The E-DATA experiment was performed between 14 and 23 November 2018 at the SDH (20.1∘ S, 68.5∘ W; 3790 ma.s.l.). This date is optimal to study evaporation due to the total absence of precipitation and high mean temperatures. The SDH is a closed basin of 1417 km2 (55 km N–S and 35 km W–E) located ∼3.8 km up and over ∼135 km from the Pacific Ocean. Note that at such altitude, the pressure level is very low compared to sea level, ∼650 hPa. Figure a shows the location of the SDH saline lake and E-DATA experiment in a vertical cross section over the western slope of the Andes Mountains. Figure b shows an overview of the surface observation installation in the vicinity of the SDH saline lake. Three SEB stations were installed over representative and homogeneous surfaces of the site: water, wet salt and desert. The first SEB station was installed above a shallow 15 cm deep lagoon (20.27∘ S, 68.88∘ W; 3790 ma.s.l.), whose surface covers 4 km N–S by 800 m W–E. The second SEB station was located over a wet-salt crust (20.28∘ S, 68.87∘ W; 3790 ma.s.l.), which is a wet soil composed of salt whose surface is covered by a mostly dry crust of slime. The third SEB station was installed in an area representative of bare rocky-soil-like desert conditions (20.35∘ S, 68.90∘ W; 3953 ma.s.l.). Figure b also shows the profiling measurement points from where radiosonde and an unmanned aerial vehicle (UAV) were launched: water and desert. The first launch site was located on the western shore of the lagoon (20.28∘ S, 68.88∘ W; 3790 ma.s.l.), covering water and wet-salt surfaces. The second point was located next to the desert SEB station (20.35∘ S, 68.90∘ W; 3953 ma.s.l.), covering the desert surface that surrounds the SDH basin. A transect of four automatic MET station deployed from 20.28∘ S, 68.90∘ W to 20.28∘ S, 68.97∘ W westward of the lagoon was utilized to characterize the advection. Finally, we also made use of a standard meteorological station placed 2 km N from the saline lake (Fig. b), which has been in continuous operation since 2015 by the Centro de Estudios Avanzados en Zonas Áridas (CEAZA).

We deployed SEB stations, complemented by additional meteorological measurements, at each of the three main surface types (water, wet salt, desert) together with a transect of four MET stations on the western slopes of the study site. Special attention is paid to the measurement of variables that are related to the drivers of E: radiation, turbulence and VPD. Table a shows the main variables and sensors utilized over each surface type organized by sensor groups. Radiation measurements and sensors differed between surfaces. The four-component radiation measurements were gathered for the water surface, whereas at the desert and wet-salt surfaces only integrated Rn measurements were available. The albedo was measured only at the water site; for the wet salt it was estimated using the net shortwave radiation and the incoming shortwave radiation measured in the lake. For the desert site we assumed the value 0.21, reported as a typical value for dry sandy soils in . Additionally, the Rn of the desert SEB station was too large so it was corrected by using incoming shortwave measurements from the water SEB station and assuming an albedo of 0.21 . We used the flux software package EddyPro version 6.2.2 from LI-COR Biosciences Inc. (Lincoln, Nebraska, USA) to calculate the turbulent fluxes of latent heat (LvE), sensible heat (H) and the friction velocity (u*) at 10 min averaging intervals. All standard data treatment and flux correction procedures were included, most notably axis rotation with the planar-fit procedure , raw data screening including spike removal , interval linear detrending and low-pass filtering correction . In addition, quality flags were determined based on . The measured ground heat flux (G) was corrected for heat storage above the heat flux plates by using the calorimetric method , and the observations were obtained from soil temperature probes buried at different depths (see Table 1) in each surface type. Note that over a shallow water layer G is stored in both the water and soil/sediment layers above the heat flux plates . We corrected for both components of the soil heat storage. Standard meteorological variables such as air temperature (T), relative humidity (RH), atmospheric pressure (P), wind speed (U) and wind direction (WD) were measured in the SEB stations and at a transect of standard meteorological stations. The details are shown in Table a. The uncertainty related to the energy balance closure at the SEB stations can be found in Appendix A.

