Climate change and lithium mining influence flamingo abundance in the Lithium Triangle

Abstract

The development of technologies to slow climate change has been identified as a global imperative. Nonetheless, such ‘green’ technologies can potentially have negative impacts on biodiversity. We explored how climate change and the mining of lithium for green technologies influence surface water availability, primary productivity and the abundance of three threatened and economically important flamingo species in the ‘Lithium Triangle’ of the Chilean Andes. We combined climate and primary productivity data with remotely sensed measures of surface water levels and a 30-year dataset on flamingo abundance using structural equation modelling. We found that, regionally, flamingo abundance fluctuated dramatically from year-to-year in response to variation in surface water levels and primary productivity but did not exhibit any temporal trends. Locally, in the Salar de Atacama—where lithium mining is focused—we found that mining was negatively correlated with the abundance of two of the three flamingo species. These results suggest continued increases in lithium mining and declines in surface water could soon have dramatic effects on flamingo abundance across their range. Efforts to slow the expansion of mining and the impacts of climate change are, therefore, urgently needed to benefit local biodiversity and the local human economy that depends on it.

1. Introduction

Technologies that reduce carbon emissions are major components of efforts to slow global climate change and minimize its effects on biodiversity [1]. Nonetheless, the development of so-called ‘green’ technologies can come with its own trade-offs, including direct wildlife mortality [2], habitat fragmentation [3,4] and loss [5] and short-term efforts to maximize the production of traditional technologies [6]. As global climate change and biodiversity loss accelerate, there is a need to better understand these trade-offs and how to balance the resource needs of humans and other species [7].

Saline lakes represent a striking example of the trade-offs initiated by the development of green technologies. Despite making up a small proportion of wetlands, saline lakes support globally important assemblages of numerous aquatic species (i.e. above 10% of a species' population [8–10]). The high salt loads of saline lakes impose physiological stresses on many organisms, however, increasing their energetic requirements [11]. When saline lake water levels decline and their salinities rise [12], most vertebrates, macroinvertebrates, and other organisms are, therefore, unable to respond, leaving ecosystems comprised only a few archaea and harmful cyanobacteria [13,14]. As a result, as saline lake water levels have declined [15], so too have the population sizes of most species relying on them [16,17].

Climate change and increasing anthropogenic water usage have been the main drivers of declines in saline lake levels [15,18], but other anthropogenic factors may also be important. For instance, in the ‘Lithium Triangle’ of Bolivia, Argentina, and Chile (figure 1), increasing demand for lithium for use in electric cars, energy storage, and cellular telephones has driven an eightfold increase in lithium production over the past two decades and is potentially threatening the more than 20 saline lakes or ‘salares’ in the region [19,20]. These salares are not only biodiverse [21] but provide considerable economic and health-related benefits for local people [22,23].

Figure 1.

Study area. (a) Map of the Lithium Triangle and Chilean salares included in this study (orange polygons). (b) Lithium mining operations (dark red) and lagunas (dark blue) within the Salar de Atacama. (c) Meta-model displaying hypothesized relationships between study system components. (Online version in colour.)

The impacts of lithium mining may be particularly extensive in arid environments like the Lithium Triangle [24]. Apart from its terrestrial footprint and other resource needs, the production of 1 ton of lithium requires approximately 400 000 l of water [25]. Some salares within the region are already at historically low water levels [26] and climate models project future precipitation declines and temperature increases [27,28]. Increased lithium mining may thus prove ecologically unsustainable if it accelerates these regional desiccation trends. Accordingly, negative changes in water quality, vegetation structure, and the distribution of local flora and fauna have already been documented, affecting local human populations via decreases in agricultural productivity and the loss of traditional jobs [29–31].

Among the most threatened and economically valuable species native to the salares are Chilean (Phoenicopterus chilensis), Andean (Phoenicoparrus andinus) and James' flamingos (Phoenicoparrus jamesi). Two of these species are endemic to the region, with one considered ‘Vulnerable’ (Andean) and the other ‘Near-threatened' (James’) [32]. Flamingos help maintain the stability of salares by regulating key trophic processes [33] and represent a major attraction for the economically important ecotourism industry [34]. Importantly, flamingos in other regions are susceptible to fluctuating water levels [35,36], and there is apprehension about the potential combined effects of climate change and mining on the flamingos of the Lithium Triangle.

