Predicting Climate Change Impacts on Sub-Tropical Fruit Suitability Using MaxEnt: A Regional Study from Southern Türkiye

Abstract

This study, conducted in Mersin, a Mediterranean sub-tropical area, examined the potential of avocado and pitaya to thrive under current and future climate conditions. Researchers utilized climate and soil data, initially selecting 14 parameters (mean annual temperature, mean minimum temperature of the coldest month, mean maximum temperature of the warmest month, mean annual precipitation, soil texture, soil depth, land use capability, soil pH, soil organic carbon, soil salinity, land cover, elevation, slope, and groundwater level) for analysis, which were narrowed down to 12 after correlation analysis. The potential distributions were projected using the MaxEnt model for current and future scenarios. Three global climate models—HadGEM3-GC31-LL, MPI-ESM1-2-HR, and GFDL-ESM4—were utilized under the SSP2-4.5 and SSP5-8.5 scenarios. Under SSP2-4.5, an average increase of 1.32%, 1.95%, and 4.02% in the “S1” class is expected. For SSP5-8.5, average gains of 1.33%, 1.58%, and 0.77% are projected. In Pitaya, the “S1” class in SSP2-4.5 is expected to increase by 0.96% compared to the first model and decrease by 7.06% and 5.71% compared to the other models, respectively. Under SSP5-8.5, the changes are determined to be 1.49%, −7.27%, and −7.28%, respectively. Our findings indicate that climate change poses a significant threat to the region; however, the application demonstrates that agricultural activities can remain sustainable despite climate change impacts.

1. Introduction

Product selection is a fundamental pillar for achieving the long-term and sustainable goals of good agricultural practices [1]. This objective is of immense significance not only for small-scale local farmers but also for large-scale plantation agriculture. Furthermore, it contributes to the development of the national economy. To establish a robust and sustainable agricultural vision, it is essential to thoroughly analyze the climate characteristics across different regions and assess the potential impacts of global climate change (global warming). Anticipated shifts in temperature and precipitation patterns will undoubtedly affect agricultural production. As a result, we can expect changes in the regions where specific crops are grown, transformations in the types of agricultural products produced, and an overall decline in yields [2]. Some regions are already witnessing these changes. Drought, in particular, is one of the most significant adverse effects of global warming in Mediterranean basin regions, severely impacting agricultural products [3,4]. Ref. [5] has shown that climate change negatively affects avocado production in the Mediterranean region. Additionally, Ref. [6] indicates that arabica and robusta coffee varieties are highly vulnerable to climate change, which has significant consequences for major coffee-producing countries like Brazil and Vietnam. Another study on Arabica coffee by the authors of [7] concluded that climate change would adversely impact current production regions, reducing yield and quality. Some studies examining climate change’s effects on agricultural products are presented in

Table 1. Studies in the literature investigating the impacts of climate change on crops.

Reference	Crop	Study Area	Goals of the Studies
[8]	Pitaya	Vietnam		Determining the impacts of climate change on agricultural crop patterns
[9]	Kiwi	Kastamonu/Türkiye
[10]	Avocado	Michoacán/Mexico
[11]	Kiwi and apple	New Zealand
[12]	Pitaya	Brazil
[13]	Coffee	Mexico
[14]	Avocado	Michoacán/Mexico
[15]	Pitaya	Vietnam
[16]	Pitaya	Central America
[17]	Kiwi	New Zealand
[18]	Kiwi	New Zealand
[19]	Coffee	Nicaragua
[20]	Coffee	Zimbabwe
[21]	Coffee and banana	Nepal
[22]	Coffee	Indonesia

.

Research shows climate change significantly threatens agricultural products, and its effects will likely increase. Therefore, to identify suitable agricultural products and policies, it is essential first to assess the climate characteristics of different regions and consider potential climate scenarios. To achieve this, various climate models and scenarios are employed (

Table 2. Models and scenarios employed in studies on climate change in the literature.

Reference	Model	Scenario
[23]	CanESM5, MPI-ESM1-2-HR, EC-Earth3, NorESM2-LM	SSP2-4.5, SSP5-8.5
[24]	CanESM2	RCP 2.6, RCP 4.5, RCP 8.5
[25]	EC-EARTH, HadGEM2-ES, MIROC5, MPI-ESM	RCP4.5, RCP8.5
[3]	CCSM4	RCP4.5, RCP8.5
[26]	CCSM4	RCP4.5, RCP8.5
[27]	Access-CM2, HadGEM, UKESM1	SSP1-2.6, SSP2-4.5, SSP5-8.5
[28]	ACCESS-ESM1, BCC-CSM-MR	SSP2-4.5, SSP5-8.5
[29]	MIROC-ES2L, BCC-CSM2-MR	SSP1-2.6, 2, SSP2-4.5, SSP5-8.5
[12]	BCC-CSM1-1	RCP2.6, RCP4.5, RCP6.0, RCP8.5
[30]	GISS-E2.1	SSP1-2.6, SSP2-4.5, SSP4-6.0, SSP5-8.5
[1]	BBC-CSM1-1, CCSM4, CNRM-CM5, GFDL-CM3, GISS-E2-R, HadGEM2-AO, HadGEM2-ES, IPSL-CM5A-LR, MIROC-ESM-CHEM, MIROC- ESM, MIROC5, MPI-ESM-LR, MRI-CGCM3, NorESM1-M	RCP2.6, RCP4.5, RCP8.5
[11]	CC-CSM1.1, CESM1-CAM5, GFDL-CM3, GISS-EL-R, HadGEM2-ES, NorESM1-M	RCP2.6, RCP8.5
[31]	CanESM2	RCP2.6, RCP4.5, RCP8.5
[32]	CESM2, GFDL-ESM4, IPSL-CM6A-LR, MIROC6, MRI-ESM2-0	SSP2-4.5, SSP5-8.5
[33]	MIROC-ESM, CCSM4	RCP2.6, RCP4.5, RCP8.5
[34]	HadGEM2ES, RegCM4.3.4	RCP4.5, RCP8.5

).

The WorldClim platform displays current and potential future climate characteristics using several climate models and scenarios. These data enable agricultural suitability analysis to be performed. In this investigation, due to the place of the study area and the recommendation of the General Directorate of Meteorology (GDM), global climate models (GCMs) within the scope of the Coupled Model Intercomparison Project Phase 6 (CMIP6) and HadGEM3-GC31-LL, MPI-ESM1-2-HR, and GFDL-ESM4 GCMs were used under the SSP2-4.5 and SSP5-8.5 scenarios from the shared socio-economic pathways (SSPs) included in the sixth Assessment Report of the Intergovernmental Panel on Climate Change (IPCC).

In order to achieve a sustainable agriculture goal, it is essential to decide the suitable places to grow agricultural products after determining the current and future possible climate characteristics of the regions. Different models, methods, and parameters are used for this purpose. Different species distribution models (SDMs) such as maximum entropy (MaxEnt) [3,26,33,35,36], the generalized linear model (GLM) [37,38], the general additive model (GAM) [39,40], and the generalized and boosted model (GBM) [41,42] have frequently opted in contemporary years. Nonetheless, machine learning (ML) methods such as extreme gradient boosting (XgBoost) [43], fuzzy logic (FL) [44], the logistic regression model (LGM) [45], random forest (RF) [46], frequency ratio (FR) [47], artificial neural networks (ANNs) [48], K-nearest neighbor (KNN) [49], and support vector machines (SVMs) [50] are also used. In addition to the models and methods mentioned, remote sensing (RS) [51] and Geographic Information System (GIS)-based Analytical Hierarchy Process (AHP) methods are also widely utilized in suitable location analysis [52,53,54,55,56,57]. Although this method is time-consuming, it has high reliability and thus is preferred [58]. In [35], MaxEnt and GLM were used to model the potential distribution of the elements (pollinators and pests) affecting the spatial distribution of avocado. In [59], in Mexico, the MaxEnt model was again preferred to determine the potential distribution of avocado sunspot viroid for avocado species. In [60], applications were carried out for 225 species collected from six different regions using methods such as MaxEnt, the GBM, XgBoost, the SVM, and RF, and the performances of these methods were analyzed. Accordingly, the MaxEnt method gave the most successful result. There are many studies in the literature that were carried out using the mentioned methods (

Table 3. Studies in the literature conducting agricultural suitability analysis.