We used two airborne instrument carriers: a radiosonde balloon and a UAV. These were equipped with similar sensor packages that provided measurements of T, RH, U, WD and P for the radiosonde and T, RH and P for the UAV (details in Table b and in ). The radiosonde balloons were launched from two locations described in Sect. 3.1. At both locations, we performed intensive campaigns on 21 November over the water surface and 22 November over the desert surface (Fig. b), where we launched balloons at 09:00, 12:00, 15:00, 18:00 and 21:00 LT. Balloons typically reached an altitude of 10 km and drifted away horizontally up to a distance of 50 km northeastward of their launching sites. Vertical profiles of θ, q, U and WD were obtained from the radiosonde to characterize and estimate the ABL height (h), using the surface pressure level of the SDH (∼650 hPa). This height was estimated through the maximum vertical gradient of θ . The UAV was flown simultaneously from the same two locations as the balloon launches (described in Sect. 3.1) from the ground to up to 500 ma.g.l. from the surface on 21 and 22 November every 30 min from 09:00 to 12:00 LT. From these flights we obtain the vertical profiles of θ to characterize the first 500 ma.g.l. of the ABL. UAV flights were, unfortunately, not possible after 12:00 LT due to high winds.

To complete the analysis of the E-DATA experiment, we reproduce the same period using the Weather Research and Forecasting (WRF) Model version 3.7 . We aim to study the atmospheric circulation that is formed daily from the Pacific Ocean to the Andes western slope. We follow the methodology suggested by , which consists of performing consecutive, short WRF runs initialized at 00:00 UTC and running for 48 h. The first 24 h of each run is used as a spin-up for the physical parameterizations and the 24–48 h to represent the weather conditions of the simulated day. Therefore, we only analysed and evaluated the period 24–28 h. This methodology ensures that each simulated day starts with its real respective initial and boundary conditions. Initial and boundary conditions are taken from ECMWF ERA-Interim reanalysis data for 20∘ S, 68∘ W with a 0.5∘ spatial resolution. By using this dataset input, every 6 h there is an update of the tendencies due to the large-scale forcing. Figure c shows the horizontal distribution of the four two-way nested model domains; detailed information can be found in Table , Appendix A. The inner domain (D04) includes all the measurements gathered in the E-DATA experiment. In the vertical direction, we imposed 61 non-equidistant grids following an exponential shape that maximizes the number of vertical levels in the boundary layer, i.e. 40 within the first 2000 m. Several physical processes such as radiation, surface and boundary layer, convection, microphysics, and land surface model are parameterized in WRF; they are also detailed in Table . A comprehensive model validation from both surface and vertical variables is presented in detail in Appendix A3.

Figure  displays the average diurnal cycles of the SEB terms, i.e. net radiation (Rn), ground (G), latent (LvE), and sensible (H) heat fluxes observed above water, desert and wet-salt surfaces. All the sites are located within in a radius of 10 km. Typical daytime values of the SEB terms are summarized in Table .

Our measurements show exceptionally high Rn levels over the water surface (∼950 Wm-2), less for the desert surface (∼700 Wm-2) and considerably less for the wet salt (∼500 Wm-2). The Rn daily cycles follow a typical sinusoidal diurnal cycle with the intermittent presence of high clouds (Fig. ). In the absence of four-component radiation measurements at the three sites we cannot provide a detailed breakdown of the short and longwave radiation components to Rn. Assuming that the incoming shortwave and longwave radiation terms are equal for all sites and taking the near-surface soil temperature (<1 cm depth) as a proxy for the longwave outgoing radiation, we can see the following (see also Table ). Maximum incoming shortwave radiation is ∼1250 Wm-2, which is close to the solar constant at the top of the atmosphere (∼1360 Wm-2) probably due to the high altitude and dry conditions of the study site. At ∼250 Wm-2 maximum longwave incoming radiation is rather small, due to the thin atmosphere and mostly-cloud-free conditions. The albedo of the desert surface is closer to the albedo of the water than the wet salt (0.21 vs. 0.12), but it is mainly the difference in surface temperature (27 vs. 22 ∘C) that leads to a larger longwave outgoing radiation loss and thus lower Rn. Compared to water, the wet-salt surface has a comparable surface temperature (20 vs. 22 ∘C), but it is the considerable difference in albedo (0.58 vs. 0.12) that leads to a larger shortwave outgoing radiation loss and thus much lower Rn.

