Untangling the importance of dynamic and thermodynamic drivers for wet and dry spells across the Tropical Andes

Rainfall seasonality in the Tropical Andes (0–25^∘S) is strongly controlled by the inland progression and retreat of the South American monsoon (SAM) system, affecting the timing of moisture transport into the Andes (e.g. Espinoza et al 2020), the majority of which originates from the Amazon basin (Trachte 2018). Locally, intra- and inter-annual SAM variability can crucially control water availability, affecting vegetation greening (Hänchen et al 2022), and increasing perceived challenges for small-scale farmers to plan crop cycles (Gurgiser et al 2016). Across the Andes, local precipitation characteristics vary considerably due to a multitude of local- to large-scale drivers and their interaction with topography (Garreaud and Aceituno 2001, Laraque et al 2007, Junquas et al 2018). In that context, the variability of the upper-level wind field has been identified as a major dynamical factor in controlling wet and dry years in particular over the Altiplano in the Southern Tropical Andes (∼14^∘–22^∘S), where the SAM reaches its southernmost extent and rainfall peaks in December to February (DJF). In that region and for the DJF peak rainy season, various studies illustrated enhanced upper-level (200 hPa) easterly wind to be consistently associated with more rainfall, linked to the development of the Bolivian high (BH) at that time of year (Vuille 1999, Garreaud et al 2003, Vuille and Keimig 2004, Thibeault et al 2012).

This significant inter-annual rainfall relationship with upper-level wind has subsequently been used as basis for predictions of potential future rainfall changes in the Southern Tropical Andes, with suggested future drying driven by a weakening of 200 hPa easterly wind anomalies in climate model projections (Minvielle and Garreaud 2011, Neukom et al 2015). At the same time, Segura et al (2020) have recently reported that the Southern Tropical Andes' rainfall relationship to upper-level wind conditions has weakened over the last two decades, while decreased atmospheric stability from moister mid-levels in the western Amazon gained of importance. Analysing four decades of reanalysis data, they suggest the latter thermodynamic driver to be independent from the dynamical 200 hPa wind effects and that it has been driving a positive rainfall trend in the region since the 2000s. Their findings thus imply that upper-level easterly winds are a favouring factor for rainfall in the Southern Tropical Andes, though not a sufficient condition in the face of moisture changes.

These results for the Southern Tropical Andes raise the question as to how rainfall sensitivities in the Central and Equatorial Tropical Andes compare to Southern conditions. Looking at drivers of wet and dry spells in the Mantaro Basin in the Central Tropical Andes, Sulca et al (2016) confirmed a relationship between upper-level easterly anomalies and rainfall as in Southern regions, but noted that even sustained strong upper-level easterlies often produce no or little rainfall. They concluded that moisture transport processes from the Amazon basin would have to be explored in more detail as additional wet spell driver. Thus, it remains an open question how upper-level wind as a dynamical driver and Amazon basin moisture as a thermodynamical driver interact to affect rainfall in the Andes.

The study aims to provide an integrated picture of the described regional findings by evaluating the relative importance of upper-level wind versus Amazon basin moisture changes for three to seven day dry and wet spells across the entire Tropical Andes mountain range. We investigate the circulation patterns that coincide with these spells during the regional rainy seasons between SAM passage and retreat, and quantify the dynamic and thermodynamic driver contribution to rainfall in the Equatorial (0^∘–5^∘S), Transition (5^∘–8^∘S), Central (8^∘–12^∘S) and Southern Tropical Andes (15^∘–18^∘S) above 2000 m altitude. Finally, we consider the effect on soil moisture (SM) and vegetation to better understand potential impacts from the evaluated spells. This study thus provides the first comparison of rainfall drivers for the entirety of the Tropical Andes and of the regional effect of multi-day spells on the ground.

