1. Introduction
Actual evapotranspiration (ET) is a major component in the hydrological cycle together with precipitation, along with being one of the major constituents of the surface energy balance (Trenberth et al. 2007, 2009). Changes in terrestrial ET can affect precipitation, streamflow and temperature among other hydroclimatological variables (Koster et al. 2004; Seneviratne et al. 2010). In addition, ET plays an important role in the climate system coupling with the water, carbon and energy cycles (Jung et al. 2010).
Being such an essential variable, its global characterization is critical to fully understand and model the main processes behind climate and life cycles on our planet. Despite global efforts to directly measure ET such as the FLUXNET project (Baldocchi et al. 2001), few long time series of ground observations exist. Additionally, while remote sensing has made major strides in recent years, satellites are still incapable of quantifying ET amounts (K. Zhang et al. 2016) directly. Instead, remotely sensed data of related variables, such as temperature, radiation, and vegetation coverage, have allowed the creation of several derived ET products in recent years. Such ET products have been often used as surrogates of ET in water balance studies (Fisher et al. 2008; Jung et al. 2010; Miralles et al. 2011; Mu et al. 2007; Yang et al. 2013; Zhang et al. 2010). Modeled ET products, and derivations using land surface models such as the ones used in the Global Land Data Assimilation (Rodell et al. 2004) have also been developed and are being used for a range of applications.
All of the abovementioned products rely on different forcing data and methods to estimate ET. This leads to different magnitudes and spatial distributions of ET on our planet. Several studies have investigated such differences in detail, showing high uncertainties in the ET estimations using different models (Chen et al. 2014; Ershadi et al. 2014; Jiménez et al. 2011; Miralles et al. 2016; Mueller et al. 2011), and McMahon et al. (2013) provide an insight of the vast disparities in evapotranspiration assessment using different theoretical methods. This is consistent with uncertainties associated with other global variable datasets, many of which are derived from satellite retrievals (Kim and Sharma 2019; Libertino et al. 2016; Zhang et al. 2019).
The one aspect that has not been looked in detail is the response of ET, as represented in these data, to rising temperatures. The dissimilarity between the trends in ET across different datasets is important especially when studying climate change, as there is consensus across modeling and observational groups worldwide that global temperatures are increasing, leading to speculation that similar increases should be reflected in observed ET as well as projected ET for future conditions (Johnson and Sharma 2010; Stephens et al. 2018). Quicker dryness of catchments following precipitation events is also noted to be occurring, again attributed to higher potential evaporation rates (Woldemeskel and Sharma 2016; Hettiarachchi et al. 2019). The observational uncertainty and dependence on models are limiting our capacity to detect changes in ET (IPCC 2014), even without considering the intrinsic variations among the datasets. This uncertainty also has considerable implications on climate change impact assessments as global ET datasets are used to compare the results of general circulation models (Mueller et al. 2011) and to investigate the drivers behind the trends in ET (Jung et al. 2010; Miralles et al. 2013; Trambauer et al. 2014).
Attempts have been made to assess the uncertainty of the trends in ET at a global scale (Vinukollu et al. 2011; Zhang et al. 2016b). However, their regional variations are less well understood. Moreover, previous works have studied the ET trends over several datasets, but have mostly focused on specific regions (Kiptala et al. 2013; Marshall et al. 2012), global totals (Jung et al. 2010; Zhang et al. 2016b) or merged products (Mueller et al. 2013). Even though the uses of merged products, constructed from a group of observations, provide useful insights over the expected changes in ET in a specific zone, they tend to mask the differences between the datasets. Namely, the trend of the merged product could be overshadowed by the presence of a dataset with considerably higher values of ET or trend magnitudes.
Considering the limitations of existing research on ET trends, this work first aims to assess the spatial differences between the ET trends across global datasets, in order to identify zones where the trends concur and differ. The second aim of this study is to analyze how many of these datasets coincide in terms of magnitude and direction of trends so as to provide assessments on the more consistent ET data to use. Therefore, in the present work, the analysis focuses on the consistency of direction, statistical significance and currency in ET trends among the datasets. In this way, we expect to provide a deeper understanding of uncertainty of the trends and identify regions where the direction of change appears inconsistent, and hence, dubious.
2. Data and method
a. Selected ET datasets and data processing
The global scale of the problem required a group of global ET datasets, with enough monthly data points to allow a trend analysis to be carried out with reliability. Considering this, as detailed in Table 1, 11 global ET datasets were selected. Based on Mueller et al. (2011), this study classifies the 11 datasets into three categories. These are “diagnostic,” derived from combinations of observations or observation-based estimates, together with relatively simple or empirical formulations; “land surface model,” derived from land surface models with observation-based forcing; and “reanalysis,” models with assimilating observations.
Global ET datasets used in the study and their main characteristics (data links are presented in the footnotes). The temporal resolutions are the ones used in this study, and finer resolutions are available for some datasets. The coverage indicates the extension in latitude with evaporation data. Acronyms used in this table are given in section 2a.
For the trend analysis in this study, the following preprocessing was carried out for the datasets. First, a monthly temporal scale (mm month−1) was used to fully utilize the available data. Second, all datasets at various spatial resolutions were resampled to the global cylindrical 36-km Equal-Area Scalable Earth, version 2 (EASEv2), grid (Brodzik et al. 2012) by using the nearest neighbor algorithm to simplify the comparison being presented, and the trend test was implemented over each cell. Note that Antarctica was not considered in the analysis because of lack of data over the area. Last, a 32-yr period from January 1981 to December 2012 was selected for the evaluation and is the longest common period covered by all datasets.
1) CSIRO
The Commonwealth Scientific and Industrial Research Organisation (CSIRO) dataset is based on the Penman–Monteith–Leuning model, using the WATCH-Forcing-DATA-ERA-Interim and the Princeton Global Forcing as meteorological forcing data (Zhang et al. 2016a,b). The leaf area index data were obtained from the Boston University dataset, and the Global Land Surface Satellite (GLASS) dataset was used for emissivity and albedo. The land cover was taken from International Geosphere–Biosphere Program data. This dataset has the particularity of being conceived to study trends in evaporation, thus its trends were validated against estimated evaporation trends in several parts of the world.
