To assess different methods that estimate Z_0C or Z_SL, we used photos collected in the frame of the citizen science programme “Vecinos de las Nieves” (CEAZA, 2021). In April 2021, we requested photos of past and current precipitation events via social networks (Figure 3A), asking for the following information: location, name of the mountain, direction of the camera, date, and hour. By sending the photo, the participants agreed that the photo could be used for scientific purposes. Nearly 70 participants sent photos taken since 2011. In addition, we gathered photos from social networks (Facebook, Twitter, and Flickr). In total, 111 photographs were selected from 28 events over 11 years (2011–2021). All the photos (Supplementary Figure S6) and the details (Supplementary Table S2) can be found in the supplementary information. We selected the photos that fulfilled the following criteria:

- Date, hour, and location of the photo are available and can be verified.

The aim of the photograph analysis is to derive a representative value of Z_SL per photo. To do so, three representative points on the mountain slope were drawn on the photo at the lowermost elevation with snow occurrence (Figure 3B). The photo was then compared to isolines in the Google Earth software to estimate the elevation of the selected points using geological features as reference (Figure 3C). The mean altitude of the three points was then calculated. Topographic isolines have a vertical distance of 12.5 m and were computed based on the Advanced Land Observation Satellite (ALOS) Digital Elevation Model (DEM) (PALSAR 2011, from Japan Aerospace Exploration Agency, JAXA, 2011).

This method might be inappropriate in cases where new snow lies above the lower extent of the current snow cover. However, this did not occur in any of the photos. We expect that such events are infrequent in this arid region, mainly because precipitation events are often separated by several days or weeks and because snow melts/sublimates quickly at low altitudes (Réveillet et al., 2020).

In the presented methodology, we compared the snowline altitude from the photos to the minimum Z_0C during an event. The photos from two precipitation events of 8–10 August 2015 and 10–12 May 2017, both of which caused floods and mass movements in Vicuña and its surroundings (Palma, 2019), were not considered in the following analyses because air temperature significantly increased during the last hours of the event, probably causing rain on snow and leaving a significantly higher Z_SL compared to the minimum Z_0C during the event (Supplementary Figure S4 in the supplementary information). It seems thus that for events with increasing air temperatures during the last hours, the minimum Z_0C is not a good estimator for the final Z_SL after the event. For such events, the Z_0C during the last relevant precipitation hours would be a better estimate. Other precipitation events with increasing air temperatures have been observed during the study period, but the increase only occurred during the final hours of the events and was accompanied by low precipitation intensities.

In addition, we excluded the photo from the event on 23–24 June 2020, where the snowline from the photo clearly lay almost 600 m above the minimum Z_0C from nearby stations. The closest station, station 46 at a distance of 9 km, showed a clear drop in air temperature, which probably did not occur at the photo site. For example, the nearby station 49 did not show this drop in air temperature. This case shows that a temperature drop at an individual station can lead to errors in the method.

We used the snow cover area maps derived from Sentinel-2 images at a spatial resolution of ca. 55 m. The maps are automatically derived by the Sentinel-2 Level 2A processing using the Normalized Difference Snow Index (NDSI) (European Space Agency, 2022). To estimate Z_SL, we used the approach presented by Gascoin et al. (2019). We considered the area of the Elqui River basin, fitting a rectangle to the catchment area. We then selected all available snow cover area maps <5 days after precipitation events 2016–2021, resulting in ten images. A DEM from the Shuttle Radar Topography Mission (SRTM) was used to segment the satellite image in elevation bands with a fixed height of 100 m. Then, the fraction of the area covered by snow in each band was computed. The simple algorithm proposed by Gascoin et al. (2019) finds the lowest elevation band at which the snow cover fraction is greater than a given fraction fs selected as 2%. The value of Z_SL is then the lower edge of that elevation band. It should be noted that Gascoin et al. (2019) selected the fraction fs of 10% and then defined the snowline in two elevation bands below that band. Here, we found that our criteria lead to a slightly higher correlation between the snowline altitude from satellites and photos. In total, we could verify the Z_SL of 22 photos during 10 events.