		Parameters **
Reference	Method	T	P	S	SpH	SS	SD	Luc	E	Sl	A	G	PA	Lu/Lc
[3]	MaxEnt	●	●
[26]	MaxEnt, GLM, GBM, GAM, ANN, RF, FDA *, CTA *, SRE *	●	●
[33]	MaxEnt, GLM, GAM, RF, BIOCLIM *	●	●
[61]	MaxEnt, CLIMEX *, SVM	●	●
[1]	CONSUS *	●	●	●	●	●				●			●	●
[62]	AHP	●	●	●	●	●
[63]	AHP	●				●	●	●	●		●
[64]	AHP	●					●	●		●		●	●
[65]	AHP, FR	●	●	●				●	●					●
[66]	RF, SVM	●	●	●	●

* FDA = Flexible discriminant analysis; CTA = Classification tree analysis; SRE = Surface response envelope; BIOCLIM = Bioclimatic Envelope Model; CLIMEX = Climate Experiment Model; CONSUS = Crops, climate change, and us. ** T = temperature; P = precipitation; S = soil; S = soil pH; SS = soil salinity; SD = soil depth; Luc = land use capacity; E = elevation; Sl = slope; A = Aspect; G = Geology; PA = protected area; Lu/Lc = land use/land cover.

).

Although the preferred methods vary, climate data (temperature and precipitation) are mainly used, and soil structure and properties are also extremely important for this analysis [11]. However, relying solely on climate data cannot yield a conclusive decision on whether a region is suitable for growing a specific agricultural product [67]. Due to challenges in accessing accurate and up-to-date soil data, numerous studies have been conducted that rely solely on climate data and have contributed to the literature.

This study conducted a comprehensive analysis using climate and soil parameters to determine potential distribution areas for avocado and pitaya cultivation in Mersin, located in the southern part of Türkiye. Unlike studies found in the literature, this study focuses on climatic factors and soil characteristics to gain an exhaustive understanding of where these fruits can thrive in the agricultural landscape of Mersin.

Mersin, situated within the Mediterranean basin, is influenced by a sub-tropical climate due to its geographical location, making it ideal for cultivating sub-tropical fruits. However, because of its semi-arid climate characteristics, this climate zone is especially vulnerable to the effects of global climate change [68,69,70]. Predictions indicate the following:

• An increase in temperatures;
• A decrease in precipitation;
• A likely reduction in surface and groundwater resources [71,72].

For these reasons, Mersin has been selected as the study area for this investigation. This study began with analyzing the region’s current climate and soil conditions, followed by projections of future climate scenarios based on anticipated changes and present soil data. Thus, a sustainable agriculture policy has been adopted. This research has been conducted due to a lack of detailed and large-scale research on Mersin.

2. Materials and Methods

2.1. Avocado

Avocado is the fruit of an evergreen tree in the Lauraceae family [64]. Avocado is a sub-tropical fruit high in vitamins A, C, B, E, and K, as well as minerals including iron, calcium, and magnesium [73,74]. Avocado, preferred for lowering blood sugar, blood pressure, and kidney stones, is also known to form a preventive barrier for liver diseases and joint disorders [75]. Furthermore, avocado oil is employed in the manufacture of skin-nourishing and healing lotions due to the acids and components it contains [76]. Additionally, it boosts collagen formation and nourishes the scalp [77]. Because of all of these characteristics, it is a highly valuable fruit.

Avocado is a sub-tropical fruit that thrives in places with warm winters and hot summers [78]. Avocado may be cultivated at temperatures ranging from −4 °C to 40 °C, with the optimal temperature being 18–26 °C [1,79,80]. It is susceptible to low temperatures (average −2 °C and below) during blooming and fruit set, temperature fluctuations, and high temperatures (over 40 °C) during the seedling phase. Regardless, powerful winds hurt avocado. Avocado is a fruit with a high water requirement; hence, places with an average annual precipitation of 800–1700 mm are ideal [81]. However, avocado requires an average precipitation of 750–1000 mm [82]. Aside from this, irrigation is required for high yield. Although it may grow in a variety of soil types, sandy and alluvial soils are ideal [78]. Deep (minimum 1 m), fertile, and well-drained soils present suitable conditions. In addition, the soil pH should be 7 or close to 7 (slightly acidic), and the groundwater level in the soil should be at most 1.5–2 m [78]. Suitable places for avocado include regions up to 2400 m above sea level (optimum altitude 0–250 m) and regions where the slope does not exceed 15% [82,83].

2.2. Pitaya

Dragon fruit is a sub-tropical fruit, also known as pitaya [84,85]. It has a thick, pink, and prickly shell. The inside contains black seeds [86]. Although it corresponds to a kiwi in shape, it tastes like strawberry or melon. The pitaya plant cannot stand straight on its own and must be supported by trees or walls [87]. Pitaya fruit is commonly used to cure wounds, remove harmful toxins from the body, and lower blood pressure. In addition, it helps moisturize dry skin, reduces the risk of heart disease and cancer, and eliminates uterine problems [88].

Pitaya is a sub-tropical fruit that cannot be grown in extreme (cold or very hot) temperatures. In other words, it becomes unproductive and damaged at −20 °C and is unable to survive at −40 °C [88]. High temperatures (40–45 °C) can be brutal for plants, leading to decreased productivity and flowering. Light sandy soils are ideal for its cultivation. Clayey soils and those with a lot of groundwater do not have acceptable qualities. Furthermore, the need for irrigation and fertilization is not very high [16].

An appropriate location analysis was performed for agricultural products expressed in the current and future conditions through climate and soil data.

To achieve this, we gathered presence data for the currently cultivated areas of the 2 products, including field studies, satellite imagery, the Global Biodiversity Information Facility (GBIF), research in the literature, and reports published by the Republic of Türkiye Ministry of Agriculture and Forestry (RTMAF). For avocado, 250 out of 299 records were derived from satellite images, 7 from GBIF [89], and 42 from the literature [90,91,92] and field studies. In the case of pitaya, 57 records were compiled, with 55 sourced from field studies and 2 from satellite imagery. Additionally, some of the presence data regarding the species not obtained from the field were checked in the area and found to be consistent.

2.3. Study Area

Mersin province, located in the southern part of Türkiye between 36 and 37° North latitude and 33 and 35° East longitude, includes the study area (Figure 1). It has a cover area of 15,853 km² and a population of 1,938,389 [93]. Mersin is divided into 13 districts, 4 of which are central, with the central districts housing more than half of the inhabitants (55.83%) [93]. Furthermore, a sizable portion of this population (86.34%) lives in city centers and coastal regions [94]. The altitude in Mersin city center and coastal districts varies from 0 to 10 m. The altitude in the Taurus Mountains, which are the roof of Mersin, reaches up to roughly 3500 m [95,96].

Due to its geographical location, Mersin is located in the Mediterranean basin and is intensively agriculture-cultivated. Agricultural activities cover 20.72% of the surface area (3284.01 km²). As a result of these activities, 1,661,600 tons of citrus fruit were produced in 2023, placing Mersin second in citrus fruit output in Türkiye [50,93]. Furthermore, Mersin ranks first in Türkiye’s production of bananas, strawberries, lemons, loquats, nectarines, peaches, squash, and broad beans and second in the production of almonds, carobs, eggplants, cucumbers, grapefruits, apricots, avocados, pomegranates, grapes, leeks, and soybeans [97]. Having a significant share in the production of agricultural products, Mersin also makes a critical contribution to the country’s economy. Its contribution to the gross national product (GNP) was TRY 88.2 billion in 2023. As of the same year, the value of plant production increased to TRY 76.3 billion.

On the other hand, the study area is in the Mediterranean basin, which has a sub-tropical climate. The mean annual temperature is 27.1 °C in the summer (June–August), 11.2 °C in the winter (December–February), and 19.3 °C on average [98]. The total annual average precipitation is 28.8 mm in the summer, 344.3 mm in the winter, and 610.9 mm in total [98]. This climate zone, influenced by specific climatic factors, exhibits a semi-arid characteristic. It is highly likely that global climate change, whose impact is increasingly felt worldwide, will also adversely affect Mersin. To achieve sustainable agriculture tailored to local conditions, it is essential to determine the crops to be cultivated, the locations for production, and the methods and quantities involved. Considering its climate characteristics and the potential repercussions of climate change, Mersin presents favorable conditions for cultivating sub-tropical fruits such as avocado and pitaya. Due to these factors, Mersin was selected as the focus of this study.

