The Impact of Climate Change on the Sustainability of PGI Legume Cultivation: A Case Study from Spain

Abstract

Legume crops are sensitive to shifting environmental conditions, as they depend on a narrow range of climatic stability for growth and nitrogen fixation. This research sought to assess the sustainability of Faba Asturiana (FA) cultivation under current and future climatic scenarios by establishing generalized linear mixed models (GLMMs). Specifically, it aimed to (1) investigate the effects of significant climatic stressors, including higher nighttime temperatures and extended drought periods, on crop viability, (2) analyze future scenarios based on Representative Concentration Pathways (RCP 4.5 and RCP 8.5), and (3) recommend adaptive measures to mitigate threats. Six spatial GLMMs were developed, incorporating variables such as extreme temperatures, precipitation, and the drought duration. Under present-day conditions (1971–2000), all the models exhibited strong predictive performances (AUC: 0.840–0.887), with warm nights (tasminNa20) consistently showing a negative effect on suitability (coefficients: −0.58 to −1.16). Suitability projections under future climate scenarios revealed considerable variation among the developed models. Under RCP 4.5, Far Future, Model 1 projected a 7.9% increase in the mean suitability, while under RCP 8.5, Far Future, the same model showed a 78% decline. Models using extreme cold, drought, or precipitation as climatic stressors (e.g., Models 2–4) revealed the most significant suitability losses under RCP 8.5, with the reductions exceeding 90%. In contrast, comprising variables less affected by severe fluctuations, Model 6 showed relative stability in most of the developed scenarios. The model also produced the highest mean suitability (0.130 ± 0.207) in an extreme projective scenario. The results highlight that high night temperatures and prolonged drought periods are the most limiting factors for FA cultivation. ecological niche models (ENMs) performed well, with a mean AUC value of 0.991 (SD = 0.006) and a mean TSS of 0.963 (SD = 0.024). According to the modeling results, among the variables affecting the current distribution of Protected Geographical Indication-registered AF, prspellb1 (max consecutive dry days) had the highest effect of 28.3%. Applying advanced statistical analyses, this study provides important insights for policymakers and farmers, contributing to the long-term sustainability of PGI agroecosystems in a warming world.

1. Introduction

Climate change poses major threats to agroecosystems worldwide [1,2,3], including—but not limited to—frequent and intense weather anomalies, altered precipitation patterns, and rising temperatures [4,5,6]. Eventually, these changes will hamper the general food security and crop yield at both local and global scales [7,8,9]. Legume crops, in particular, are highly sensitive to changing conditions, as they rely on a narrow range of climatic stability for growth and nutrient fixation [10,11,12,13]. Climatic shifts have already led to losses due to increased heat stress and prolonged droughts, as observed in the Mediterranean region, known for its high climatic variation [14,15,16,17]. The recent Sixth Assessment Report (AR6) by the Intergovernmental Panel on Climate Change (IPCC) [18] suggests that overlapping drought and heat events will affect rainfed agricultural systems [19,20]. Such changes threaten the viability of traditional crops, such as those with a Protected Geographical Indication (PGI) [21,22]. Developing adaptation alternatives for the production of legumes is important to mitigate these adverse effects [23,24,25]. In particular, a spatially explicit assessment of vulnerabilities is crucial for determining region-specific adaptation requirements [26,27,28,29]. Therefore, evaluating the adaptive capacity and physical limitations of traditional agroecosystems requires the use of high-accuracy and robust climate modeling techniques [30,31].

PGI legumes can be considered to be more than conventional agricultural products, as they also provide cultural and economic benefits to local communities. These crops are able to grow under adverse geographical and agronomic conditions [32,33,34]. Although they perform well in extreme environments where other crops fail, they are particularly vulnerable to climatic variability and extreme weather events. Ensuring the sustainability of PGI legumes depends on preserving their distinct quality attributes, which are strongly linked to the climatic conditions [35,36,37]. However, the relationship between climate change and the agroecological needs of PGI legumes is poorly understood, particularly concerning future climate projections [38,39,40]. Environmental changes can reduce the crop yield and, in some cases, compromise a crop’s certification status by affecting quality standards [41,42,43]. Therefore, it is imperative to monitor and analyze these changes to ensure the long-term sustainability of PGI legumes and to identify the most influential climatic variables affecting their production [44,45,46,47].

In global policy discussions, the complex interrelationships between agricultural systems, environmental sustainability, and human nutrition have gained considerable attention [48,49,50,51,52]. The inclusion of these interconnections in the Sustainable Development Goals (SDGs)—particularly Zero Hunger (SDG 2), Good Health and Well-Being (SDG 3), and Clean Water and Sanitation (SDG 6)—emphasizes the recognition of systemic interdependencies [53], which, in turn, increases their strategic importance.

The 2016 Sustainable Food Systems Conference at Tel Aviv University was a turning point, as experts emphasized the connections among climate-resilient agriculture, dietary practices, and ecological well-being [54].

These discussions proved to be anticipatory, as later confirmed by one of the reports published in 2019 by the Food and Agriculture Organization of the United Nations (FAO) showing the rapid loss of agrobiodiversity [55]. The EAT–Lancet Commission on Food, Planet, Health (EAT–Lancet Commission) further assessed these links and demonstrated that sustainable food systems must simultaneously enhance their nutritional quality and address ecological degradation [56]. Additionally, the Farm to Fork Strategy, which was established as part of the European Green Deal in 2020, highlights specific actions to promote food production’s sustainability while tackling climate change and biodiversity loss [57].

In this context, PGI climate-exposed legumes have been promoted as essential systems for assessing sustainability initiatives [58,59,60,61].

Numerous studies have examined the impact of climate change on staple crops [62,63,64,65,66], and in some cases, PGI legumes have been considered [67,68,69]. Understanding the responses of these crops to environmental changes requires conducting in-depth and context-specific studies [69,70,71]. When the existing models fail to take into account spatial autocorrelation and multicollinearity between climatic variables, they often make biased predictions [72,73]. Furthermore, extreme climate indicators such as warm nights and consecutive dry days are still not widely used in suitability models, despite evidence of their physiological effects on legumes [74]. Research on the relationship between PGI legumes and climate change remains limited. Detailed, location-specific models are needed for us to better understand how these crops respond to local climate conditions [75,76]. Such models can help reveal complex crop–environment interactions [77]. They also support the development of effective regional adaptation strategies [78].