In this context, there is an urgent need to (1) determine the degree to which lithium mining, climate change, and other anthropogenic factors, like human population growth, may be affecting salar water levels in the Lithium Triangle and (2) identify how those changes—along with such direct mining-related impacts as disturbance—may be affecting flamingo populations. We hypothesized that climatic and anthropogenic factors could influence flamingo abundance within a salar both directly and indirectly—via their influence on salar water levels and, subsequently, food availability for flamingos (figure 1). We then predicted that decreases in surface water levels and increases in lithium mining would negatively affect flamingo abundance. We tested these predictions by combining more than 30 years of on-the-ground flamingo surveys with analyses using remote sensing data to monitor long-term changes in surface water levels in five salares in the Chilean portion of the Lithium Triangle (figure 1). Our study is the first investigation of the effects of lithium mining on these keystone species and will have important ramifications not only for biodiversity conservation, but also global economic sustainability, as lithium mining efforts are projected to increase worldwide [22].

2. Methods

(a) . Study species and study area

Chilean, Andean and James' Flamingos are long-lived waterbirds that breed and—to varying degrees—remain resident in the salares of the Lithium Triangle (figure 1; [32,37,38]). Chilean Flamingos are the most widespread and abundant, with their range extending across southern South America [39]. Andean and James’ Flamingos breed exclusively within the region, but some individuals spend the non-breeding season in saline lakes east of the Andes. All three species breed during the Austral summer, with reproduction peaking in January [37].

The Chilean portion of the Lithium Triangle comprises five complexes of salares (salt playas) and lagunas (wetlands-ponds)—Huasco, Surire, Pujsa, Tara and Atacama—located along a north–south elevational (4585–2300 m) and precipitation gradient (3–20 mm annually; 1958–2019). The salares are largely groundwater and low-run-off fed endorheic basins that contain shallow lakes reaching salinities greater than 40 g l⁻¹ and supporting some of the region's only permanent water [40].

Lithium mining in the Chilean portion of the Lithium Triangle began in the Salar de Atacama in 1984 and has remained focused there [41,42]. Chile, however, now controls approximately 1/2 of the global lithium reserves [43] and production in the Atacama is predicted to increase to 270 000 megatons of lithium bicarbonate equivalent by 2026, a threefold increase above 2018 levels [20].

(b) . Flamingo surveys

We used data from comprehensive, simultaneous surveys (1985–1988, 1997–2019) of flamingos across the five salares during the breeding (Jan) and non-breeding seasons (Jul). These surveys were carried out by the Corporación Nacional Forestal de Chile (CONAF) and Grupo para la Conservación de Flamencos Altoandinos using standardized double-observer point counts [44].

Within the Salar de Atacama—a Ramsar wetland of international importance—we also used ‘local’ quarterly surveys (Jan/Feb, Apr/May, Jul/Aug and Oct/Nov) at seven lagunas (figure 1) performed by CONAF in agreement with the Sociedad Química y Minera (2002–2009, 2011–2013). Three of these lagunas—Salada, Saladita and Interna—were treated as a single complex given their proximity and the difficulty of associating count data with specific lagunas.

The regional and local counts consisted of adult flamingos—juveniles were not identified to species or included here. The regional and local counts are also not fully comparable because of the different timespans, frequencies, dates and teams conducting the surveys.

(c) . Surface water change

To quantify potential variation in flamingo habitat over time, we conducted time series analyses from 1984–2018 by measuring surface water change following Donnelley et al. ([18]; electronic supplementary material, appendix S1). We used Landsat 5 Thematic Mapper (1984–2011) and Landsat 8 Operational Land Imager (2013–2018) satellite imagery to monitor salar and laguna surface water area. We measured surface water area using constrained spectral mixture analysis (SMA) that allowed proportional estimates of water contained within a continuous 30 × 30 m pixel grid. We masked areas containing clouds, cloud shadow, snow and ice using the Landsat CFMask band and incorporated all unmasked pixels in the visible, near-infrared, and short wave infrared bands into the SMA with the exception of the Landsat 8 coastal aerosol band. We extracted training data for the SMA model from satellite imagery as spectral end-members unique to the individual images classified.

For each salar, we used our model to calculate the total surface water area for 90 days preceding and including the month of the two flamingo censuses. Additionally, in the Salar de Atacama, we calculated surface water area for each smaller laguna during the four seasonal flamingo censuses. In these cases, surface water area was calculated as the mean of each count season (winter, Jun–Aug; spring, Sep–Nov; summer, Dec–Feb; autumn, Mar–May).