2) GLEAM
The Global Land Evaporation Amsterdam Model (GLEAM) has two different datasets that were based on reanalysis (v3.3a) and remote sensing data (v3.3b), respectively (Martens et al. 2017; Miralles et al. 2011). In this study, we used the first one, GLEAM v3.3a. GLEAM v3.3a uses various data as 1) European Centre for Medium-Range Weather Forecasts (ECMWF) interim reanalysis (ERA-Interim) data for radiation and air temperature; 2) Multi-Source Weighted-Ensemble Precipitation v1.0 for precipitation; 3) Global Snow Monitoring for Climate Research (GLOBSNOW) L3A v2 and National Snow and Ice Data Center (NSIDC) v01 for snow water equivalents; and 4) Land Parameter Retrieval Model–based vegetation optical depth.
3) GLDAS
Four Global Land Data Assimilation System (GLDAS), version 1.0 (GLDAS-1), datasets and two version-2.0 (GLDAS-2) datasets were used in this study, each one based on a different land surface model (LSM) (Rodell et al. 2004). GLDAS-1 datasets share the same forcing data, which change over time. To estimate ET from 1979 to 1993 a bias-corrected ERA dataset was used. From 1994 to 1999, the utilized dataset was a bias-corrected National Center for Atmospheric Research reanalysis; for year 2000, the National Oceanic and Atmospheric Administration Global Data Assimilation System (GDAS) atmospheric analysis fields were used; and from 2001 to the present, a mixture of GDAS atmospheric fields, Climate Prediction Center Merged Analysis of Precipitation (CMAP) precipitation and observation-based downward shortwave and longwave radiation fields derived using the method of the Air Force Weather Agency’s Agricultural Meteorological modeling system (AGRMET) were combinedly used to estimate ET among other variables. It should be noted that the GLDAS-1 was decommissioned in June 2020 and replaced with GLDAS-2.0, 2.1, and 2.2. Whereas GLDAS-1 used multiple different meteorological datasets as mentioned above, GLDAS-2 is entirely forced with the Princeton meteorological forcing dataset (Sheffield et al. 2006) and therefore provides a temporally consistent series from 1948 through 2014 (Rui and Beaudoing 2019). Taking into account that both GLDAS-1 and -2 have different forcing datasets in terms of type and temporal continuity, the differences in time trends were detailed and discussed in section 3.
Considering that each version of GLDAS datasets uses the same forcing variables, the results of these LSMs are expected to be similar and highly correlated, as their differences arise from their LSM configuration. The Mosaic LSM estimates evaporation through an energy and water balance on the surface layer of the model (Koster and Suarez 1996; Rodell et al. 2007d); meanwhile the Variable Infiltration Capacity (VIC) model uses the Penman–Monteith equation for the potential evaporation as the basis of its estimation, limiting its effects through water balances in every layer of the model (Liang et al. 1994, 1996; Rodell et al. 2007a). The Common Land Model (CLM) uses a modified version of the Biosphere–Atmosphere Transfer Scheme (Dickinson et al. 1993) and the Philip (1957) method to calculate the ground evaporation (Dai et al. 2003; Rodell et al. 2007c). The basis for the evaporation estimation in the Noah model is the Penman equation (Chen et al. 1996; Koren et al. 1999; Rodell et al. 2007b). The Catchment Land Surface (CLS) model (Koster et al. 2000) uses the Mosaic model’s energy balance calculation and canopy interception reservoir formulation.
4) ERA-Interim
This study also used the ERA-Interim (Dee et al. 2011) dataset that represents the synoptic monthly means of evaporation. The revised land surface Hydrology Tiled ECMWF Scheme for Surface Exchanges over Land (HTESSEL) is adopted for the land component that interacts with the atmosphere component. The ET is estimated based on an energy balance using the ERA-Interim forecasts of near-surface conditions (i.e., temperature, humidity, pressure, and wind speed) and downward energy and water fluxes (i.e., precipitation and solar/thermal radiation) (Balsamo et al. 2009; Trambauer et al. 2014).
5) MERRA2
We also used the monthly evaporation land (EVLAND) reanalysis data from the second Modern-Era Retrospective Analysis for Research and Applications (MERRA2) products (Gelaro et al. 2017). MERRA2 is successor of the original MERRA reanalysis (Rienecker et al. 2011), produced by NASA’s Global Modeling and Assimilation Office using an upgraded version of the Goddard Earth Observing System Model, version 5 (GEOS-5), data assimilation system.
6) JRA55
Last, we used monthly evaporation (EVP) reanalysis data from the Japanese 55-year Reanalysis (JRA55) produced by the Japan Meteorological Agency (Kobayashi et al. 2015). JRA55 is an advanced version of its predecessor, JRA25, which considerably mitigated the major biases reported in JRA25 and also improved the temporal consistency of temperature.
b. Methods
1) Seasonal Kendall test
The Mann–Kendall (MK) test (Hirsch et al. 1982; Kendall 1948; Mann 1945) is used to reject the null hypothesis H0 or accept the alternative hypothesis H1 of a two-sided test, which are H0 means that the data (x1, …, xn) are a sample of n independent and identically distributed random variables and H1, means that the distribution of xk and xj are not identical for all k, j with k ≠ j, respectively.
ET global trends by dataset. The trends were estimated using the series of monthly global evaporation on each dataset.
For the two-sided SK test, the null hypothesis H0 is rejected at a significance level α if
2) Data concurrence index
The DCI shows how these datasets behave as a group. By definition, the upper limit (in absolute terms) of the DCI is NS/ND, when all the significant trends found throughout the datasets in a specific location have the same sign. This implies that a difference between the value of NS/ND and the DCI in a specific location must be caused by inconsistencies on the directions of the detected trends. In addition, to investigate how the DCI distribution is temporally varied, a dynamic trend analysis with a 10-yr moving window (120 months) was also carried over all the datasets over the 32-yr study period (e.g., the first 10-yr moving window includes monthly datasets from January 1981 to December 1990).
3) Pearson correlation, CSI, and bias across trends of compared datasets
To deepen the analysis of how the spatially distributed trends differ from each other, the Pearson correlation coefficient, the critical success index (CSI), and bias between the trends of the datasets were calculated.
The Pearson correlation measures how much the datasets look alike in term of magnitudes of estimated trends. These correlation coefficients were calculated only in the common area of the two datasets analyzed, and they consider only the cells with significant trends in both of the compared datasets. For further comparisons, we also added Table S1 in the online supplemental material, presenting the Pearson correlation coefficients for all calculated trends regardless of their statistical significance.