The ability to represent spatial dependencies and climatic variability in agricultural systems has been significantly enhanced by recent developments in spatial generalized linear mixed models (GLMMs), particularly for crops distributed within geographically constrained cultivation zones [79,80]. GLMMs facilitate the identification of fine-scale environmental stressors affecting crop performance by accounting for both fixed climatic drivers and spatially structured random effects. This is particularly important for statistical analyses that focus on legume species, as local agroecological conditions are closely linked to water needs and sensitivity to temperature. For example, a previous study using GLMMs to examine Vicia faba L. in Atlantic and Mediterranean regions illustrated how spatial modeling can help explain the effects of heat waves, warm nights, and variability in spring rainfall on phenology and yield stability [81]. These findings are of importance in PGI contexts, where it is essential to consider spatial variation in the climate conditions when evaluating potential threats to quality characteristics linked to certification. Similarly, in the case of PGI Asturian Faba (AF), stressors such as warm nights, prolonged droughts [82], and extreme precipitation events can interact in varying ways, even across short distances [83,84]. This clearly highlights the indispensable role of spatial GLMMs in vulnerability assessments [85].

Located in northern Spain, the Asturias region has an oceanic climate (Cfb according to the Köppen classification) with mild temperatures, high humidity, and regular rainfall [86]. The Asturian Faba (Phaseolus vulgaris L.) is a large white bean variety, oval and flattened in shape, with a very thin skin and a buttery texture after cooking. It is renowned for its ability to absorb the flavors of the ingredients it is cooked with, making it the main ingredient of the traditional ‘fabada asturiana’, a dish typical of the local gastronomy. Its cultivation is closely tied to the territory, traditional farming practices, and specific pedoclimatic conditions, which determine its quality and uniqueness [87]. Within these characteristic climatic conditions, the AF has successfully developed and adapted into its current form [88]. However, projections of higher nighttime temperatures and an increased drought frequency now [49] threaten traditional agricultural practices [89]. The existing climate records from Asturias already indicate a clear warming trend, raising concerns about changes in legume growth cycles and the reliability of yields [90]. In this context, we focus on a legume granted Protected Geographical Indication (PGI) status in 1990, as documented in issue no. 17 of the Spanish Official State Gazette (Boletín Oficial del Estado, BOE), which officially recognized its name and defined its production standards [91]. The crop is subject to a framework of European [21], national, and regional rules [92], including those set by the Asturias region [32], which require strict production methods such as a complete ban on chemical herbicides [41]. The land must be prepared using mechanical techniques without chemical treatment, creating suitable conditions for healthy plant development [93]. In addition to establishing a clear link between agricultural quality and the environmental conditions, these regulations serve as a guideline for integrating ecological considerations into local farming practices [94].

The aim of this study was to assess the sustainability of AF cultivation under current and future climate scenarios using GLMMs. To achieve this objective, three stages were implemented: (1) combining spatial autocorrelation analysis with high-resolution climate projections; (2) employing rigorous variable selection to identify key climatic stressors (e.g., drought periods, warm nights) and analyzing future scenarios based on Representative Concentration Pathways (RCP 4.5 and RCP 8.5); and (3) offering practical adaptation strategies for PGI systems. Representative Concentration Pathways (RCPs) are scenarios used in climate models to describe different future trajectories of greenhouse gas emissions and their effects on the climate. Each RCP is defined by its radiative force (in watts per square meter, W/m²) projected for the year 2100. RCP 4.5 is an intermediate scenario in which global emissions peak around 2040 and then decline, thanks to climate policies and sustainable technologies. In this case, the radiative force stabilizes at 4.5 W/m² by 2100. RCP 8.5 represents a high-emissions scenario, in which no significant policies are adopted to reduce greenhouse gases. Emissions continue to rise throughout the century, resulting in a radiative force of 8.5 W/m² by 2100 [95,96].

Climate change is hypothesized to significantly affect the agroclimatic adaptation of PGI legume cultivation, with major consequences for the yield, crop quality, and their geographical distribution. Specifically, shifts in temperature regimes, precipitation patterns, and the frequency of extreme weather events are expected to disrupt the delicate balance of environmental conditions required for optimal growth of PGI legumes. These changes may reduce the suitability of traditional cultivation areas, meaning that the implementation of adaptive strategies will be required to maintain production stability.

Moreover, the unique bioclimatic and cultural elements supporting PGI systems may encounter exceptional challenges, threatening agricultural sustainability and regional socio-economic resilience. By applying advanced statistical approaches, this study quantifies these impacts, identifies vulnerability thresholds, and recommends adaptation measures. The findings are expected to contribute to the protection of agricultural sustainability and the socio-economic value of these communities in the face of changing climatic conditions.

2. Materials and Methods

2.1. Study Area

This study was conducted in the northwest of Spain, in the Asturias region. Specifically, it focused on a legume with a Protected Geographical Indication (PGI) produced in this area: the AF. The climate of the region is oceanic, characterized by abundant rainfall distributed throughout the year, moderate temperatures, and high relative humidity [97,98]. These conditions support the growth of dense vegetation consisting mainly of permanent meadows, mixed forests, and forage crops. The predominant agricultural model in the region is extensive and family-run, with the strong integration of crop cultivation and livestock farming and a particular emphasis on quality and sustainability in local production systems [98,99]. This study area was selected because the area is geographically delimited by the production regulations for the PGI AF, which includes municipalities located mainly in river valleys and inland hilly areas [99]. The geographical coordinates of the agricultural parcels were obtained from the data provided at the regional level and subsequently mapped through georeferencing. Figure 1 displays the geography of the land and cadastral parcels where the legume under study is cultivated.

2.2. Data Collection

First, cadastral parcels within the study area were identified, from which geographical coordinates at the parcel level (X, Y) and data on the surface area of agricultural zones (m²) were extracted utilizing GIS tools and official cadastral records [100].

Climate data were retrieved from the official platform for Adaptation to Climate Change in Spain (AdapteCCa) [101], which provided access to climate variables for the current period (1971–2000) and two different scenarios, RCPs (Representative Concentration Pathways) 4.5 and 8.5, for the near future (2011–2040) and medium- (2041–2070) and long-term (2071–2100) periods.

2.3. Data Preprocessing and Variable Selection

A two-stage filtering procedure was employed in AdapteCCa to remove the least significant climate factors. This reduced the problem of multicollinearity and avoided redundancy among the variables. First, Pearson’s correlation analysis was performed using the Uncertainty Analysis for Species Distribution Models (usdm) package [102] on 31 climate variables in the AdapteCCa database. Only variables with high biological significance were retained, and pairs of variables with an observed correlation of less than 0.7 (r > 0.7) were eliminated. In the second step, variance inflation factor (VIF) analysis was performed using the R programming language version 4.4.2 [103], and the usdm package [102] was employed to assess the multicollinearity among the environmental variables. Variables with a VIF < 3 were considered and then used in the modeling process (

Table A1. VIFs of the remaining variables.