(d) Salar/laguna productivity

Cyanobacteria are a key flamingo food source [45]. To understand flamingo food availability, we calculated the normalized difference vegetation index (NDVI), which correlates with cyanobacterial fluctuations [46,47]. Using the Google Earth Engine Platform, we derived average NDVI measures across each census season using our model generated surface water area polygons. We calculated NDVI with the normalized difference function using the NIR (near-infrared) and red bands with the standard equation (NIR − red)/(NIR + red). This resulted in a number between −1 and 1, with larger numbers denoting higher NDVIs.

(e) . Climate change

We identified climatic variables that could drive variation in salar surface water area—precipitation, minimum and maximum temperature, run-off and potential evapotranspiration. Using the Google Earth Engine Platform, we extracted these variables monthly for each watershed from the TerraClimate dataset [48], which has a approximately 4-km (1/24th degree) spatial resolution (see www.climatologylab.org/terraclimate.html). Watershed boundaries were constructed from the HydroBASINS dataset [49], which has global watershed boundaries and sub-basin delineations at a 15-s resolution (see www.hydrosheds.org).

(f) . Estimating human water use

We incorporated two different measures of potential human water usage. First, we derived the human population (a proxy for water use) surrounding the salares from the 250-m population grid within the Global Human Settlement Layer [50] for 1975, 1990, 2000 and 2015. We estimated the population from the intervening years with a spline. Second, we assessed the potential water use of lithium mining operations within the Salar de Atacama. We determined the area of the mining ponds constructed for lithium refinement as a surrogate for lithium production and the water used in industrial processes. To calculate changes in mining pond area, we digitized the maximum pond extent for each year of our surface water area analysis and summed the areas for that year. Although pond area is not a direct measure of the water used for lithium extraction, it is a good measure of relative change in potential water use [51].

(g) . Estimating temporal trends

All statistical analyses were performed in the R programming environment [52]. First, to understand how environmental conditions and flamingo abundance may have changed during the study, we evaluated temporal relationships among climatic variables, surface water area, NDVI, and flamingo abundance at the regional and local scales. For this, we used linear regressions for all salares together and separately. For the regional analyses, we only used summer data because coverage was poor during winter; for the local analyses, we used data from all quarterly surveys.

To include time series data within our linear models, we tested for stationarity using Augmented Dickey Fuller (ADF) tests with the package ‘tseries’ [53]. Most variables in our regional dataset were stationary (except human population size; electronic supplementary material, appendix S2), enabling us to use mixed-effect models while accounting for serial autocorrelation (see below). On the other hand, all variables in the local dataset were non-stationary (electronic supplementary material, appendix S2). We thus took the first difference of all series (i.e. X_t − X_t−1), effectively removing trends and serial autocorrelation, and allowing variables to be related in a single analysis [54]. Prior to differencing, we log-transformed flamingo abundance.

(h) . Mixed-effect modelling

To evaluate the hypothesized pathways by which climate and anthropogenic activities could affect surface water availability, NDVI and flamingo abundance (figure 1), we used mixed-effect modelling as a preliminary step to inform our structural equation modelling (SEM) [55,56]. We fit separate linear mixed-effect models with the packages ‘nlme’ (surface water area and NDVI; [57]) and ‘MASS’ (flamingo abundance; [58]) to the components of our system, with predictor variables (climatic and hydrological variables, and human population size) included as fixed effects. For surface water area and NDVI, we used linear mixed models with a Gaussian error term. We modelled regional flamingo abundance with generalized linear mixed models that included salar as a random effect, a Poisson error term, and a quasi-likelihood penalization with the function glmmPQL. For the Salar de Atacama, we also modelled flamingo relative (i.e. differenced) abundance with linear mixed models that included laguna as a random effect and a Gaussian error distribution. In all models, we included a temporal correlation term with the function corCAR1 from the package ‘nlme’ to account for potential autocorrelation between consecutive years and counts. Because the number of counts from each salar/laguna was limited, we subjected predictor variables to variance inflation factor analyses in the ‘car’ package [59] and removed predictors with variance inflation factor values greater than 4 from our global model (regional: maximum temperature; local: maximum temperature, run-off and human population size).