2.4. Materials

To accomplish the main purpose of this study, a total of 14 parameters were utilized, including climate (temperature, precipitation) and soil data. Current and future climate data, namely, mean annual temperature (MAT), mean minimum temperature of the coldest month (MMTCM), mean maximum temperature of the warmest month (MMTWM), and mean annual precipitation (MAP), were received from the WorldClimv2.1 platform world [99]. These data were cut according to the study area boundary, and the desired temperature and precipitation values were produced. Soil texture (S), soil depth (SD), and land use capability (LUC) data were obtained from the Republic of Türkiye Ministry of Agriculture and Forestry Agricultural Land Assessment Portal (RTMAF TadPortal) and digitized [100]. Soil pH (SpH) and soil organic carbon (SOC) parameters were obtained from the SoilGrids database [101]. Soil salinity (SS) was obtained from the FAOSoil portal [102]. As with the climate data, these data were also cut according to the study area. Land cover (Lc) and elevation (DEM) data were acquired from the Corine Land Cover (CLC) dataset in the Copernicus Land Monitoring Service (CLMS) and were cut according to the study area boundaries [103]. The slope (Sl) parameter was generated from the elevation data. Eventually, the groundwater level (GL) data (3845 points) used in the agricultural suitability analysis were obtained from the General Directorate of State Hydraulic Works (GDSHW) and interpolated for the study area using the IDW method. The protected area (PA) data were not directly employed in the analysis and were evaluated as an exclusion zone parameter. Furthermore, data belonging to some classes in parameters (MAT, MMTCM, MAP, elevation, slope, soil texture, soil depth, soil pH, protected area, and LUC) were excluded from the analysis (

Table 4. Parameters used for agricultural suitability analysis.

Parameters	Scale/Resolution → Final Resolution	Data Type	Source
Climate
(1) MAT (°C)	30 arc second → 25 m	Raster	WorldClim [99]
(2) MMTCM (°C)	30 arc second → 25 m	Raster
(3) MMTWM (°C)	30 arc second → 25 m	Raster
(4) MAP (mm)	30 arc second → 25 m	Raster
Soil
(5) DEM (m)	25 m	Raster	CLMS [103]
(6) Sl (%)	25 m	Raster	Production from DEM
(7) S	1/100,000 → 25 m	Vector	RTMAF TadPortal [100]
(8) SD (cm)	1/100,000 → 25 m	Vector
(9) SpH	2 arc second → 25 m	Raster	SoilGrids [101]
(10) SS (dS/m)	2 arc second → 25 m	Raster	FAOSoil [102]
(11) SOC (dg/kg)	1:100,000 → 25 m	Raster	SoilGrids [101]
(12) GL (m)	25 m	Vector	GDSHW
(13) LUC	1/100,000 → 25 m	Vector	RTMAF TadPortal [100]
(14) Lc	1/100,000 → 25 m	Raster	CLMS [103]

).

Given the location and characteristics of the study area, conducting an agricultural suitability analysis is critical. The resolution of the parameters employed plays a crucial role in the accuracy of the detection process. Consequently, all parameters were reproduced according to the highest-resolution DEM (final resolution: 25 m) data, and the analyses were conducted accordingly.

The climate data utilized in this study were sourced from the WorldClim dataset, initially at a resolution of 30 arc second (838.95 m~900 m). Subsequently, these data were resampled to a resolution of 25 m within the ArcGIS environment to ensure compatibility with the soil layers employed in the analysis.

The study area (Mersin) exhibits significant topographic and ecological diversity, particularly between its coastal and mountainous regions. This heterogeneous structure greatly influences microclimate conditions and, consequently, agricultural suitability. Since sub-tropical species such as avocado and pitaya are typically grown in small agricultural plots of a few acres; analysis at a 25 m resolution yields more meaningful and accurate results for agricultural practices at this scale. In contrast, using a resolution of approximately 900 m may overlook these local variations, jeopardizing the reliability of the analysis. Indeed, recent studies indicate that high-resolution parameters enhance model accuracy and spatial discrimination [60,104,105].

The rescaling process inevitably involves certain uncertainties, notably the assumption of homogeneity within a grid cell of approximately 900 m. It is important to note that this study only geometrically reduced the data without generating new information. While climate data presented the same value for each 25 m pixel, the model could capture micro-scale spatial differences due to the high resolution of the topographic data. Furthermore, the model was evaluated at both 900 m and 25 m resolutions. Although the general distribution of results was similar across both resolutions, a significantly higher level of discrimination was achieved at the 25 m resolution, particularly in fragmented agricultural areas and sloped regions. This finding is also corroborated by existing literature [106,107]. Consequently, although the original climate data resolution was lower, all parameters were adjusted to a 25 m resolution, aiming for a more precise analysis.

When determining the potential distribution areas of the aforementioned species (avocado and pitaya), the 14 used and excluded PA parameters were generally classified into four classes (S1, S2, S3, and N) based on the climate and soil requirements required for cultivating these agricultural products (

Table 5. Classification of parameters used for agricultural suitability analysis.

Parameters	Sub-Classes	Suitability Level
Climate
(1) MAT (°C)	14.59–19.55	S1
	9.63–14.59	S2
	9.63–4.68	S3
	−0.28/4.68	N
(2) MMTCM (°C)	−14.30/−7.48	N
	−7.48/−2.76	N
	−2.76/2.05	S3
	2.05/8.00	S3
(3) MMTWM (°C)	17.70–22.08	S1
	22.08–26.45	S2
	26.45–30.83	S3
	30.83–35.20	S3
(4) MAP (mm)	454.00–540.25	N
	540.24–625.50	S3
	625.50–712.75	S2
	712.75–799.00	S1
Soil
(5) DEM (m)	0–867.27	S1
	867.27–1740.52	S2
	1740.52–2400.00	S3
	>2400.00	N
(6) Sl (%)	0–8	S1
	8–16	S2
	16–28	S3
	>28	N
(7) S	A, K, L, O, R, S	S1
	E, M, T	S2
	N, P, Y	S3
	Unclassified	N
Soil
(8) SD (cm)	A→90+	S1
	B→50–90	S2
	C→20–50	S3
	D→0–20, E→ Lithosolic, Unclassified	N
(9) SpH	5–6.5	S1
	4.5–5/6.5–7.7	S2
	4.3–4.5	S3
	<4.3	N
(10) SS (dS/m)	0–3.0	S1
	3–3.1	S2
	3.1–3.3	S3
	3.3–3.5	S3
(11) SOC (dg/kg)	0–155.25	S3
	155.25–310.50	S3
	310.50–465.75	S2
	465.75–621.00	S1
(12) GL (m)	0–5	S1
	5–10	S2
	10–20	S3
	>20	S3
(13) LUC	I, II, III, IV	S1
	V, VI, VII	S3
	VIII, Unclassified	N
(14) Lc	Agricultural land, Vegetation, Water body	S1
	Barren land	S3
	Settlement	N
Protected area (PA)	WIR, Grade I., II., III. Archeological Site	N
	SPEA, Urban site, Ruin	N
	Natural Monument, Natural Park	N

). Data from the “N” class were eliminated from the analysis. The appropriate temperature, precipitation, and soil values for fruit growth were collected from relevant studies in the literature and reports published by RTMAF, experts, and local farmers. Based on these data, appropriate locations were identified for the aforementioned products.

2.5. Methods

To achieve sustainable agriculture in the study area, the aim was to determine the agricultural product pattern according to climate and soil characteristics. First, a suitable location analysis was conducted for sub-tropical (avocado and pitaya) products for Mersin using the current (1971–2000) values and soil characteristics. Secondly, suitable locations were determined for the projection period (2021–2100) using the changing climate characteristics and soil data. The future period was divided into 20-year periods as (2021–2040), (2041–2060), (2061–2080), and (2081–2100); then, analysis was carried out for each period. Thereby, the application was submitted until the end of the 21st century, and a perspective for agricultural activities was developed (Figure 2). The MaxEnt model was used to meet the objective of this study.

2.5.1. Maximum Entropy (MaxEnt) Method

The maximum entropy (MaxEnt) approach is widely used when modeling species distribution. It enables exceptionally successful results to be produced by utilizing existing species records and climate characteristics [3,108,109]. Furthermore, this analysis method improves the modeling of species distribution by incorporating soil parameters (soil depth, texture, organic matter, salinity, and pH).

The MaxEnt method is utilized to calculate the distribution of a probability situation P(x). The input parameters have maximum entropy as a result of the defined restrictions, and the distribution expressed in this way can be calculated (Equation (1)).

$$H\left(P\right)=-\sum _{x\in X}P\left(x\right)logP\left(x\right)$$

(1)

where X represents the data space (set), and P(x) reflects the probabilistic distribution of the data space.