The Coquimbo Region contains a dense network of automatic meteorological stations maintained by the CEAZA research institute (Supplementary Table S1). Air temperature, precipitation and other variables are recorded at 10 min intervals and then averaged to hourly data. Most stations were installed between 2004 and 2010, and their elevations range between sea level and 4,774 m asl. Precipitation and air temperature data were used to estimate Z_0C through three different traditional methods and for the period 2011–2021.

The first method (M1, Figure 1B) is based on data from a single meteorological station and the assumption that temperature decreases with altitude following a certain constant lapse rate (Eq. 1). We evaluate different typical free air temperature lapse rates under moisture saturated conditions, ranging from −5 to −6.5°C km⁻¹ (Blandford et al., 2008; Minder et al., 2010; Garreaud, 2013; González and Garreaud, 2019; Ibañez et al., 2020; Mardones and Garreaud, 2020). In addition, in hydrological models, a threshold temperature T_SL of around 1–2°C is usually used to discriminate between snow and rain at certain elevation bands (Schaefli et al., 2005, Eq. 2). This threshold temperature corresponds to the air temperature at which snowflakes can reach the ground below Z_0C without melting (at the snowline) and therefore, in theory, relates to the offset Z_D between Z_0C and Z_SL, see Figure 1. The air temperature at the altitude of the snowline is, however, substantially uncertain and depends on a variety of variables such as humidity, precipitation rate, soil temperature or vegetation type. The equations are as follows: Z 0 C = Z S t − T S t LR (1) and Z S L = Z 0 C − Z D   = Z S t − T S t − T S L LR , (2) where Z_0C is the altitude of the 0°C isotherm, Z_SL is the snowline altitude, Z_St and T_St are station elevation and air temperature, T_SL is the threshold air temperature at the snowline altitude, and LR is the air temperature lapse rate.

In detail, the following steps were performed to compare Z_0C from Eq. 1 to Z_SL from photos taken within a certain time after precipitation end (T₁): For every station within a certain distance (D_st) to the photo, Z_0C was estimated with Eq. 1 and different LRs (−5, −5.5, −6, and −6.5°C km⁻¹) for all rainy hours (≥0.2 mm h⁻¹) within a defined time window before the photo capture (T₂). From the rainy hours, we then selected the hour with the lowest Z_0C, assuming that this value is most decisive for the final snowline observed in the photo. For each photo and LR, we derived n values of Z_0C depending on the number of stations within the searching distance. For each LR, the mean Z_0C of the n values was compared to the Z_SL for all photos. From all these data pairs per LR, mean offset, standard deviation (SD), and coefficient of determination (R ² ) were computed. The highest R ² values indicate the most suitable LRs, and the coefficient of determination is an indicator of the uncertainty. The mean offset corresponds in part to the difference Z_D between Z_0C and Z_SL, even though it might also be influenced by local conditions at the station or nearby the snowline and needs careful evaluation. Nevertheless, this offset was used to estimate T _SL (T _SL = Z _D ‧LR), which is required for Eq. 2. This equation was then applied to stations within 50 km with the aim to better understand the behavior and representativeness of different stations to estimate Z_SL with M1 in the function of distance and elevation (Section 4.2.2).

The following decisions were made for the method M1: 1) D_St: Stations within a 10 km distance to the photo location were selected. If the distance is shorter, a considerable number of photos is not close enough to stations and needs to be excluded. If the distance is larger, the representativity of the stations decreases. 2) T₁: We included only photos taken ≤12 h after precipitation end. Some of the selected photos were taken up to 24 h after precipitation end and showed, in some cases, a relatively high snowline compared to Z_0C since snow might have melted/sublimated already. If we include photos taken up to 48 h after precipitation end, there might be a positive bias in snowline altitude estimates. 3) T₂: The time window is 36 h. A simple analysis has shown that the time window does not influence much the results. The window only defines which hours are included to search for the coldest wet hour, and in most cases, changing the window does not affect the value for the coldest wet hour.

