A key Earth system process is the circulation of evaporated
moisture through the atmosphere. Spatial connections between evaporation and
precipitation affect the global and regional climates by redistributing
water and latent heat. Through this atmospheric moisture recycling,
land cover changes influence regional precipitation patterns, with
potentially far-reaching effects on human livelihoods and biome
distributions across the globe. However, a globally complete dataset of
atmospheric moisture flows from evaporation to precipitation has been
lacking so far. Here we present a dataset of global atmospheric moisture
recycling on both 0.5
Atmospheric moisture flows from evaporation to precipitation are a fundamental component of the hydrological cycle on Earth (Brubaker et al., 1993; Savenije, 1996; Van der Ent et al., 2010). It is estimated that 36 % of global rainfall over land evaporated from land (Van der Ent et al., 2014) and that most evaporation is transported for several hundreds to thousands of kilometers before precipitating (Dirmeyer and Brubaker, 2007; Van der Ent and Savenije, 2011; Wei and Dirmeyer, 2019), although the regional orography can influence this significantly (Tuinenburg et al., 2012). This means that land cover changes that affect evaporation have regional consequences. Indeed, it is increasingly recognized that anthropogenic land use changes affect atmospheric moisture flows and our dependencies on them (Keys et al., 2016, 2019; Wang-Erlandsson et al., 2018; Staal et al., 2020). However, the remote effects of evaporation changes on precipitation patterns are not fully understood, partly due to limited systematic documentation of atmospheric connections.
Often, studies on the atmospheric component of Earth's hydrological cycle
use moisture tracking models based on atmospheric reanalysis data to
simulate the flows of moisture (Dirmeyer et al., 2009;
Gimeno et al., 2010; Van der Ent and Savenije, 2011; Tuinenburg et al.,
2012). The recent development of the ERA5 reanalysis data
(Copernicus Climate Change Service, C3S) allows for
unprecedented high-resolution estimates of moisture flows between any pair
of areas across the globe. However, the required data input for high-quality
moisture flow simulations with ERA5 can be high (Tuinenburg and
Staal, 2020), which may preclude wide application of these simulations. A
readily usable and comprehensive global-scale dataset of pairwise links of
atmospheric moisture connections could satisfy much of the demand, but it
has been lacking until now. Here we present such a global-scale dataset of
moisture flows between all pairs of land and ocean surface at 0.5
The dataset was generated with “UTrack-atmospheric-moisture”, hereafter “UTrack”, which is a Lagrangian (trajectory-based) moisture tracking model developed by Tuinenburg and Staal (2020). In the simulations, large numbers of moisture particles were released at random locations and heights within each cell, after which their locations were tracked based on the reanalysis data for wind speed and direction. Thus, whereas the presented output is grid-based, the simulations behind them are not. The UTrack model behind this dataset resulted from an elaborate sensitivity analysis to test the effects of a range of assumptions and uncertainties on the accuracy of moisture tracking simulations and hydrologically relevant statistics (Tuinenburg and Staal, 2020).
Our dataset is the first to present global moisture flows across the globe, including both land and ocean. Although the simulations were carried out using “forward tracking” from evaporation to precipitation, it implicitly includes the backward trajectories of moisture flows. Because our dataset offers a complete picture of moisture flows across the globe including the oceans, it can be used to construct both forward footprints and “evaporation sheds”, the downwind areas of precipitation receiving evaporation from regions of interest (Van der Ent and Savenije, 2013), as well as backward footprints or “precipitation sheds”, the upwind areas of evaporation contributing to precipitation in regions of interest (Keys et al., 2012). Furthermore, it is the first to use the new ERA5 atmospheric reanalysis data and the most detailed to date.
This paper is structured as follows. In Sect. 2 we outline the methods behind the data, where we summarize the UTrack model in Sect. 2.1, detail the global simulations and the format of their output in Sect. 2.2 and explain our data validation procedure in Sect. 2.3. In Sect. 3 we present a number of results regarding the global synthesis of atmospheric moisture flows. First, in Sect. 3.1 we show a number of metrics obtained from the data: evaporation moisture recycling (land and basin recycling) and average traveled distance (longitudinal and latitudinal direction). Next, in Sect. 3.2 we present validating tests of the dataset: sum of precipitation originated from any grid cell, which should add up to 1, and comparison of total precipitation of UTrack and ERA5. We briefly relate this dataset to existing work and discuss its potential applicability in Sect. 4. Section 5 presents the conclusions and Sect. 6 gives the instructions to download the data and the scripts to process them.