CSI represents the number of significant trends present in the same location with the same sign on both datasets, divided by the total number of significant trends detected on each dataset. Consequently, the CSI measures the similitude of two datasets in terms of trend detection.
3. Results and discussion
a. Consistency of trends in global mean ET
As a more general assessment of trends among the datasets, the monthly global mean ET time series were calculated and analyzed. The results are summarized in Table 2.
During the study period between 1981 and 2012, the trends of the datasets presented a wide range of differences in their values and signs. The two diagnostic datasets, CSIRO and GLEAM v3.3a, presented positive global trends during the period, while the four GLDAS-1 products and MERRA2 showed negative trends whereas VIC presented the most negative value among them. In contrast to the GLDAS-1 datasets, the two GLDAS datasets, CLSv2 and Noahv2, presented positive global trends and confidence intervals that are similar to those of the two diagnostic datasets. The two remaining reanalysis datasets, ERA-Interim and JRA55 displayed insignificant trends. The series of the mean ET across the datasets showed a negative significant trend, clearly influenced by the values of the GLDAS-1 group (Mosaic, VIC, CLM, and Noah).
b. Spatial distribution of concurrent trends
Figure 1 shows the spatial distribution of the results obtained from the DCI analysis and Table 3 presents a summary of the DCI analysis. The first interesting result in Table 3 is that more than 40% of the land surface has a ratio of NS/ND higher than 0.6. This suggests that over a large part of the world, there is a high concurrence among the datasets indicating significant evaporation trends during the studied period. This phenomenon is globally well represented except for Canada/Alaska, northern Eurasia, North Africa, and Western Australia (Fig. 1a). For example, as presented in Fig. 1b (marked with two yellow squares), noteworthy is the presence of two defined zones in the world with high values of DCI, implying a high consistency in both the magnitude and the sign of significant trends. The first zone is around the region joining the United States and Mexico, where the vast majority of the datasets show significant downward trends of the actual evaporation (i.e., negative DCI). The second zone is located in China, where the datasets coincide in significant increasing evaporation trends (i.e., positive DCI). Even where the DCI is not as high as in the two zones described above, a positive or negative value implies some concurrence of significant trends among the datasets. In general terms, as summarized in Table 3, the significant trends found across the datasets globally during the studied period resulted in a positive DCI covering 48.8% of the land area (excluding zero-DCI regions) and negative DCI covered 43.6%. The remaining 9.4% of the continental surface has a DCI of zero, which is larger than the area covered with no significant trends found in the datasets (i.e., NS/ND = 0, 2.1%).
Spatial distributions of (a) NS/ND, (b) DCI, (c) NS/ND minus DCI, and (d) estimated trends of mean monthly ET throughout 11 datasets.
Citation: Journal of Hydrometeorology 22, 1; 10.1175/JHM-D-20-0059.1
Summary of DCI analysis by land area percentages in range classes, where NS = number of significant trends, ND = number of datasets, and DCI = data concurrence index.
Positive DCI values are widely spread around the globe (Fig. 1b), a pattern that is greater and more diverse than what was expected from Miralles et al. (2013). It seems that the datasets were more prone toward positive trends (DCI > 0.5) over these 32 years in eastern Europe, the western Sahara, southern Africa, the Tibetan Plateau, central and eastern China, and northern Eurasia. Meanwhile, zones in North and South America, Central Africa/Arabian Peninsula, and northern China present negative DCI (<−0.5), indicating a general agreement between the datasets on decreasing trends.
High coincidences of significant trends throughout the datasets are scarce, covering only 11.3% of the surface area with NS/ND higher than 0.8 as shown in Table 3. Meanwhile, DCI over 0.6 is present over just 10.5% of the land area, mostly concentrated in the areas described above. On the other hand, the two lowest quintiles in Table 3 cover most of the world for DCI, with a value of 73.7%. This concentration of high DCI scores contrasted with the widespread low scores implying that some phenomena affecting the trends in these places are being modeled consistently by all the datasets, but for most of the world, the main drivers behind the evaporation trends are not congruently represented. The reduction of the area covered by the highest quintile of DCI and the increase of the two lowest quintiles is caused by inconsistent significant trends that lower the score. This problem, while present all over the globe, is more acute in some areas [NS/ND − abs(DCI) > 0.5], as presented in Fig. 1c. The northeastern United States, central Amazonia, Central Africa, and Central and East Asia are the most affected zones with contradictory trends throughout the datasets. The selection of a specific dataset to work in these zones should be done carefully considering that the results may be substantially sensitive to the selected dataset.
In the zones with high (negative or positive) DCI scores, the selection of a specific dataset is less problematic since the trends are consistent. However, as shown in Fig. 2, differences appear when the magnitudes are compared. The box plots in Fig. 2a and 2b present of magnitudes of trends over two 10° × 10° tiles showing strong negative and positive DCI scores, respectively. As marked with two squares in Fig. 1b, the first tile covers grid cells with strong negative DCI scores located at North America (30°–40°N, 100°–110°W), and the second tile is for positive DCI scores at China (25°–35°N, 105°–115°E).
Boxplots of estimated trends across datasets in (a) a decreasing trend zone (North America; 30°–40°N, 100°–110°W) and (b) an increasing trend zone (China; 25°–35°N, 105°–115°E), the locations of which are marked as yellow-outlined squares in Fig. 1b.
Citation: Journal of Hydrometeorology 22, 1; 10.1175/JHM-D-20-0059.1
For the decreasing trends in Fig. 2a, all datasets generally show negative trend values but with different distributions of trend magnitudes. It is noticeable that the four GLDAS-1 datasets show strong negative values with similar interquartile ranges to each other, the remaining datasets present different distributions of trend magnitudes, but the strong negative trends considerably ease in the two GLDAS-2 datasets by showing similar distribution shapes with the two diagnostic datasets, CSIRO and GLEAM. The three reanalysis datasets show different shapes of distributions. Especially in the case of JRA55, it produces greater and more disperses trends than the others, followed by ERA-Interim and MERRA2. On contrary to the decreasing trend zone, as shown in in Fig. 2b, the datasets produce considerably different patterns from each other over the increasing trend zone although they generally show positive trend values. In general terms, the diagnostic and reanalysis datasets have more conservative and less dispersed significant trends than the GLDAS-1 and -2 datasets. Especially GLDAS-1 produces greater and more dispersed trends than GLDAS-2. These results suggest that even when the directions are consistent, the selection of a dataset can still have a great impact on the results of a study conducted in the area.