To model the distribution, the input parameters must have some restrictions [109,110]. For this, the restrictions are mostly determined by an $f_i\left(x\right)$ function (Equation (2)).

$$\begin{gathered} \sum _{x\in X}P\left(x\right)f_i\left(x\right)={\widehat{f}}_i, {\forall }_i \\ \sum _{x\in X}P\left(x\right)=1 \end{gathered}$$

(2)

where $f_i\left(x\right)$ is the expected value and ${\widehat{f}}_i$ is the observed value, and they are expected to match each other. Thus, the distribution P(x) that maximizes the entropy under the constraints is found by the MaxEnt model (Equation (3)).

$$P\left(x\right)={\displaystyle \frac{1}{Z}}exp\left(\sum _i{\mathsf{\lambda }}_i{\mathrm{f}}_i\left(x\right)\right)$$

(3)

where ${\mathsf{\lambda }}_i$ indicates the Lagrange multipliers. In other words, they are optimized parameters designed to provide constraints. They are used to determine a distribution’s maximum and minimum points using parameters. exp is the expansive form, which allows the distribution to be obtained using the entropy approach. Z represents the normalized constant (Equation (4)).

$$X=\sum _{x\in X}exp\left(\sum _i{\mathsf{\lambda }}_i{\mathrm{f}}_i\left(x\right)\right)$$

(4)

The MaxEnt v3.4.4 program, which works according to this method, was used to determine current and future agricultural suitability. For the analysis, 14 parameters were converted to ASCII format and transferred to the program, and the application was carried out. In addition, correlation analysis was performed for the parameters to obtain higher accuracy and prevent multicollinearity [111,112]. For this purpose, principal component analysis in SDMToolbox v2.6 developed for ArcGIS10.5 software was employed. In the obtained correlation matrix, values with a correlation coefficient ≥ ±0.85 were excluded [3,26,33,113,114]. Thus, the average minimum temperature (°C) and altitude (m) parameters in winter were removed from the analysis, and 12 variables were used. A total of 70% of the asset data for 2 species whose agricultural product patterns were to be determined were used as training and 30% as test data [33]. The cross-validation technique was preferred for the models to have high accuracy and consistency [3,115]. Cloglog was selected as the output type. In addition, 10,000 background points were used, and 15 repetitions were made to ensure the reliability of the models. With the help of presence data and 12 parameters, suitable regions for both current and future cultivation of 2 different agricultural products were determined. In the analyses carried out for the projection period, climate parameters were taken from 3 GCMs (HadGEM3-GC31-LL, MPI-ESM1-2-HR, and GFDL-ESM4) and SSP2-4.5 and SSP5-8.5 climate scenarios, different for each period, and were used in the agricultural suitability analysis in this way. The achieved species distribution models include a value between 0 and 1, with the lowest being 0 and the highest being 1. Accordingly, the places where the species are least likely to be found take values close to 0, while the regions where the probability of being found is highest take values close to 1 or 1 [3,26,33].

2.5.2. Validation

The validation procedure is critical to the dependability, performance, and consistency of scientific investigations [60,116]. After the analysis, the receiver operating characteristic (ROC) curve was used, as in much research, for the validation and performance evaluation of the models [3,26,33,117]. The area under the curve (AUC) value shows the estimation accuracy [118,119]. The y-axis in the ROC curve represents the true positive rate, and the x-axis represents the true negative rate. The AUC value varies between 0 and 1. Values close to 1 indicate that the model has better performance and high reliability. Another critical performance metric for model validation is the true skill statistic (TSS), which is often favored, particularly in evaluating species distribution models developed using ML methods [120,121]. To calculate the TSS, a threshold value is first necessary. This threshold is typically established by maximizing sensitivity and specificity [3,120,122]. Thus, the TSS is calculated using the threshold value and the confusion matrix with the presence and absence data. The TSS ranges from −1 to 1, with values approaching 1 indicating a successful estimation by the model. In this study, the validation process of the analysis performed with the MaxEnt method was performed using the ROC curve and TSS. AUC and TSS values were obtained from 15 repetitions. In addition, the Jackknife test was used to determine the contribution of the 12 variables used in the analysis.

After validation, the models were classified. The FAO land evaluation approach was utilized as a reference for the classification, and the models were divided into four categories: S1 (very suitable), S2 (moderately suitable), S3 (marginally suitable), and N (unsuitable) [123].

3. Results

Twelve parameters and the MaxEnt model were utilized to determine the agricultural product pattern. First, correlation analysis was performed on the parameters. Correlation matrices were developed for the present and 21st century based on existing and predicted climate values and soil parameters. The average correlation matrix was given (Figure 3), as the matrix values were consistent except for the second (MMTCM) and fifth (DEM) parameters.

Following the correlation analysis, an agricultural suitability analysis with 12 variables was conducted for avocado and pitaya. The contributions of each independent variable were then determined using the Jackknife test. The Jackknife test was performed on each agricultural product’s current and future periods. Because the results were consistent as in the correlation analysis, the final test results derived by taking their averages were given individually for avocado and pitaya (Figure 4 and Figure 5).

Another metric employed to evaluate the models’ performance is the TSS. Average TSS values were obtained for avocado and pitaya (Figure 6).

The average AUC value of the models produced for avocado is 0.890, and the TSS value is 0.912. Accordingly, it is possible to state that the predictive power of the models is extremely high. When the results of the jackknife tests are analyzed, the highest contributions are made by 10 (49.3%), 1 (32.4%), and 6 (6.4%), respectively (

Table 6. Contributions of variables in models produced for avocado.

Parameters	Contribution (%)
(10) SS (dS/m)	49.3
(1) MAT (°C)	32.4
(6) Sl (%)	6.4
(7) S	4.0
(14) Lc	3.0
(11) SOC (dg/kg)	2.5
(12) GL (m)	1.6
(9) SpH	0.3
(3) MMTWM (°C)	0.2
(8) SD (cm)	0.1
(4) MAP (mm)	0.1
(13) LUC	0.1

). The environmental variable that improves the gain the most when utilized alone and decreases the gain the most when removed is 10. Therefore, it would not be wrong to state that it offers the most informative and detailed information regarding the analysis of the potential distribution performed.

Furthermore, when the response curves of the most important factors are evaluated, it is clear that the avocado plant often prefers areas where soil salinity (3.0–3.1 dS/m) is relatively low, slope (0–8%) is not high, and the annual average temperature is 14.59–19.55 °C. While soil salinity and slope increase negatively affect avocado, increasing temperature positively affects it.

Models were also created for pitaya, another agricultural product. The average AUC for the models was 0.802, and the TSS value was 0.903. As with the avocado models, the models had a predictive capacity of more than 80%. The first (44.6%), tenth (28.5%), and sixth (9.5%) parameters contributed the most to the appropriateness analysis out of the 12 parameters employed (

Table 7. Contributions of variables in models produced for pitaya.

Parameters	Contribution (%)
(1) MAT (°C)	44.6
(10) SS (dS/m)	28.5
(6) Sl (%)	9.5
(8) SD (cm)	5.2
(11) SOC (dg/kg)	4.5
(14) Lc	3.8
(4) MAP (mm)	1.9
(7) S	0.6
(12) GL (m)	0.6
(3) MMTWM (°C)	0.5
(9) SpH	0.2
(13) LUC	0.1

). The first parameter provides the most relevant information.

The response curves provide information on the characteristics that have the most impact on the pitaya models (mean annual temperature, soil salinity, and slope), as well as the magnitude of those impacts. The places with temperatures ranging from 14.59 to 19.55 °C are better suited. However, it is possible to say that pitaya prefers areas where the slope is low (0–8%). Furthermore, like with avocado, regions with soil salinity (3.0–3.1 dS/m) are more favorable.

3.1. Potential Distribution Areas for Current Period

Following correlation analysis and confirmation, the potential distributions of the two species for the present and future eras were calculated. First, avocado and pitaya maps for the current period are presented (Figure 7).

When the possible distribution maps of the species developed based on the climatic and current soil data of the reference period were analyzed, it was discovered that the coastal and southern regions were predominantly in the “S1” class (4.16%) for the avocado species. On the other hand, the mountainous and northern slopes at high altitudes were mostly classified as “N” (78.27%). When the agricultural suitability analysis results for pitaya were evaluated, it was discovered that coastal locations were extremely suitable (“S1”, 7.51%), as with avocado. However, this species can also be grown in the Göksu River valley (“S2” and “S3”). Other locations are categorized as unsuitable (“N”, 69.71%).

Table 8. Potential distribution areas for the current period.

Class	Avocado		Pitaya
	Area (km2)	(%) Area	Area (km2)	(%) Area
S1	660.00	4.16	1191.10	7.51
S2	1558.88	9.83	1457.33	9.19
S3	1226.18	7.73	2152.67	13.58
N	12,407.94	78.27	11,051.90	69.71
Total	15,853	100	15,853	100

shows the prospective distribution classes of the species and the areas and percentages of the regions in these classes.