The UTrack atmospheric moisture tracking model.
The dataset is generated using the Lagrangian atmospheric moisture tracking model UTrack by Tuinenburg and Staal (2020) (Fig. 1). The model tracks parcels of moisture through the atmosphere from their locations of evaporation to those of precipitation, the first to be based on ERA5 atmospheric reanalysis data. This means that the model post-processes these atmospheric data. The moisture tracking consists of three steps: (1) the release of moisture evaporated from the land surface into atmospheric moisture parcels, (2) the calculation of trajectories through the atmosphere for each parcel and (3) the allocation of moisture present in the parcels to precipitation events at the location of the parcel.
In the first step, ERA5 total evaporation is determined and divided into
atmospheric moisture parcels. For each millimeter of evaporation, 100 parcels are
released 50 hPa above the surface height at random spatial locations within
each
In the second step, the parcels are tracked forward in time through three-dimensional space, where the location and moisture content of each parcel are updated at every time step of 0.1 h. This forward-trajectory calculation is based on interpolated three-dimensional ERA5 wind speed and wind direction data, which consist of 25 pressure layers in the atmospheric column. This step is partly randomized as follows. The main uncertainty of atmospheric moisture tracking is the redistribution of moisture in the vertical direction (Tuinenburg and Staal, 2020), for example due to parameterized processes including convective up- and downdrafts, reevaporation, and microphysics. To account for the uncertainty in the vertical redistribution, we employ a probabilistic scheme in which moisture parcels are randomly distributed along the local vertical moisture profile (Tuinenburg and Staal, 2020). The scheme is run at every time step, with such a probability that every parcel is redistributed along the local moisture profile on average once per 24 h.
In the third and final step in the moisture tracking model, a fraction of
the moisture present in a moisture parcel may be allocated to rainfall
events at the location of the parcel. For this, ERA5 hourly total
precipitation (
Precipitation
Further information about the justifications of the model settings can be found in Tuinenburg and Staal (2020).
For all ERA5 moisture evaporated between 2008 and 2017, we determined the
downwind precipitation location using UTrack. These simulations were done
globally for each 0.25
In order to save storage space, the output is treated as follows. First, we
take the natural logarithm of the downwind locations and we subsequently
multiply this by
The procedure to retrieve the normalized downwind precipitation locations of
evaporation at location
Due to the imprecision of the storage in integers, the global sum of
We calculate evaporation moisture recycling (land and basin recycling), precipitation moisture recycling (land and basin recycling), downwind distance between evaporation and precipitation location (longitudinal and latitudinal directions), global simulated precipitation, and the sum of precipitation originated from one source cell.
Land evaporation recycling (in short land recycling) is defined as the
fraction of evaporation that precipitates over land areas
The distance between evaporation and precipitation locations is determined for each grid cell from its evaporation footprint, determined by taking the mean over all downwind locations (weighted by moisture flow). For this mean distance, we present its latitudinal and longitudinal components. Because flows in opposite directions cancel out one another, we also present the mean absolute distances, which are more informative of the actual distances between evaporation and precipitation.
The above metrics are all related to the characteristics of the relation between evaporation and its downwind location, which result from the forward simulations that form the basis of the dataset. In addition to these, we employ the fact that the database contains the downwind information for evaporation from all locations globally. By multiplying the (normalized) downwind precipitation locations with the evaporation in the source location, we get the absolute amount of downwind precipitation. In this “backward analysis”, if the absolute downwind precipitation resulting from all global evaporation locations is summed, the total global precipitation should result. Moreover, we can determine the evaporation sources of precipitation in each individual grid cell by multiplying the normalized values in the database with the actual evaporation in each grid cell and only select the precipitation grid cell of interest from the downwind precipitation footprints. This will then result in the precipitation in the grid cell under consideration and its evaporation origins. Thus, for the evaporative sources of precipitation, the precipitation moisture recycling (land and basin recycling) is determined.