The second method (M2, Figure 1C) is based on a linear regression of hourly air temperature records for stations at different altitudes. For both study regions, CEAZA stations were selected according to their elevation, neglecting stations below 500 m asl. or close to the coast (<10 km) since these stations might be influenced by the maritime climate (see green stations in Figure 2B). To assess the sensitivity to high elevation stations (such as described by Ibáñez et al., 2020), we estimated Z_0C considering all stations (M2) and stations below 2,400 m asl. (M2-low). Then, hourly temperature lapse rates and Z_0C were estimated by fitting a linear regression to observed air temperature from all selected stations. Only the hours with at least three stations recording precipitation (≥0.2 mm h⁻¹) and with R ² ≥ 0.5 were considered.

The third method (M3, Figure 1D) estimates Z_0C from the linear interpolation of air temperature recorded at the highest and lowest stations recording positive and negative air temperatures, respectively. Z_0C is computed separately for Elqui and Limarí/Choapa.

For the methods M2 and M3, we used stations at a certain distance (D_St) to identify hours with at least one station recording precipitation (≥0.2 mm h⁻¹) and then selected the hour with the lowest Z_0C. The following decisions were made: 1) the search distance D_St is 25 km. As in the method M1, we considered 2) only photos taken ≤12 h after precipitation end (T₁) and 3) the time window is 36 h (T₂).

Hourly Z_0C values were retrieved from ERA5, the most recent reanalysis product of the European Centre for Medium-Range Weather Forecasts (ECMWF), with an hourly temporal resolution and at a spatial resolution of ∼31 km (Copernicus Climate Change Service, 2017). Z_0C is directly downloadable, and there is no need for interpolation of air temperature at different pressure levels. Since the variable from ERA5 is given in m above ground, it must be added to the ERA5 orography to get Z_0C in meters above sea level.

Although the radiosonde data at the launching point are assimilated in ERA5 (Hersbach et al., 2020), there might be systematic biases between the two data sets that need to be considered (Virman et al., 2021). With the aim to assess the representativity of ERA5 data over the coast, we first compared Z_0C from ERA5 to Z_0C from the radiosonde data. The closest upper air sounding data are available at Santo Domingo (33.6547°S, 71.6144°W, 73 m asl.), about 170 km away from the southern edge of the Coquimbo Region. Radiosondes are released twice a day (00 and 12 UTC), and data are provided by the Integrated Global Radiosonde Archive (IGRA). We used freezing height from the radiosonde data (FRZHGT), defined as the height above the surface where the temperature first reaches the freezing point when the balloon moves upward from the surface. We then compared radiosonde freezing height to Z_0C from ERA5 at the same time. The drift was not considered, since data of the maximum balloon drift distance for a 2 year period (July 2007–June 2009) is less than 14 km (DriftPlotter software, Seidel et al., 2011). The SYNOP messages from the same location (Santo Domingo, 33.39°S, 71.37°W, 75 m) were used to check if it was raining or not during the sounding launch. We found that the correlation between Z_0C from ERA5 and radiosonde data is very high, both for all days (R ² = 0.95 for 00 UTC and R ² = 0.96 for 12 UTC) and for rainy conditions (R ² = 0.95, mean offset = 75 m), Supplementary Figure S1 in the supplementary information. This result is in line with a notable correlation between daily averages of Z_0C from the Climate Forecast System Reanalysis (CFSR) and upper-air sounding at Santo Domingo (Mardones and Garreaud, 2020). We, hence, assumed that ERA5 also represents well the upper air conditions over the coastal zone of the Coquimbo Region.