3.2. Potential Distribution Areas for Future Period

3.2.1. Avocado (SSP2-4.5)

In the continuing phase, a suitability study for the projected term was performed. For this aim, potential distributions of species were determined for three GCMs, namely, HadGEM3-GC31-LL, MPI-ESM1-2-HR and GFDL-ESM4, and SSP2-4.5 and SSP5-8.5 climate scenarios. Analyses were conducted utilizing projected climatic data and current soil parameters. First and foremost, studies were conducted for avocado based on the SSP2-4.5 climate scenario and the stated GCMs (Figure 8).

3.2.2. Avocado (SSP5-8.5)

The implementations performed according to the SSP2-4.5 climate scenario were also performed for SSP5-8.5, and the probable distribution areas of the avocado species were determined under this scenario (Figure 9).

3.2.3. Pitaya (SSP2-4.5)

Similar approaches were used for another species, pitaya. The likely future distribution zones of pitaya were evaluated using SSP2-4.5 (Figure 10) and SSP5-8.5 (Figure 11) climate scenarios, as well as three GCMs.

3.2.4. Pitaya (SSP5-8.5)

The identical operations were carried out following SSP2-4.5 and according to SSP5-8.5 (Figure 11).

Figure 11. Potential distribution areas for pitaya obtained using (1) HadGEM3-GC31-LL, (2) MPI-ESM1-2-HR, and (3) GFDL-ESM4 according to the period (2021–2100) and SSP5-8.5.

Figure 11. Potential distribution areas for pitaya obtained using (1) HadGEM3-GC31-LL, (2) MPI-ESM1-2-HR, and (3) GFDL-ESM4 according to the period (2021–2100) and SSP5-8.5.

As a result of the analyses made according to the climate and available soil data of the projection period, it is possible to access information on the distribution of species in different periods. Accordingly, under the SSP2-4.5 climate scenario, which limits global warming to almost 3 °C, there will be a 3.80% increase in the “S1” class in the HadGEM3-GC31-LL model in the first 20-year (2021–2040) period compared to the current situation. A 1.47% decrease is estimated for the next 40-year period (2041–2060), and a 0.64% decrease is estimated for the (2061–2080) period. By the end of the century, it is expected that the “S1” class will cover an area of 1229.35 km², and there will be an increase of 3.59% compared to the current situation. However, there will probably be much change in the unsuitable (“N”) class over the four different periods. At the end of the 21st century, it is expected to decrease by 3.99% in this class and regress to 11,775.40 km² (

Table 9. Distribution of avocado for the projection period according to the models (SSP2-4.5).

	Avocado (SSP2-4.5)
	HadGEM					MPI				GFDL
		S1	S2	S3	N	S1	S2	S3	N	S1	S2	S3	N
(2021–2040)	km2	1263.15	1493.18	1402.01	11,694.66	1200.19	1255.97	1344.09	12,052.74	1630.57	2056.19	1323.17	10,843.07
	%	7.97	9.42	8.84	73.77	7.57	7.92	8.48	76.03	10.29	12.97	8.35	68.40
(2041–2060)	km2	427.41	1436.59	1553.09	12,435.90	1370.34	1191.97	1209.61	12,081.08	1186.02	1113.01	113.79	12,440.18
	%	2.70	9.06	9.80	78.45	8.64	7.52	7.63	76.21	7.48	7.02	7.03	78.47
(2061–2080)	km2	558.62	1784.16	1623.77	11,886.45	1218.31	1195.97	1393.16	12,045.56	1172.66	1249.90	1723.98	11,706.46
	%	3.52	11.25	10.24	74.98	7.69	7.54	8.79	75.98	7.40	7.88	10.87	73.84
(2081–2100)	km2	1229.35	1362.57	1485.68	11,775.40	87.73	1603.87	1823.11	12,338.29	12,202.37	1265.46	1695.70	11,689.46
	%	7.75	8.60	9.37	74.28	0.55	10.12	11.50	77.83	7.58	7.98	10.70	73.74

).

When the findings of the MPI-ESM1-2-HR global climate model are examined, it is expected that the “S1” class will increase by 3.41%, 4.48%, and 3.52% in the first 60-year period, respectively. However, there will be a decrease of 3.61% in the last 20-year period. As in the HadGEM3-GC31-LL model, there is no significant change in the “N” class. A decrease of 0.44% by the end of the century seems to be a strong possibility. The highly suitable (“S1”) class will reach its maximum value of 1370.34 km² (8.64%) in the (2041–2060) period. However, the minimum value of the “N” class is 12045.56 (75.98%). This value will be reached in the (2061–2080) period. Table 9 shows the expected areal values in the classes of the models for the projection period.

It is also possible to access the values of the findings of the other climate model used in this study, GFDL-ESM4. According to these model values, increases of 6.12%, 3.32%, 3.23%, and 3.42% are expected in the “S1” class, in turn. There will be a decrease in the unsuitable (“N”) class. There will be an increase of 0.20% only in the (2041–2060) period. The “S1” class will have the largest value in the (2021–2040) period, with its value corresponding to 10.29% (1630.57 km²) of the study area (

Table 10. Distribution of avocado for the projection period according to the models (SSP5-8.5).

	Avocado (SSP5-8.5)
	HadGEM					MPI				GFDL
		S1	S2	S3	N	S1	S2	S3	N	S1	S2	S3	N
(2021–2040)	km2	544.72	1835.99	1787.78	11,684.51	1242.92	1199.05	1139.05	12,271.98	1175.34	1198.68	1724.84	11,754.14
	%	3.44	11.58	11.28	73.71	7.84	7.56	7.19	77.41	7.41	7.56	10.88	74.14
(2041–2060)	km2	533.32	1767.17	1678.37	11,874.14	91.03	1565.41	1484.48	12,712.08	1180.28	1284.67	1672.34	11,715.70
	%	3.36	11.15	10.59	74.90	0.57	9.87	9.36	80.19	7.45	8.10	10.55	73.90
(2061–2080)	km2	12,56.94	1470.03	1402.02	11,724.01	1191.76	1035.40	1090.49	12,535.36	676.02	1215.50	1148.58	12,812.90
	%	7.93	9.27	8.84	73.95	7.52	6.53	6.88	79.07	4.26	7.67	7.25	80.82
(2081–2100)	km2	1150.22	1517.66	1423.23	11,761.88	1116.41	977.93	956.63	12,802.03	96.51	123.52	1464.11	14,168.86
	%	7.26	9.57	8.98	74.19	7.04	6.17	6.03	80.75	0.61	0.78	9.24	89.38

). Based on this, the smallest value of the “N” class is 68.40% (10843.07 km²) and is also in this period.

Furthermore, when the prospective distribution maps developed using three GCMs within the context of the SSP2-4.5 climatic scenario were analyzed, it was revealed that avocado species would be better suited for cultivation in the research area’s southern and coastal regions. It is expected that the high, mountainous, and northern regions will be mostly in the unsuitable (“N”) class.

According to the SSP5-8.5 climate scenario, which has high radiative forcing and limits warming to about 5 °C, the changes are expected to be extreme. According to the HadGEM3-GC31-LL model for the avocado species, there will be decreases of 0.73% and 0.80% in the “S1” class in the first 40 years, respectively. However, increases of 3.77% and 3.09% are expected in the next 40 years. It is estimated that the area covered by the very suitable (“S1”) class will be the largest in the period (2061–2080) with 1256.94 km². However, there is a decrease for each period in the unsuitable (“N”) class. The period with the largest decrease is the period (2021–2040) with 4.56%. Accordingly, the lowest value of the “N” class is the (2021–2040) period with 11,684.51 km² (73.71%). The “S1” class has the highest value with 1256.94 km² (7.93%) in the (2061–2080) period (Table 10). The period in which the “N” class has the smallest area (11,684.51 km², 73.71%) is (2021–2040).

In the MPI-ESM1-2-HR global climate model, compared to HadGEM3-GC31-LL, a decrease (3.59%) is expected only for the “S1” class for the period (2041–2060). Apart from this, it is possible to express increases of 3.68% (2021–2040), 3.35% (2061–2080), and 2.88% (2081–2100). In the “N” class, there will be a decrease (0.86%) for the first period and increases (1.92%, 0.80%, and 2.49%, respectively) for the other periods. In this case, the period (2021–2040) is the period in which the “S1” class has the largest area with 1242.92 km² (7.84%) and the “N” class has the smallest area with 12,271.98 km² (77.41%) (Table 10).

