were calculated by the linear ordinary least squares method. To extract the interdecadal variation (IDV), all variables were first detrended by removing the linear trend, and we performed the 9-year running mean on the detrended annual mean (the average values in January–December, namely Year) evaporation anomalies and seasonal mean evaporation anomalies from spring to winter (the average values in March–April–May, June–July–August, September–October–November, and December–January–February, namely MAM, JJA, SON, and DJF) over the tropical Indian Ocean and their main factors. The statistical significances of correlation coefficients and trends were based on a two-tailed Student’s t-test. In this study, to keep the same number of years of data used to calculate the averages, we used 38 years (1981–2018) of data in Year, MAM, JJA, and SON and 38 years (1981–2018) of data in DJF. To consider the influence of 9-year running mean on degrees of freedom, the effective number of degrees of freedom was calculated following previous studies [29,30]:

$$\frac{1}{{\mathrm{N}}^{\mathrm{eff}}}\approx \frac{1}{\mathrm{N}}+\frac{2}{\mathrm{N}}{\displaystyle \sum }_{\mathrm{j}=1}^{\mathrm{N}}\frac{\mathrm{N}-\mathrm{j}}{\mathrm{N}}{\mathsf{\rho }}_{\mathrm{XX}}\left(\mathrm{j}\right){\mathsf{\rho }}_{\mathrm{YY}}\left(\mathrm{j}\right)$$

(3)

where N is the sample size and ${\mathsf{\rho }}_{\mathrm{XX}}\left(\mathrm{j}\right)$ and ${\mathsf{\rho }}_{\mathrm{YY}}\left(\mathrm{j}\right)$ are the autocorrelations of two sampled time series X and Y at time lag j, respectively.

3. Results

3.1. Trend of Evaporation over the Tropical Indian Ocean

Figure 2 shows the linear trend and interdecadal variation of annual mean ${\mathrm{EVP}}^{\prime }$ over the tropical Indian Ocean in ERA5. An increasing ${\mathrm{EVP}}^{\prime }$ was observed with a significant linear trend of 0.085 mm/d/decade with a 99% confidence level. A clear interdecadal variation was observed, showing a negative ${\mathrm{EVP}}^{\prime }$ from the 1980s to the mid-1990s and from 2010 onwards and positive ${\mathrm{EVP}}^{\prime }$ from the mid-1990s to the 2000s. The results are consistent with previous studies [22,31]. We further investigated the main factors affecting the increasing trend of ${\mathrm{EVP}}^{\prime }$ by the evaporation decomposition method.

Table 1. The trends of annual and seasonal mean evaporation anomalies and the decomposition terms. Unit: mm/d/decade.

	${\mathbf{EVP}}^{\prime }$	NC	S*	RH*	U*	CE*
Year	0.085	0.026	−0.004	0.031	0.026	0.001
MAM	0.098	0.022	0	0.035	0.034	0.004
JJA	0.081	0.03	0.003	0.036	0.01	0.002
SON	0.078	0.027	−0.014	0.021	0.029	0.013
DJF	0.086	0.025	−0.006	0.032	0.033	0.001

Note: the values in bold font and bold italic font are significant with the 95 and 99% confidence level, respectively.

shows the trends of annual and seasonal mean ${\mathrm{EVP}}^{\prime }$ and the related decomposition terms. From the annual mean ${\mathrm{EVP}}^{\prime }$, the significant trends of NC, RH*, and U* were 0.026, 0.031, and 0.026 mm/d/decade, respectively. These results indicate that the NC, U*, and RH* were the major contributors to the increasing trend of ${\mathrm{EVP}}^{\prime }$, being the latter slightly more important.

Similar to the annual mean ${\mathrm{EVP}}^{\prime }$, the NC, RH*, and U* play major roles in the increasing trend of seasonal mean ${\mathrm{EVP}}^{\prime }$ (Table 1). The ${\mathrm{EVP}}^{\prime }$ increase reached its maximum (0.098 mm/d/decade) in MAM, which may be affected by the magnitude of the U* trend. The ${\mathrm{EVP}}^{\prime }$ increase showed a minimum (0.078 mm/d/decade) in SON, which is affected by the magnitude of the RH* trend and a negative S* trend. In JJA, the increasing trend of ${\mathrm{EVP}}^{\prime }$ was mainly due to a positive trend in NC associated with an SST warming and a positive trend in RH*. The positive U* trend was not significant with a 95% confidence level in MAM and JJA. Although the positive U* trends had a large magnitude in SON and DJF, they were significant with a 95% confidence level and not significant with a 99% confidence level. Therefore, the upward trends of the seasonal mean ${\mathrm{EVP}}^{\prime }$ were mainly affected by the NC, U*, and RH*, and their magnitude was significantly affected by the S*, especially in SON. Under greenhouse warming, the increased downward longwave radiation warmed the tropical Indian Ocean, leading to SST warming and evaporation increasing [8]. The increased saturated specific humidity in global warming could tend to reduce relative humidity, which may further promote the increase in evaporation [16,32]. The wind speed showed an increasing trend over the past two decades, which was likely associated with the increasing trend of evaporation [14,33,34].