(i) . Structural equation models

We used piecewise SEM in the package ‘piecewiseSEM’ [60] to combine our linear mixed-effect models and evaluate the hypothesized associations among the components of our study system. Piecewise SEM allows for the inclusion of random effects and variance structures from each linear mixed-effects model in a global SEM. We used Shipley's test of d-separation [61] to assess model fit and determine whether paths were missing. Following Grace et al. [62], we added two suggested paths to our local SEM because a plausible connection existed between the variables minimum temperature and rainfall. To better assess the effects of predictor variables on overall flamingo abundance (i.e. the combined abundance of all species), we fit a multi-group SEM with species as the grouping variable. This allowed us to ask whether or not relationships among variables differed by species. To improve interpretability, we also centred and scaled each predictor variable before calculating its path coefficient.

3. Results

(a) . Flamingo abundance at the regional scale

Regionally, flamingo counts fluctuated among years but did not decline over time (figure 2 and electronic supplementary material, appendix S2). Nonetheless, within the Salar de Tara and the Salar de Atacama, all species declined (both p < 0.01; figure 2 and electronic supplementary material, appendix S2). NDVI (β = −0.002 ± 0.000, CI = −0.002, −0.001; electronic supplementary material, appendix S2) and surface water area (β = −11.636 ± 1.589, CI = −14.755, −8.518; electronic supplementary material, appendix S2) both declined as well. We found no correlation between surface water area and flamingo abundance (β = 0.365 ± 0.304, CI = −0.234, 0.963). However, surface water area correlated positively with NDVI (β = 3.502–05 ± 0.571 × 10⁻⁵, CI = 2.382 × 10⁻⁵, 4.621 × 10⁻⁵; electronic supplementary material, appendix S2), as did flamingo abundance (β = 5635.380 ± 2537.453, CI = 641.364, 10 629.395; electronic supplementary material, appendix S2).

Figure 2.

Flamingo abundance. Austral summer abundance of flamingos in: (a) Atacama, (b) Huasco, (c) Pujsa, (d) Surire and (e) Tara. (f) Illustration of the three flamingo species, reproduced with permission from 'Aves de Chile. Sus islas oceánicas y Península Antártica' [63]. (Online version in colour.)

Climatic variables also varied across years and salares (electronic supplementary material, appendix S2). Both maximum and minimum temperatures tended to increase (maximum temperature: β = 0.013 ± 0.006, CI = 0.000, 0.024; minimum temperature: β = 0.018 ± 0.004, CI = 0.008, 0.027), whereas rainfall tended to decrease (β = −0.116 ± 0.065, CI = −0.244, 0.011) across time. Although run-off did not vary across time (β = −0.004 ± 0.005, CI = −0.013, 0.006), surface water area decreased by approximately 30% over the study period (β = −12.051 ± 3.941, CI = −19.807, −4.300; figure 3).

Figure 3.

Trends in human activities within the Salar de Atacama. (a) Chilean lithium production (metric tons LCE; data from US Geological Survey); Atacama (b) mining pond area, (c) human population and (d) surface water area. Note that lithium production and mining pond area were strongly correlated and we only included mining pond area in analyses. (Online version in colour.)

Our multigroup SEM indicated (Fisher's C = 10.442, df = 14, p = 0.729; figure 4a) that surface water area was negatively influenced by potential evapotranspiration (β = −0.535, p < 0.001), but positively influenced by run-off (β = 0.137, p = 0.046). Moreover, surface water area positively affected NDVI (β = 0.182, p < 0.001), which, in turn, positively influenced flamingo abundance (β = 0.255, p < 0.019). Coefficients did not differ across species, meaning all species responded similarly to variation in environmental and anthropogenic variables (electronic supplementary material, appendix S2).

Figure 4.

Factors influencing flamingo abundance. Path diagrams of factors influencing regional (a) and local (b) abundances of flamingos. Black and red paths represent positive and negative influences, respectively; dashed grey paths represent non-significant influences. Path thickness is proportional to the standardized regression coefficient. Grey boxes surround abiotic variables for watersheds (averages of the 12-month period prior to and including flamingo censuses) and lagunas (averages of the three-month period prior to and including flamingo censuses). PET = potential evapotranspiration. Flamingo illustrations reproduced with permission from [63]. (Online version in colour.)