Note that the spatial resolution of ERA5 is about 31 km, which is too coarse to resolve the complex topography over the Andes (see Figure 11A). If Z_0C (given in height above ground) is below the model surface, the value of Z_0C is constrained by the topography. This leads to unrealistic West–East profiles of Z_0C. We, therefore, decided to take the upwind value (50 km west) of the photo location for a more robust comparison. For the comparison of Z_0C during rainy hours (≥0.2 mm h⁻¹) from ERA5 with Z_SL from photos, we did the same as described in M2 and M3 and only considered hours when at least one nearby station (within 25 km) indicates rainy conditions.

To explore the mesoscale lowering of Z_0C, we first assessed West–East cross-sections of extrapolated Z_0C from the coast to the slopes of the Andes along the Elqui River basin, together with profiles from ERA5. On the slopes of the Andes, we estimated Z_0C using four mountain stations (stations 34, 41, 44, and 47) applying the method M1 with a representative moist adiabatic temperature lapse rate of −6.5°C km⁻¹. On the coast, we estimated Z_0C using hourly data of the two surface stations (stations 15 and 17) and wet temperature lapse rates calculated with radiosonde data launched at the Santo Domingo station. For this site, we estimated LRs with linear regression from the surface to the first level with negative air temperatures, obtaining LRs for 250 cases (Supplementary Figure S2 in the supplementary information). The SYNOP messages from the radiosonde station were used to check if it was raining or not during the sounding launch (00 and 12 UTC).

To create the cross-sections, we only selected days with rainy radiosonde records (from 2 days before or 1 day after) and with at least five of the six mentioned stations along the Elqui catchments indicating wet conditions (>2 mm day⁻¹). For two precipitation events that last more than one calendar day (11–12 May 2017 and 12–13 July 2015), we manually selected the day with more significant precipitation intensities and where we, therefore, expect the lowering (12 May 2017 and 12 July 2015), which results in 11 study events. For the same dates, we plotted cross-sections of ERA5 data for the latitude of −29.95°S for the hour when the air temperature was lowest at the station Nr. 17.

Assuming lapse rates from a somewhat distant radiosonde release and using a unique value of the temperature lapse rate in the method M1 may lead to uncertainties, but this approach uses only observations and will give us a first look at this mesoscale atmospheric phenomenon. To account for uncertainties related to the assumption of constant lapse rates for the mountain area, we also assessed the cross-sections for a lapse rate of −5°C km⁻¹.

Since we found that ERA5 adequately represents Z_0C for rainy conditions at the coast in Santo Domingo (Supplementary Figure S1), we assumed that ERA5 also adequately represents upwind Z_0C in the Coquimbo Region. We, therefore, assumed that the comparison between ERA5 and Z_0C from M2 might indicate lowering. For this, we compared the daily minimum value of Z_0C from ERA5 at the coast (upwind of the Andes) with Z_0C from linear regression (method M2) on the upslope zone of the Andes range, which were first assessed with photographic evidence. This analysis was performed for Elqui and Limarí/Choapa separately, selecting ERA5 grid points −30.2°S, −71.25°W and −31.2°S, −71.25°W, respectively. We only considered hours where a representative reference station indicates rainy hours (reference station 44 for Elqui and station 39 for the Limarí/Choapa). To this comparison, we added daily mean precipitation to evaluate if precipitation intensity may influence the lowering of Z_0C over the mountains.

For every photo, we plotted the temporal evolution of hourly precipitation and Z_0C from M1 (from stations within 25 km and the coastal station Nr. 17 and 6 for Elqui and Limarí/Choapa, respectively), M2, M2-low, M3, and ERA5 during the 48-h prior the photo capture. In Figures 8A,B, we present two case studies (13 July 2015 and 14 October 2015, respectively) for events with snow observed in Pisco Elqui (∼1,200 m asl.) and with some types of precipitation triggered hazards in Vicuña and surroundings (Palma, 2019). The selection represents similar findings as for the other events that are not shown here.