3.2. Interdecadal Variation of Evaporation over the Tropical Indian Ocean

Next, we investigated the dominant factors of interdecadal variation of ${\mathrm{EVP}}^{\prime }$ over the tropical Indian Ocean. Figure 3 shows the IDV of the annual mean ${\mathrm{EVP}}^{\prime }$ and the related decomposition terms. The IDV of the ${\mathrm{EVP}}^{\prime }$ showed a small downward trend in the 1980s, an upward trend from the 1990s to the early 2000s, and a downward trend from the early 2000s onwards. Note that the RES is small and bears little resemblance to the IDV of ${\mathrm{EVP}}^{\prime }$, suggesting that the evaporation decomposition method has captured the main features in the IDV of the ${\mathrm{EVP}}^{\prime }$. The IDV of the U* increased from the 1980s to the early 2000s and decreased since about 2005, similar to the results of Yu [22]. The correlation coefficient between the IDV of U* and ${\mathrm{EVP}}^{\prime }$ was 0.81 with the 95% confidence level. The IDV of C_E* decreased in the 1980s, corresponding to the downward trend of the IDV of ${\mathrm{EVP}}^{\prime }$ (Figure 3b). The IDV of C_E* showed high correlation with the IDV of ${\mathrm{EVP}}^{\prime }$ (0.78, above the 95% confidence level), suggesting that C_E has an important influence on the IDV of ${\mathrm{EVP}}^{\prime }$. Although the IDV of RH*, similar to the IDV of ${\mathrm{EVP}}^{\prime }$, the correlation between the IDV of RH* and ${\mathrm{EVP}}^{\prime }$ was not significant with the 95% confidence level (the effective number of degrees of freedom was 5). The results indicate that the U* and C_E* play a major role in the IDV of ${\mathrm{EVP}}^{\prime }$. The C_E* contributed to the IDV of ${\mathrm{EVP}}^{\prime }$ in the 1980s and the U* contributed to the IDV of ${\mathrm{EVP}}^{\prime }$ after the late 1990s.

Furthermore, the IDV of ${\mathrm{EVP}}^{\prime }$ and the related decomposition terms from MAM to DJF are shown in Figure 4. The IDVs of MAM and SON ${\mathrm{EVP}}^{\prime }$ were similar to the IDV of the annual mean ${\mathrm{EVP}}^{\prime }$. The IDV of ${\mathrm{EVP}}^{\prime }$ featured an upward trend from the 1980s to the early 2000s and the downward trend from the early 2000s onwards in JJA. The IDV of DJF ${\mathrm{EVP}}^{\prime }$ showed a downward trend in the 1980s and an upward trend in the 1990s. Compared with the IDV of the annual mean ${\mathrm{EVP}}^{\prime }$, the IDV of C_E* showed low correlation with the IDV of ${\mathrm{EVP}}^{\prime }$ in MAM, and the IDV of RH* also had a significant correlation (0.86) with the IDV of ${\mathrm{EVP}}^{\prime }$ in JJA. The correlation between the IDV of RH* and ${\mathrm{EVP}}^{\prime }$ exceeded that of the IDV of U* and ${\mathrm{EVP}}^{\prime }$ in JJA. The IDV of C_E* had a high correlation (0.9) with the IDV of ${\mathrm{EVP}}^{\prime }$ in DJF with a 95% confidence level, suggesting that the C_E* contributed largely to the IDV of ${\mathrm{EVP}}^{\prime }$ in DJF. Overall, the IDV of ${\mathrm{EVP}}^{\prime }$ was not consistent in different seasons. The U* played an important role in the IDV of ${\mathrm{EVP}}^{\prime }$ in all seasons, the RH* contributed to the IDV of ${\mathrm{EVP}}^{\prime }$ in JJA, and the C_E* exerted an important influence on the IDV of ${\mathrm{EVP}}^{\prime }$ in JJA, SON, and DJF.