	Variables	VIF
1	prmax24_reference-period	1.863074
2	prspella1_reference-period	1.660775
3	prspellb1_reference-period	2.780308
4	tasdrp99_reference-period	1.798192
5	tasmaxhwdmax_reference-period	1.534910
6	tasmaxnap90_reference-period	1.002059
7	tasminna20_reference-period	2.069492
8	tasminnap90_reference-period	1.003214
9	tasminnb0_reference-period	2.091630

in Appendix A). The selected variables included important climate indicators such as temperature- and precipitation-related extreme events and the number of warm nights/days and consecutive dry/wet days. The nine variables that satisfied these criteria and were used in the analyses are listed in

Table 1. Climatic variables used in the analysis and their corresponding codes.

Variable Code	Variable Description
prmax24	Max precipitation (24 h)
prspella1	Max consecutive wet days
prspellb1	Max consecutive dry days
tasdrp99	99th percentile of temp range
tasmaxhwdmax	Max duration of heat waves

.

The selected variables, taken together, provided a comprehensive description of the temperature extremes, rainfall intensity, rainfall continuity, and main stress factors influencing legumes’ physiology and productivity. At this stage, Moran’s I test [104] was conducted and applied with the help of the Spatial Dependence (spdep) package [105] to assess whether the selected climatic variables were spatially correlated, i.e., in the geographical area of interest. This analysis was performed using Monte Carlo simulations [106] to determine the level of spatial autocorrelation of each variable. Moran’s I test was performed with 999 permutations for each variable using the function moran.mc(), and statistical significance was determined at the p < 0.05 level (

Table A2. Results of Moran’s I test using Monte Carlo simulations.

Variable	Moran’s_I	Expected_I	SD	Z_Score	p_Value	MC_Statistic	MC_p_Value
tasminNa20_ reference-period	0.57522	−0.0026	0.05516	10.475	5.623 × 10−26	0.57522	0.001
tasminNb0_ reference-period	0.14038	−0.0026	0.03013	4.745	1.042 × 10−6	0.14038	0.001
prmax24_ reference-period	0.02025	−0.0026	-	-	-	0.02025	0.006
tasmaxhwdmax_reference-period	0.01167	−0.0026	-	-	-	0.01167	0.026
prspella1_ reference-period	0.00729	−0.0026	-	-	-	0.00729	0.061
prspellb1_ reference-period	0.00716	−0.0026	-	-	-	0.00716	0.04
tasdrp99_ reference-period	0.0068	−0.0026	-	-	-	0.0068	0.067
tasmaxNap90_ reference-period	0.00113	−0.0026	-	-	-	0.00113	0.138
tasminNap90_ reference-period	0.00113	−0.0026	-	-	-	0.00113	0.128

in the Appendix A). Based on the results of the Monte Carlo simulation of Moran’s I test, five variables that showed significant spatial autocorrelation were retained for modeling: the days with a minimum temperature > 20 °C (tasminNa20), days with a minimum temperature < 0 °C (tasminNb0), maximum precipitation (24 h) (prmax24), maximum duration of heat waves (tasmaxhwdmax), and maximum number of consecutive dry days (prspellb1) (Table A2 in Appendix A).

2.4. Spatial Modeling Approach

Once the preprocessing steps had been performed, the spaMM package [107] was used to establish spatial generalized linear mixed models (spatial GLMMs) using six previously identified significant climate variables. These models were structured with a binomial distribution and a logit link function to explain the distribution of the binary response variable graphs of “productive (1)” and “unproductive (0).”

The construction of the spatial GLMMs allowed us to assess the effects of climatic and environmental variables on the cultivation of PGI legumes. In accordance with the recommendations of Zuur et al. [108], a structured protocol was followed to construct the model, which included the assessment of collinearity, stepwise inclusion of significant climate variables, and careful control of spatial autocorrelation. Using this approach, five climate variables were retained after excluding those affected by multicollinearity, based on the VIF thresholds and correlation analysis.

Spatial autocorrelation is a fundamental requirement in spatial ecological modeling guidelines [108], which is why a Matérn(1|X + Y) spatial correlation structure was incorporated into each model. This structure allowed for flexible modeling of the spatial dependence using the Euclidean distances between the observation points. Models were estimated using the fitme() function, and predictions were generated using the predict() function in the spaMM package.

The model performance was evaluated using receiver operating characteristic (ROC) analysis, the associated Area Under the Curve (AUC) values, and classification metrics such as the Kappa, sensitivity, and specificity. The AUC threshold of 0.84 was exceeded by all six models, which demonstrated acceptable levels of classification accuracy. Following the guidelines of Zuur et al. (2009) [108], we retained all six environmental model formulations rather than selecting only a subset based on model selection criteria. Among these, we compared the models using the commonly employed Akaike information criterion (AIC) by balancing the goodness of fit with the model complexity [109]. This approach allowed for a comprehensive exploration of how different combinations of temperature- and rainfall-related variables influence legume productivity under varying climatic stress conditions. The six spatial GLMMs were formulated as shown in

Table 2. Climatic models used to evaluate crop productivity: factors, formulas, and description of models.

Model Number	Factors Considered	Formula	Description of Model	Model Limitations
1	Tropical nights, heat waves, and dry periods	response~tasminNa20 + tasmaxhwdmax + prspellb1 + Matern(1|X + Y)	This model enables analysis of the combined effect of high night temperatures (tropical nights), prolonged heat waves, and prolonged drought periods on the probability of the productivity of the plots.	This model does not consider cold temperature stress (e.g., frost), which could be relevant in high-altitude or transitional zones.
2	Temperature extremes: frost and heat waves	response~tasminNa20 + tasminNb0 + tasmaxhwdmax + Matern (1|X + Y)	This model covers both cold and hot extremes, including the number of freezing days, tropical nights, and the heat wave duration, highlighting the variability of stress due to temperature.	This model does not include any precipitation-related variables, which limits its ability to assess water stress.
3	Extreme cold and heavy precipitation	response~tasminNa20 + tasminNb0 + tasmaxhwdmax + Matern (1|X + Y)	This model analyzes how the coexistence of frost events and intense short-duration precipitation influences agricultural productivity.	It does not account for prolonged drought periods, which may interact with cold stress in real scenarios.
4	Extreme cold and prolonged drought	response~tasminNa20 + tasminNb0 + prspellb1 + Matern (1|X + Y)	This model analyzes the impact of low-temperature stress (frost) in combination with prolonged drought periods, representing a double threat to the sustainability of legumes.	This model does not include high-temperature extremes (e.g., heat waves), which are increasingly relevant under climate change.
5	Short-term extreme precipitation and heat waves	response~tasminNa20 + prmax24 + tasmaxhwdmax + Matern (1|X + Y)	This model evaluates the influence of short-term precipitation and prolonged high-temperature events on field productivity.	It does not incorporate cold-related variables such as frost, limiting its scope to warm-season events.
6	Combined climate stress: full interaction	response~tasminNa20 + tasminNb0 + prmax24 + tasmaxhwdmax + prspellb1 + Matern (1|X + Y)	This model integrates all major thermal and precipitation stressors to represent the cumulative and interactive effects of climate extremes.	Its complexity increases the risk of overfitting and may reduce its interpretability in small or data-scarce regions.

