Analysis of Suitable Cultivation Sites for Gastrodia elata Using GIS: A Comparison of Various Classification Methods

Abstract

Gastrodia elata has been a valuable medicinal resource in the East for approximately 3000 years. In South Korea, G. elata is cultivated in open-fields or greenhouses near residential areas. However, due to severe continuous damage, cultivation sites need to be frequently relocated, leading to a shortage of available cultivation areas. Alternatively, farmers are focusing on mountain cultivation. This study analyzed suitable cultivation sites for G. elata in mountainous areas using a geographic information system (GIS) and applied various classification methods to identify their characteristics and similarities. The analysis showed that the Natural Breaks (Jenks) classification method maximized the differences between grades, whereas the Quantile method reclassified the area of suitable sites to a relatively high proportion. In contrast, the Equal Interval method reclassified the areas of suitable and unsuitable sites to a lower proportion, whereas the Geometric Interval method best demonstrated extreme-temperature regions as unsuitable sites. Among the classification methods, the Natural Breaks (Jenks) and Geometric Interval methods yielded the most similar results. These findings provide critical methodological outcomes for G. elata cultivation and sustainable agriculture and forestry. Future empirical research and the application of climate change scenarios are necessary to enhance the sustainability of the G. elata cultivation industry.

1. Introduction

Gastrodia elata Blume (GE) is a perennial orchidaceous vascular plant. However, its leaves and roots are degenerate, making independent growth impossible. Photosynthesis and carbon assimilation are temporary, and they grow parasitically by obtaining nutrients through a symbiotic relationship with the basidiomycete fungus Armillaria [1,2,3,4,5,6]. This allows them to thrive even in low-light forest environments. The stems and leaves of GE temporarily grow above the ground in spring (March to May) in South Korea when temperatures are between 15 and 18 °C, but they perish within a month [1]. GE grows primarily in the high-altitude forested valleys in North Asia, including South Korea, China, and Japan, and in South Asia, including Bhutan and India. It is commonly found in oak-dominated forests rich in humus and sloped terrain [1,7,8,9].

GE has been widely used in oriental medicine for many centuries owing to its medicinal properties. It has been valued as a medicinal resource in the East for approximately 3000 years. Tubers, stalks, and seeds, which resemble potatoes, are used to prevent and treat ailments such as hypertension, headaches, paralysis, dizziness, and diabetes. It is regarded as one of the most effective herbs for treating neurological conditions such as stroke and apoplexy [1,10,11]. In particular, GE has shown high efficacy in inhibiting cancer cell motility due to its anticancer effects [12]. Additionally, gastrodin, a major component of GE, is highly effective against neurological disorders owing to its antioxidant, anticonvulsant, anti-inflammatory, antiepileptic, anti-obesity, and anxiolytic properties. Vanillyl alcohol has been reported to be effective against epilepsy, heart attacks, and lung cancer [7,10].

GE thrives in shaded, humid environments and typically grows in forests through a symbiotic relationship with mycorrhizal fungi. GE production in South Korea and China relied primarily on wild plants to meet the market demand until the 1960s. However, since the development of artificial cultivation methods in the 1970s, the cultivation of GE has become increasingly common. In South Korea, the large-scale artificial cultivation of GE began in earnest in the 1990s [10]. Cultivation methods for GE are diverse, but fundamentally, they involve mimicking the natural life cycle of GE by planting Armillaria fungus, logs (oak species), and seed tubers (immature GE) together in the soil to establish a symbiotic relationship [7]. The cultivation types of GE can be broadly divided into artificial and facilitated cultivation. Artificial cultivation, which refers to open-field cultivation, involves inoculating oak logs with Armillaria fungus, burying them in the soil, and then inoculating them with seed tubers. Currently, this method is used by most farmers. Facilitated cultivation involves setting up greenhouses and cultivating GE using the same method as open-field cultivation [7].

As per the forest product survey in South Korea on medicinal plants, the production of GE was 343,891 kg (USD 277,000) in 2014 when the survey began. By 2022, annual production increased to 504,668 kg (USD 27,780,000), approximately 1.5 times higher. Muju County, which has maintained its position as the main production area for GE in South Korea, accounted for 22% of the national production in 2014 and 63% in 2022. Production in Muju County expanded from 75,110 kg (USD 62,000) in 2014 to 340,887 kg (USD 17,499,000) in 2022, an increase of more than 4.5 times [13,14,15,16,17].