2.1. Atmospheric conditions

2.2. Local DRS

2.3. Wet and dry spells

2.4. Vegetation response

Focusing on the dynamical driver, we first evaluate the pattern of the identified local DRSs across our full analysis domain. Figures 1(a) and (b) reveals significant seasonally-dependent cloud/wind relationships over the full latitudinal extent of the Tropical Andes (0^∘–25^∘S). In contrast, we find no significant relationship in the adjacent Amazon lowlands (figure 1(b)), suggesting that the sensitivity to the upper-level flow is coupled to orographic effects and induced circulations. Considering seasonal timing, figures 1(a) and (b) reveals a strong north–south gradient in the timing and length of correlation periods, in line with the seasonal movement of the SAM. While regions nearest to the equator show significant cloud/wind co-variability over ten months, with a break in July–August, the relationship is only evident for ∼5 months from November to March in the Southern region. These results illustrate that, while seasonally varying, the rainfall sensitivity to the 200 hPa wind dynamical driver exists across the Tropical Andes region, albeit specifically confined to the mountain range and its vicinity. The identified CTT/wind relationship is exclusively positive, i.e. easterly wind anomalies drive colder clouds (figure 1(c)). In addition, the relationship is strongest in the western Andes and weakens towards the moister eastern slopes, potentially reflecting increased importance of this dynamical driver to allow moisture transport to the drier western Andes. Given our focus on drivers for Andes rainfall, all following analyses only consider pixels with significant upper-level wind correlation that lie above 2000 m altitude (shaded pixels within black contours in figures 1(b) and (c)).

Zoom In Zoom Out Reset image size

Next, to explore the relation of the 200 hPa easterly wind inland progression and SAM rainfall seasonality over the Andes, we consider four different latitudinal regions across our Andes domain (figure 1(c)). These are characterised by a temporally bimodal rainfall seasonality in the Equatorial and Transition regions, and a unimodal season in the Central and Southern regions (e.g. Segura et al 2020). This comparison intentionally only covers the regions that experience mean upper-level easterly winds at some point in the seasonal cycle (figure 1(a), dark blue shading), which reach as far south as 18^∘S, imposed by the anticyclonic circulation of the well-known BH during the peak monsoon (December–March). However, we find that an anticyclonic circulation can in fact be identified for all months except May–July, positioned in the central Amazon during the transition months before strengthening and moving south-westward to form the BH (cf supplementary file figure 3). Mean upper-level easterly winds over the Andes move with this circulation over the seasonal cycle, and relative maximum rainfall over the Andes occurs co-located with the southernmost boundary of the easterlies (figure 1(a), blue line). This seasonal movement of easterly wind and rainfall is fully contained within our local DRS periods for all regions (figure 1(a), light blue shading), with a break in August when rainfall stays close to the Equator.

We thus showed that we can define a DRS based on CTT sensitivity to upper-level wind conditions, which encompasses the easterly wind and SAM rainfall progression over the Andes (figure 1(a)). Easterly wind anomalies are associated with colder CTTs across all regions, which translates into increased rainy day probabilities (figure 1(d)). However, while from the Equatorial to Central Andes region the mean wind direction during the DRS is easterly, it is on average westerly in the Southern region. Indeed, 17% of equatorial and 42% of southern rainy pixels occur under 200 hPa westerly wind conditions (figure 1(e)), illustrating that prevailing upper-level easterly winds favour rainy days in all regions, but are not a necessary nor a sufficient condition.

We now investigate the differences in upper-level wind and Amazon basin moisture conditions between three day wet and dry spells for the local DRSs as identified in figure 1. We first focus on these shortest and most frequent spells, assuming that this allows to detect shorter- as well as longer-lived spell drivers. Figures 2(a) and (b) show the differences between wet and dry spell conditions, composited on the first spell day (day₀) at 1000 LT to capture the atmospheric conditions that facilitate spell initiation, aiming to reduce anomaly patterns that are a consequence of rather than a cause for spell development.