.

By applying the stepwise, spatially contextual modeling approach proposed by Zuur and colleagues [108], these models provide both a statistically rigorous and ecologically interpretable framework. This is essential for understanding how multiple climate-related stressors interact to shape the spatial variability of PGI legumes’ productivity.

2.5. Application Based on Future Projections

The six basic models developed in this study were based on climate conditions from the reference period (1971–2000) and served as baseline scenarios. These models also allowed for the assessment of the future sustainability in the areas of PGI-AF cultivation. To this end, climate data for the near- (2011–2040), medium- (2041–2070), and far-future (2071–2100) periods were evaluated considering two Representative Concentration Paths (RCPs), an intermediate scenario (RCP 4.5) and a high-emissions scenario (RCP 8.5), using climate projections obtained from the AdapteCCa platform.

The six models were applied to future data sets without altering their structure, allowing comparative assessments to be made. Specifically, the following were analyzed: (1) the changes in the environmental threshold values identified by each model, (2) the spatial contraction or expansion of productive agricultural areas, and (3) the differences between the CPR scenarios.

2.6. Ecological Niche Modeling

In parallel with the generalized linear mixed model (GLMM) analyses, ecological niche models were developed to estimate the spatial distribution pattern of PGI legumes. The study area was delineated by first identifying the center coordinates of the plots. Around each central point, a buffer zone of 2 degrees was applied. Based on the extent of these buffered zones, a rectangular boundary box was constructed to define the overall study area. The center coordinates of the plots were collected as presence data; all records were georeferenced according to the WGS84 coordinate system, and their accuracy was checked using ArcGIS (v10.7, ESRI, Redlands, CA, USA). To reduce sampling bias and ensure a homogeneous sampling effort [110,111,112,113], a buffer zone with a 5 km radius was defined around each observation point, and the data was thinned using the spThin (v0.2.0; [114]) package in R version 4.4.2 [103], allowing only one point to be randomly selected within these areas. Initially, a total of 598 georeferenced point records corresponding to the midpoints of the plots were obtained, and these records were spatially thinned to 49 points. During the modeling process, current and future climate data were sourced from the AdapteCCa data set, and the six climate variables used in the GLMM analyses were also incorporated. Ecological niche models were developed to estimate the current and future habitat suitability of PGI legumes using the Species Distribution Modeling (sdm) package [115] running in the R environment. Five different algorithms were applied in the modeling: BIOCLIM [116], a generalized linear model (GLM; [117]), Boosted Regression Trees (BRTs; [118]), Random Forest (RF; [119]), and the Maximum Entropy (MaxEnt; [120]). All algorithms were run with 10,000 randomly selected pseudo-absence data points and default parameters in the sdm.

A 5-fold cross-validation method was employed to evaluate the model’s accuracy, with 100 repetitions performed for each algorithm. This process produced a total of 500 models (5 algorithms × 1 resampling method × 100 repetitions) to determine the current habitat suitability of the PGI legumes. To create ensemble models for each scenario, only the best models, ones that met the thresholds of a TSS > 0.7 and AUC > 0.9, were selected. To assemble these models, the predicted presence/absence values were averaged. Specifically, the predicted probabilities were first converted to binary (presence/absence) values using a specific threshold value, and then the average of these values was calculated. The selected models were projected onto both current and future climate scenarios. The model outputs represented the habitat suitability on a scale ranging from 0 (unsuitable) to 1 (suitable). The resulting ensemble suitability maps were converted into binary maps representing suitable environmental conditions. Visualization of the results was carried out using the RasterVis package [121].

3. Results

This study analyzed the sustainability of PGI AF in relation to reference period climate conditions and future climate scenarios using spatial generalized linear mixed models (spatial GLMMs). To identify the most parsimonious and efficient models, we tested different spatial GLMMs that incorporated various combinations of variables related to the temperature, precipitation, and drought. A model comparison was carried out using the Akaike (AIC) and Bayes (BIC) information criteria. Based on these criteria, the six models considered the most representative were chosen (

Table A3. SpaMM model comparison (Zuur-based).

Model	Num_Vars	AIC	BIC	logLik	Converged
tasminNa20_reference-period + tasmaxhwdmax_reference-period + prspellb1_reference-period	3	739.166990956824	750.74	−362.58	True
tasminNa20_reference-period + tasminNb0_reference-period + tasmaxhwdmax_reference-period	3	739.447062239629	751.02	−362.72	True
tasminNa20_reference-period + tasminNb0_reference-period + prmax24_reference-period	3	739.472813210192	751.05	−362.74	True
tasminNa20_reference-period + tasminNb0_reference-period + prspellb1_reference-period	3	739.482841176638	751.06	−362.74	True
tasminNa20_reference-period + prmax24_reference-period + tasmaxhwdmax_reference-period	3	739.536024287047	751.11	−362.77	True
tasminNb0_reference-period + prmax24_reference-period + tasmaxhwdmax_reference-period	3	740.115133947254	751.69	−363.06	True
tasminNb0_reference-period + prmax24_reference-period + prspellb1_reference-period	3	740.124032091255	751.7	−363.06	True
tasminNb0_reference-period + tasmaxhwdmax_reference-period + prspellb1_reference-period	3	740.142324533998	751.72	−363.07	True
tasminNa20_reference-period + prmax24_reference-period + prspellb1_reference-period	3	740.186315918666	751.76	−363.09	True
tasminNa20_reference-period + prmax24_reference-period + tasmaxhwdmax_reference-period + prspellb1_reference-period	4	741.160344975525	757.13	−362.58	True
tasminNa20_reference-period + tasminNb0_reference-period + tasmaxhwdmax_reference-period + prspellb1_reference-period	4	741.163557022221	757.13	−362.58	True
tasminNa20_reference-period + tasminNb0_reference-period + prmax24_reference-period + tasmaxhwdmax_reference-period	4	741.244205182667	757.21	−362.62	True
tasminNa20_reference-period + tasminNb0_reference-period + prmax24_reference-period + prspellb1_reference-period	4	741.358564865126	757.33	−362.68	True
prmax24_reference-period + tasmaxhwdmax_reference-period + prspellb1_reference-period	3	742.002566380944	753.58	−364.0	True
tasminNb0_reference-period + prmax24_reference-period + tasmaxhwdmax_reference-period + prspellb1_reference-period	4	742.044244663033	758.01	−363.02	True
tasminNa20_reference-period + tasminNb0_reference-period + prmax24_reference-period + tasmaxhwdmax_reference-period + prspellb1_reference-period	5	743.148614563252	763.51	−362.57	True