When the analysis findings using GFDL-ESM4 are examined, a decrease of only 3.55% is expected for the “S1” class for the period (2081–2100). Apart from this, there will be increases of 3.25%, 3.28%, and 0.10% for the other periods, respectively. In the “N” class, decreases (4.12% and 4.37%) for the first 40 years and increases (2.55% and 11.11%) for the next 40 years are highly probable. The (2041–2060) period is noteworthy as the period in which the “S1” class has the largest area (1284.67 km²) and the “N” class has the smallest area (11,715.70 km²) (Table 10).

Avocado can be grown primarily in the lowland and southern regions, according to agricultural suitability evaluations based on the SSP5-8.5 climate scenario and three GCMs. It can, however, be grown, albeit to a lesser extent, in the Göksu River valley (“S3”). The study area’s mountainous and high regions stand out as unsuitable locations.

The other sub-tropical fruit is pitaya, and similar analyses were carried out for it. First, the application was carried out using the HadGEM3-GC31-LL model under the SSP2-4.5 climate scenario. Accordingly, an increase is expected in the first (2.11%) and last (2.17%) 20-year periods compared to the current period in the “S1” class, and a decrease is expected in the middle 40 years (0.23% and 0.23%). Accordingly, this class will have the largest surface area (1535.79 km²) in the period (2081–2100) (Table 9). There is a decrease in the “N” class for all periods (5.63%, 1.46%, 1.00%, and 5.66%). The period when this class has the smallest area is the end of the century with 10,154.35 km² (64.05%).

Analysis of the MPI-ESM1-2-HR model indicates a significant decreasing tendency in the “S1” class (7.23%, 6.41%, 7.19%, and 7.38%, respectively), but a notable growing trend is evident in the “N” class (8.65%, 7.37%, 14.87%, and 26.90%, respectively). In the (2041–2060) period, the “S1” class has an area of 174.31 km² (1.10%), which is the highest value of the class at all times (

Table 11. Distribution of pitaya for the projection period according to the models (SSP2-4.5).

	Pitaya (SSP2-4.5)
	HadGEM					MPI				GFDL
		S1	S2	S3	N	S1	S2	S3	N	S1	S2	S3	N
(2021–2040)	km2	1526.32	1459.75	2707.01	10,159.93	44.63	325.31	3059.27	12,423.79	1002.82	1328.20	2806.38	10,715.60
	%	9.63	9.21	17.08	64.09	0.28	2.05	19.30	78.37	6.33	8.38	17.70	67.59
(2041–2060)	km2	1154.83	1362.54	2514.91	10,820.72	174.31	752.85	2706.24	12,219.60	44.65	325.30	3059.27	12,423.78
	%	7.28	8.59	15.86	68.26	1.10	4.75	17.07	77.08	0.28	2.05	19.30	78.37
(2061–2080)	km2	1154.75	1341.93	2462.81	10,893.51	51.10	32.39	2064.55	13,409.96	49.97	333.06	2090.80	13,379.16
	%	7.28	8.46	15.54	68.72	0.32	2.07	13.02	84.59	0.32	2.10	13.19	84.40
(2081–2100)	km2	1535.79	1504.98	2657.88	10,154.35	20.55	40.30	476.33	15,315.81	49.07	292.70	1226.63	14,284.61
	%	9.69	9.49	16.77	64.05	0.13	0.25	3.00	96.61	0.31	1.85	7.74	90.11

). Parallel to this, the “N” class also has the lowest value (12,219.60 km²) in this period.

The analysis of the GFDL-ESM4 climate model is similar to the MPI-ESM1-2-HR model. However, the “S1” class experiences the smallest reduction of 1.19% (2021–2040). The other periods will see a drop of 7.23%, 7.20%, and 7.20%, respectively. In the “N” class, only the first period (2.12%) shows a reduction, while the remaining 60-year period is expected to increase. This growth is expected to be 8.65%, 14.68%, and 20.39%, respectively. The period in which the very suitable (“S1”) class has the largest area is (2021–2040) with 1002.82 km² (6.33%) (Table 11). This is the period with the least area covered by the unsuitable (“N”) class (107,515.60 km², 67.59%).

According to the analysis of the determination of potential cultivation areas of pitaya under the SSP2-4.5 climate scenario utilizing three GCMs, it was determined that the coastal areas of the study area were more suitable. Regardless, there were also points where the models differed from each other. In the HadGEM3-GC31-LL model, it was determined that the coastal areas with low altitudes and the southern parts were suitable places; however, the Göksu River valley also delivered relatively suitable conditions. According to the other two models, it was determined that fewer parts allowed the cultivation of this species. It was seen that the southwest and southeast parts of the study area provided suitable opportunities. For all three models, the unsuitable places were determined as the high, mountainous, and northern parts of the study area.

Similar analyses were carried out under the SSP5-8.5 climate scenario, as in SSP2-4.5, and attempts were made to determine potentially suitable areas for pitaya cultivation. According to the HadGEM3-GC31-LL model, except for the (2081–2100) period (0.74%), increases of 2.55%, 1.77%, and 2.45% are predicted in the “S1” class in other periods, respectively. Accordingly, this class will have the largest area, with 1595.00 km² (10.06%), in the (2021–2040) period (Table 10). Nevertheless, a continuous lowering is expected in the unsuitable (“N”) class (6.09%, 1.78%, 5.93%, and 3.07%, in turn). The first 20-year period, in which the “S1” class covers the largest area, also draws attention as the period in which the “N” class has the smallest surface area (10,086.64 km²).

The MPI-ESM1-2-HR model shows a drop in the “S1” class (7.24%, 7.42%, 7.23%, and 7.21%, respectively). The “N” class is predicted to see gains of 8.64%, 14.91%, 8.68%, and 27.81% in turn. The period (2081–2100) is when the highly suitable (“S1”) class covers the largest territory, with 48.70 km² (0.31%), while the “N” class covers the smallest area, with 12265.85 km² (77.37%) (

Table 12. Distribution of pitaya for the projection period according to the models (SSP5-8.5).

	Pitaya (SSP5-8.5)
	HadGEM					MPI				GFDL
		S1	S2	S3	N	S1	S2	S3	N	S1	S2	S3	N
(2021-2040)	km2	1595.00	1406.12	2765.23	10,086.64	43.86	324.78	3057.84	12,426.52	44.26	322.85	3063.91	12,421.97
	%	10.06	8.87	17.44	63.63	0.28	2.05	19.29	78.39	0.28	2.04	19.33	78.36
(2041-2060)	km2	1460.70	12,080.65	2413.77	10,769.87	14.69	36.56	341.69	15,460.06	47.40	334.62	2054.67	13,416.31
	%	9.21	7.62	15.23	67.94	0.09	0.23	2.16	97.52	0.30	2.11	12.96	84.63
(2061-2080)	km2	1579.87	1566.97	2595.05	10,111.11	44.45	332.19	3068.55	12,407.81	43.96	325.57	3055.57	12,427.48
	%	9.97	9.88	16.37	63.78	0.28	2.10	19.36	78.27	0.28	2.05	19.28	78.39
(2081-2100)	km2	1073.06	1553.93	2660.28	10,565.74	48.70	432.97	3105.47	12,265.85	14.69	36.56	341.69	15,460.06
	%	6.77	9.80	16.78	66.65	0.31	2.3	19.59	77.37	0.09	0.23	2.16	97.52

).

The GFDL-ESM4 climate model analysis is consistent with the MPI-ESM1-2-HR results. It was determined that the “S1” class would tend to deteriorate. By the end of the century, the reduction will have reached a peak of 7.42%. This is followed by 7.24% (2021–2040), and 7.23% (2061–2080). The smallest reduction occurs between 2041 and 2060 at 7.21%. Table 10 shows that the “S1” class will have the highest area (0.30% or 47.40 km²) in the second 20-year timeframe. The “N” class is projected to have the smallest area at 78.36% (12,421.97 km²) between 2021 and 2040. Changes (increases) in this class from the current period are 8.64%, 14.91%, 8.68%, and 27.81%, respectively.

The analyses conducted using the SSP5-8.5 climate scenario and three GCMs revealed that the research area’s coastal areas were better suited. However, there are several areas where the models differ. The HadGEM3-GC31-LL model prefers southern locations with low altitudes. In addition, it was determined that the valley where the Göksu River flows was suitable in some areas. The MPI-ESM1-2-HR 2 and GFDL-ESM4 models determined that only a small portion of the research region was suitable. The southwest–southeast axis of the study region is generally suitable. Unsuitable locations in all three models include mountainous, harsh, high-altitude, and northern regions.

