A twenty year GIS-based assessment of environmental sustainability of land use changes in and around protected areas of a fast developing country: Spain

Abstract Spain has experienced massive recent socioeconomic changes that have had an influence on biodiversity and landscapes through land use-land cover (LULC) changes. Protected areas (PAs) seek to conserve biodiversity by establishing a legal and, sometimes, managerial regime that forbids or restricts LULC changes that are damaging to biodiversity. Here, we used CORINE Land Cover (CLC) data between 1987 and 2006 to assess differences in LULC changes and processes of change as metrics of effectiveness in four PA networks of clear legal and managerial characteristics in Spain: Nature reserves (NRs), Nature parks (NPs), Sites of Community Importance (SCIs) and Special Protection Areas (SPAs). We also compared LULC changes and processes of change around each PA network applying a modified Before-After-Control-Impact (BACI) research design with two increasingly distant control areas and two models of increased validity. The four PA networks were more environmentally sustainable than their surrounding areas although an effectiveness gradient was shown: NRs > SCIs > SPAs > NPs, suggesting little influence of PA management on LULC changes overall. Another gradient of environmental sustainability of control areas was evident: SCIs > SPAs > NPs > NRs. Proximal controls were more sustainable than distant ones. The main LULC increases inside PAs affected agro-forestry areas and transitional woodland-shrub, whereas artificial surfaces, permanently irrigated lands and burned areas prevailed in the proximal and distant controls. Three main LULC processes of change inside and around Spanish PAs outstood: forest succession, land development, and new irrigated areas, the two former chiefly affecting surrounding areas and posing serious threats to effective biodiversity conservation.


Introduction
Global socioeconomic improvements in the last decades (United Nations, 2015) have run in parallel to an unprecedented decline of biodiversity and ecosystem services (Butchart et al., 2010), despite substantial international efforts to maintain healthy ecosystems (UNESCO, 1971(UNESCO, , 1972CMS, 1979;EEC, 1979;CBD, 1992;EEC, 1992;OSPAR, 1992;United Nations, 1994). Protected areas (PAs) have been widely advocated and used as the main global policy to stop the loss of biodiversity for a long time (CBD, 1992), although evidence of their effectiveness at conserving species and natural habitats varies (Gaston et al., 2008;Craigie et al., 2010;Geldmann et al., 2013;Davis et al., 2014;Spracklen et al., 2015;Gray et al., 2016). Moreover, active discussions are in place on accurate techniques for measuring the have become more intensive, resulting in greater water, fertilizer and pesticide use in search of profitability (Zorrilla-Miras et al., 2014;Custodio et al., 2016). Finally, increasing population and tourist figures boosted intense'anthropisation' of the Spanish countryside especially around big cities and along the coast (Barbero-Sierra et al., 2013;García-Ayllón, 2015). These artificial changes have been especially intense in the 2000-2008 period, where massive residential and infrastructural construction took over across the country (Jiménez, 2012) as a result of abundant credit, deficient regulations and expansive mindsets (in't Veld et al., 2014).
The administrative responses to the biodiversity crisis in Spain have been uneven. In terms of in situ protection of biodiversity, PA coverage has expanded greatly in recent years. The 1958 nationally designated PAs and 1865 Natura 2000 sites (Sites of Community Importance, Special Areas of Conservation and Special Protection Areas; EEC, 1992) cover approximately 28% of the terrestrial area of the country, the largest absolute area of any country in the European Union (European Commission, 2016). That national PA coverage largely exceeds global PA coverage targets, at 17% by 2020 (CBD, 2010), and nearly doubles global figures at 14.7% (Bhola et al., 2016). Although efforts are being made to make adequate management plans for such areas, both PA management and evaluation still need much improvement in most PAs in the country (Rodríguez-Rodríguez et al., 2015). Research attention has also been paid to areas surrounding PAs given their influence on PAs' conservation outcomes (Radeloff et al., 2010). Thus, the environmental sustainability trends of Spanish PAs and the wider countryside remain an active research subject (Martínez-Fernández et al., 2015;Hewitt et al., 2016;Martínez-Vega, 2017, 2018a).
Satellite images are ideal sources of information to analyse LULC changes in and around PAs through a BACI design (Rodríguez-Rodríguez and Martínez-Vega, 2018a). Their recurrent spatial coverage (Chuvieco and Huete, 2010) is highly appreciated by PA managers (Dimobe et al., 2017;Ramachandran et al., 2018) and planners to achieve sustainable development. Additionally, they facilitate the making of spatially-explicit models, which are useful tools to quantify driving forces of change and simulate future land uses under different scenarios (Veldkamp and Lambin, 2001;Gallardo and Martínez-Vega, 2018). The BACI design (Green, 1979) has been long applied to statistically evaluate environmental and ecological impacts (Smith, 2002). However, despite its highly valid outcomes, it has scarcely been used by the remote sensing community (Blackman, 2013;Meroni et al., 2017;Staentzel et al., 2018).
In this study we used remote sensing products to: 1) assess environmental sustainability trends in and around different PA networks with clear legal and managerial characteristics in Spain during a period of intense environmental transformation in the country : 1987: -2006: (Jiménez, 2012; 2) compare sustainability trends among PA networks; 3) test the validity of different spatial-statistical techniques at ascertaining causality; and 4) make recommendations for the adequate planning, management and conservation of the country's protected biodiversity.