in Appendix A).

The six spatial GLMMs (M1-M6) showed similar behavior under the reference period climate conditions (1971–2000). The AUC (Area Under the Curve) values ranged from 0.840 to 0.887, with Model 5 (AUC = 0.887) showing the highest observed predictive accuracy. The sensitivity and specificity indices were also high, ranging between 76% and 91% for all the models, indicating strong performance in classifying the production potential (

Table 3. Reference period climate (1971–2000) conditions under which spatial GLMMs (M1–M6) were evaluated based on AUC, sensitivity, specificity, and Cohen’s Kappa metrics. All models showed acceptable predictive performance, though M5 ranked highest in all metrics. Values are rounded to three decimal places.

Model	AUC	Sensitivity	Specificity	Kappa
M5	0.887	0.792	0.911	0.631
M2	0.864	0.785	0.844	0.572
M4	0.860	0.766	0.860	0.561
M3	0.854	0.797	0.799	0.553
M6	0.842	0.785	0.765	0.511
M1	0.840	0.785	0.743	0.493

).

Models and Climatic Stress Variables

The models were structured to accommodate different combinations of climate variables to reflect specific aspects of climate stress. Model 1 integrated high-impact stress indicators such as warm nights (tasminNa20) and the number of consecutive dry days (prspellb1). Model 2 focused on frost events (tasminNb0) and drought. Model 3 incorporated extreme precipitation and drought, while Model 4 included an alternative combination of temperature variables and a different precipitation regime. Model 5 was based on moderate temperature and drought indicators, and Model 6 incorporated low-impact stress variables, such as temperature fluctuations. These structural variations were reflected in their different responses. For example, models sensitive to temperature and drought indicators (especially M1 and M2) predicted higher rates of decline, while models incorporating fewer stressors (especially M6) produced more robust forecasts. All the models shown included tasminNa20, which highlights the importance of high night temperatures in determining the climatic suitability for growing AF.

Analysis of the six spatial GLMMs showed that the climate variables influenced the predicted suitability in a distinct and model-specific manner (

Table 4. Standardized coefficient estimates for fixed effects included in spatial GLMMs (M1–M6) under reference period climate conditions. Negative values indicate negative influence on predicted suitability. All predictors were standardized prior to modeling (tasminNa20 = average minimum temperature over last 20 days; tasminNb0 = number of days with minimum temperature below 0 °C; taxmaxhwdmax = maximum temperature during longest heat wave; prspellb1 = duration of the first dry spell; prmax24 = maximum 24 h precipitation).

Model	Intercept	tasminNa20	tasminNb0	taxmaxhwdmax	prspellb1	prmax24
M1	0.859	−1.158	-	0.224	−0.072	-
M2	0.765	−0.735	0.020	0.019	-	-
M3	0.888	−0.583	0.026	-	-	−0.001
M4	0.822	−0.645	0.023	-	0.002	-
M5	0.815	−1.158	-	0.110	-	−0.010
M6	0.875	–1.069	0.006	0.186	–0.053	–0.003

).

The variable tasminNa20 (number of nights with a minimum temperature above 20 °C) was present in all the models and exerted a strong negative effect, with coefficients ranging from −0.58 to −1.16, highlighting the detrimental role of warm nights. Drought-related variables such as prspellb1 (number of consecutive dry days) showed weaker but still negative effects, particularly in M1 and M6. Precipitation-related predictors (prmax24) were included in M3, M5, and M6 but had coefficients close to zero, suggesting that this variable does not significantly explain the trend of the phenomenon. Frost-related variables (tasminNb0) showed smaller positive effects in M2-M4 and M6, indicating that occasional cold events may slightly enhance the suitability in some specific contexts. Models with more balanced or moderate coefficients (e.g., M6) also showed greater resilience in future scenarios. The models constructed suggest that the nighttime temperature and drought are the dominant limiting factors and that having the ability to manage or moderate their effects will be essential to ensure the productive sustainability of legumes in the face of climate change.

Under the reference period climate conditions (1971–2000), the six spatial GLMMs showed similar levels of environmental suitability for the cultivation of PGI AF, with mean probabilities of around 0.71. However, the projections under future scenarios based on Representative Concentration Pathways (RCP 4.5 and RCP 8.5) showed substantial differences in the model responses, as illustrated in Figure 2. The figure shows the percentage change in the suitability for the cultivation of AF for each of the six spatial GLMMs (M1–M6) under six future climate scenarios, compared to the reference period climate conditions.

A comparison of the six models revealed some notable differences regarding future climate scenarios. Specifically, Model 1, which focused on extreme temperatures and drought persistence, showed relative stability in RCP 4.5, Far Future, with an increase in the suitability of 7.9%. A substantial decrease of 78% in RCP 8.5, Far Future, also indicated high sensitivity to long-term warming and water stress. Models 2, 3, and 4 indicated a more substantial decrease, especially under RCP 8.5, Far Future, where in each the suitability decreased by more than 93%. Even in the medium-future projections (2041–2070) under RCP 4.5, these models showed notable losses, such as a 45.6% reduction in Model 2. Model 5, which included moderate stress indicators, demonstrated intermediate performance. Although its predicted suitability decreased by 21.9% in RCP 8.5, Far Future, and by 93.7% in RCP 8.5, Far Future, it remained relatively stable in RCP 8.5, Near Future, (2011–2040) with only a 4.2% decrease. The most resilient model was Model 6. It maintained a slight increase in suitability in RCP 4.5, Far Future (1.8%), and RCP 8.5, Near Future (5.6%), with only a small decrease in RCP 8.5, Mid-Century (4.9%), suggesting that its predictors were less sensitive to extreme climate variations. Overall, models that were more sensitive to temperature and drought indicators tended to show a sharper decline in suitability, especially under RCP 8.5. In contrast, Model 6 offered a more optimistic outlook, which is useful for designing adaptation strategies for sustainable legume cultivation under future climate change. The mean suitability values and respective standard deviations for each model under the different climate scenarios are presented in

Table 5. Mean predicted suitability values (±standard deviation) for each spatial GLMM (M1–M6) under reference period and future climate scenarios. Predictions were made by spatial GLMMs using PGI Asturian Faba cultivation data. RCP = Representative Concentration Pathway.