The spatial distribution of the trends in the mean ET throughout the datasets (Fig. 1d) shows some resemblance to the DCI: zones with positive DCI generally fit zones with positive trends and negative DCI generally fit zones with negative trends, in the mean series. The midlatitude area, 30°N–30°S, generally presents negative trends except for arid zones like the Sahara and the Arabian Peninsula, whereas the high-latitude area presents positive trends. The steepest ET trends are observed over southern Africa/Tibetan Plateau with the positives (>0.3 mm month−1 yr−1) and Central and West Africa/Southeast Asia with the negatives (<−0.7 mm month−1 yr−1). Regions such as the middle of the Americas, East Africa, and eastern Australia also present strong negative trends, even when none of these places show high concurrence between the datasets (DCI > 80%). In fact, the zones with high DCI scores do not show particularly great trends in the mean series. Therefore, high trends in the ET mean series do not necessarily represent high agreement among the datasets in terms of trends direction or detection and vice versa.
c. Correlation, CSI, and bias among datasets
As described in section 2b(3), we calculated the Pearson correlation, CSI, and bias for the datasets in Tables 4–6, respectively. The Pearson correlation between the trends of the datasets is generally low except for the GLDAS datasets that belong to the same data version. The main reasons behind these high correlations are the shared forcing variables and methods used. It is interesting to note the moderate correlations between JRA55 and ERA-Interim (i.e., 0.439). The rest of the correlations are low, implying that there is no substantial relationship between the location of the trends and their value across the datasets. Table S1 in the online supplemental material shows the Pearson correlation for all trends regardless of their statistical significance. Here, the pattern of the values in Table S1 is similar to that of Table 4, but their average value is about 0.05 lesser than that of Table 4.
Comparison of Pearson correlation coefficients between significant trends of the datasets over their common areas, where CS is CSIRO, GL is GLEAM v3.3a, MOS is GLDAS Mosaic, VIC is GLDAS VIC, CLM is GLDAS CLM, NOAH is GLDAS Noah, CLv2 is GLDAS CLSv2, NOv2 is GLDAS Noahv2, EI is ERA-Interim, ME is MERRA2, and JRA is JRA55.
Comparison of CSI (%) between significant trends of the datasets over their common areas. Higher values of CSI imply similar higher similitudes in the number of the significant trends, their location, and their direction.
Comparison of significant trends with focus on bias (%) between datasets. Biases lower than 100% indicate that the datasets in the columns present less significant trends than the ones in the rows, whereas values greater than 100% imply the contrary.
Even if the Pearson correlation gives a good idea about how much the significant trends on two datasets are alike in terms of magnitude, it does not clearly reflect that trends with different signs separated by a constant value represent opposite phenomena. This aspect is better represented using the CSI and bias results in Tables 5 and 6. Note, however, that these results of CSI and bias are influenced by their various spatial resolutions and the resampling method that we applied. Given what the limitations are, an indication of the similarities and differences across the datasets on a regional basis is provided in the results as follows.
Looking at the bias in Table 6, the datasets that generally produced more significant trends (i.e., mean bias > 100) are GLEAM, GLDAS-1 except for CLM, GLDAS-2 CLS, and ERA-Interim. The biases of the trends for CSIRO, GLDAS-1 CLM, and MERRA2 datasets were relatively lower (i.e., mean bias < 100). Especially, the trends of GLDAS-1 MOSAIC and MERRA2 are contrasted with their notably high and low bias values, respectively. Last, GLDAS-2 NOAH and JRA55, in particular, were less biased among their pairs (i.e., mean bias ≈ 100). The bias also echoes on the CSI scores (Table 5), lowering it in those comparisons when the bias was more pronounced. This is particularly visible in the scores obtained between the GLDAS datasets in each version, where the number of significant trends between the datasets is the most relevant reason behind the CSI scores. In other cases, the main driver is the difference in the directions of the trends.
d. Variation of trends with the time period considered
Figure 3a presents the distribution in time of DCI over the land surface, showing that high concurrence (DCI > 0.8 or DCI < −0.8) among the datasets is rare throughout the entire available period. The major changes in coverage happen in the range between 0.2 and 0.6, for both positive and negative DCI. Probably this is an effect of the correlation between the GLDAS datasets, changing at the same time in relative discordance with the remaining datasets. Also interesting is that the global DCI is getting negative by 1997 but the pattern rebounds since that time, positive DCI increases till the end of the period.
Changes of DCI in time using a dynamic trend analysis with a 10-yr moving window. (a) Time series of mean precipitation (mean P; mm month−1; blue) and coefficient of variation of precipitation (CVP; red) in the top panel and the land covered by percentile of DCI, differentiating positive (browns) and negative (blues) values of DCI in the bottom panel. The x axis represents the end year of each 10-yr period. The land covered with a DCI of zero was not considered. Here, periods 1 and 2 were selected for further investigations to be compared with Jung et al. (2010). (b) The number of times DCI changed sign between 1990 and 2012.
Citation: Journal of Hydrometeorology 22, 1; 10.1175/JHM-D-20-0059.1
The general shape of the series in Fig. 3a shows a very dynamic DCI distribution. These changes in distribution make the selection of the time period critical to the analysis of evaporation trends. To illustrate where these variations are more relevant, the total number of changes of sign on each cell was calculated and presented in Fig. 3b. In there, it can be seen that certain points of the world are more prone to present important variations in the direction of the trends, as southern Africa, India, North America, and Europe. Studies of evaporation trends in these regions should be particularly cautious regarding the studied time period. We further discuss this point in the next section.
e. How to interpret and use DCI
For investigating the dynamics of DCI, we noted the refraction of the ET trend between two periods: 1982–97 and 1998–2008 (Jung et al. 2010). The change in DCI toward both positive and negative directions in the period of 1998–2007 (period 2) in Fig. 3a seems to correspond to the cessation in the ET increase from 1998 to 2008 as reported in Jung et al. (2010). However, the strong negative DCI in the period of 1988–97 (period 1) is contradictory to the increasing global ET trend from 1982 to 1997. For further examining this contradiction and similarity, we investigated the distributions of ET trends over three 10° × 10° tiles as presented in Fig. 4. Those represent Russia (60°–70°N, 60°–70°E), South Africa (17°–27°S, 10°–20°W), and Western Australia (20°–30°S, 120°–130°E) showing contrasting changes in the ET trends, also highlighted by Jung et al. (2010).