GE research has been conducted in various fields for several years. Recent studies have primarily focused on the pharmacological efficacy of GE, the extraction and characterization of bioactive compounds [18,19], cultivation techniques [20,21,22,23], and physiological properties [5,6,24,25]. Pharmacologically, GE has been proven to have neuroprotective effects, enhance learning and memory, provide cardiac protection, regulate blood vessels, and have antidepressant and anticancer effects [26,27,28,29,30]. Furthermore, studies on the potential distribution simulation of GE under various climate change scenarios [31] and geographical tracking studies for habitat distribution prediction, resource exploration, and the conservation of GE [32] have been conducted in China. However, studies exploring the optimal cultivation sites for GE, considering its growth characteristics, are lacking.

According to an interview with officials from the Forestry Department of Muju County on 21 May 2024, diseases due to continuous cropping damage are increasing at existing GE cultivation sites, such as open fields and greenhouses near urban areas. Due to a shortage of suitable cultivation sites, GE farmers are turning their attention to mountainous areas for new cultivation opportunities. The continuous cultivation of GE causes soil-borne diseases, leading to significant ecological problems that drastically reduce the yield and quality [5,33,34]. To avoid these issues, it is recommended to leave the land fallow for at least three years [35]. Considering the increasing need to reduce continuous cropping damage at existing open-field or greenhouse cultivation sites, it is crucial to analyze suitable cultivation sites for GE in mountainous areas using geographic information systems (GIS). Mountain cultivation can serve as an alternative to address this issue, and overlay analysis using GIS can help scientifically identify new cultivation areas. Using the native habitat information of GE, powerful ecological niche modeling methods such as MaxEnt (Maximum Entropy Reinforcement Learning), Random Forest, and ENFA (Ecological Niche Factor Analysis) can predict the potential habitat of GE. Given the lack of native habitat information for GE and the need for identifying suitable sites for artificial cultivation, GIS overlay analysis can be a valuable alternative technique. It offers convenience in the procedural setup and aids in the rapid exploration of reliable potential areas [36].

Therefore, this study used GIS overlay analysis to analyze suitable cultivation sites for GE in mountainous areas. Specifically, this study reclassified a composite suitability map that synthesized the cultivation factors for GE using various reclassification methods and compared their characteristics. This study aimed to scientifically identify the optimal cultivation sites for GE in mountainous areas and evaluate the effectiveness and characteristics of different reclassification methods. This study pioneered a new research area by analyzing suitable cultivation sites for GE in mountainous areas, moving away from the traditional open-field or greenhouse cultivation sites. By utilizing various reclassification methods to analyze these sites and comparing the characteristics of each method, this study presents a unique and different approach from previous research. Additionally, it provides practical and applicable results by reflecting the actual trends of cultivation site relocation and addressing continuous cropping damage issues. This approach promotes the efficient use of land resources, contributes to sustainable agriculture and forestry practices, enhances agricultural productivity, and protects the environment.

2. Materials and Methods

2.1. Study Overview and Selection of the Study Area

The final step in analyzing suitable cultivation sites for crops in mountainous areas using GIS involves generating overlay operation data (composite suitability map) with various environmental factors relevant to optimal growth conditions and then reclassifying these data. The novelty and originality of this study can be attributed to the analysis of suitable cultivation sites for GE using various reclassification methods applicable to this reclassification process and to the comparison of the results obtained from each method. To this end, this study explored new GE cultivation sites in mountainous areas using various methods, focusing on Muju County, which is the main GE production area in South Korea. As of 2022, the total land area of Muju County is 63,207 ha, of which 51,165 ha is forested, resulting in a forest cover rate of approximately 81% [37] (Figure 1).

2.2. Selection of Factors for Suitable GE Cultivation Site Analysis and Assignment of Weights Based on Criteria

The environmental factors necessary for analyzing suitable cultivation sites for GE, that is, the optimal growth conditions for GE, were selected based on the Korea Forest Service’s Standard Cultivation Guidelines for Forest Products [38], Rural Development Administration’s Medicinal Mushroom Guidelines [39], and Muju GE Project’s Cultivation Manual of Gastrodia elata [40]. The individual factors and criteria required for the artificial cultivation of GE, as provided in the above-mentioned literature, were integrated to select the final factors. The final factors selected to analyze suitable cultivation sites for GE included elevation, aspect, slope, forest type (tree species), effective soil depth, soil texture, soil moisture, organic matter content, drainage, growing season temperature (April–November), active growth period temperature (June–October), summer temperature (June–August), and winter temperature (December–February). For temperature factors, the growing season and active growth period temperatures were averaged using monthly average temperatures. However, the summer and winter temperatures were averaged using the monthly average maximum and minimum temperatures over 10 years. This is because GE’s growth is inhibited when the summer temperature exceeds 30 °C and perishes when it exceeds 35 °C. Additionally, GE is sensitive to extreme temperatures, as it freezes to death when the winter temperature falls below −15 °C. Therefore, regions with these temperature conditions were defined as extreme-temperature areas, unsuitable for GE cultivation, and were excluded from the composite suitability map (