Two tests are done on the dataset. First, to check for numerical consistency we determine whether the sum of precipitation originating from each source cell adds up to the evaporation from that cell. Note that the total sum can deviate from 1 due to the storage format, as described in Sect. 2.2. Second, for each grid cell, we multiply the (normalized) downwind precipitation location with the monthly evaporation for that grid cell, yielding the actual downwind precipitation from that grid cell. The sum of this actual downwind precipitation over all the grid cells releasing evaporation should add up to the global precipitation amount and pattern again. The global precipitation should be similar to ERA5 total precipitation as the dataset is forced with those data. Therefore, to check the internal consistency of the dataset, the global precipitation is calculated from the database and compared to ERA5 total precipitation.
Seasonal variability of evaporation recycling ratio (as a
fraction) over land. Evaporation recycling ratio is the fraction of
evaporation that subsequently rains down over land.
We find that, globally, 70 % of all terrestrial evaporation precipitates over land (i.e., the global average land recycling ratio is 0.70). However, the land recycling ratio varies considerably across the globe (Fig. 2). For example, in southern South America the land recycling ratio is low due to strong transport of moisture towards the ocean there. For inland regions, land recycling ratios are close to 1. Furthermore, the land recycling ratio varies throughout the year. The largest percentage of terrestrial evaporation recycling occurs in June and July (74 %) and the smallest in January and December (62 %) (Fig. 3). The Northern Hemisphere contains more land surface, and therefore, there is more convection during summer months in the Northern Hemisphere, which has an important contribution to the evaporation recycling ratio (Fig. 2).
Monthly variability of evaporation (red) and precipitation (blue) global land recycling ratios. Evaporation recycling ratio is the fraction of evaporation that subsequently rains down over land. Precipitation recycling ratio is the fraction of precipitation that has evaporated from land.
Only 17 % of moisture evaporated from the ocean rains out over land. In general, the land recycling ratio for the oceans (the fraction of evaporation from an ocean location that rains out over land) does not exceed 0.2. Exceptions include the high land recycling around the Equator as a result of the trade winds that transport oceanic moisture towards land. Also, in both the northeastern and southeastern Pacific Ocean and the northern Atlantic Ocean, land recycling exceeds 0.2 due to the westerlies (Fig. 2a). Throughout the year there is little variation in the relative amount of moisture that evaporates from the ocean and rains out over land. There is a small deviation from the annual mean in the months April (16 %), May (15 %) and June (16 %). All other months show no deviation from the annual mean (17 %).
With the implicit backward analysis we find that 51 % of global precipitation over land and 8 % of precipitation over the ocean has evaporated from land. Figure 4 shows the seasonal variation in the ratio of precipitation with moisture evaporated from land. In the Northern Hemisphere, there is a clear seasonal cycle as, here, during spring and summer (AMJ and JAS) the land recycling ratio is larger as a result of convection, over both oceanic and terrestrial regions. For the Southern Hemisphere the seasonal differences are smaller. However, over Australia, the land recycling ratio is also larger for spring and summer (OND and JFM). As the seasonal cycle is stronger for the Northern Hemisphere, the global precipitation over land and ocean that evaporated from land shows maximum values for June (57 % and 10 %) and July (57 % and 10 %) (Fig. 3). Minima occur during the Northern Hemisphere winter. The part of global precipitation over land that evaporated from land is lowest in January (36 %) (Fig. 3); the part of global precipitation over the ocean that evaporated from land is lowest in January, February and March (7 %).
Seasonal variability of precipitation recycling ratio over land.
Precipitation recycling ratio is the fraction of precipitation that has
evaporated from land.
Average evaporation and precipitation basin recycling ratios for the 26 largest basins of the world.
Out of the 26 largest river basins of the world (Table 1), the evaporation recycling ratio in almost all cases exceeds the precipitation recycling ratio (Fig. 5). Only for the Murray and Mackenzie basins do we find a larger proportion of recycled precipitation than recycled evaporation (12 % vs. 11 % for the Murray basin and 34 % vs. 33 % for the Mackenzie basin). The highest evaporation recycling ratios occur in the Amazon and Congo basins, with respective ratios of 63 % and 60 %, which can be explained by their size and shape, high forest evapotranspiration, and strong convection over the tropical rainforests (Spracklen et al., 2012; Staal et al., 2018). The Congo basin has a higher precipitation recycling ratio than the Amazon: in the Congo, 47 % of precipitation has last evaporated in the same basin, against 36 % for the Amazon.