While Rn shows a clear sinusoidal diurnal cycle, the SEB heat fluxes show two distinct regimes. The first occurs in the morning (07:00–12:00 LT) and is characterized by very low values of LvE (<30 Wm-2), almost zero H, over the water surface for instance. As a result, most of the radiative available energy is used to heat up the lake water and underlying soil sediment (G≈Rn, with values up to 600 Wm-2). The second regime occurs in the afternoon to early evening (12:00–20:00 LT). It begins with a rapid (2 h) rise in LvE and to lesser extent also in H at the expense of G, which diminishes in the afternoon to the point at which it becomes negative and provides additional energy, in addition to the decreasing Rn, to the turbulent fluxes H and LvE. Focusing on E, its behaviour is atypical for surfaces where water is plentiful and E is mainly driven by the available energy (energy-limited system). Here, our analysis shows that in the morning E is very small even though the levels of Rn are very high, which indicates that it is limited either by turbulence or VPD (see Sect. 2). In turn, in the afternoon, the E regime changes to the typical Rn-limited type to the point at which it requires additional energy from the soil (G becomes negative even before 15:00 LT).

On the wet-salt and desert surfaces, two similar surface flux regimes are observed, indicating that this feature dominates the entire study site and is not only specific to the water surface. However, there are interesting differences between the wet-salt and desert surfaces with respect to the water surface. In the wet-salt surface all the heat fluxes are much lower, reflecting the limited amount of Rn available (about half of that of water, as shown in Table ). Furthermore, the roles of H and LvE are reversed; i.e. it is H that suddenly increases when the afternoon regime commences (water and wet-salt surface Bowen ratio of 0.2 and 4, respectively). The salt crust reduces the soil evaporation of the wet-salt surfaces, in addition to the salt lowering E in general (Fig. b). In the desert, LvE is zero all day and Rn is balanced between G and H (Fig. c). The two regimes are clearly visible and show similarities to the wet-salt regime, with the difference that in the morning regime G and H are similar while in the afternoon regime H is dominant.

In the next section we further analyse the mechanisms that explain the two-regime behaviour in the local SEB fluxes and link them to a description of the local boundary layer profiles.

Figure a shows the mean daily cycle of wind speed (U) and direction (WD) over the water surface. Over wet-salt and desert surfaces, a similar diurnal variability is observed (Appendix B). The morning regime with low turbulent fluxes is related to conditions of very low wind speed (U<1 ms-1) and variable wind direction between the S and SW. The afternoon regime with high turbulent fluxes is related to high wind speeds (U>10 ms-1) and a well-defined wind direction from the west. This wind pattern is typical of this season and has been observed regularly in 2015, 2016, and 2017 as well.

Figure b shows that as a result of the low wind speed in the morning the aerodynamic resistance is very high (ra>400 sm-1; turbulent kinetic energy, TKE ∼0 m2s-2), and E in SDH can be regarded as turbulence-VPD-limited (see Eq. ). Note that in the absence of any wind the water surface is extremely smooth , and subsequently the surface roughness does not assist in generating shear. Additionally, H over the water is nearly zero as well, meaning that the high ra is the result of the absence of both shear and buoyancy-generated turbulence. In contrast, for the desert surface, this occurs when the winds are equally low but the temperature gradient is steep enough to sustain a mainly-buoyancy-driven H of about 200 Wm-2. In the afternoon, when the strong wind starts, ra drops dramatically and TKE increases in the same manner (4 m2s-2; see inset Fig. a), which results in the onset of the fluxes, when the E regime goes from a turbulence-VPD-limited to a radiation-limited E regime.

We now connect the gradients of temperature (linked to buoyancy forced turbulence) and moisture (linked to the VPD) between the surface of the water and the atmosphere at 1 m height, as well as how these affect E. Figure c shows the daily cycle of near-surface temperature (Ts), ∼1 m height air temperature (T) and surface–1 m thermal gradient (dT), over the water surface. The early morning (03:00–07:00 LT) displays low values of dT, where both air and water surface set below 0 ∘C and stay nearly constant due to the formation of water ice. In the late morning, Ts and T increase rapidly, and mild thermal gradients corroborate the low H (Fig. a) and no buoyancy-generated turbulence. In the afternoon, dT increases to about 7 ∘C and then falls in accordance with the available radiation. Note that there is a lag between Ts and T peaks, where T decreases earlier than Ts. This behaviour is explained by the effect of the wind and cold air advection, which is stronger at 1 m than at the surface. The latter is corroborated by the H>0 Wm-2 shown in Fig. a from 12:00 LT.