Model	Reference Period	RCP 4.5 Near	RCP 4.5 Mid	RCP 4.5 Far	RCP 8.5 Near	RCP 8.5 Mid	RCP 8.5 Far
M1	0.709 ± 0.047	0.760 ± 0.080	0.538 ± 0.324	0.766 ± 0.199	0.774 ± 0.082	0.722 ± 0.213	0.156 ± 0.236
M2	0.709 ± 0.047	0.629 ± 0.080	0.387 ± 0.203	0.466 ± 0.164	0.639 ± 0.089	0.417 ± 0.167	0.040 ± 0.085
M3	0.709 ± 0.047	0.617 ± 0.072	0.399 ± 0.177	0.460 ± 0.152	0.631 ± 0.084	0.419 ± 0.155	0.049 ± 0.087
M4	0.709 ± 0.047	0.617 ± 0.077	0.382 ± 0.183	0.446 ± 0.157	0.629 ± 0.088	0.402 ± 0.159	0.040 ± 0.080
M5	0.709 ± 0.047	0.680 ± 0.101	0.410 ± 0.280	0.554 ± 0.197	0.682 ± 0.105	0.487 ± 0.201	0.054 ± 0.115
M6	0.709 ± 0.047	0.735 ± 0.080	0.516 ± 0.308	0.723 ± 0.194	0.749 ± 0.083	0.675 ± 0.206	0.130 ± 0.207

.

After analyzing the results shown in this table, we can establish that in the present-day climate conditions (1971–2000) all the models produced very similar suitability estimates, with the mean values centered around 0.709 and the standard deviations ranging from 0.047 to 0.048, indicating coherence in the models. However, if we instead consider the future scenarios, we notice notable differences between the models, in particular in the case of RCP 8.5, Far Future (2071–2100). Models 2, 3, and 4 showed the steepest declines, with mean predicted suitability values that fell to 0.040–0.049 and showed greater variability (e.g., SD = 0.085–0.087), which demonstrates greater uncertainty. These models appeared to be highly sensitive to the intensification of climate stressors and effectively highlighted the influence of frost conditions, extreme precipitation, and prolonged drought.

Model 1 performed well under the RCP 4.5 scenarios, maintaining high suitability in both the near and far future (0.760 ± 0.080 and 0.766 ± 0.199, respectively). However, a significant deterioration in its performance was highlighted under RCP 8.5, Far Future (0.156 ± 0.236), indicating vulnerability to extreme warming and water stress.

Model 5 exhibited moderate reductions across all the future projections but maintained intermediate fitness in RCP 4.5 scenarios. The best-fitting estimation model, Model 6, consistently provided the most reliable estimates in all the scenarios. It maintained high mean values (e.g., 0.735 ± 0.080 in RCP 4.5, Near Future, and 0.723 ± 0.194 in RCP 4.5, Far Future), and even under RCP 8.5, Far Future, it produced the highest mean (0.130 ± 0.207) among all the models.

This result demonstrates that models incorporating stressors due to extreme temperatures and drought tended to predict greater losses of environmental suitability under future climatic conditions. In contrast, the stable performance of Model 6 suggests that the inclusion of more stable climate variables that are less subject to extreme fluctuations can lead to more reliable projections, making it a valuable basis for adaptive agricultural planning under climate change.

Comparing the future climate scenarios with the reference period climate (1971–2000),

Table 6. Percentage change in mean predicted suitability values under six future climate scenarios relative to reference period climate (1971–2000) for each of six spatial GLMMs. Positive values indicate improved suitability; negative values indicate losses. RCP = Representative Concentration Pathway.

Model	RCP 4.5 Near	RCP 4.5 Mid	RCP 4.5 Far	RCP 8.5 Near	RCP 8.5 Mid	RCP 8.5 Far
M1	+7.0%	−24.3%	+7.9%	+9.0%	+1.7%	−78.0%
M2	−11.4%	−45.6%	−34.4%	−9.9%	−41.3%	−94.3%
M3	−13.1%	−43.8%	−35.2%	−11.1%	−40.9%	−93.1%
M4	−13.1%	−46.1%	−37.1%	−11.5%	−43.4%	−94.3%
M5	−4.2%	−42.3%	−21.9%	−3.9%	−31.4%	−93.7%
M6	+3.5%	–27.3%	+1.8%	+5.6%	+5.0%	–81.7%

presents a summary of the percentage change in the mean expected suitability values for every one of the six spatial GLMMs. Positive values indicate a possible loss, while negative values suggest an improvement in the crop suitability.

The results indicate that each model applied to each climate scenario responded differently to changing climatic conditions. Models 2, 3, and 4 exhibited the most severe declines across all future scenarios, with a decline in the projected suitability of more than 90% under RCP 8.5, Far Future (2071–2100). The proposed models demonstrated a decline of more than 40% under RCP 4.5, Mid-Century (2041–2070), suggesting strong sensitivity to both long-term warming and aridity. Model 1 performed well under RCP 4.5, Near and Far Future, showing modest increases of 7–8%. However, its projected suitability dropped significantly by 78% under RCP 8.5, Far Future, indicating a gradual decline under enhanced climate stress.

Model 5 showed intermediate sensitivity. It remained relatively stable in RCP 4.5, Near Future (−4.2%), but experienced a significant reduction in 8.5F (−93.7%). Model 6 consistently showed the smallest changes across the future periods. Its predicted suitability increased slightly under RCP 4.5, Near and Far Future, and under RCP 8.5, Near Future, and never declined by more than 5% in the mid-century scenarios. This trend suggests that the combination of predictors in Model 6 provides the most robust basis for projecting suitability under future climate change conditions.