(b) . Flamingo abundance at the local scale

In the Salar de Atacama, we found significant trends in some climatic and anthropogenic variables during the period for which we had laguna-specific count data (2002–2013; electronic supplementary material, appendix S2). Rainfall increased (β = 0.064 ± 0.022, CI = 0.021, 0.107), but summer (Dec–Feb) surface water levels did not exhibit a trend (β = 2.050 ± 2.308, CI = −2.503, 6.609) while winter (Jun–Aug) surface water area declined by more than 40% (−8.870 ± 1.130, CI = −11.113, −6.626; figure 3). At the same time, both mining pond area and human population increased (figure 3 and electronic supplementary material, appendix S2).

Our multi-group SEM (figure 4b; Fisher's C = 9.771, df = 10 p = 0.461) indicated that minimum temperature (β = −0.220, p < 0.001) negatively influenced surface water area. Surface water area, in turn, positively influenced the abundance of Chilean (β = 0.259, p < 0.001) and James' flamingos (β = 0.166 p < 0.001), but not Andean flamingos (β = −0.018, p = 0.812). Minimum temperature also negatively affected Chilean (β = −0.206, p = 0.005) and James’ flamingos (β = −0.244, p < 0.001), but positively affected Andean flamingos (β = 0.187, p = 0.010; figure 4b). Moreover, surface water area (β = 0.390, p < 0.001) and rainfall (β = 0.254, p < 0.001) had direct positive effects on NDVI, while NDVI had a positive effect on flamingo abundance (β = 0.083, p = 0.033). Our results also showed a direct negative effect of mining pond area on Andean (β = −0.377, p < 0.001) and James' flamingos (β = −0.158, p = 0.020), but not Chilean flamingos (β = −0.104, p = 0.135; figures 4 and 5).

Figure 5.

Effect of mining pond area on flamingo abundance. Predictions and 95% confidence intervals from linear mixed models of temporal changes in (a) Andean and (b) James' flamingos, which are significantly affected by mining pond area in our local SEM (figure 4b). Note that flamingo abundances were log-transformed prior to differencing, hence the y-axis indicates the relative change in flamingo abundance.

4. Discussion

Global lithium demand is rapidly increasing [43], raising concerns about possible effects on the arid ecosystems in which lithium is mined and produced [19]. In the Chilean portion of the Lithium Triangle, we found that variation in surface water levels and primary productivity were strongly correlated with fluctuations in the abundance of three economically important species of flamingos. Additionally, within the Salar de Atacama, our results suggest that lithium mining is negatively correlated with the abundance of two of these species. Although flamingos are not yet declining at the regional level, lithium mining is projected to expand both within the Salar de Atacama and across the Lithium Triangle [64]. Thus, in combination with climate change-related declines in surface water, lithium mining could soon have strong negative impacts on flamingo abundance across their range. Efforts to mitigate the effects of mining activities and continued reductions in surface water are, therefore, urgently needed to conserve local biodiversity and human economies.

(a) . Mining, climate change and changes in surface water area

Saline lake water levels are declining globally [15]. The salares of the Lithium Triangle are no exception—some are currently at a 600-year low [26] and the five salares in our study have shrunk by more than 30% since 1984. In our study, these declines were at least partly driven by increasing rates of potential evapotranspiration. The fact that rates of potential evapotranspiration did not change as dramatically as surface water levels, however, points to the additional role of groundwater declines. This has important implications for salar stability: salares are mainly groundwater-fed and groundwater levels generally change over longer timescales, meaning they cannot rapidly recharge via short-term increases in run-off or precipitation [65].

Lithium mining also largely relies on groundwater and as the size of mining operations increases, demand for groundwater will as well. Groundwater pumping in the Salar de Atacama, for example, increased from 0–1.8 m³ s⁻¹ between 1986 and 2018 as lithium production increased (electronic supplementary material, appendix S2). Although the salar's summer surface water levels did not change during our study, its winter surface water area declined 2.7 ha per year. As such, mining activities, along with declining snowpacks [26], pose a threat to regional groundwater levels.

Superimposed on these long-term surface water declines is the extreme variability of the region's climate. Annual precipitation varied by three- to fivefold during our study, while rates of run-off differed by 150-fold (electronic supplementary material, appendix S2). Such fluctuations are not uncommon in regions with saline lakes, but the combination of long-term declines and short-term instability can lead to frequent drying events, when a salar temporarily loses all surface water or reaches critically high salinities. In other saline lakes, such events have led to regime shifts in the benthos with stark reductions in biodiversity [13,14], including mass flamingo die-offs [36].