In general, the lowest recycling ratios are found for the low midlatitudes, with values between 0 and 0.4 (except for mountainous regions). Due to the Hadley cells and the Ferrel cells, there is subsidence in the low midlatitudes. As the cool air sinks it warms adiabatically, which results in a decrease in relative humidity. Because of this, the air is dry and precipitation is less likely to occur, which explains the lower recycling. Furthermore, differences in precipitation recycling can be expected due to regional and temporal differences in the strength of transport from moisture evaporated from the ocean, such as in monsoon regions (Wei et al., 2012).
Average evaporation
Average forward moisture transport from evaporation to
precipitation.
Figure 6 shows the annual mean of the traveled distance for evaporated
moisture across the globe. For latitudinal transport there is an equatorward
flow for the polar regions and a poleward flow around the Equator. For the
midlatitudes an alternating pattern of northward and southward moisture
transport is visible, with maxima up to 18
Average backward moisture transport from precipitation to
evaporation.
The sum of precipitation originating from any grid cell can be derived and used for its validation. For all source cells, the sum of their downwind footprints (in fraction of evaporation) adds up to a value of around 1.05, due to the conversion from floats to integers. This means that the values should be renormalized and converted back to floats, so their global sum again equals 1 (see Eq. 6). While this process reduces the data size considerably, it introduces some uncertainty in the footprints.
To further validate the dataset, we compare the result of global precipitation (Fig. 8a) with total precipitation from the ERA5 dataset (Fig. 8b). Because only ERA5 data are used to force the UTrack model, global precipitation should be similar between the two datasets. Figure 8c shows the absolute difference between global precipitation derived with UTrack and from ERA5, and Fig. 8d shows the relative deviation of UTrack to ERA5 precipitation. For most locations the error is small in comparison to the global precipitation. The largest deviations occur over the poles, mountainous regions and dry areas. The drier areas show a smaller absolute deviation, although the relative deviations are large due to the small amounts of precipitation in these areas. Due to the relatively small error, the precipitation patterns are similar for both datasets and both maxima and minima are located in the same regions. However, the exact values of extreme precipitation levels show some differences. For example, precipitation at the coast of Colombia is higher for the ERA5 data.
We present the first globally comprehensive dataset for atmospheric moisture
flows at 0.5
No independent observations exist to validate the source-to-sink relations in our database directly. Some studies validate moisture recycling using tracers in atmospheric models or stable water isotope measurements (Koster et al., 1993; Kurita and Yamada, 2008). However, tracers in atmospheric models are still very much dependent on model physics, especially the treatment of the vertical transport of moisture (see Tuinenburg and Staal, 2020). Furthermore, because observations and interpretation of stable water isotopes are not available on a global scale, we do not use these here. However, this database provides the opportunity to analyze local stable water isotopic measurements in light of the local moisture sources. Due to this inherent difficulty in validating the dataset, we provide an internal validating check regarding the global precipitation distribution. This showed that the patterns of total precipitation are similar to the total precipitation in ERA5 and are thus internally consistent. However, due to the fact that the ERA5 climatological mean precipitation is about 1.5 % higher than the evaporation, there is an added uncertainty of that order due to this non-closure of the moisture balance. The current dataset only stores the 2D downwind precipitation location and not the altitude at which the atmospheric moisture condensates. In future datasets, this could be stored and the resulting vertical latent heating profiles could be compared against the ERA5 latent heating profiles as a further validating check.
We analyzed a number of hydrologically relevant statistics of the global atmospheric moisture flows for 2008–2017. On average, 70 % of global land evaporation rains down over land. This fluctuates across the year, with a minimum of 62 % in December–January and a maximum of 74 % in June–July. This evaporation recycling is higher than precipitation recycling: 51 % of global land precipitation has evaporated from land, varying between 36 % in January and 57 % in June–July. This estimate is significantly higher than the estimate of 36 % by Van der Ent et al. (2014), which was based on other input data and a coarser, Eulerian moisture tracking model. Furthermore, Dirmeyer et al. (2014) found similar variability in the patterns and values of precipitation recycling throughout the year. However, their results show some differences, including lower precipitation recycling in parts of South America. The highest basin recycling occurs in the Amazon and Congo basins, with evaporation and precipitation recycling of 63 % and 36 % for the Amazon basin and 60 % and 47 % for the Congo basin. These numbers allow us to further compare our estimates of moisture recycling to previous ones. A number of studies have estimated the precipitation recycling ratio for the Amazon. These estimates tend to vary between 24 % and 35 % (Brubaker et al., 1993; Eltahir and Bras, 1994; Costa and Foley, 1999; Trenberth, 1999; Zemp et al., 2014; Staal et al., 2018), meaning that our new estimate is slightly above other ones, although a possible recycling ratio of 41 % has also been reported (Burde et al., 2006). A recent estimate using a very similar Lagrangian moisture tracking method but forced with ERA-Interim data instead of ERA5 (Staal et al., 2018) estimated 32 % evaporation recycling for the Amazon. This could mean that previous estimates based on the ERA-Interim dataset tend to underestimate atmospheric moisture recycling, although the study periods were not equal. Keune and Miralles (2019) found evaporation basin recycling values for the Danube of around 35 % for JJA. For these months, we find a similar number, although the annual mean basin recycling rate is somewhat lower.