4. Discussion

4.1. Avocado (SSP2-4.5)

It is feasible to make inferences from the findings of the agricultural suitability analyses conducted for avocado and pitaya for the future period, which is expressed as the projection period. Firstly, the analyses performed for the avocado species were interpreted. For this purpose, the findings of the application conducted according to the HadGEM3-GC31-LL climate model under the SSP2-4.5 climate scenario were preferably evaluated. Accordingly, when the potential distribution areas of avocado in the current and future periods are examined, it is seen that dissimilarities occur. It was determined that there will be fluctuations in the very suitable (“S1”) class but that the study area will offer more suitable conditions than the current situation by the end of the century as a result of the possible effects of climate change. In this respect, according to the results of this model, it would not be wrong to say that, despite the predominantly negative effect of global warming, avocado, a sub-tropical fruit, will have a positive effect on growth in the study area. By the way, it is possible to state that the selected product will adapt to the possible climate and soil characteristics of the study area. This condition is, furthermore, supported by the general decrease in the unsuitable (“N”) class.

According to the analysis findings carried out using the MPI-ESM1-2-HR global climate model, it is possible to state that there will be an increase in the “S1” class except for in the last 20-year period. In this respect, it differs from the HadGEM3-GC31-LL model. The model indicates that, by the end of the century, the study area will have a limited capacity for avocado production, covering only 87.73 km², representing 0.55% of the total area. This suggests that the conditions for cultivating avocados in that region will be largely unsatisfactory. Regardless, the continuous decrease in the “N” class, albeit at different amounts, shows that the study area offers reasonable conditions for this species in general (“S2”, “S3”). Although there was a decrease in the “S1” class (2081–2100) period, the increase in the moderately suitable (“S2”) and marginally suitable (“S3”) classes (0.28% and 3.77%, respectively) supports this situation. As a result, although a decrease in the “S1” class is predicted compared to the first model, this model also shows that the study area offers suitable conditions for avocado in the future.

In the last analysis, to determine the potential distribution of avocado, the GFDL-ESM4 model was opted for. According to this model, it is not possible to mention any decrease in the “S1” class, unlike the other two models. In addition, when the areas covered by this class are taken into account, it is similar to the MPI-ESM1-2-HR model, except for the last period. In addition to the increases in the “S1” class, with the decreases in the “N” class, it is foreseen that avocado cultivation can be completed in the study area according to this model as a result of the potential effects of global climate change.

4.2. Avocado (SSP5-8.5)

Analyses were conducted using three GCMs under the SSP5-8.5 climate scenario, and their findings were interpreted. First, when the HadGEM3-GC31-LL model was evaluated, it was observed that a decrease and then an increase were expected in the “S1” class corresponding to the current period. In this respect, it differs from the results of this model in the SSP2-4.5 scenario. In particular, it was predicted that the study area would have suitable conditions for avocado for the year 2061 and beyond. The decrease in the “N” class also confirms the situation already stated. In this case, it is possible to state that climate change will positively contribute to the cultivation of this species according to the values of this model under SSP5-8.5.

Considering the findings of the MPI-ESM1-2-HR global climate model, the period in which the “S1” class will decrease is only the second 20-year period. However, under the SSP2-4.5 scenario, the decrease is seen in the last 20-year period. In this respect, they are different from each other. The expectation of an increase in suitable places as time progresses shows that the probable climate characteristics and soil structure of the study area have suitable opportunities for this species.

The analysis findings performed via GFDL-ESM4 show that the “S1” class will be on an increasing trend except for the last period. Compared to the two models under this scenario, a reduction is estimated only in this model at the end of the 21st century. Nonetheless, it has been determined that the decrease in the “S2” class and the increase expected in the “S3” and “N” classes will lead to a serious decrease in suitable areas, especially in 2061 and later. When the findings of this model are evaluated, avocado cultivation in the study area will probably be affected first positively and then negatively as a result of global climate change.

4.3. Pitaya (SSP2-4.5)

Secondly, agricultural suitability analysis was carried out for pitaya, and as in other analyses, first SSP2-4.5 and then SSP5-8.5 findings were evaluated. According to the results of the HadGEM3-GC31-LL model under the SSP2-4.5 scenario, while notable increases occurred in the “S1” class in the first and last periods, a decrease is expected in the second and third periods (0.23% and 0.23%, respectively). Considering that the increase is higher, and the decrease is proportionally smaller, these model outputs show that the study area will provide suitable opportunities for pitaya in the future. The continuous decrease in the “N” class also sustains the stated situation.

In the analysis completed by choosing the MPI-ESM1-2-HR model, a decrease in the “S1” class and an increase in the “N” class are expected in general. Especially, it is estimated that there will be few very suitable (“S1”) areas (20.55 km², 0.13%) at the end of the century. Based on this, it is predicted that pitaya will be exposed to negativity in the study area over time as the effects of climate change increase.

The GFDL-ESM4 model is also parallel to the results of this model in that the “S1” class decreases proportionally in all periods, similar to MPI-ESM1-2-HR. It varies from the second model in that the reduction happens in the “N” class during the first period. Although the suitability decreases, it is still viable to conduct agricultural operations with this species in the research region in the future.

4.4. Pitaya (SSP5-8.5)

Other analyses for modeling the potential distribution of pitaya were conducted under the SSP5-8.5 scenario. Firstly, the HadGEM3-GC31-LL model was interpreted. According to this model, the “S1” class tends to increase except for the last period. The “N” class is also in a decreasing trend for all periods in parallel with this increase. In this respect, it is consistent with the same model results under SSP2-4.5. When the values obtained are examined, suitable places for pitaya will be found in the future under this scenario and this model in the study area.

The MPI-ESM1-2-HR model provides results close to the values obtained by employing the SSP2-4.5 scenario. Again, the “S1” class tends to decrease in all periods, while the “N” class tends to increase. Therefore, according to the results of this model, it is obvious that the study area will not offer very good opportunities for pitaya in the future. However, it would not be correct to say that it is completely unsuitable for cultivation, since there are significant increases (5.78% and 6.01%, respectively) in the marginally suitable (“S3”) class starting from the period (2061–2080). This situation also shows that the study area offers suitable areas for this species for the projection period, although small.

The GFDL-ESM4 model results are close to the values under the SSP2-4.5 climate scenario. Nevertheless, it is also quite analogous to the second model, MPI-ESM1-2-HR, in the SSP5-8.5 scenario. The point where both models differ is the decrease rates in the “N” class (2081–2100) period (7.66% and 27.81%, respectively). Apart from this, extremely consistent values were obtained. In this respect, the study area for pitaya also offers few opportunities, according to GFDL-ESM4.

The inferences drawn from this study’s agricultural suitability analysis using several climate models and climate scenarios have been given above. However, making comparisons to similar studies in the literature on the research topic is critical for a scientific publication.

In this study, as in many studies, MaxEnt was preferred among ML methods [35,59,108]. When evaluated in terms of the methods used in determining the potential distribution areas, Ref. [3] used MaxEnt in determining the potential distribution of olives in their study for Türkiye. In addition, it is noticed that this method was also preferred in the investigations conducted by the authors of [26,33]. In this context, it would not be wrong to state that this study used a current and reliable method compatible with the literature. However, only one method was preferred in this study. In this respect, the three methods (MaxEnt, CLIMEX, and SVM) used in the study conducted by the authors of [61] differ from the four methods (RF, MaxEnt, and GBM) in the research conducted by the authors of [124].

In this study, two metrics, AUC and TSS, were employed for the validation and performance evaluation of the models developed using the specified method. AUC is commonly used in species distribution analyses, as highlighted in the studies by the authors of [3,33,108,125]. The research conducted by the authors of [26,120] also utilized TSS alongside AUC. Thus, in our study, both AUC and TSS were utilized to assess the consistency and performance of the models. In this regard, our study aligns with the current literature.

In the investigation, climate and soil parameters were used in the suitability analysis carried out for sub-tropical products (avocado and pitaya). Although many parameters can be preferred for the analysis expressed, the application is mainly made with climate data. Refs. [3,26,33,126] preferred to use just climate data consisting of temperature and precipitation in their studies and conducted their analyses in light of these data. However, since the analyses made only with climate data do not include soil properties, it is difficult to say that they directly give definite and accurate results. Hence, in this study, in addition to climate properties, soil properties such as soil depth, texture, organic matter amount, salinity and pH, groundwater level, LUC, Lc, altitude, and slope variables were utilized. Analysis was made through the expressed parameters. This study differs from other studies in this respect and has originality. Similar to the parameters, other than climate, used in this study, Ref. [108] used elevation and slope parameters, and Ref. [127] used elevation, slope, and aspect criteria. Ref. [125] used Lu/Lc, elevation, and the Human Influence Index (HII), Ref. [66] used soil texture and pH, and Ref. [128] used soil depth and texture. In addition, this study differs from the studies in the literature in terms of the variety and number of preferred parameters (12). Ref. [129] preferred to use 5 parameters, Ref. [62] preferred to use 5 parameters, Ref. [130] preferred to use 4 parameters, Ref. [131] preferred to use 3 parameters, and Ref. [127] preferred to use 11 parameters. Ultimately, this study, using 12 variables, offers a new contribution to the literature in this respect.

