We make use of a unique high-quality, long-term observational dataset on a tropical lake to assess the effect of rainfall on lake surface temperature. The lake in question is Lake Kivu, one of the African Great Lakes, and was selected for its remarkably uniform climate and availability of multi-year over-lake meteorological observations. Rain may have a cooling effect on the lake surface by lowering the near-surface air temperature, by the direct rain heat flux into the lake, by mixing the lake surface layer through the flux of kinetic energy and by convective mixing of the lake surface layer. The potential importance of the rainfall effect is discussed in terms of both heat flux and kinetic energy flux. To estimate the rainfall effect on the mean diurnal cycle of lake surface temperature, the data are binned into categories of daily rainfall amount. They are further filtered based on comparable values of daily mean net radiation, which reduces the influence of radiative-flux differences. Our results indicate that days with heavy rainfall may experience a reduction in lake surface temperature of approximately 0.3 K by the end of the day compared to days with light to moderate rainfall. Overall this study highlights a new potential control on lake surface temperature and suggests that further efforts are needed to quantify this effect in other regions and to include this process in land surface models used for atmospheric prediction.
The works published in this journal are distributed under the Creative Commons Attribution 4.0 License. This licence does not affect the Crown copyright work, which is re-usable under the Open Government Licence (OGL). The Creative Commons Attribution 4.0 License and the OGL are interoperable and do not conflict with, reduce or limit each other. © Crown copyright 2018
Lakes are important features of the terrestrial environment for
physical, ecological, economic and recreational reasons. Physically,
lake–atmosphere interactions can influence the local weather and
climate. Thus their representation in Earth system modelling has
increased in complexity in recent years. Lake water surface
temperature (LWST) is of particular relevance to atmospheric modelling
due to the contrast in temperature, and hence in boundary-layer
fluxes, that often exists between lakes and their surroundings
At high latitudes, correct prediction of freezing temperatures, and thus
ice-cover periods, is important to obtain accurate boundary-layer fluxes. In
the tropics, varying temperature contrasts between lakes and the surrounding
land may be associated with cycles of severe weather. For instance, remote
sensing data highlight an important impact of the African Great Lakes on the
diurnal precipitation and thunderstorm cycle, especially over Lake Victoria
and Lake Tanganyika
Adequately observing and modelling tropical lake–atmosphere
interactions often remains a challenge, even though efforts have been made
to quantify these exchanges
Lakes interact with the atmosphere via a variety of processes.
Physical lake models developed for use in a meteorological context
have thus far concentrated on lake–atmosphere interaction through
turbulent and radiative fluxes. The effects of rain on LWST, both
directly from thermal perturbation and indirectly from changing the
lake stratification, are little understood or represented to date.
Evidence of the significance of rain effects, particularly in the
tropics, is beginning to emerge however
To summarise the structure of the following sections, in Sect.
The African Great Lakes are of utmost importance for regional economies, as
well as being essential to the survival of the local population. As the
largest reservoir of freshwater in the tropics, they provide numerous
ecosystem services to local communities, such as fishing grounds, drinking
water and electricity. Lake Victoria alone directly supports 200 000
fishermen operating from its shores and sustains the livelihood of more than
30 million people living on its shores
Lake Kivu (01
Maps of Lake Kivu geography and situation. The lower left corner of
the large map is at 2.6
Rain may have an effect on the lake surface in four ways: (i) evaporative cooling of the near-surface air during precipitation, which induces an additional upward sensible heat flux from the lake towards the atmosphere; (ii) direct rain heat flux into the lake; (iii) mixing of the lake surface layer through the flux of kinetic energy; and finally (iv) convective mixing of the lake surface layer. The first of these ought to be parametrised by atmospheric models, as with related atmospheric effects like the reduction of insolation by cloud cover. The others lie mainly in the lake modelling domain. Hereafter, we discuss points (i)–(iv) above in some more detail.
Raindrops falling into unsaturated air will cool through evaporation.
Their passage through the air leads to heat transfer from the air,
hence cooling the air. As the rainfall continues, the air will tend to
saturation, and both rain and air will approach the air's original
wet-bulb temperature. Thermodynamically, further quantification and
parametrisation of this process requires consideration of various
factors such as the atmospheric moisture and temperature profile,
drop-size distribution, drop concentration, etc.
At the extreme end of intense convection, atmospheric cooling and
momentum transfer from rain may produce cold convective downdraughts,
which transport cold air to the surface from higher levels
The specific heat capacity of water is approximately
On seasonal to decadal timescales, the sensible heat contribution by rainfall
is deemed small
As well as the direct effects of an additional heat flux, rainfall may produce a perturbation of LWST by the mechanical and convective mixing of the near-surface portion of the lake.