Study area
Spain's land territory covers 505,370 km 2 that expand across four biogeographic regions: Alpine, Atlantic, Mediterranean and Macaronesian (EEA, 2016). Four PA networks with clear legal and managerial characteristics across the country were analysed (Table 1): NRs, NPs, SCIs and SPAs. Fig. 1 shows the spatial distribution of those networks. The protected area assessed exceeds 20% of the Spanish land territory, and amounts to approximately 64% of the terrestrial protected area of the country in 2017.

Data collection and analysis
We used CLC data as the longest, most accurate and comparable source of LULC data broadly accessible for the whole country (Pérez-Hoyos and García-Haro, 2013). LULCs from level-3 CLC data were compared at two time points: 1987points: and 2006points: (IGN, 2016. To simplify comparisons, we grouped the 44 CLC level 3 LULCs in 22 classes (Table 2).
PA boundaries were retrieved from the Spanish Ministry of Environments' digital repository updated by December of 2014 (Spanish Government, 2015). We joined the initial four PA layers and spatially homogeneous protected polygons (PPs) were produced, as done previously (Rodríguez-Rodríguez and Martínez-Vega, 2018a). PPs account for total or partial overlaps among PA categories. In order to apply a multiple, BACI research design (Smith, 2002), we assigned the oldest designation category to each overlapping PP, as that was the date when that land was initially protected (i.e., the "Impact").
The actual time when each CLC data were taken was considered by spatially overlapping each PA network and all the CLC satellite scenes for Spain for 1987and 2006. CLC-1990and 2006 scenes were provided by Tragsatec Ltd. Those scenes specifying the exact dates when remote data were taken for each territorial 'window' allowed us to select only those PPs that had been designated after their overlapping CLC scenes for the initial time point (t1; "Before", around 1987) and those PPs that had been designated at least 3 years before the second time point (t2; "After", around 2005). That way, we made sure that baseline LULC data was consistently older than the analysed PPs, and that enough time was given to PAs designated before 2005 to have some environmental effect, thus minimising temporal confounding. Two buffer zones ("Controls") were generated for each PP: a proximal buffer of 1 km from PPs boundaries, and a distant buffer of 5 km. Both controls were assigned the PA category of the PP they were created from (e.g. 1 km, NR buffer).
We analysed our data using two models of increasing validity. Model 0 (M0) considered all PP and buffer areas. PAs from different networks (e.g. Biosphere Reserves) designated until the end of 2006 or other regulated areas that restrict LULC changes, including the public hydraulic domain (IGN, 2012) and the public coastal domain (IGN, 2015), were excluded from control areas in Model 1 (M1), as their protection status was likely to introduce confusion at ascertaining causality (Rodríguez-Rodríguez and Martínez-Vega, 2018a).
Cross-tabulation matrices (Pontius et al., 2004) were created for each PA network and related controls, for both models. We aimed at Note: NRs (Nature Reserves), NPs (Nature Parks), SCIs (Sites of Community Importance), SPAs (Special Protection Areas). a Based on Atauri et al. (2008).
identifying the most relevant LULC change processes by class between 1987 and 2006. Twenty-seven such processes were analysed (Table 3). In addition to the descriptive statistics generated by the matrices: absolute and relative change, class persistence, gains, losses, net change, etc., we calculated the Annual Rate of Change (rt) for each reclassified LULC class in cases and controls. For this, we used the following equation (Rodríguez Eraso et al., 2013):  Where A 1 y A 2 are the areas (in ha.) of the assessed LULC class in the initial year (t1) and final year (t2), respectively. We considered only LULC changes that affected at least 0.1% of each PA network area or control zone area ('noticeable changes'). Among these LULC changes, we considered 'relevant changes' those affecting at least 1% of each PA network or control zone. The environmental interpretation of LULC changes is straightforward only in few cases (e.g., changes towards artificial LULCs; Rodríguez-Rodríguez and Martínez-Vega, 2017). Many other LULC transitions are difficult to interpret from an environmental sustainability point of view (Rey-Benayas et al., 2007;Queiroz et al., 2014), especially CLC intra-class transitions (Gregor et al., 2016;Hewitt et al., 2016). In order to facilitate the interpretation of LULC changes, we followed the interpretation of LULC changes used previously (Rodríguez-Rodríguez and Martínez-Vega, 2017). We assumed that those changes and processes that increased 'naturalisation' (i.e., the trend towards ecosystem complexity and climax) were positive. Therefore, some processes such as abandonment of crops and related forest succession were deemed positive, even though agricultural areas and other semi-natural ecosystems provide key habitats for a number of endangered species in Spain (Araújo et al., 2007;Rodríguez-Rodríguez and Martínez-Vega, 2018b). The twenty-seven resulting LULC change processes were aggregated in 16 main ones for eased explanation as shown in Table 4.
The steps taken in the study are summarised in Fig. 2. Spatial analyses were made with ArcGIS v10.3 whereas matrix data analyses were made using Microsoft Excel.