Zoom In Zoom Out Reset image size

The spatial differences between day₀ wet and dry spell conditions in figure 2(a) reveal similar patterns across the regions: During wet spells, 850 hPa northerly wind and atmospheric moisture are significantly increased along the eastern slopes of the Andes bordering the respective spell region. Notably, there is no dominant large-scale low-level moisture increase in the Amazon basin in any of the regions, suggesting that the positive lowland moisture anomalies preceding Andes wet spells are critically fostered by the low-level flow encountering the topographic barrier. In the Central and Southern regions, this is accompanied by a tendency for 200 hPa easterly winds reaching further south, while there is little to no difference in easterly wind position for the Equatorial and Transition regions. However, the regional cross-sections in figure 2(b) show significantly enhanced upper-level divergence (black bars) positioned over the mountain range for all regions, indicating enhanced lifting. At the same time, the northerly wind associated with the South American low-level jet (LLJ) along the Andes eastern slopes is strengthened, partly due to a downward expansion compared to dry spell conditions (supplementary file figure 4). In line with figure 2(a), increased low-level humidity is maximised at the Andes foothills, and follows the eastern slopes with a secondary peak in anomalous humidity between 500 and 600 hPa stretching from the east over the Andes crest (figure 2(b)). In all regions, three day wet spell initiation is hence associated with increased low-level moisture within the LLJ and lifting along the eastern slopes, with apparent convective moistening of the mid-levels. The latter is further strengthened for seven day spells, and patterns remain similar when instead of spell differences, spell anomalies from a seasonal climatology are considered (supplementary file figure 4).

The temporal evolution of three day wet spells in figure 2(c) shows similarities between regions, with 200 hPa easterly wind anomalies gradually increasing over a period of 4–6 days, covering start and end of the spell (green lines). The increase in low-level moisture (light blue) at the Andes eastern slopes similarly starts before the spell, but peaks on days$_{0,1}$, accompanied by the strengthening of the northerly LLJ (yellow lines). This increase precedes a co-located increase in total column water vapour (TCWV, dark blue), indicating the moistening of the mid-levels on day₀, which continues throughout the wet spell. This suggests that in addition to favourable upper-level wind conditions, low-level moisture injection into the latitudinal band of the wet spell during periods of a strengthened LLJ is of importance and a common feature across regions for their development.

4.1. Contribution of dynamic and thermodynamic drivers

To further quantify the control of Andes basin moisture as compared to upper-level wind conditions, we bin daily regional-mean Andean rainfall (${\gt}$2000 m) as a function of 200 hPa zonal wind as well as Andes' foothill TCWV (calculated from between the vertical lines in figure 2(b)) in figure 3. For all regions, average rainfall intensities increase with stronger easterly wind and higher TCWV, which together explain 67% (Equatorial) to 83% (Southern) of the binned rainfall variability. However, while the scaling rate of daily rainfall with zonal wind is very stable across regions with 0.12–0.14 mm per m s⁻¹, rainfall scales at a rate of 0.35–0.38 mm per mm TCWV only from the Equatorial to Central region, while it drops to 0.19 mm in the south. This may be linked to the considerably lower mean relative humidity of ∼64% at 550 hPa in the Southern region, implying a higher necessary increase in moisture for a similar increase in precipitation compared to the equator-ward regions (71%–74% mean relative humidity). In line with that, the differences in average wind and TCWV values between same-length dry and wet spells are most pronounced in the Southern region (red and blue markers in figure 3(a)). Notably, however, wet spells in all regions show an increase in upper-level easterly wind as well as in Andes foothill TCWV compared to dry spells.

Zoom In Zoom Out Reset image size

We now want to estimate the separate contribution of TCWV and wind changes to regional rainfall changes for wet and dry spell situations. For that, we first calculate the day₀ differences between same-length wet and dry spells for TCWV and 200 hPa zonal wind, which we translate into rainfall changes using regression coefficients from figure 3(a). The result is presented in figure 3(b), showing the contribution difference between wind and TCWV (Δ%) relative to the combined ($\Delta{u} + \Delta{\mathrm{TCWV}} = 100\%$) rainfall change, i.e. 0% indicates equal contribution from both drivers. Across the four regions, we find TCWV changes to dominate rainfall changes for three day spells, with a contribution of more than 90% in the equatorial region. Upper-level wind changes gradually increase in importance for longer spells over 5–7 days, reaching a maximum contribution of 58% in the Southern region. When considering these relationships since the 2000s, the contribution of TCWV increases further in all regions (supplementary file figure 5), in line with previously suggested increases in thermodynamic driver importance for recent decades in the Southern Andes (Segura et al 2020). Overall, the increase in wind contribution going from 3- to 7-day spells may reflect the synoptic time scale to which the upper-level wind changes are linked and is in line with the temporal spell evolution in figure 2(c), where the upper-level wind anomalies are longer-lived than anomalies in atmospheric moisture and may represent a synoptic constraint on maximum spell length.