To investigate the impacts of climate modes on the IDV of ${\mathrm{EVP}}^{\prime }$, we calculated the correlation between the IDVs of major climate modes (PDO, AMO, and NAO) and the annual mean ${\mathrm{EVP}}^{\prime }$ and the related decomposition terms (Figure 5). The correlations between the IDV of RH*, U*, and AMO were 0.88 and 0.89 (above the 95% confidence level), respectively. The results indicate that the AMO contributed to the increased ${\mathrm{EVP}}^{\prime }$ in the 1990s and the decreased ${\mathrm{EVP}}^{\prime }$ after the early 2000s by affecting the RH* and U* (Figure 5a). The correlation between the IDV of C_E* and NAO was −0.8 (above the 95% confidence level), suggesting that the NAO contributed to the decreased ${\mathrm{EVP}}^{\prime }$ in the 1980s and the increased ${\mathrm{EVP}}^{\prime }$ in the 1990s by affecting the C_E* (Figure 5b). The correlation between the IDVs of ${\mathrm{EVP}}^{\prime }$ and PDO was −0.76, which was significant with the 95% confidence level. However, the IDVs of other terms were not significantly correlated with the IDV of PDO (Figure 5d). The 9-year running correlation between the IDVs of U* and PDO was comparable to that of the IDVs of RH*, U*, and AMO in the 1990s, but showed a shape decrease after the early 2000s (Figure 5c). The results indicate that the PDO contributed to the increased ${\mathrm{EVP}}^{\prime }$ in the 1990s by affecting the U*. Therefore, the NAO plays an important role in the decreased ${\mathrm{EVP}}^{\prime }$ in the 1980s and the increased ${\mathrm{EVP}}^{\prime }$ in the 1990s, and the AMO played an important role in the increased ${\mathrm{EVP}}^{\prime }$ in the 1990s and the decreased ${\mathrm{EVP}}^{\prime }$ after the early 2000s. In addition, the PDO may have an important influence on the IDV of ${\mathrm{EVP}}^{\prime }$ by affecting the U* in the 1990s.

4. Conclusions and Discussion

This study focuses on the trend and interdecadal variation of evaporation in the tropical Indian Ocean. The results show that the RH* plays a slightly more important role in the trend of annual mean ${\mathrm{EVP}}^{\prime }$. Gao et al. [16] emphasized the important role of humidity difference in the latent heat flux trends. Li et al. [15] suggested that the increasing trend of latent heat flux is strongly associated with the SST warming and strengthening U during 1977–2006. The IDV of annual mean ${\mathrm{EVP}}^{\prime }$ is mainly affected by the U and C_E. Previous studies pointed out that the U plays the important role in the IDV of ${\mathrm{EVP}}^{\prime }$ [22,31]. The Atlantic warming induces the wind anomalies over the Indo-western Pacific [35]. The PDO may induce wind anomalies to change the subsurface ocean heat content in the Indian Ocean by the atmospheric bridge [36]. Robertson et al. [25] showed that the variations of C_E* are smaller than the other terms in global ocean evaporation. Morioka et al. [37] suggested that the C_E* contributes to the ${\mathrm{EVP}}^{\prime }$ over the south Indian Ocean in December, corresponding to the result of the IDV of ${\mathrm{EVP}}^{\prime }$ in DJF. In addition, the C_E* may be affected by the U [26].

In this study, we investigated the trend and interdecadal variation of evaporation over the tropical Indian Ocean in ERA5 for the period of 1980–2018 by the evaporation decomposition method. A significant increasing trend appeared from 1980 to 2018, with a linear trend of 0.083 mm/d/decade. The key terms controlling the increasing trend of annual mean ${\mathrm{EVP}}^{\prime }$ were the NC, RH*, and U*, with the linear trends of 0.026, 0.032, and 0.026 mm/d/decade, respectively. The result indicates that the increasing trend of ${\mathrm{EVP}}^{\prime }$ was associated with the SST warming and the increasing U*. The decomposition analysis of seasonal mean ${\mathrm{EVP}}^{\prime }$ indicates that the C_E* significantly affected the magnitude of the increasing trend of ${\mathrm{EVP}}^{\prime }$ in SON.

The IDV of annual mean ${\mathrm{EVP}}^{\prime }$ showed a decrease in the 1980s, an increase from the 1990s to the early 2000s, and a decrease from the early 2000s onwards. The decreased ${\mathrm{EVP}}^{\prime }$ in the 1980s was attributed to the C_E, the increased ${\mathrm{EVP}}^{\prime }$ from the 1990s to the early 2000s, and the decreased ${\mathrm{EVP}}^{\prime }$ from the early 2000s onwards were associated with the U. For the IDV of seasonal mean ${\mathrm{EVP}}^{\prime }$, the U was the important factor in the IDV of ${\mathrm{EVP}}^{\prime }$ in all seasons. The RH* played a key role in the IDV of ${\mathrm{EVP}}^{\prime }$ in JJA and the C_E* had an important influence on the IDV of ${\mathrm{EVP}}^{\prime }$ in JJA, SON, and DJF. Furthermore, the IDV of the annual mean ${\mathrm{EVP}}^{\prime }$ was significantly correlated with the PDO, AMO, and NAO. The AMO contributed to the increased ${\mathrm{EVP}}^{\prime }$ in the 1990s and the decreased ${\mathrm{EVP}}^{\prime }$ after the early 2000s by affecting the RH* and U*. The NAO contributed to the decreased ${\mathrm{EVP}}^{\prime }$ in the 1980s by affecting the C_E*. The PDO may affect the increased ${\mathrm{EVP}}^{\prime }$ in the 1990s by the U*.