Results
All the results shown here refer to those from Model 1. Results from Model 0 are shown only in the Appendices on space grounds.

LULC changes
There was high persistence (P) of existing LULCs in all PA networks (> 90%). NRs were the most dynamic PA category (P = 93.06). Fig. 3 summarises net LULC changes for the four PA networks and their control areas. Appendix 1 Supplementary data shows all the results of cross-tabulations for all PA networks, control areas, and models. Relevant net positive changes occurred in transitional woodland-shrub and agro-forestry areas in all PA networks. Noticeable net positive LULC changes in artificial surfaces and water bodies occurred in all PA networks except in NRs. The LULCs with the greatest losses and net negative changes inside PAs were: shrub, agricultural land with vegetation and non-irrigated arable land.
In proximal control areas, persistence of LULCs was smaller than in PAs, although greater than 90% in all networks. The smallest persistence occurred in NRs' proximal buffers. In this network, permanently irrigated land increased very relevantly, whereas in the other networks, relevant increases of artificial areas occurred, except in the SCI network where those areas increased very noticeably. In proximal buffer areas, noticeable losses of forests, transitional woodland shrub, shrubs, grasslands and sparsely vegetated areas also occurred.
In distant control areas, persistence was lower than in proximal buffer areas, although still high, around 90%, except around NRs (P = 81.65%). In these areas, former sparsely vegetated areas turned to irrigated lands very relevantly (> 11%). Artificial areas experienced relevant or noticeable net positive changes in all networks' distant controls (e.g. 1.84% around NPs). Burnt areas increased very noticeably around managed PAs (> 0.28%). Non-irrigated arable lands experienced net negative relevant changes in all distant control areas, whereas Table 3 Cross tabulation matrix between the simplified land-use land-cover classes of 1990 (in rows) and 2006 (in columns). The main processes of LULC change are indicated with an alphabetic and color key. (For interpretation of the references to colour in this Table legend,  forest areas, shrubs and grasslands experienced noticeable negative changes in those areas.
Artificial surfaces, mine and dump sites and green urban areas had the greatest rt inside PAs and in their control areas, except in NRs.

LULC change processes
Natural succession trends prevail inside PAs, especially inside NRs.
However, forest regression processes have also affected SCIs relevantly. Actually, forest regression outstands as the prevailing negative LULC change process in NPs, SCIs and SPAs. In NRs' control areas, the dominating process was new irrigated areas, affecting distant controls very relevantly. Land development was the prevailing LULC change process around NPs and SPAs, and the second most important process around NRs. In contrast, positive processes towards forest succession dominated around SCIs. Burned areas were only noticeable inside SCIs,  with no such process occurring in NRs or in their proximal controls. In all networks, there was generally an increasing gradient of burned areas from PAs' boundaries, except for SPAs. Forest regression exceeded forest succession around NPs. Table 5 summarises the main LULC change processes that have taken place in PAs and their control areas.