An early study by
Several subsequent studies of artificial rainfall have examined
rainfall-generated turbulence in slightly more detail. Artificial heavy
rainfall has been observed to produce turbulent mixing over depths of
10–20 cm in the study of
The kinetic energy flux of real rainfall has also been estimated in the
context of soil erosion studies
Mechanical surface forcing by wind upon
lakes is usually modelled through matching of stress, so that the
aqueous friction velocity at the lake surface
Thus, setting
For cold rain falling onto a relatively warm lake, convective effects will
presumably add to the mixing strength and depth. While
Automatic weather station on Lake Kivu after its installation, 8 October 2012 (© Wim Thiery). Position (a) indicates the location of the temperature, relative humidity and radiation sensors at 4.40 m above the lake surface. Position (b) shows the location of the wind vane at 7.20 m above the lake surface. Position (c) indicates the container on which the station was mounted.
AWS Kivu is installed on the research platform of the Rwanda Energy Company,
approximately 3 km offshore of the cities of Gisenyi (Rwanda) and Goma (Democratic Republic of the Congo; see Fig.
AWS Kivu consists of sensors for air temperature (
Variables are sampled every 15 s, from which 30 min averages are calculated and stored. In the case of precipitation, accumulated values are stored, and for wind speed both mean and maximum values are recorded. Moreover, short periods of high-frequency radiation measurements enable an assessment of the potential effect of platform movements. Through a General Packet Radio Service (GPRS), the KU Leuven Regional Climate Studies group receives the observations directly from the station, allowing for remote problem detection.
The time span of measurements used here was between 13 September 2012 and 14 August 2017; however most of the analysis is based on 4 calendar years of data from 1 January 2013 to 30 December 2016.
AWS Kivu sensor specifications.
The data indicate that there is a remarkable uniformity of the lake climate.
The annual air temperature range is around 14 K. The daily rainfall totals
for 4 calendar years show a generally uniform spread but with a slightly
drier period around July (Fig.
Daily rainfall totals for 4 years of the observational
record, beginning on 1 January 2013. The red line marks the 8 mm
point, which is used to partition rain days between WET (
The temporal variations in weather may be examined further using power
spectra of rainfall and wind speed (Fig.
We also note the presence of some sub-daily peaks in the wind speed spectrum, the most dominant at a frequency corresponding to a period of approximately 8 h, and the next two corresponding to periods of approximately 6 and 12 h. (As will be shown later, the sub-daily wind fluctuations giving rise to these peaks are evident on plots of mean daily wind speed. These fluctuations are probably due to local circulations caused by lake or land breezes, and the largely bimodal distribution of the wind direction, also shown later, appears to support this interpretation.)
To examine the effect of heavy rain, 4 years of data will be analysed
(1 January 2013 to 30 December 2016). This amounts to 1457 days, as three
days are omitted due to missing data. These data are referred to as ALL data
in the following analysis. Based on daily rainfall totals, they may be
divided into DRY, WET and VWET days. DRY days are days with no rainfall. The
remaining days are partitioned into WET or VWET depending on whether the
rainfall total is respectively less or greater than a threshold of 8 mm
(Fig.
Regarding the distribution of hourly rainfall over the 4 years, 2.4 %
of hours had a rainfall total greater than 1 mm, and 0.6 % of hours had
a rainfall total greater than 5 mm. Figure
The average rainfall for each hour of the day, for VWET days
(cyan), WET days (green) and DWET days (purple). In this and later
plots, time is shown as UTC (coordinated universal time), which is 2 h
behind LT (local time), i.e. LT
The effect of daily weather on LWST is summarised in Fig.
Mean diurnal cycles of
Atmospheric forcing of LWST is usually characterised in terms of turbulent or
radiative fluxes, with turbulent fluxes depending on mean wind speed,
lake–air temperature difference and near-surface humidity. For the categories
described here, the choice of partitioning threshold between WET and VWET
days coincidentally produces extremely similar graphs of mean wind speed.
This has the effect of removing an important potential source of variation
between these categories. The distributions of wind directions and speeds are
also quite uniform; see Figs.
Histograms of half-hourly mean wind direction for
2013–2016. The colours are the same as in
Figs.
Histograms of half-hourly mean wind speed for
2013–2016. The colours are the same as in
Figs.
Figure
Comparison of the differences in net radiation. WET minus VWET net radiation is the solid line, and DWET minus VWET net radiation is the dashed line.
WET days have a higher mean net radiation than VWET days
(Fig.
Percentages of the types of day which came immediately before and
after each type,
broken down by observation year.
Note that calculating percentages to the nearest percent occasionally produces sets that do not sum to 100 exactly.