The meteorological synoptic conditions associated with the two case studies are illustrated in Supplementary Figure S5 in the supplementary information. Both events were produced by cold fronts and their associated closed-low or trough pressure systems and moisture plumes with strong water vapor transported from the Pacific Ocean. The first event (12–13 July 2015) had stronger synoptic forcings, with lower pressure depression and stronger moisture transports, and produced higher precipitation accumulation than those of the second event 2 (14 October 2015). Although both events had different precipitation and synoptic forcing intensities, both represent typical synoptic conditions with frontal systems and Pacific-originated moisture plumes reaching the subtropical coastal latitudes of the Coquimbo Region (∼30°S).

In Figure 8, we can observe a temperature decline during the two events for almost all meteorological stations located in valleys (especially 39 and 44), leading to temporal drops in Z_0C from M1. The drop generally appears around the hours with the highest hourly precipitation intensities and seems less pronounced for events with low hourly intensities. In all events with snow in Vicuña, this drop appears, and this temporal drop is less clear or not visible in Z_0C from M1 applied to the coastal stations. It probably occurs due to mesoscale air cooling on the upslope sector of the Andes caused mainly by diabatic cooling from the melting of orographically enhanced precipitation rates and adiabatic cooling for the upslope ascending of the air (Minder et al., 2011). But more studies are needed to identify and quantify the processes that lead to this drop in air temperature.

Z_0C from linear regression and low altitude stations (<2,400 m asl., M2-low) shows similar minimum values as Z_0C from M1 applied to valley stations. Z_0C from linear regression, including high altitude stations, is generally higher, and the drop of Z_0C is, in most cases, less pronounced. The drop is also visible in the timeline of Z_0C from M3. In contrast, ERA5 reanalysis data indicate higher Z_0C values compared to M1 and M2 and the temporal drop of Z_0C is not clearly indicated.

Even if the M1 method seems suitable in most cases for nearby stations (Figures 6A-D), uncertainties are expected to increase for meteorological stations at larger distances or for certain stations due to non-elevational factors. But there is a lack of understanding about the representativeness of stations depending on their characteristics, such as elevation or surrounding topography. Also, the magnitude of the uncertainty in the function of these characteristics is not well known. Therefore, we applied a lapse rate of −6.5°C km⁻¹ and the estimated offset Z_D from Figure 6D to stations up to 50 km away from the photo using Eq. 2. Figure 9 shows the difference between Z_SL from M1 and from photos, in the function of 1) distance between meteorological stations and photos and 2) station elevation.

In Figure 9A, we can observe that the Z_SL from M1 corresponds well to Z_SL from photos for close stations <10 km. The difference between the estimated and observed Z_SL, however, significantly increases with a larger distance between the meteorological station to the photo. For stations >10 km away from the photographic evidence, the estimated Z_SL lies in many cases more than 200 m above or below the observed Z_SL, which is considerable for hydrological applications.

Uncertainty also depends on station elevation (Figure 9B). As already observed in the study cases (Figure 8), low coastal stations with elevations up to ∼500–700 m asl. tend to overestimate Z_SL. The most representative stations seem to be located at ∼700–1,500 m asl. with most of the errors within ±200 m (see the interquartile range and relatively small median error in Figure 9B). These stations are located below the typical snow–rain transition zone. Comparatively, the spread of the uncertainty probably increases for higher stations (see whiskers for standard deviation), especially for stations >1,000 m asl. The study cases in Figure 8 (and more study cases not shown here) indicate that there is a strong temporal drop in Z_0C for these stations. The relatively low temperatures in specific valleys and events might thus be accurate to estimate Z_SL in the same valleys, but not for other sites. The relatively high-altitude stations 47 (1,696 m asl.) and 49 (2,500 m asl.) tend to overestimate Z_SL. The highest station TASC (3,427 m asl.)—located clearly above the snow–rain transition zone in most cases and at large distances—probably lies too high or far for representative Z_0C estimations.