Comparison of both spatial-statistical models
The environmental balance of all LULC changes around PAs was less negative for M0 than for M1 in all PA networks except in SCIs (Table 6).

Main LULC changes and processes of change
The major LULC increases inside PAs were experienced by agroforestry areas and transitional woodland-shrub, whereas artificial surfaces, permanently irrigated land and burned areas prevailed in the proximal and distant controls. Conversely, the area of non-irrigated arable land decreased inside and around Spanish PAs. In addition, all the LULC classes associated with forest areas -forests, transitional woodland-shrub, shrubs, natural grasslands and sparsely vegetated areas-registered negative net changes in the proximal and distant controls. These land use dynamics relate to similar ones that have occurred in Spain (Jiménez, 2012;Martínez-Fernández et al., 2015) and in other Mediterranean (Brunori et al., 2016)  The main LULC processes of change inside and around Spanish PAs: forest succession, land development, and new irrigated areas largely relate to the three major LULC processes across Spanish landscapes in recent times (Stellmes et al., 2013): farmland abandonment (Vidal-Macua et al., 2018), intensification of farming practices (Custodio et al., 2016) and land development (Jiménez, 2012). LULC change processes related to land development and farming were far more environmentally sustainable inside Spanish PAs than outside (Martínez-  The LULC change processes are expressed in % with regard to the total area of each PA network or control area. NRs (Nature Reserves), NPs (Nature Parks), SCIs (Sites of Community Importance), SPAs (Special Protection Areas), B1k (1 km-buffer), B5k (5 km-buffer).
D. Rodríguez-Rodríguez et al. Int J Appl Earth Obs Geoinformation 74 (2019) 169-179 Fernández et al., 2015;Rodríguez-Rodríguez and Martínez-Vega, 2018a). Such processes are very relevant territorial-wise. Agricultural land covered half the terrestrial territory of Spain, contributed 2.75% of the country's GDP, and accounted for 75% of consumed water in 2005 (Jiménez, 2007). Unsustainable agricultural processes such as intensification of non-irrigated arable land, crops reconversion and, especially, new irrigated areas generally proliferated much less in PAs than in their surroundings, as expected due to regulation (Martínez-Fernández et al., 2015). Our results also show lesser abandonment of crops inside PAs, which appear to better maintain agricultural practices thus likely contributing to endangered species' conservation (Rodríguez-Rodríguez and Martínez-Vega, 2018b) and also probably being more socieconomically sustainable for the agricultural guild than outer unregulated areas, in contrast to common European farmers' claims (Kati et al., 2015;Blicharska et al., 2016). Notwithstanding worrisome land development processes in most PA networks, the good territorial performance of PAs and other sectoral legislation at reducing land development in Spain (Rodríguez-Rodríguez and Martínez-Vega, 2018a) seem to have partially offset insufficient territorial planning (Jiménez, 2010) in a country that increased its artificial areas almost double than the European Union-15's mean in the 1987-2000 period (Jiménez et al., 2005).