There are missing values for Antarctica and Greenland regarding precipitation and recycling ratios, which affect the metrics of traveled distance, resulting in large gradients. These errors are the result of positive values in the ERA5 evaporation data, indicating condensation, which may represent errors in that dataset.
Recently, Link et al. (2020) developed another dataset for the
destinations of global land evaporation. This current dataset differs in a
number of important ways from that of Link et al. (2020).
First, we present moisture connections among all global grid cells including
the oceans, allowing the user to quantify the complete upwind evaporation
source area across the globe. Second, our moisture tracking model
(Tuinenburg and Staal, 2020) uses the most detailed forcing data
that are currently available: hourly evaporation and precipitation and
hourly wind speed and three-dimensional wind directions for 25 atmospheric
layers in the troposphere at 0.25
The drawback of our high-resolution simulations is the heavy data load that
results from them. This is the reason that we reduced the output resolution
to 0.5
We foresee a range of applications of the dataset from fundamental Earth
system science to applied global change research. For example, the moisture
flows can be analyzed as a network where its nodes refer to areas that are
connected by the moisture flows among them (Zemp et al.,
2014; Krönke et al., 2020). Network analysis is becoming an increasingly
important means to study global weather patterns and the Earth system (e.g., Boers et al., 2019). Our simulations
potentially provide a mechanistic basis for empirical relations among remote
weather patterns, such as drought correlations and land cover. There could
be a benefit in classifying the weather patterns in terms of the length
scale difference between the upwind and downwind parts of the atmospheric
moisture cycle. Finally, they can be used to assess the effects of land
cover changes on precipitation patterns at any scale between 0.5
The dataset (Tuinenburg et al., 2020) is available from the
PANGAEA archive in both the full-resolution version at 0.5
Over the last decades, the interest in the atmospheric moisture cycle and
the explicit links between evaporation and precipitation locations has
increased. Moisture recycling rates are relevant for land use change
studies, hydrology and atmospheric predictability. Many moisture recycling
studies, including this study, determine these explicit relations between
evaporation and precipitation location by forcing a moisture tracking model
with atmospheric reanalysis data, effectively calculating moisture recycling
as a post-processing scheme. Here, we have calculated the explicit relation
between evaporation and precipitation locations, based on the
state-of-the-art, high-resolution ERA5 atmospheric reanalysis. The presented
global database consists of the downwind precipitation location for all
evaporation globally, on a 0.5
This dataset can be used to assess source–target relations of moisture fluxes into and out of the atmosphere. With these relations, the downwind precipitation and 2D latent heating effects of land use changes could be determined. Furthermore, due to the global domain of the dataset, the same can be done for the reverse direction, determining the evaporative moisture sources of precipitation in a certain location. These upwind moisture source locations could be used in atmospheric predictability studies.
OAT and AS conceived and designed the study. OAT and JJET carried out the study. All authors wrote the paper.
The authors declare that they have no conflict of interest.
The authors would like to thank the participants of the “Natural water production” theme for the fruitful discussions and financial support.
Obbe A. Tuinenburg received support from the research program Innovational Research
Incentives Scheme 016.veni.171.019 by the Netherlands Organisation for
Scientific Research (NWO). Arie Staal was supported from the European
Research Council project Earth Resilience in the Anthropocene (743080 ERA).
Part of this work was performed in the cooperation framework of Wetsus,
European Centre of Excellence for Sustainable Water Technology (
This paper was edited by David Carlson and reviewed by Jiangfeng Wei and one anonymous referee.