Over two 10-yr periods, (left) 1998–97 (period 1) and (right) 1998–2007 (period 2), (a),(e) the spatial distributions of DCI and boxplots of estimated trends across datasets for (b),(f) Russia (60°–70°N, 60°–70°E); (c),(g) South Africa (17°–27°S, 10°–20°W); and (d),(h) Western Australia (20°–30°S, 120°–130°E), as marked with the three magenta-outlined squares in (a) and (e).
Citation: Journal of Hydrometeorology 22, 1; 10.1175/JHM-D-20-0059.1
The negative tendency of DCI in period 1 is well depicted in space as shown in Fig. 4a, where the world shows a negative DCI in general except for some regions such as the northwestern United States, the Arabian Peninsula, and Western Australia. This overall negative DCI is contradicting the increasing ET trend over the period from 1982 to 1997 found in Jung et al. (2010) whereas the spatial pattern of DCI over period 2 (Fig. 4d) is relatively well matched to that of Jung et al. (2010) over the period from 1998 to 2008. When it comes to the DCI distributions over the three regions, the contradiction in period 1 is well presented. Jung et al. (2010) reported that the trend change in Russia is not considerable. However, as shown in Figs. 4b and 4f, the GLDAS, especially GLDAS-1, datasets present a clear trend reversal from negative to positive while the diagnostic and reanalysis datasets are holding small trend changes. On the other hand, the reported trend change over South Africa (from negative to positive) and Western Australia (from positive to negative) is clearly reflected through all datasets as presented in Figs. 4c and 4g and Figs. 4d and 4h.
As a possible reason for the discrepancy in the ET trend, we noticed that the temporal pattern of DCI (the bottom panel of Fig. 3a) is well correlated with the spatial variation of precipitation (in the right y axis of the top panel of Fig. 3a) along with the moving windows. For this, we used Global Precipitation Climatology Project, version 2.3, monthly precipitation data (Pendergrass et al. 2020) over the study period, from January 1981 to December 2012. Here, the spatial variation of precipitation is represented by the coefficient of variation CVP of temporally averaged monthly precipitation data in each 10-yr period. In addition to this, Fig. 5 shows temporal changes of the ET trend–precipitation correlation in space (rTrend−meanP) for the 11 ET datasets. Here, for simplicity, the 11 datasets were aggregated into four groups: diagnostic (2 members), GLDASv1 (4 members), GLDASv2 (2 members), and Reanalysis (3 members), and then the temporal trend lines and confidence intervals were simply represented by medians and standard errors, respectively, at each time step.
Temporal changes of spatial correlation between ET trends of the 11 datasets and mean precipitation (rTrend−meanP) over regions with (a) weak and (b) strong DCI scores. Here, strong DCI = abs(DCI) > 0.7 and weak DCI = abs(DCI) < 0.3. Also shown are (b),(d) the cross correlations between time series for (a) and (c), respectively.
Citation: Journal of Hydrometeorology 22, 1; 10.1175/JHM-D-20-0059.1
As presented in Fig. S1 in the supplemental material, these changes are separately plotted for two regions with weak [abs(DCI) < 0.3] and strong [abs(DCI) > 0.7] DCI scores referred from Fig. 1b. Over the weak DCI regions (Figs. 5a,b), interestingly, strong inverse behaviors of rTrend−meanP for the four GLDAS-1 datasets relative to other datasets are observed, especially for the CSIRO, GLDAS-2, and JRA55 datasets. This tendency exhibits the greatest contrast during period 1 (1988–97), which showed the ET trend contradiction in Fig. 4a and tends to be somewhat eased in period 2 (1998–2007). In case of the strong DCI regions (Figs. 5c,d), these inverse behaviors reverse or subside over the strong DCI regions in both periods (Figs. 5c,d). Interestingly, the high DCI tends to appear over the supply limited regions identified in Jung et al. (2010) where ET is more positively correlated with precipitation than temperature, but the low DCI is likely to be distributed over the demand limited regions in which ET is more positively correlated with temperature than precipitation. Given that the tendency of the four GLDAS-1 datasets seems an important reason for the contradiction noted in period 1, it can be inferred that the performance of GLDAS-1 over the demand limited regions tends to be sensitively affected.
Consistent with this, DCI locally agrees with the real phenomena or not depending on the spatiotemporal behaviors of the datasets considered. Therefore, note that DCI is not to be used for indicating how the datasets are close to the truth but rather for how the datasets are different from each other, especially when the number of datasets used is not statistically sufficient.
4. Summary and conclusions
The present work analyzed significant trends in monthly evaporation from 11 global datasets, during their common period, between 1981 and 2012. In this period, the global ET trends of the mean throughout the datasets decreased with a pace of −0.072 mm month−1 yr−1, and the datasets showed a wide range of differences in magnitudes and directions their global trends. While CSIRO and GLEAM v3.3a showed positive trends during the period, the four GLDAS products and MERRA2 presented negative trends and ERA-Interim and JRA55 displayed insignificant trends. The scales of the global trends throughout the datasets were considerably different. The trends found in the GLDAS group were substantially greater than the rest, while the remaining datasets presented smaller magnitudes.
Only 2.1% of the world presented a high coincidence of no significant trends. For around half of the world, there was a general disagreement among the trends of the datasets in the presence and location of significant trends, represented by low to moderate DCI scores (0.2–0.6). The datasets tend to be positive trends (DCI > 0.5) over eastern Europe, the western Sahara, southern Africa, the Tibetan Plateau, China, and northern Eurasia. Over North and South America, Central Africa/Arabian Peninsula, and northern China, negative trends, where DCI <−0.5, were observed. Also, it was found that the magnitudes of the trends on the series composed by the mean ET across the datasets, were not representative of the agreement between them in term of direction and detection of trends. The significant trends between the datasets were weakly correlated, except for those belonging to the same group, as GLDAS. However, even when highly correlated, their general behavior was considerably different, as the low CSI scores demonstrate. The main reasons behind these differences are the tendency to present more or less significant trends (bias) and the presence of contradictory trends in a specific location. To study the similarities of the trends in the datasets through time, a dynamic trend analysis was performed, using a 10-yr window moving over the datasets. The results show a very dynamic system, with important changes over time. However, high concurrences are rare in every point in time, and the most significant variations can be attributed to the correlated GLDAS datasets, changing simultaneously in relative discrepancy with the rest. Some places like southern Africa, India, North America, and Europe are more prone to revert the direction of the significant trends in many datasets with time.