Environmental sustainability of PA networks
The four PA categories were more environmentally sustainable than their surroundings in the 1987-2006 period, as expected and shown previously (Jiménez, 2012;Martínez-Fernández et al., 2015). This is reflected by their greater LULC persistence and more positive LULC change balances. However, there was a gradient in the overall environmental sustainability of LULC change processes among the four PA networks: NRs > SCIs > SPAs > NPs. This result adds to evidence that different PA categories perform differently according to their legal and managerial characteristics (Seiferling et al., 2012;Linardi et al., 2013;Terra et al., 2014;Castro et al., 2015;Martínez-Fernández et al., 2015) and that grouping PA categories based on untested assumptions may lead to inaccurate results in PA effectiveness studies (Rodríguez-Rodríguez and Martínez-Vega, 2018a).
The relatively low effectiveness of NPs, where negative LULC change processes were dominant, indicates that PA management does not seem to have played a clear role in the sustainability of LULC changes in Spanish PAs, in contrast to common assumptions (Hockings et al., 2006;Dudley, 2008) and suggestions (Martínez-Fernández et al., 2015). In turn, multiple-use 'paper parks' (i.e., unmanaged SCIs and SPAs) performed moderately well environmentally, and better than NPs in general terms. The combination of legal stringency and management (as for NRs) resulted in high sustainability values, as shown previously for land development (Rodríguez-Rodríguez and Martínez-Vega, 2018a). In spite of good PA performance, environmentally negative LULC changes, including destructive, irreversible land development processes (McKinney, 2002), exceeded positive changes in all PA networks except in NRs, which cast reasonable doubts on the long-term conservation of biodiversity in those areas, should the drivers of those changes remain.
Negative agricultural LULC change processes were the smallest in SCIs and the largest in SPAs. Our results also show lesser abandonment of crops inside PAs, especially in managed PAs, which appear to best maintain farming practices and thus be also more socially sustainable than Natura 2000 sites, at least for the primary sector. Martínez-Fernández et al. (2015) found that Natura 2000 sites occupied more farmland area than nationally designated PAs and experienced greater agrarian abandonment in the studied period. The periurban location of many NPs is likely to have influenced greater persistence of agrarian activities with regard to more rural Natura 2000 sites (Rodríguez-Rodríguez and Martínez-Vega, 2018a), which are probably more affected by decreasing labour force due to emigration and ageing (Collantes and Pinilla, 2004).
Although LULC persistence was the lowest in NRs, LULC changes in that network were overwhelmingly positive, thus increasing its environmental sustainability. Good performance of legally stringent reserves (IUCN management categories Ia or Ib; Dudley, 2008) at conserving natural habitats have been shown in tropical areas (Linardi et al., 2013;Sims, 2014;Terra et al., 2014) and in temperate areas, including Spain (Castro et al., 2015;Rodríguez-Rodríguez and Martínez-Vega, 2018a), when compared to multiple-use PAs. In all the other networks assessed here, including managed NPs, environmental sustainability decreased in the assessed period. Natura 2000 sites (SCIs and SPAs) behaved similarly in terms of LULC persistence, sign and magnitude of LULC balance, and main LULC change processes, although SCIs were notably more sustainable than SPAs. In both categories forest succession and regression processes prevailed, respectively. However, largely destructive and irreversible land development and reservoir construction processes (McKinney, 2002) were much greater in SPAs, whereas less environmentally negative intensification of non-irrigated arable land and forest fires were much more relevant in SCIs. Forest fire-related processes were far greater in both Natura 2000 sites than in NPs and, especially, NRs, which suggests positive managerial influence on this important threat to Spanish PAs (Rodríguez-Rodríguez and Martínez-Vega, 2017).

Environmental sustainability of areas surrounding PAs
Control areas experienced chiefly unsustainable LULC change patterns, mirroring the developmental trends of the whole country in that period (Jiménez, 2012). However, LULC persistence values and LULC change balance values were consistently less negative in proximal controls than in distant ones, suggesting a 'protection spillover' from Spanish PAs (Sánchez-Azofeifa et al., 2003;Rodríguez-Rodríguez and Martínez-Vega, 2018a) through which the four PA networks 'irradiated' environmental sustainability to the rest of the territory.
New irrigated areas, land development and forest regression were the dominant LULC change processes in both control zones, although forest succession dominated around SCIs. This is worrisome, as surrounding pressures are likely to affect PAs' environmental effectiveness (Radeloff et al., 2010). Unsustainable territorial trends have been shown around the Spanish network of national parks (Rodríguez-Rodríguez and Martínez-Vega, 2017) and suggest the need for more intense environmental measures around Spanish PAs. A gradient of environmental sustainability of control areas was evident: SCIs > SPAs > NPs > NRs. Interestingly, the areas surrounding NRs and NPs were much more unsustainable than those around Natura 2000 sites, which tend to be located in rural areas. NPs and NRs were found to be the closest and most distant PA categories to main cities in Spain, respectively (Rodríguez-Rodríguez and Martínez-Vega, 2018a). This likely explains unsustainable LULC change processes around NPs. Interestingly, land development is also widespread around more rural NRs, whereas the other relevant LULC processes of change affect individual sites, for instance, expansion of new irrigated areas on the coast around Punta Entinas-Sabinar NR (Almeria), or expansion of nonirrigated arable land around Hoces del Cabriel NR. Endowing areas surrounding PAs some sort of territorial regulation buffering external impacts emerges as a meaningful recommendation to territorial planners. Existing regulations for buffering external impacts to PAs in Spain exist for National Parks (Spanish Government, 2014). Those buffer areas experienced fewer LULC changes than unprotected areas although forest fires affected both zones similarly (Rodríguez-Rodríguez and Martínez-Vega, 2017). Such external protection would most likely enhance sustainability and also increase connectivity across the heavily fragmented Spanish landscape (Torres et al., 2016), as nature conservation regulations state (EEC, 1992;Spanish Government, 2007).