We found that applying other lapse rates (−5, −5.5, and −6°C km⁻¹) and corresponding offsets from Figures 6A–C does not change the overall pattern of the results in Figure 9 (not shown here). Our findings confirm that uniform lapse rates applied to representative and nearby (<10 km) stations might be suitable to estimate the local snowline altitude for rainy conditions if the station is located below the snow–rain transition. In Chile, wet-weather SLTR values from −5.8 to −6.2°C km⁻¹ were found for the western subtropical Andes (Ibañez et al., 2020) and ∼−6.5°C km⁻¹ for the coastal mountains (González and Garreaud, 2019). Even if the assumption of a uniform and constant surface lapse rate of −6.5°C km⁻¹ is a poor one for annual mean lapse rates (Gardner and Sharp, 2009; Minder et al., 2010), we can confirm the recommendation by Ibañez et al. (2020) of using a value between −5.5 and −6.5°C km⁻¹ to estimate minimum Z_0C during rainy conditions.

In Figures 6E and F, we observed that a linear regression method using surface stations at different elevations seems accurate, but the result strongly depends on the inclusion of high-elevation stations. If only stations in valley bottoms and at low altitudes (<2,400 m asl.) are considered, the extrapolated Z_0C lies much lower, and in many cases, the snowline lies above Z_0C. This might indicate that the method is not very precise in these cases. To understand the elevational distribution of air temperature during precipitation events, we examined the daily minimum air temperature during the six snow events in the period 2014–2021 in Pisco Elqui (Figures 10A-F). The minimum temperature is recorded at almost the same hour for all stations during these days. LR and Z_0C are then calculated through the M2 method with linear regression for stations above and below 1,500 m asl. For all six events, Z_0C estimated with high elevation stations lies clearly above Z_0C from low elevation stations and LRs of the higher zone are less steep compared to the lower elevations. Hence, there seems to be a break in LR around the snow–rain transition, together with an offset of up to 430 m. Our findings confirm that stations from a wide range of elevations (especially high elevations) are crucial, such as suggested by Lute and Abatzoglou (2020) or Ibañez et al. (2020).

The difference between the two Z_0C values could be interpreted as a zone of an air temperature of near 0°C, called the isothermal layer (Findeisen, 1940; Unterstrasser and Zängl, 2006), which occurs if air mass in the valley becomes decoupled from the environmental flow. In this case, the latent heat required for melting the falling snow is continuously removed from the valley, thereby lowering the snow–rain transition and increasing Z_D. The formation of an isothermal layer of similar magnitude has been observed and simulated for other mountainous regions (Thériault et al., 2012). This process could explain a drop in air temperature of stations in steep valleys such as observed in the study cases (Figure 8), for example, Pisco Elqui (station 44). An isothermal layer might also explain why M2 leads to higher Z_0C values if high elevation stations above the isothermal layer are included. More research is needed to study the formation of isothermal layers during storm events in this region. It is also important to note that near-surface air temperature might also be altered by other local conditions. Further research could focus on the sensitivity of estimated lapse rates to single stations.

The interpolation technique using two stations (one below and one above Z_0C) is accurate but relies on high elevation stations above the typical range of the snow–rain transition. In some cases, there is an air temperature inversion with positive air temperatures at higher elevations than negative air temperatures below. In these cases, the linear interpolation between air temperature recorded at two stations are not accurate. Similar as in M2-low, the comparison with the snowline from photos has shown many cases where the snowline lies above Z_0C. This indicates that a local cooling of the atmosphere could lead to temperature drops at certain stations, which then lead to lower Z_0C estimations that are not representative of the conditions of nearby valleys. But we need further studies to better understand these cases. We suggest that this method should only be applied if reliable stations are available and if local effects of temperature variability are well understood.

The cross-barrier plot in Figure 11A shows that the 11 events present a lowering of extrapolated Z_0C toward the Cordillera, with values up to 766 m. Assuming that a constant lapse rate of −6.5 C km⁻¹ is associated with important uncertainties in the estimation of the lowering, there might also be cases where conditions at low elevations are dry and saturated aloft, leading to different LRs. To account for these uncertainties, we did the same cross-sections assuming a lapse rate of −5°C km⁻¹ for the mountain stations (34, 41, 44, and 47), which leads to a less pronounced lowering of up to 487 m (not shown here). Still, 7 of 11 study cases present a decrease of Z_0C over the mountain area compared to the coast. In contrast, cross-sections of ERA5 data do not show the lowering of Z_0C. Figure 11A also shows the topography of ERA5, which seems coarse for this complex topography lying some hundreds of meters above the location of the meteorological stations.