The results of this work show how different approaches to estimate the evaporation can lead to important inconsistencies in the location and direction of its trends. These inconsistencies are a big concern for climate change studies that require accuracy to evaluate models or estimate actual and future impacts. Therefore, it is important to address these discrepancies and generate more reliable global datasets with validated trends, to improve the performance of our models and provide better information for decision makers. Studies in regions with highly inconsistent trends should be especially careful when selecting an evaporation dataset to work with and be aware that its results may be greatly sensitive to the dataset selected. The same can be said of the places with tendency to revert the direction of their trends. In these locations, the selected time period could also change the results of a trend study drastically. No single dataset can be used as a general representation of the rest to analyze trends in evaporation. Even within the groups of datasets that use similar methods to estimate ET, the differences are big enough to recommend the use of several datasets instead of just one as a representative choice.
Acknowledgments
This work has been supported with a Linkage Project (LP160100620) funded by the Australian Research Council, WaterNSW, and Sydney Water. The authors of this work express their gratitude to the authors of the datasets used here for providing such important information.
REFERENCES
Anabalón, A., and A. Sharma, 2017: On the divergence of potential and actual evapotranspiration trends: An assessment across alternate global datasets. Earth’s Future, 5, 905–917, https://doi.org/10.1002/2016EF000499.
Baldocchi, D., and Coauthors, 2001: FLUXNET: A new tool to study the temporal and spatial variability of ecosystem-scale carbon dioxide, water vapor, and energy flux densities. Bull. Amer. Meteor. Soc., 82, 2415–2434, https://doi.org/10.1175/1520-0477(2001)082<2415:FANTTS>2.3.CO;2.
Balsamo, G., A. Beljaars, K. Scipal, P. Viterbo, B. Hurk, M. Hirschi, and A. K. Betts, 2009: A revised hydrology for the ECMWF model: Verification from field site to terrestrial water storage and impact in the integrated forecast system. J. Hydrometeor., 10, 623–643, https://doi.org/10.1175/2008JHM1068.1.
Brodzik, M. J., B. Billingsley, T. Haran, B. Raup, and M. H. Savoie, 2012: EASE-Grid 2.0: Incremental but significant improvements for Earth-gridded data sets. ISPRS Int. J. Geoinf., 1, 32–45, https://doi.org/10.3390/ijgi1010032.
Chen, F., and Coauthors, 1996: Modeling of land surface evaporation by four schemes and comparison with FIFE observations. J. Geophys. Res., 101, 7251–7268, https://doi.org/10.1029/95JD02165.
Chen, Y., and Coauthors, 2014: Comparison of satellite-based evapotranspiration models over terrestrial ecosystems in China. Remote Sens. Environ., 140, 279–293, https://doi.org/10.1016/j.rse.2013.08.045.
Dai, Y., and Coauthors, 2003: The Common Land Model. Bull. Amer. Meteor. Soc., 84, 1013–1024, https://doi.org/10.1175/BAMS-84-8-1013.
Dee, D. P., and Coauthors, 2011: The ERA-Interim reanalysis: Configuration and performance of the data assimilation system. Quart. J. Roy. Meteor. Soc., 137, 553–597, https://doi.org/10.1002/qj.828.
Dickinson, E., A. Henderson-Sellers, and J. Kennedy, 1993: Biosphere–Atmosphere Transfer Scheme (BATS) Version 1e as coupled to the NCAR Community Climate Model. NCAR Tech. Note NCAR/TN-387+STR, 88 pp., https://doi.org/10.5065/D67W6959.
Donaldson, R., R. M. Dyer, and M. J. Kraus, 1975: An objective evaluator of techniques for predicting severe weather events. Preprints, Ninth Conf on Severe Local Storms, Norman, OK, Amer. Meteor. Soc., 321–326.
Ershadi, A., M. McCabe, J. P. Evans, N. W. Chaney, and E. F. Wood, 2014: Multi-site evaluation of terrestrial evaporation models using FLUXNET data. Agric. For. Meteor., 187, 46–61, https://doi.org/10.1016/j.agrformet.2013.11.008.
Fisher, J. B., K. P. Tu, and D. D. Baldocchi, 2008: Global estimates of the land–atmosphere water flux based on monthly AVHRR and ISLSCP-II data, validated at 16 FLUXNET sites. Remote Sens. Environ., 112, 901–919, https://doi.org/10.1016/j.rse.2007.06.025.
Gelaro, R., and Coauthors, 2017: The Modern-Era Retrospective Analysis for Research and Applications, version 2 (MERRA-2). J. Climate, 30, 5419–5454, https://doi.org/10.1175/JCLI-D-16-0758.1.
Helsel, D. R., and R. M. Hirsch, 1992: Statistical Methods in Water Resources. Studies in Environmental Science, Vol. 49, Elsevier, 546 pp.
Hettiarachchi, S., C. Wasko, and A. Sharma, 2019: Can antecedent moisture conditions modulate the increase in flood risk due to climate change in urban catchments? J. Hydrol., 571, 11–20, https://doi.org/10.1016/j.jhydrol.2019.01.039.
Hirsch, R. M., J. R. Slack, and R. A. Smith, 1982: Techniques of trend analysis for monthly water quality data. Water Resour. Res., 18, 107–121, https://doi.org/10.1029/WR018i001p00107.
Hogan, R. J., and I. B. Mason, 2011: Deterministic forecasts of binary events. Forecast Verification, 2nd ed. I. T. Jolliffe and D. B. Stephenson, Eds., John Wiley and Sons, 31–59.
Huizhi, L., and F. Jianwu, 2012: Seasonal and interannual variations of evapotranspiration and energy exchange over different land surfaces in a semiarid area of China. J. Appl. Meteor. Climatol., 51, 1875–1888, https://doi.org/10.1175/JAMC-D-11-0229.1.
IPCC, 2014: Climate Change 2013: The Physical Science Basis, Cambridge University Press, 1535 pp.
Jiménez, C., and Coauthors, 2011: Global intercomparison of 12 land surface heat flux estimates. J. Geophys. Res., 116, D02102, https://doi.org/10.1029/2010JD014545.
Johnson, F., and A. Sharma, 2010: A comparison of Australian open water body evaporation trends for current and future climates estimated from class A evaporation pans and general circulation models. J. Hydrometeor., 11, 105–121, https://doi.org/10.1175/2009JHM1158.1.