Figure d shows the daily cycle of saturated specific humidity (qs), 1 m height specific humidity (q) and the surface–1 m humidity gradient (dq), over the water surface. In the morning, dq values are small due to qs being constant according to Ts (ice on water). Note that the gradient is taken between 1 m and z0h (very close to the surface); this does not seem to warrant the system being labelled VPD-limited. However, the absolute q of the IRGA (sensors in Table ) is sensitive to calibration issues; therefore, we hypothesize that the gradient very close to the surface could have been smaller, to such a degree that the lack of turbulence results in a thin, water-saturated layer that prevents the creation of a gradient, and as a result leads to very small values of E. During the late morning qs increases according to Ts, and q shows a sudden drop of about 1 gkg-1, just before the change in the wind regime. Finally, during the afternoon, qs reaches its peak and then falls according to Ts. Likewise, q increases by 2 gkg-1 revealing, together with T, an advection of cold and slightly moister air into the study site. The advection of heat and moisture is discussed below in the vertical profile measurements and WRF modelling results.

These surface gradients are very dependent on the diurnal evolution of the ABL. Here, we present the vertical profiles at the water and desert surfaces as observed in the morning (Fig. ). During the morning in the desert, the vertical structure of potential temperature, θ, and specific humidity, q, follows the evolution of a prototypical dry convective ABL (Fig. a and b). The morning starts (09:00 LT profile) with a shallow unstable layer, corresponding to the unstable surface layer, followed by a stable layer until 1000 ma.g.l. Driven by the surface sensible heat flux (H=100 Wm-2), the ABL rapidly develops into a deep, well-mixed ABL (12:00 LT profile) where the boundary layer is capped by an inversion at h=1800 m. The entrainment of dry, warm air from above the ABL supports its growth to 12:00 LT. On the basis of the high warming observed from 09:00 to 12:00 LT (Fig. a), we have estimated a non-local contribution of warm air close to 140 Wm-2.

Contrary to this, in the early morning over the water surface, we observe for both the θ and q profiles (Fig. c and d, 09:00 LT profiles) a transition from a stable to a close-to-well-mixed profile. The stable profile at 09:00 LT is quantified in 0.026 Km-1 and starts to decrease its stability to 0.016 Km-1 at 10:00 LT, 0.001 Km-1 at 11:00 LT, and reaching a well-mixed type profile at 12:00 LT (∂θ/∂t with >0.001 Km-1). From 11:00 to 12:00 LT the θ profile shows an entire well-mixed boundary layer higher than 500 m, which is probably attributable to the desert convective ABL that is dominant on the study site (Fig. a). In the absence of wind and significant heat fluxes in the morning, the ABL is not driven by surface processes, and weak, local (mesoscale) flows are likely to be dominant. Figure  shows the time series of a typical day of U and T of a westward, upslope transect of meteorological stations (see Fig. ). Here, in the early morning (03:00–06:00 LT) a WNW flow is visible, in which cold air accumulates at the lowest station. Figure  also shows that in the course of the morning the wind direction veers 180∘ to ESE. The night downslope and morning upslope circulations are indicative of a katabatic (early morning) and anabatic (late morning) circulation between the low-lying saline lake and the surrounding mountain ridges. The anabatic circulation interacting with the top of the boundary layer potentially exhibits return flow that leads to a compensated subsidence over the lake , which would explain the eroding of the stable ABL in the course of the morning (Fig. c), as well as the warming observed in Fig. a. In Sect. 4.3 we return to the observational evidence by combining it with the analysis of the WRF results in order to determine the diurnal variability of these local circulations.

Figure a and b depict the wind profiles for the entire day as measured at the desert site. These profiles are very similar to those measured over the water. We therefore assume them as being representative of the entire study site. In the morning the winds are weak (<2 ms-1) coming from different directions through the height, similarly to the ones represented in Fig. a. In the afternoon the westerly wind increases strongly all across the boundary layer but is especially concentrated in a shallow jet near the surface (between 0∼250 m) with maximum wind speeds of ∼15 ms-1 at z∼80 m. These observations confirm that the surface winds are coupled to boundary layer dynamics, which in turn are determined by the regional circulation flows.