Using this additional constraint, the number of DWET days in the 4-year
period is 425 or 73 % of the WET days. Table
Mean diurnal cycles of
Diurnal behaviour of the difference in mean values of LWST in
the DWET and VWET cases, normalised by the standard error of the
difference, SED; see Eq. (
The average daily rainfall on DWET days is 2.31 mm, compared to 2.33 mm on
WET days and 17.99 mm on VWET days. Thus, the contrast in rainfall amount is
largely preserved by this resampling. The diurnal evolution is also plotted
in Fig.
The average diurnal evolution for the categories of ALL, DRY, DWET and VWET
is shown in Fig.
Considering the reliability of this difference, it may be noted that a
difference of 0.3 K against a background at approximately 300 K is
equivalent to a difference in upwelling longwave radiation of approximately
1.8 W m
In terms of significance, the standard error of the difference between
the mean DWET and VWET values is given by
Finally, the effect of rain on the sensing of LWST should also be considered
as a possible cause of observed LWST differences. From an atmospheric
modelling viewpoint, the sensed surface temperature is the important quantity
in many cases, as has been recently discussed in the context of the
introduction of a “skin” temperature into the FLake lake model
Lake Kivu has a remarkably stable tropical lake climate, and AWS Kivu has yielded a high-quality multi-year over-lake observational record, which is rare and perhaps unique in the tropics and well suited to the present research question. This study is the first such use of these data.
Data over 4 years from AWS Kivu have been categorised by daily rainfall
amount and net radiation to investigate the possible effects of rainfall on
lake water surface temperature (LWST), which may be particularly significant
in the tropics
The possible pathways by which this effect may arise are (i) cooling of the air due to contact with evaporatively cooled raindrops, and a subsequent increase in atmospheric sensible heat flux from the lake, (ii) negative heat flux directly to the lake from rain impingement, (iii) mechanical mixing of the lake surface layer by the kinetic energy of rain impact and (iv) convective mixing of the lake surface layer due to the negative heat flux from rain. Of these, the first is the most likely to be a parametrised process in a general circulation model of the atmosphere, although it could be considered the most indirect of the four. The rain heat flux is likely to be proportional to the difference between the air wet-bulb temperature and LWST. We have indicated with scaling arguments that the mechanical mixing due to heavy rain may be comparable to that of a strong wind. The convective mixing will depend on the near-surface temperature structure of the lake, and hence on its recent history.
Unfortunately, the available data do not cover several other process-related quantities that would be useful to have, such as turbulent heat fluxes, rain temperature, fine-scale lake temperature profiles or lake turbulence measurements. Thus, the processes producing this effect are not directly measured. However, through our indirect analysis of the processes, it seems likely that cooling by rain combined with mechanical and convective mixing from droplet impact may have an effect on LWST, in addition to the effect from the more widely studied pathway of evaporative cooling.
Potential avenues of future work would be to examine these processes
more closely in a targeted campaign of observations, including the
quantities listed above, and to consider how lake models may be
modified to include their representation. An intermediate step in the
latter might be to re-examine previous modelling studies to explore
correlations between lake model errors and rainfall records. Since,
as discussed earlier, rainfall may affect not only the surface
temperature but potentially also the temperature or depth of any upper
mixed layer, some or all of these quantities could be susceptible to
rainfall effects. For models that predict vertical fluxes through the
water column, comparison of these with any available flux or TKE
measurements would be a possible way to estimate rain penetration in
real lakes. There is an indication in the data of
In large tropical lakes, it is possible that a surface temperature
difference of the order of 0.5 K may suppress or enhance the strength
of local air circulations, such as lake breezes, and hence have some
effect (or even feedback) on the evolution of the local weather
The data are available on request from the dataset owners, Wim Thiery and Nicole van Lipzig.
The authors declare that they have no conflict of interest.
This article is part of the special issue “Modelling lakes in the climate system (GMD/HESS inter-journal SI)”. It is a result of the fifth workshop on “Parameterization of Lakes in Numerical Weather Prediction and Climate Modelling”, Berlin, Germany, 16–19 October 2017.
WT was supported by an ETH Zurich postdoctoral fellowship (Fel-45 15-1). The Uniscientia Foundation and the ETH Zurich Foundation are thanked for their support of this research. The Belgian Science Policy Office (BELSPO) is acknowledged for the support through the research project EAGLES (CD/AR/02A). We thank Stijn Bruggen, who analysed the AWS data in his master's thesis and thereby supported the design of this study and the analysis presented here. GGR thanks John M. Edwards for a helpful discussion of TKE budgets. Edited by: Miguel Potes Reviewed by: two anonymous referees