Methodological remarks
As expected (Rodríguez-Rodríguez and Martínez-Vega, 2018a), M1 LULC change balances in control areas were more negative than in M0, after erasing additional territorial protection, which confirms previous claims on the improvable validity of default surrounding buffer areas used as controls in PA effectiveness studies (Andam et al., 2008;Spracklen et al., 2015).
Our study provides substantial improvements regarding previous studies on LULCs in PAs in Spain. Firstly, our study discriminates a larger number of CLC classes (22), possible LULC changes (484), and processes of change (15) than previous studies (Martínez-Fernández et al., 2015;Rodríguez-Rodríguez and Martínez-Vega, 2017). Here, we also distinguished effects of four specific PA networks rather than aggregated PA categories, which likely masks results. Moreover, we applied a more accurate method, both spatially (through the PP approach) and temporally, leading to more valid results. We also compared results from PAs with two control areas using two models of increased validity. However, some LULC changes are difficult to interpret from an environmental sustainability point of view, as trade-offs among species and habitats are expected (Gregor et al., 2016;Hewitt et al., 2016;van der Zanden et al., 2017).
The CORINE Land Cover project can be considered the largest operational application of remote sensing at European level for the detection of LULC changes (EEA, 1993;Bossard et al., 2000). Due to its spatial-temporal dimension, it is appropriate to apply a BACI approach. Thus, our work could be replicated in other study areas using the same procedure. However, CLC data may result in some inaccuracies mostly due to its small scale (Catalá Mateo et al., 2008;Díaz-Pacheco and Gutiérrez, 2013). For instance, apparent LULC transitions from water to farmland or other LULCs are common in dam shores, lake shores and river banks because no unique geographic baselines defining such entities have been established (e.g. upper water mark levels) and because the remote images used to produce CLC data were not normally taken at the same time of the year or, if they were, they may compare hydrologically distinct years. In spite of limitations, CLC is the most widely used source of information on LULCs that is available at European scale, allowing studies to be replicated in other locations and providing a long historical series, dating from 1987 (Pérez-Hoyos and García-Haro, 2013).

Conclusions
Spanish NRs, NPs, SCIs and SPAs experienced more environmentally sustainable LULC changes and processes of change than their surrounding areas in the 1987-2006 period. However, their effectiveness, assessed as persistence of LULCs and balance between positive and negative LULC change processes, differs, being the greatest in NRs and the smallest, in NPs, which suggests paramount importance of regulation stringency and limited influence of management on LULC changes (Rodríguez-Rodríguez and Martínez-Vega, 2018a). Agro-forestry areas and transitional woodland-shrub experienced the greatest increases inside PAs, whereas artificial surfaces, permanently irrigated land and burned areas dominated in the proximal and distant controls, indicating worrisome unsustainable LULC processes in those areas (Martínez-Fernández et al., 2015) such as new irrigated areas, land development, forest regression or burned areas that are not unique to the four PA networks studied here (Rodríguez-Rodríguez and Martínez-Vega, 2017). However, not all control areas were equally unsustainable. An environmental sustainability gradient emerged: SCIs > SPAs > NPs > NRs. Also, proximal controls were more sustainable than distant ones. Greater effort should be made to increase environmental sustainability in Spanish PAs, mainly in NPs, as well as in their surrounding areas. Methodologically, interpreting the environmental sustainability of some LULC changes and processes is challenging due to trade-offs (Gregor et al., 2016). Also, environmental regulations affecting control areas should be considered when assessing PA effectiveness, as not doing so affects result validity (Andam et al., 2008;Rodríguez-Rodríguez and Martínez-Vega, 2018a).