We now examine the importance of the multi-day spells on the ground by considering SM and vegetation responses. Feldman et al (2021) found rapid VOD increases 0–1 days after a rainfall pulse in wet regions, with subsequent drying along with SM. They linked this to immediate plant rehydration while slower responses, mostly in dry regions, may reflect vegetation growth. Figure 4(a) first illustrates the direct response of SM to wet and dry spells by compositing and averaging 24-day timeseries around the local per-pixel spells for each region (Northern = Equatorial + Transition regions) from which the climatological seasonal cycle is subtracted. The anomalies, depicted as deviations from day₋₂ before spell start, illustrate an SM signal of within ±1.5%–2% for the seven day wet and dry spells in all regions, a deviation from an average SM of 26%–27% for all regions. Any response in the vegetation metrics VOD and physiological SIF yield $\phi_\mathrm{F}$ (figures 4(b) and (c)) generally co-occurs immediately with SM on day₀ of each spell. However, compared to SM, VOD and $\phi_\mathrm{F}$ show clearer inter-regional differences.

Zoom In Zoom Out Reset image size

VOD responses are weak in the humid Northern region (figure 4(b)) and correlation between SM and VOD is non-existent (supplementary file figure 6), suggesting that the spells have limited effect on vegetation water content here. Correspondingly, the $\phi_\mathrm{F}$ signal remains noisy and is not affected by wet spells in this region. In the semi-arid Southern region on the other hand, we find a larger increase in VOD during wet spells and a more pronounced drying over the dry spells, reaching 0.016 and −0.012 over seven days, respectively, which, in terms of magnitude, corresponds to ∼24% of the average seasonal change in VOD in this region (annual ΔVOD = 0.05). In the Central and Southern region, the physiological signal from $\phi_\mathrm{F}$ agrees with the VOD signal, implying that both vegetation structure and physiology respond to 3–7 day wet and dry spells. In all regions, anomalies rebound and stabilise after 2–3 days after spell end, suggesting that spell effects are short-lived, even for seven day spells. However, prolonged increase in VOD after the wet spells and decrease after dry spells beyond that timeframe in the Southern region, and to a lesser degree in the Central region, may suggest biomass buildup and loss, respectively, continuing beyond the pure plant hydration signal. This indicates that, in line with the climatological differences of these Andean regions, wet and dry spells seem to have the most important and longest-lasting effects on vegetation in the drier, semi-arid Central and Southern regions, reaching more than 20 days.

Based on 33 years of reanalysis and remote sensing data, we examined the control of upper-level wind and Amazon basin moisture on 3 to 7 day wet and dry spells across four different zones in the Tropical Andes and quantified their contribution to regional rainfall during the SAM. This was motivated by our relatively extensive knowledge on drivers of interannual rainfall conditions in the Southern Tropical Andes (Altiplano region). In particular, the Amazon basin was previously suggested as crucial moisture source for wet periods in the Southern region, with enhanced moisture advection linked to upper-level easterly winds controlling the majority of interannual rainfall variability (Garreaud et al 2003, Thibeault et al 2012). Here, we show that this rainfall sensitivity to upper-level winds exists on a multi-day basis across the entire Tropical Andes, however with pronounced seasonal dependence in line with the progression of the SAM. We found easterly wind anomalies to accompany wet spells in all regions compared to dry spells. However, in the mean, light westerly winds prevail in the Southern region even during 3–5 day wet spells. Indeed, we find 42% of wet days to occur under upper-level westerlies in the Southern region. On the other hand, mean winds are easterly for seven day wet spells, suggesting that the most intense and wider-spread wet spells are likely associated with mean easterly winds even in the Southern region, as was found for intense wet periods by Falvey and Garreaud (2005) and is in line with findings on the interannual timescale for the Altiplano (Garreaud and Aceituno 2001, Vuille and Keimig 2004).