Figure c shows the evolution of the ABL depth as determined from the θ profiles over the desert and water surface. After the strong convective growth in the morning (∼530 mh-1), we observe that the boundary layer height decreases rapidly in the afternoon, from 1600 ma.g.l. at 12:00 LT to 750 ma.g.l. at 17:00 LT over the water and from 1800 to 650 ma.g.l. over the desert. We attribute this decrease to a change in the wind regime, which allows the entrance of air masses with different temperature, moisture, and stability (Fig. a and b). The mixed ABL values at 15:00 LT are cooler (decrease of ∼55 K) and moister (increase of 3 gkg-1) than those observed at 12:00 LT (Fig. a and b). Although the advected air is moist, compared to the desert conditions (q∼0.5 gkg-1), it is still characterized by a very low specific humidity (q∼3 gkg-1) considering the above-water conditions (qs>15 gkg-1). Hence, these moisty air mass does not significantly contribute to the E (see VPD subterm in Eq. ). Moreover, the ABL during the afternoon at the desert site is characterized by a strong inversion capping at ∼500 m above ground, in which at 18:00 LT θ jumps Δ4 K (Fig. a) and q jumps Δq 2 gkg-1 (Fig. b). Likewise, the ABL formed in the afternoon (after regional flow arrival) over the water presents a higher inversion capping that the desert (750 ma.g.l.), but lower θ jumps, Δ1 K (Fig. c), and higher q jumps, Δq 3 gkg-1 (Fig. d). Returning to the surface fluxes presented in Fig. a, we can now identify two mechanisms that increase H in the afternoon. The first is wind-enhanced turbulence, which increases the mixing efficiency between the surface and the atmosphere. Second is advection of cool air that increases the θ gradient and the subsequent near-surface instability of the atmosphere. Based on the turbulent heat fluxes (Fig. ) and the ABL height (Fig. b), and using the second term of Eqs. () and () as a residual, we quantify in Table  the local (surfaces fluxes), non-local (entrainment) and regional (advection) contributions to the mixed-layer tendencies of θ and q between 15:00–18:00 LT. It is not surprising that with LvE=0, the increase in humidity of 0.2 gkg-1h-1 is entirely accounted for by regional advection (see computation details in Appendix C). Here the overall trend is relatively small (+0.33 Kh-1), given the relatively large H=400 Wm-2 (17:00 LT), due to the cool-air advection, which largely cancels the local heating.

The afternoon profiles over water (Fig. c and d) show that on the arrival of the afternoon wind regime, the stably stratified boundary layer up to 500 m present at the end of the morning (Fig. c and d) becomes progressively eroded. In contrast to the eroding shallow mixed layer in the morning, in the afternoon, the destruction of the existing boundary layer structure is driven by the surface processes. This process is explained by (a) enhanced mechanical turbulence from the strong winds of the near-surface jet (Fig. a), (b) higher surface temperature (Fig. c) by the wind-induced mixing of the shallow water layer and (c) enhanced instability due to the cold air advection. This results in a shallow unstable layer, ranging from 0 to ∼150 m above the water surface between 15:00 and 18:00 LT. These levels are similar to the depth of the jet shown in Fig. a. Regarding the moisture budget, the arrival of the wind flow in the afternoon moistens the unstable layer, while wind shear mixes it, resulting in steadily-better-mixed q profiles. At 21:00 LT, an around 400 m deep well-mixed boundary layer has developed over the water surface. Considering the budgets of local and non-local vs. regional contributions to the q and θ mix layer tendencies over the water surface (Table ), we quantify a major local q contribution of about 0.86 gkg-1h-1 between 15:00 and 18:00 LT and a small non-local contribution of 0.006 gkg-1h-1. This moisture contribution exceeds the q tendency observed, which can be only balanced by the negative regional contribution (-0.59 gkg-1h-1). The negative regional contribution of q confirms that even though the advected air is moist compared to the desert conditions, this is still dry for the water surface conditions. The θ tendency behaves similarly to that of q, whose local contribution of heat is equivalent to double the tendency value, but it is compensated for by cold regional flow (negative θ contribution).

In the previous sections, the measurement results indicate that E in the SDH is largely controlled by small-scale local circulations during the night and morning. This E pattern changes in the afternoon by the formation and arrival of regional mesoscale circulations. In order to better quantify how this circulation influences E at the SDH, we analyse WRF model results of the atmospheric conditions surrounding the SDH, using the regional-scale model WRF. We focus on two issues. The first concerns evaluating whether our measurements are influenced by small flows from katabatic–anabatic effects that dominate night-time and morning boundary layer in the absence of strong local or regional forcing. The second and more important one is the quantification of insights into the mechanism that generates the strong winds in the afternoon and produce the enhancement of E.