(b) . Factors affecting flamingo population dynamics

Flamingo abundance varied greatly from year-to-year. In fact, the number of breeding flamingos could fluctuate locally and regionally by many thousands of individuals over a few years. At both spatial scales, these fluctuations were strongly predicted by the effects of precipitation, minimum temperature, potential evapotranspiration and surface water levels via their effects on food availability (i.e. NDVI). Years with higher precipitation and lower potential evapotranspiration—and thus higher surface water levels—were associated with both more food and flamingos. The effects of these variables were, in turn, largely similar across species and spatial scales, indicating that water and food availability are key drivers of flamingo abundance.

Within the Salar de Atacama, we also found evidence that lithium mining is negatively correlated with the abundance of the two endemic species. Although Chilean flamingos were unaffected—possibly because they were buffered by their larger global population—counts of globally threatened James' and Andean flamingos declined by approximately 10 and 12%, respectively, in just 11 years. Importantly, our structural equation models indicated that the negative relationship between mining and flamingo abundance was direct and not mediated by other environmental factors. This likely suggests that increases in disturbances such as noise and vehicular traffic from industrial activities, and not just the effects of increasing mining pond area on the amount of surface water in the salar, altered flamingo breeding behaviour [66]. Irrespective of mining's exact effects, fluctuations in surface water, primary productivity and mining can all act separately or in concert to influence flamingo abundance.

We did not, however, find evidence of flamingo declines at the regional level. Given the high adult survival of flamingos, this suggests that the region is not an isolated metapopulation [67]. Instead the five Chilean salares are part of a larger network of potential breeding sites including salares in Argentina and Bolivia, with flamingos moving among them in response to fluctuating water levels and food availability [37]. The effects of lithium mining in the Salar de Atacama may, therefore, be buffered regionally so long as the other salares remain healthy enough to support their current flamingo breeding densities.

(c) . The future of flamingos in the Chilean Lithium Triangle

Surface water, food availability and lithium mining can all separately and interactively influence flamingos. This necessitates that multifaceted efforts be instituted to mitigate the impacts of recent surface water declines and increases in lithium mining before flamingo populations begin to decline regionally and not just locally. There may yet be time: lithium mining in Chile is still largely limited to the Salar de Atacama and the Chilean government recently recommended the Salar de Huasco be set aside for long-term protection [68]. The Chilean government, however, has also announced it will open a tender for the additional exploration and production of 400 000 tons of metallic lithium outside the Salar de Atacama [64]. Moreover, lithium mining is expanding to other portions of the Lithium Triangle, including the Salar de Uyuni, a major flamingo breeding area in Bolivia. If mining limits the ability of flamingos to optimally track environmental conditions, it could have severe negative impacts on regional flamingo populations via repeated reproductive failure and/or declines in adult survival [69,70]. Therefore, to sustain flamingo populations over the long term, transnational efforts are needed that slow the spread of lithium mining, improve groundwater conservation, and minimize other forms of anthropogenic disturbances.

More broadly, there remains a tension between the demand for green technologies and the effects their production has on local ecosystems and economies. In addition to our findings, other studies have already indicated that lithium mining is impacting the livelihoods of local people across the Lithium Triangle [29,31]. Ultimately, our study illustrates that even ‘green’ technologies can involve significant negative environmental trade-offs at global scales of production and that these trade-offs can manifest via complex socio-ecological pathways to impact both people and biodiversity [71].

Supporting information

Extended methods for surface water analyses (electronic supplementary material, appendix S1) and results (electronic supplementary material, appendix S2) are available online. The authors are solely responsible for the content and functionality of these materials. Queries (other than data absence) should be directed to the corresponding authors.

Data accessibility

The datasets supporting this article have been uploaded as part of the electronic supplementary material [72].

Authors' contributions

J.G.: conceptualization, data curation, formal analysis, investigation, methodology, resources, visualization, writing—original draft, writing—review and editing; J.N.M.: conceptualization, data curation, formal analysis, investigation, methodology, resources, visualization, writing—review and editing; J.P.D.: data curation, formal analysis, methodology, writing—review and editing; C.D.: resources; J.G.N.: resources, writing—review and editing; N.R.S.: conceptualization, formal analysis, investigation, methodology, writing—original draft, writing—review and editing.

All authors gave final approval for publication and agreed to be held accountable for the work performed therein.

Competing interests

We declare we have no competing interests.

Funding

This research was funded by Junta de Extremadura (Consejería de Economía, Ciencia y Agenda Digital) and ERDF through grant GR21081. J.S.G. was supported by Junta de Extremadura (grant TA18001).