Jung, M., and Coauthors, 2010: Recent decline in the global land evapotranspiration trend due to limited moisture supply. Nature, 467, 951–954, https://doi.org/10.1038/nature09396.
Kendall, M. G., 1948: Rank Correlation Methods. Charles Griffin, 160 pp.
Kim, S., and A. Sharma, 2019: The role of floodplain topography in deriving basin discharge using passive microwave remote sensing. Water Resour. Res., 55, 1707–1716, https://doi.org/10.1029/2018WR023627.
Kiptala, J. K., Y. Mohamed, M. L. Mul, and P. Van der Zaag, 2013: Mapping evapotranspiration trends using MODIS and SEBAL model in a data scarce and heterogeneous landscape in Eastern Africa. Water Resour. Res., 49, 8495–8510, https://doi.org/10.1002/2013WR014240.
Kobayashi, S., and Coauthors, 2015: The JRA-55 reanalysis: General specifications and basic characteristics. J. Meteor. Soc. Japan, 93, 5–48, https://doi.org/10.2151/jmsj.2015-001.
Koren, V., J. Schaake, K. Mitchell, Q.-Y. Duan, F. Chen, and J. M. Baker, 1999: A parameterization of snowpack and frozen ground intended for NCEP weather and climate models. J. Geophys. Res., 104, 19 569–19 585, https://doi.org/10.1029/1999JD900232.
Koster, R. D., and M. J. Suarez, 1996: Energy and water balance calculations in the Mosaic LSM. NASA Tech. Memo. 104606, Vol. 9, 60 pp., http://gmao.gsfc.nasa.gov/pubs/docs/Koster130.pdf.
Koster, R. D., M. J. Suarez, A. Ducharne, M. Stieglitz, and P. Kumar, 2000: A catchment-based approach to modeling land surface processes in a general circulation model: 1. Model structure. J. Geophys. Res., 105, 24 809–24 822, https://doi.org/10.1029/2000JD900327.
Koster, R. D., and Coauthors, 2004: Regions of strong coupling between soil moisture and precipitation. Science, 305, 1138–1140, https://doi.org/10.1126/science.1100217.
Liang, X., D. P. Lettenmaier, E. F. Wood, and S. J. Burges, 1994: A simple hydrologically based model of land surface water and energy fluxes for general circulation models. J. Geophys. Res., 99, 14 415–14 428, https://doi.org/10.1029/94JD00483.
Liang, X., D. P. Lettenmaier, and E. F. Wood, 1996: One-dimensional statistical dynamic representation of subgrid spatial variability of precipitation in the two-layer variable infiltration capacity model. J. Geophys. Res., 101, 21 403–21 422, https://doi.org/10.1029/96JD01448.
Libertino, A., A. Sharma, V. Lakshmi, and P. Claps, 2016: A global assessment of the timing of extreme rainfall from TRMM and GPM for improving hydrologic design. Environ. Res. Lett., 11, 054003, https://doi.org/10.1088/1748-9326/11/5/054003.
Mann, H. B., 1945: Nonparametric tests against trend. Econometrica, 13, 245–259, https://doi.org/10.2307/1907187.
Marshall, M., C. Funk, and J. Michaelsen, 2012: Examining evapotranspiration trends in Africa. Climate Dyn., 38, 1849–1865, https://doi.org/10.1007/s00382-012-1299-y.
Martens, B., and Coauthors, 2017: GLEAM v3: Satellite-based land evaporation and root-zone soil moisture. Geosci. Model Dev., 10, 1903–1925, https://doi.org/10.5194/gmd-10-1903-2017.
McMahon, T. A., M. C. Peel, L. Lowe, R. Srikanthan, and T. R. McVicar, 2013: Estimating actual, potential, reference crop and pan evaporation using standard meteorological data: A pragmatic synthesis. Hydrol. Earth Syst. Sci., 17, 1331–1363, https://doi.org/10.5194/hess-17-1331-2013.
Miralles, D. G., T. R. H. Holmes, R. A. M. De Jeu, J. H. Gash, A. G. C. A. Meesters, and A. J. Dolman, 2011: Global land-surface evaporation estimated from satellite-based observations. Hydrol. Earth Syst. Sci., 15, 453–469, https://doi.org/10.5194/hess-15-453-2011.
Miralles, D. G., and Coauthors, 2013: El Niño–La Niña cycle and recent trends in continental evaporation. Nat. Climate Change, 4, 122–126, https://doi.org/10.1038/nclimate2068.
Miralles, D. G., and Coauthors, 2016: The WACMOS-ET project – Part II: Evaluation of global terrestrial evaporation data sets. Hydrol. Earth Syst. Sci., 20, 823–842, https://doi.org/10.5194/hess-20-823-2016.
Mu, Q., F. A. Heinsch, M. Zhao, and S. W. Running, 2007: Development of a global evapotranspiration algorithm based on MODIS and global meteorology data. Remote Sens. Environ., 111, 519–536, https://doi.org/10.1016/j.rse.2007.04.015.
Mueller, B., and Coauthors, 2011: Evaluation of global observations-based evapotranspiration datasets and IPCC AR4 simulations. \Geophys. Res. Lett., 38, L06402, https://doi.org/10.1029/2010GL046230.
Mueller, B., and Coauthors, 2013: Benchmark products for land evapotranspiration: LandFlux-EVAL multi-data set synthesis. Hydrol. Earth Syst. Sci., 17, 3707–3720, https://doi.org/10.5194/hess-17-3707-2013.
Pendergrass, A., J.-J. Wang, and NCAR Staff, Eds., 2020: The Climate Data Guide: GPCP (Monthly): Global Precipitation Climatology Project v2.3. UCAR, accessed 5 June 2020, https://climatedataguide.ucar.edu/climate-data/gpcp-monthly-global-precipitation-climatology-project.
Philip, J. R., 1957: Evaporation, and moisture and heat fields in the soil. J. Meteor., 14, 354–366, https://doi.org/10.1175/1520-0469(1957)014<0354:EAMAHF>2.0.CO;2.
Rienecker, M. M., and Coauthors, 2011: MERRA: NASA’s Modern-Era Retrospective Analysis for Research and Applications. J. Climate, 24, 3624–3648, https://doi.org/10.1175/JCLI-D-11-00015.1.
Rodell, M., and Coauthors, 2004: The Global Land Data Assimilation System. Bull. Amer. Meteor. Soc., 85, 381–394, https://doi.org/10.1175/BAMS-85-3-381.