The resolution of the parameters used is just as important as their number. Data collected or generated from various sources with differing resolutions can significantly impact the analysis. Therefore, it is essential to convert all parameters to a common resolution before conducting the analysis [132,133]. Ref. [134] conducted an agricultural suitability analysis using climate data (0.5° × 0.5°) and soil data (0.05°) at varying resolutions. Ultimately, all parameters were converted to a 0.5° × 0.5° resolution for the analysis. In a separate study by the authors of [1], 13 climate and soil datasets at different resolutions were converted to 30 arc seconds for analysis. Similarly, Ref. [135] performed an agricultural suitability analysis utilizing soil data (1:25,000) and climate data (1.0° × 1.0°) at different resolutions. In this research, data with different resolutions—comprising four climate parameters and eight soil parameters—were converted to raster format with a 25 m resolution for analysis. Thus, it can be concluded that this study shares similarities with existing literature. Furthermore, while the original resolution of the climate data used in this study was lower, the analysis was likely compatible, ecologically meaningful, and beneficial for decision-makers when all parameters were adjusted to a resolution of 25 m.

In this study, analyses were carried out for the future period by assessing the current distribution of the relevant species as well as the potential impacts of climate change. In this respect, it differs from the studies of [136,137,138], which only applied according to the current situation. In this study, SSP2-4.5 and SSP5-8.5 climate scenarios were used to determine suitable future locations for the species. These scenarios were preferred because they belong to the current WorldClimv2.1 and are more consistent and reliable than the representative concentration pathway (RCP) scenarios presented in the previous version (v1.4) [135]. This issue is also emphasized in the study conducted by the authors of [26], and the use of SSP scenarios is encouraged. Due to these stated features, the SSP scenarios were preferred in this study and are thought to have original value in this respect.

In the investigation performed by the authors of [126], two different GCMs (CNRM-CM5 and GFDL-CM3) were used under RCP4.5 and RCP8.5 scenarios. In the study conducted by the authors of [125], analyses were performed using HadGEM2-AO, CCSM4, and BCC-CSM1-1 GCMs together with RCP2.6, RCP4.5, and RCP8.5 climate scenarios. Ref. [139] conducted their study using one climate model (BCC-CSM1-1) and scenario (RCP2.6). In the studies conducted by the authors of [27,28], SSP scenarios were preferred because they were stated to be more up-to-date and accurate. In the first study, ACCESS-ESM1 and BCC-CSM-MR climate models were used under SSP2-4.5 and SSP5-8.5, while in the second study, three different scenarios (SSP1-2.6, SSP2-4.5, and SSP5-8.5) and GCMs (Access-CM2, HadGEM, and UKESM1) were selected. At this point, the use of appropriate climate models and scenarios is extremely important in studies to determine the potential distribution of species as a result of climate change [26]. In this context, HadGEM3-GC31-LL, MPI-ESM1-2-HR, and GFDL-ESM4 global climate models were used under SSP2-4.5 and SSP5-8.5 scenarios due to the location of the study area and the recommendation by GDM. Analyses were carried out according to these.

It is conceivable to say that the climate scenarios, models, and the quality and number of preferred parameters used in this study are different from similar studies in the literature and offer a new perspective. However, there are also points where the research is limited. (a) The spatial resolution of the parameters used, particularly the climate data, is approximately 900 m (838.95 m), which directly impacts the accuracy of the analyses performed. (b) Nonetheless, climate scenarios (SSP2-4.5 and SSP4-8.5) were used in the analyses performed for the future period. Although these are expressed as consistent and reliable, they contain estimated values. (c) The soil properties utilized in the analyses for the future period reflect current conditions. However, it is important to acknowledge that climate change may lead to alterations in soil structure. As such, it is reasonable to consider that the potential distribution maps generated for the projection period might have certain limitations. (d) Additionally, global warming can vary depending on many natural and artificial factors. Modeling with a limited number of criteria, the elimination of some factors or reasons such as failure to calculate leads to the updating of the scenarios expressed over time. The revision of the RCP scenarios in the current situation also supports this. (e) In this study, analyses were conducted using three different GCMs. Although their findings are predominantly similar, there are differences. It would not be wrong to say that the differences obtained negatively affect the consistency of the analysis conducted. However, it seems that these models can be used because they are the models recommended by GDM. (f) The analyses conducted in this study include methodological findings. This study does not include the evaluation of socio-economic factors, such as the economic value of species (avocado and pitaya) and farmer behaviors. It is possible to express the points where this study is limited in this way in general.

5. Conclusions

This study was conducted at Mersin, which is in the Mediterranean basin and has a sub-tropical climate. First of all, the determination of suitable agricultural products for the study area, which is extremely vulnerable to the negative effects of global warming due to its location and the climate it affects, was carried out. For this goal, the current and future climate (temperature and precipitation) characteristics of the study area were determined. Furthermore, the characteristics of the soil were also identified. In addition to the climate and soil characteristics of the study area, an agricultural suitability analysis was carried out for sub-tropical products (avocado and pitaya) that will contribute to Türkiye’s sustainable agriculture target with their current and potential economic values. The analysis consisted of two steps. First, the potential distribution areas of the species were determined with the MaxEnt model through the reference period (1971-200) climate and soil data for the current situation. Secondly, the same processes were repeated with HadGEM3-GC31-LL, MPI-ESM1-2-HR, and GFDL-ESM4 climate models under SSP2-4.5 and SSP5-8.5 scenarios using possible future climate characteristics and current soil data for the projection period (2021–2100). When examining the anticipated quantitative changes for the future compared to the current period, a 2.43% increase in the avocado within the “S1” class is projected under the SSP 2-4.5 scenario, based on the average of all models and periods analyzed. In the SSP5-8.5 scenario, which represents a high radiative forcing and limits warming to approximately 5 °C, the expected increase is 1.23%. In contrast, the analysis of pitaya reveals a decline of 3.93% under the first scenario and a decrease of 4.35% under the second scenario. These findings suggest species have distinct temperature and precipitation thresholds, indicating varying ecological tolerances.

From a qualitative perspective on future species distributions, it is noteworthy that avocado and pitaya find suitable growing conditions primarily in the coastal and southern regions of the study area. Conversely, highland and mountainous areas are largely deemed unsuitable for cultivation. In addition, a long-term decreasing trend is anticipated for pitaya in the “S1” class due to rising temperatures and irregularities in the precipitation regime. However, increases observed in the “S2” and “S3” classes demonstrate that the study area may support the cultivation of this species in the future.

As a result, a current situation and a future perspective for the study area were drawn. Therefore, an alternative was presented for the sustainable continuation of agricultural activities even under the possible negative effects of climate change.

Based on the experiences, analyses, and results evolved throughout the planning of this study, literature research, and draft writing, it is advised that researchers working on determining agricultural appropriateness pay attention to several points.

(1)

Agricultural suitability determination studies should utilize criteria that are appropriate for the product’s growth conditions. Using irrelevant, under, or over variables has a direct impact on the correctness of the analysis.

(2)

The resolution of the parameters employed has an enormous effect on the reliability and consistency of the analysis. Furthermore, even if the parameters retrieved or developed have varied resolutions, it is important to ensure that they are all at the same resolution when creating the final distribution map. Otherwise, the analysis may be negatively impacted.

(3)

Preferred methods are just as vital as parameters. In recent years, ML approaches have been widely used. They have been utilized because they are modern and reliable. Nonetheless, RS- and GIS-based multi-criteria decision-making (MCDM) techniques have been employed, too. The approaches are time-consuming; thus, they have drawbacks. However, the reliabilities are reasonable.

(4)

This study supplies a basic foundation for engineers, agricultural experts, climate scientists, and managers involved in the decision-making process. Especially, it is hoped that it will have a substantial impact on future decisions regarding the type of agricultural activities. It is anticipated that the preferred method and parameters can be used in different regions without modification for similar products (sub-tropical). However, at this stage, it is critical to select climate models appropriate for the research area’s location in the analyses to be conducted throughout the future period.