The findings make it clear that while some models declined sharply under enhanced warming (particularly under RCP 8.5—Far Future), others remained fairly stable, providing insights into potential pathways for climate-wise planning.

When considering future scenarios (RCP 4.5 and RCP 8.5), the model responses diverged significantly (Figure 2, Table 4 and Table 5). The suitability predicted by M1 remained relatively stable under RCP 4.5, Far Future, but declined sharply (−78%) under RCP 8.5, Far Future. M2, M3, and M4 consistently showed strong suitability declines, exceeding −90% under RCP 8.5, Far Future, and –40% under RCP 4.5, Mid-Century. M5 showed moderate stability in the short term but lost over 90% of its predicted suitability in the most extreme scenarios. M6 was the most resilient model, with the variations in the predicted suitability mostly remaining within ±5% in all the periods except 8.5, Far Future, where it decreased by 81.7%.

This table presents the differences between the spatial GLMMs under the reference period climate conditions and their responses to the future climate scenarios. These results changed significantly depending on which climate variables were chosen and how they were set. In particular, the models incorporating indicators related to increased temperatures and drought, Models 1 to 4, tended to show a marked decline in suitability in high-emission environments, particularly RCP 8.5, Far Future. In contrast, Model 6, which included more moderate and less volatile predictors, generated projections that were more stable, suggesting greater resilience to climate stressors. The consistent negative influence of warm nights in all the models demonstrates the increasing threat posed by high night temperatures. Overall, the results highlight the importance of developing specific adaptation strategies that account for the different sensitivities of production systems. Furthermore, they underscore the value of integrating robust predictors into forward-looking agricultural planning. In the case of the PGI AF, a combination of climate-resilient practices and cultivar selection will be essential to sustain productivity under rapidly changing environmental conditions.

Overall, our ecological niche models (ENMs) performed well, with an average AUC of 0.991 (SD = 0.006) and an average TSS of 0.963 (SD = 0.024), as shown in

Table 7. Comparative performance metrics of five ecological niche modeling algorithms (AUC = Area Under the Curve; COR = Correlation Coefficient; TSS = True Skill Statistic).

Method	AUC	COR	TSS	Deviance
Bioclim	0.92	0.38	0.84	0.03
GLM	0.98	0.24	0.95	0.03
BRT	0.96	0.25	0.91	0.04
RF	0.98	0.15	0.95	0.04
Maxent	0.99	0.38	0.96	0.08

.

According to the modeling results, among the variables influencing the current distribution of PGI AF, prspellb1 (max consecutive dry days) had the greatest effect of 28.3%. This indicates that prolonged dry periods affect the distribution of the species. The second most influential variable was the maximum duration of heat waves (tasmaxhwdmax), accounting for 23.8% of the variance. This shows that periods of extreme heat also influence the habitat suitability of the species. Other significant variables included tasminnb0 (days with a min temp < 0 °C) with 21.5%, tasminna20 (days with a min temp > 20 °C) with 16.9%, and prmax24 (max precipitation in 24 h) with 9.4%. These results suggest that both temperature-related extreme climate events and drought characteristics are the main climate variables determining the distribution of PGI AF. In particular, the intensity of heat waves and the dominant effects of prolonged drought periods may play a critical role in determining the potential future impacts of climate change. The modeling also showed that areas with high habitat suitability for the PGI legume species are currently limited and largely concentrated in and around the existing cultivation zones (see Figure 3).

However, future climate scenarios—particularly RCP 8.5—indicate substantial shifts in the spatial distribution of suitable habitats (Figure 4).

There is clear evidence of declining levels of suitability, and it is anticipated that some regions may become entirely unsuitable. Moreover, the areas currently classified as suitable may become unsuitable in the future considering the increasing drought and high-temperature stress (see Figure 5).

These results suggest a hypothetical decrease in cultivation areas, both in terms of the extent and actual likelihood of occurrence. Considering the future projections across the near, medium, and distant time horizons under both the RCP 4.5 and 8.5 scenarios, there will be a decrease in the appropriate areas and the fragmentation of high-quality areas, leading to a loss of ecological connections between habitats (see Figure 5). This trend is especially pronounced under RCP scenario 8.5. While climate change is expected to reduce the suitability across many regions, certain small areas may become more suitable in the future. These isolated zones could play a critical role in the conservation and continued cultivation of the species. In summary, not only is the geographical distribution of cultivation likely to shift, but the overall quality is expected to decline, making it difficult to maintain the species.

4. Discussion

The accelerating pace of climate change is expected to influence PGI AF in northern Spain. By developing six models incorporating different climatic variables, the results of this study have revealed both the sensitivity of this crop to future climate change and the comparative robustness of the model structures. All the models showed good predictive performance under historical conditions (AUC = 0.840–0.887), which supports the reliability of the model structures and the historical suitability of the region. The findings align with those of Ramirez-Villegas et al. [122], who emphasized that the selection of climate variables, model architecture, and environmental characterization influence crop–climate predictions and their interpretability. Among the variables examined, the warm-night frequency (tasminNa20) proved to be the most reliable and adverse factor, consistently present in all the models with major adverse coefficients (−0.58 to −1.16). This finding is supported by previous studies that indicate that high nighttime temperatures in legumes enhance respiration rates, diminish carbohydrate accumulation, and disrupt seed formation [123,124,125]. Similar effects have been seen in many legume species (e.g., Vicia faba, Phaseolus vulgaris), where nighttime temperature stress has been found to decrease the grain-filling periods and pod growth, especially under increased-humidity climatic conditions [126,127,128,129]. A recent study by Manning et al. [126] revealed that high nighttime temperatures significantly impaired pod development and yield components in legumes [36,37]. This phenomenon was attributed to disruptions in reproductive development during critical phenological stages. The consistent significance of tasminNa20 across all the models highlights both its physiological relevance and its critical role in limiting legume productivity under high-temperature stress.