This is summarised in figure 5, which furthermore highlights our findings on the importance of positive low-level moisture anomalies along the Andes foothills at wet spell start in all regions. The moisture anomaly is accompanied by a surface-ward expansion and strengthening of the South American LLJ, tapping into moisture that is lifted above the Andes crest level by convection along the eastern slopes. As a consequence, Amazon basin mid-level moisture gradually increases over wet spells. During dry spells on the other hand, the LLJ is weaker, holds a lifted position and air masses at both low- and mid-levels are significantly drier. By quantifying rainfall changes as a function of upper-level wind and Amazon basin moisture, we find moisture to dominate three day wet spells with a maximum in the Equatorial region of 91% contribution to mean rainfall, while upper-level wind control increases for 5–7 day wet spells. These results indicate that while moisture changes along the eastern Andes slopes dominate rainfall changes on daily time scales associated with three day spells, upper-level dynamics play a more important role at the synoptic time scale of 5–7 day spells. This is true in particular in the Southern region, where upper-level wind conditions are less frequently favourable for moisture advection and where wind changes thus contribute 55%–58% to mean rainfall changes. Considering effects on the ground, the immediate effect of 3–7 day spells on vegetation remain short lived, visible to ∼2 days after spell end in all regions. However, prolonged changes after 5–7 spells in the Central and Southern Tropical Andes with increased VOD/$\phi_\mathrm{F}$ after wet and decreased VOD/$\phi_\mathrm{F}$ after dry periods out to day 20 after spell start suggest potential effects on vegetation growth, which shows the importance of spells at these timescales and may also translate into crop growth sensitivity in these semi-arid regions.

Zoom In Zoom Out Reset image size

Our findings highlight strong similarities in drivers of wet and dry periods across all of the Tropical Andes, and emphasize the need to consider both dynamic and thermodynamic changes when evaluating effects on Andes rainfall. For the Southern Tropical Andes, Segura et al (2020) already identified increased southward moisture transport via the LLJ and associated mid-level moistening as an important and independent driver of rainfall trends there. This renders future rainfall predictions exclusively based on upper-level wind changes (e.g Minvielle and Garreaud 2011, Neukom et al 2015) questionable, particularly considering likely increases in TCWV with higher future temperatures, which may be able to buffer easterly wind weakening. Emerging multi-decadal convection-permitting simulations for the region will help to better quantify in particular contributions from moisture transport and changes therein, which requires both improved representation of topography and of convective processes, both of which current coarse-scale climate models cannot sufficiently capture.

The research leading to these results was conducted within the AgroClim-Huaraz project (https://agroclim-huaraz.info/), which received funding from the Österreichische Akademie der Wissenschaften. CK also acknowledges funding from the NERC-funded LMCS project (NE/W001888/1). We thank the providers of key data sets used here. GRIDSAT-1B is available from www.ncdc.noaa.gov/gridsat, ECMWF ERA5 reanalysis data is available from the Copernicus Data Store (https://cds.climate.copernicus.eu/), CHIRPS data are available from www.chc.ucsb.edu/data/chirps, VOD data from https://doi.org/10.5281/zenodo.2575599, soil moisture data from www.esa-soilmoisture-cci.org/v06.1_release, BESS data (for PAR) from www.environment.snu.ac.kr/bess-rad, CSIFv2 (for SIF) data from https://osf.io/8xqy6/ and MODIS MCD43A4v061 (for NIRv) data were acquired via AppEEARS https://appeears.earthdatacloud.nasa.gov/. Special thanks go to the providers of the python packages matplotlib/cartopy, xarray, salem, scipy, metpy, and their dependencies.

No new data were created or analysed in this study.

CK and FM conceptualised the study, with input from EP and CJ. All authors contributed to the methodological design, analyses were conducted by CK and LH, and data was provided by BH. CK led the writing of the manuscript with contributions from all authors.