The local conditions that dominate the E in the SDH are analysed in Fig. , which depicts the wind flow and temperature in the study site at 07:00 LT calculated with a grid resolution of 1 km, i.e. an effective resolution of approximately 3 km. The circulation is characterized by a downward flow from the surrounding mountains (z>4500 ma.s.l.) around the lowlands (z≈3800) where the saline lake is located, which tends to accelerate over pronounced slopes and closely follows the shape of the terrain. However, the lowest temperatures shown at the bottom of the valley in our observations (Fig. ) are less clearly recognizable in the model (Fig. a), where low temperatures occur in the surroundings of the lake. This downslope flow is responsible for the stratified layers observed over the water surface at 09:00 LT (Fig. c). This katabatic flow progressively decreases in the course of the morning, whereas a transition from a stable to well-mixed layer occurs above the water from 09:00 to 12:00 LT (Fig. c). Here, we observe two processes that are responsible for the local circulation and the low surface fluxes over the water during the morning (Fig. a). The first, between 09:00–10:00 LT, is an anabatic radial flow from the lake to its surroundings (Fig. b). The second one is a downward flow produced by the interaction between the anabatic flow with the thermally driven wind during the morning–afternoon transition. This flow shown in Fig. 9c produces a compensated subsidence at the western margin of the SDH valley, which explains the morning stratification over the lake shown in Fig. c.

To characterize the recurrence of the wind pattern and its robustness at larger spatial scales, Fig.  shows averages over 10 d (E-DATA period) of zonal wind speed (U), temperature (T), and specific humidity (q) in the morning (10:00 LT) and afternoon (16:00 LT) in a SW–NE vertical cross section of the Andes Mountains obtained by the WRF model. In the morning we identify two main zones with clear U, T and q conditions. The first corresponds to the coast (z<1 km) over the ocean (70.3∘ W), where the marine boundary layer (MBL) is characterized by low westerly winds of 2 ms-1 (Fig. a), a thermal inversion capping at ∼1 km height (Fig. b), and a quite-well-mixed MBL with a moisture ranging between 7 and 10 gkg-1 (Fig. c). The second zone corresponds to the western slope of the Andes (70.0 to 68.5∘ W) above z>1 km. This zone presents a very low U (∼1 ms-1) that increases to 2 ms-1 at the surface upslope, producing a small local circulation in the SDH basin (see red square in Fig. a). Likewise, there is a thermal contrast between the land and the top of the MBL (5 K) and incipient heating in the surface (70.0∘ W) together with a vertical thermal stratification of the atmosphere of 0.6 K per 100 m. Finally, low values of moisture are observed at middle altitude lands (∼4 gkg-1), with a variation ∼-1 gkg-1 per kilometre ascended on the slope (Fig. c).

During the afternoon, the morning conditions rapidly intensify. The U increases at the surface >10 ms-1) along the slope, with a steep variation in its vertical profile, i.e. the weakest zonal winds are between 2 and 4 kma.s.l. (∼1 ms-1). Above ∼4 kma.s.l., typical synoptic southwesterly winds are found with speeds around 5 ms-1 (Fig. d). The thermal contrast between the MBL top and the inland desert surface increases up to 10 K in the afternoon, in association with intense land warming. This strong wind circulation is characterized by higher values of the specific humidity (4–6 gkg-1) from the top MBL (z: 1–2 km and longitude 70.3∘ W) along the side of the Andes slope (Fig. f). This strong advection follows two paths: one reaches the SDH and increases the specific humidity from 1 to 3.5 gkg-1, and the other one returns back westward at ∼2 kma.s.l. Two additional mechanisms on the western slope of the Andes that enhance the surface wind flow are also reproduced by the numerical experiment in WRF. The first mechanism is an anabatic flow formed at the midlands (70.0∘ W) driven by the high sensible heat fluxes, which corresponds to 73 % of Rn. The second mechanism that is superimposed on the anabatic flow is a surface flow acceleration along the slope, which we recognize as flow channelling. This channelling is given by the shape of the topography and the subsidence produced by the SE subtropical anticyclone over the SE Pacific Ocean and the western slope of the Andes. The flow is then channelled down into the SDH basin from the SW, producing local subsidence (Fig. c). In summary, the origin of the strong wind that controls the evaporation in the Salar del Huasco originates in the regional daily atmospheric circulation from above the MBL to the Atacama Desert.