Each of the six models developed for this study produced different quantitative results under future emission scenarios. Model 2, which combined tasminNa20 with the drought frequency (prspellb1), showed suitability losses exceeding 90% under RCP 8.5, highlighting the compounding impact of thermal and hydric stressors. These findings confirm the sensitivity of legumes to simultaneous stress exposure. This was discussed by Lobell and Gourdji [130], who provided important information on how consecutive drought years adversely affect the yield potential of pulse crops [15,20,131]. Model 3 tested potential compensation for warm-night effects by mid-season rainfall variability. However, contrary to earlier assumptions, the suitability did not remain stable under all the projections. The model results show that rainfall irregularity does not mitigate the physiological impacts of thermal stress. These findings align with those of Fraga et al. [132], who documented the destabilizing effects of irregular precipitation on Mediterranean legumes. Model 4, which focused on temperature seasonality and short-term extreme rainfall (prmax24), provided complementary findings. This model first projected dispersed refugia under RCP 4.5, but the suitability was entirely lost under RCP 8.5. This implies the increasing influence of climate extremes. These sharp decreases are in line with the findings of Vadez et al. [133], who proposed that phenological failure in legume ecosystems results from sudden hydroclimatic anomalies instead of gradual change. Models 5 and 6 provided a methodological contrast. Model 5 combined minimum temperature anomalies with the dry spell length to generate uneven trend lines and small AUC values. As a result, the model had a constrained predictive capability over heterogeneous surfaces. This supports the recommendation of Urban et al. [134] regarding the indiscriminate use of generalized climate indicators in spatially complex environments. By contrast, Model 6—based on long-term climatic averages—emerged as the most stable structure, exhibiting suitable variation within ±5% for both RCP 4.5 and RCP 8.5. Its explanatory strength was not affected by its simplicity, which is consistent with the findings of Schauberger et al. [135], who identified the higher predictive value of climatological persistence over episodic extreme events. Subsequently, Model 6 appears to be especially suitable for the identification of resilient central production areas and may serve as a basis for long-term agricultural planning in the context of increasing uncertainties regarding the future climate [1,23].

Despite the widespread application of geographical GLMMs and the inclusion of several climate indicators, the variable selection limits the interpretive and predictive power of this study. A prominent limitation is the exclusion of non-climatic variables, such as soil properties, agricultural management practices, and socio-economic factors, which can influence legume production. A significant limitation lies in the temporal resolution of the climate predictions used. Although the RCP 4.5 and 8.5 scenarios are widely accepted, they remain assumptions that may fail to account for local extremes, interannual variability, or decadal anomalies [136,137,138].

The TSS value of 0.963 and the mean ensemble AUC of 0.991 resulting from the ecological niche modeling demonstrate excellent predictive accuracy. The strong performance of the model shows that its results exceed the set criteria for assessing the model quality in species distribution modeling [139]. This exceptional outcome aligns closely with previous modeling efforts focused on legumes. Ramírez et al. [140] applied a mechanistic modeling approach for P. vulgaris in the Andean and Mesoamerican regions, suggesting analogous patterns of regional climatic adaptability changes in response to rising temperatures. Their application of climatic thresholds and process-informed variables produced forecasts that aligned with recognized agricultural patterns. Similarly, De Mattos et al. [141] emphasized the importance of incorporating climatic extremes, such as maximum temperatures and the drought length, into niche models for crops cultivated in Mediterranean climates. Their results showed that models that included such extremes outperformed those based solely on long-term averages. These results support our use of the drought length and heat wave-related indices as the main predictors of suitable areas for AF cultivation and imply that our model configurations are ecologically and statistically suitable for the present conditions.

More importantly, our models identified the drought duration (prspellb1) and heat wave duration (tasmaxhwdmax) as the most influential variables explaining the current and future habitat suitability for AF. The significant contributions of these variables are consistent with findings from agronomic and physiological studies on beans. Cortés et al. [142] employed genome–environment association analysis to validate the adaptive importance of heat and water deficits, revealing that populations of wild P. vulgaris thriving in drought-prone regions had specific genotypic patterns linked to these stressors. Stefanov et al. [143] investigated the physiological responses of common bean cultivars to combined heat and drought, finding substantial changes in their chlorophyll fluorescence and photosystem stability following short-term stress exposure. These physiological limitations are known to reduce the yield and seed quality, particularly when challenges occur during flowering or the early stages of pod formation [125]. Therefore, the dominance of temperature- and drought-related indices in our models reflects genuine biological constraints rather than a bias introduced by the variable selection. Our interpretation that AF cultivation will face increasing challenges under intensifying climatic extremes is further supported by the alignment between the model outputs and findings from experimental studies.

The expected decrease in suitable agricultural land for AF becomes particularly concerning under the projected RCP 8.5 scenarios. Under RCP 4.5, some high-altitude or coastal areas may remain somewhat suitable; almost all the models showed significant range contraction under high-emission scenarios. The primary causes of suitability losses in this study were extreme heat events and prolonged periods of consecutive dry days, conditions that compromise plants’ water balance and reproductive development. In a similar way, Lobell et al. [130] pointed out how common bean yields in sub-Saharan Africa are highly sensitive to temperature rises at night and the timing of dry periods. This suggests that even small changes in precipitation regimes might result in unstable yields. Although AF is usually grown under Atlantic climatic conditions with mild thermal regimes, its expected vulnerability under warming scenarios reflects larger trends observed in bean-producing areas worldwide. This suggests that, particularly in conventional farming areas, the physiological thresholds for thermal and hydric stress are narrow and can be easily exceeded under high-emission scenarios. By combining five algorithms (GLM, MaxEnt, Random Forest, BRT, and BIOCLIM) into an ensemble, our modeling technique improved the model reliability and characterization of environmental aspects. The integration of statistical and machine learning techniques helps to effectively establish complementary patterns in species–climate interactions [141]. In future climate scenarios, the ensemble models proved more effective in identifying potential refugia and defining appropriate range limits. Our study highlighted areas that consistently showed relevance across different models and confirmed that an ensemble approach reduces the uncertainty associated with individual modeling techniques. These consistently suitable areas could be prioritized in long-term agricultural planning aimed at developing climate-adaptable crops.

5. Conclusions

This study, focusing on Protected Geographical Indication (PGI) Faba Asturiana cultivated in northern Spain, analyzed the vulnerability of the legume in relation to the effects of climate change, highlighting in particular its sensitivity to factors such as increased nighttime temperatures, the duration of drought periods, and precipitation instability. Integration of generalized linear mixed models (GLMMs) with ecological niche models (ENMs) allowed for the prediction of possible reductions in crop suitability under future high-emission scenarios. The six GLMMs developed showed different responses depending on the climatic factors considered, underscoring the complexity in predicting the future crop suitability. Despite the robustness of the analysis, some limitations remain due to the exclusion of non-climate variables and the temporal resolution of the RCP scenarios. To mitigate risks, adaptive strategies such as modification of planting calendars, use of reflective mulch, nighttime irrigation, and introduction of heat- and drought-resistant varieties are recommended. From a regulatory perspective, PGI certification could benefit from flexible updates to incorporate environmental changes. In the future, there are plans to extend this modeling approach to other PGI legume species grown in different geographical settings.