Rodell, M., H. K. Beaudoing, and NASA/GSFC/HSL, 2007a: GLDAS VIC Land Surface Model L4 Monthly 1.0 × 1.0 degree V001. GES DISC, accessed 7 June 2020, https://data.nasa.gov/Earth-Science/GLDAS-VIC-Land-Surface-Model-L4-Monthly-1-0-x-1-0-/tg9t-zgur.
Rodell, M., H. K. Beaudoing, and NASA/GSFC/HSL, 2007b: GLDAS Noah Land Surface Model L4 Monthly 1.0 × 1.0 degree V001. GES DISC, accessed 7 June 2020, https://data.nasa.gov/Earth-Science/GLDAS-Noah-Land-Surface-Model-L4-Monthly-1-0-x-1-0/fxur-u3e7/data?pane=feed.
Rodell, M., H. K. Beaudoing, and NASA/GSFC/HSL, 2007c: GLDAS CLM Land Surface Model L4 Monthly 1.0 × 1.0 degree V001. GES DISC, accessed 7 June 2020, https://data.nasa.gov/Earth-Science/GLDAS-CLM-Land-Surface-Model-L4-Monthly-1-0-x-1-0-/8xxw-6ya2.
Rodell, M., H. K. Beaudoing, and NASA/GSFC/HSL, 2007d: GLDAS Mosaic Land Surface Model L4 Monthly 1.0 × 1.0 degree V001. GES DISC, accessed 7 June 2020, https://data.nasa.gov/Earth-Science/GLDAS-Mosaic-Land-Surface-Model-L4-Monthly-1-0-x-1/392n-bnvz.
Rui, H., and H. Beaudoing, 2019: README document for NASA GLDAS version 2 data products. GES DISC Doc., 32 pp., https://hydro1.gesdisc.eosdis.nasa.gov/data/GLDAS/README_GLDAS2.pdf.
Sen, P. K., 1968: Estimates of the regression coefficient based on Kendall’s Tau. J. Amer. Stat. Assoc., 63, 1379–1389, https://doi.org/10.1080/01621459.1968.10480934.
Seneviratne, S. I., and Coauthors, 2010: Investigating soil moisture–climate interactions in a changing climate: A review. Earth-Sci. Rev., 99, 125–161, https://doi.org/10.1016/j.earscirev.2010.02.004.
Sheffield, J., G. Goteti, and E. F. Wood, 2006: Development of a 50-year high-resolution global dataset of meteorological forcings for land surface modeling. J. Climate, 19, 3088–3111, https://doi.org/10.1175/JCLI3790.1.
Stephens, C. M., T. R. McVicar, F. M. Johnson, and L. A. Marshall, 2018: Revisiting pan evaporation trends in Australia a decade on. Geophys. Res. Lett., 45, 11 164–111 172, https://doi.org/10.1029/2018GL079332.
Theil, H., 1992: A rank-invariant method of linear and polynomial regression analysis. Henri Theil’s Contributions to Economics and Econometrics: Econometric Theory and Methodology, B. Raj and J. Koerts, Eds., Springer, 345–381.
Trambauer, P., E. Dutra, S. Maskey, M. Werner, F. Pappenberger, L. P. H. van Beek, and S. Uhlenbrook, 2014: Comparison of different evaporation estimates over the African continent. Hydrol. Earth Syst. Sci., 18, 193–212, https://doi.org/10.5194/hess-18-193-2014.
Trenberth, K. E., L. Smith, T. Qian, A. Dai, and J. Fasullo, 2007: Estimates of the global water budget and its annual cycle using observational and model data. J. Hydrometeor., 8, 758–769, https://doi.org/10.1175/JHM600.1.
Trenberth, K. E., J. T. Fasullo, and J. Kiehl, 2009: Earth’s global energy budget. Bull. Amer. Meteor. Soc., 90, 311–324, https://doi.org/10.1175/2008BAMS2634.1.
Vinukollu, R. K., R. Meynadier, J. Sheffield, and E. F. Wood, 2011: Multi-model, multi-sensor estimates of global evapotranspiration: Climatology, uncertainties and trends. Hydrol. Processes, 25, 3993–4010, https://doi.org/10.1002/hyp.8393.
Vourlitis, G. L., N. P. Filho, M. M. S. Hayashi, J. de S. Nogueira, F. T. Caseiro, and J. H. Campelo Jr., 2002: Seasonal variations in the evapotranspiration of a transitional tropical forest of Mato Grosso, Brazil. Water Resour. Res., 38, 1094, https://doi.org/10.1029/2000WR000122.
Woldemeskel, F., and A. Sharma, 2016: Should flood regimes change in a warming climate? The role of antecedent moisture conditions. Geophys. Res. Lett., 43, 7556–7563, https://doi.org/10.1002/2016GL069448.
Yang, Y., D. Long, and S. Shang, 2013: Remote estimation of terrestrial evapotranspiration without using meteorological data. Geophys. Res. Lett., 40, 3026–3030, https://doi.org/10.1002/grl.50450.
Zhang, K., J. S. Kimball, and S. W. Running, 2016: A review of remote sensing based actual evapotranspiration estimation. Wiley Interdiscip. Rev.: Water, 3, 834–853, https://doi.org/10.1002/wat2.1168.
Zhang, R., S. Kim, and A. Sharma, 2019: A comprehensive validation of the SMAP Enhanced Level-3 Soil Moisture product using ground measurements over varied climates and landscapes. Remote Sens. Environ., 223, 82–94, https://doi.org/10.1016/j.rse.2019.01.015.
Zhang, Y., R. Leuning, L. B. Hutley, J. Beringer, I. McHugh, and J. P. Walker, 2010: Using long-term water balances to parameterize surface conductances and calculate evaporation at 0.05° spatial resolution. Water Resour. Res., 46, W05512, https://doi.org/10.1029/2009WR008716.
Zhang, Y., J. Pena Arancibia, T. McVicar, F. Chiew, J. Vaze, H. Zheng, and Y. P. Wang, 2016a: Monthly global observation-driven Penman-Monteith-Leuning (PML) evapotranspiration and components, v2. CSIRO, accessed 5 June 2020, https://doi.org/10.4225/08/5719A5C48DB85.
Zhang, Y., and Coauthors, 2016b: Multi-decadal trends in global terrestrial evapotranspiration and its components. Sci. Rep., 6, 19124, https://doi.org/10.1038/srep19124.