Figure 11B suggests that the differences in the mountainside and upwind Z_0C may accentuate with higher precipitation intensities, but more data are needed to further investigate this relationship. A dependency on precipitation intensity would be in line with observations (Marwitz 1983) and analytical models (Unterstrasser and Zängl, 2006) that have been used to understand how higher precipitation intensities lead to more latent cooling from melting precipitation and deeper isothermal 0°C layers.

An alternative interpretation of this result is that above the location of valley stations, an isothermal layer forms with temperatures slightly above 0°C and with very low air temperature gradients within that layer (Figure 1, Findeisen, 1940; Unterstasser and Zaengl, 2006), which is supported by the findings in Figure 10. At stations 44 and 47, temperatures slightly above 0°C are recorded for most precipitation events, despite the station elevation difference of 456 m. When we apply constant lapse rates of, for example, −6.5°C km⁻¹ to these low air temperatures, the extrapolated Z_0C lies a few 100 m above ground at both locations. However, if there is an isothermal layer with air temperatures slightly above 0°C and Z_0C at its top, a constant LR of, for example, −6.5°C km⁻¹ is only valid up to the bottom of the isothermal layer, and the top of the isothermal layer lies at a much higher altitude. This would then result in a much less significant lowering of Z_0C over the mountains. If this is true, we underestimate Z_0C significantly using ground stations located below an isothermal layer and standard LR, even if Z_SL estimations seem accurate (Figure 6).

Based on this analysis, we cannot conclude whether this drop in extrapolated Z_0C is due to a mesoscale lowering of Z_0C or an isothermal layer, or both. In any case, it indicates that during precipitation events, there are either mesoscale lowering in Z_0C or locally changed lapse rates due to isothermal layers that need to be considered to estimate the snow–rain transition altitude, especially if we base the estimation on radiosonde stations from the coast or reanalysis data. Due to these effects, the local snow–rain transition seems to be much lower than estimated based on the conditions at the coast, although an offset could be added to account for the lowering or the isothermal layer.

In Section 4.3.1, we were unable to quantify the mesoscale drop in Z_0C, as constant LRs (M1) might underestimate Z_0C if an isothermal layer forms. In contrast, Z_0C estimates using linear regression (M2) with in situ stations might lead to more robust Z_0C estimates. Since we found that ERA5 Z_0C is well correlated with radiosonde records, we assume that these data may also be representative along the coast of the Coquimbo Region. A large difference between ERA5 and mountainside Z_0C might therefore indicate a mesoscale lowering of Z_0C over the Cordillera.

Figure 12 shows that the two values (Z_0C from M2 representing the mountainside and ERA5 data representing the coast) are well correlated (R ² = 0.69 and 0.66 for Elqui and Limarí/Choapa), but there is an offset in the range of 150–160 m, which may indicate a mesoscale lowering. For the snow events in Pisco Elqui (see snowflake symbols in Figure 12A), the offset even increases to 297 m. The results are in line with observations from the Sierra Nevada, where an average drop of the snow–rain transition of 170 m was identified using 3 years of profiling radar observations over the windward slopes (Minder and Kingsmill, 2013). For example, Thériault et al. (2015) have shown using WRF model simulations that Z_0C decreased by 400 m within 60 min until reaching the ground in the Pacific Coast Ranges. The difference between the two regions might be due to the different properties of the incoming flow or topography, which all have effects on the mesoscale lowering (Minder et al., 2011). Since the lowering is the result of a complex interplay between humidity, wind speed, stratification, air temperature, and topography, more research is needed to understand regional differences along the Chilean Andes.