Table 1. Factors selected for suitable GE cultivation site analysis and weight assignment based on these criteria.

Factor	Criterion	Rank	Weight
Elevation (m)	700≤	1	1
	400 ≤ elevation < 700	2	0.67
	<400	3	0.33
Aspect	Southeast	1	1
	Others	2	0.5
Slope (°)	<15	1	1
	15≤	2	0.5
Forest type (tree species)	Quercus acutissima, Q. mongoica,	1	1
	Q. variabilis, Another Quercus spp.
	Others	2	0.5
Effective soil depth (cm)	60≤	1	1
	30 ≤ depth < 60	2	0.67
	<30	3	0.33
Soil texture	Loan, Sandy loam	1	1
	Others	2	0.5
Soil moisture	Moderately moist	1	1
	Slightly dry, slightly moist	2	0.67
	Dry, wet	3	0.33
Organic matter content (%)	4≤	1	1
	2 ≤ organic content < 4	2	0.67
	<2	3	0.33
Drainage	Good, very good	1	1
	Moderate	2	0.67
	Poor	3	0.33
Growing season temperature (April–November) (°C)	15 ≤ temperature < 30	1	1
	Other	2	0.5
Active growth period temperature (June–October) (°C)	20 ≤ temperature ≤ 25	1	1
	Other	2	0.5
Summer temperature (June–August) (°C)	30≤	–	Exclusion x
Winter temperature (December–February z) (°C)	≤−15	–	Exclusion

^z Winter in South Korea spans December of the current year to February of the following year. ^x Regions identified as extreme-temperature regions are excluded from the composite suitability map.

).

Meanwhile, the composite suitability map was generated by applying weights to the criteria of various factors and performing mathematical operations on these factors. The weights were standardized (evenly divided) to values between 0 and a maximum of 1 based on the ranks of the criteria for each factor. The ranks of the criteria for each factor were assigned by evaluating their importance based on the frequency of mentions in the GE cultivation guidelines [38,39,40]. Equation (1) is a reformulated formula for converting weights into standardized scores based on the ranks of the criteria for each factor [41], with differentiated weights shown in Table 1.

$${Weight}_k={\displaystyle \frac{1}{n}}\times \left[n-\sum _k^n(k-1)\right]$$

(1)

where n represents the number of criteria for each factor and k represents the rank of the corresponding criterion.

2.3. Generation of Thematic Maps for Environmental Factors

To generate the composite suitability map, thematic maps for environmental factors such as elevation, aspect, and slope were created using a digital elevation model (DEM) with a spatial resolution of 10 m × 10 m, generated from a 1:5000 digital topographic map in shapefile format provided by the National Geographic Information Institute of South Korea. The forest type (tree species) was derived from the 1:5000 large-scale forest-type map in shapefile format provided by the Korea Forest Service, and regions unsuitable for GE cultivation, such as non-stocked forest land, grassland, and orchards, were removed before creating the thematic map. To obtain precise results, the soil depth and texture were analyzed using a 1:5000 large-scale forest soil map in shapefile format provided by the Korea Forest Service. For factors not included in the large-scale forest soil map in shapefile format, such as soil moisture, organic matter content, and drainage, thematic maps were created using the 1:25,000 5th forest soil map provided by the Korea Forest Service. All thematic maps were generated in the raster format with a spatial resolution of 10 m × 10 m.

To create thematic maps for the temperature factors, this study calculated and utilized the geo-temperature index (gT; °C). WorldClim versions 1 [42] and 2 [43], and CHELSA (Climatologies at High Resolution for the Earth’s Land Surface Areas) version 1.2 [44] are excellent datasets for understanding the climate of terrestrial areas worldwide. However, with a spatial resolution of approximately 1 km², they are deemed suitable only for large-scale regional analysis. In this study, we focused on a small-scale area and thus generated and used thematic maps for temperature factors with a relatively higher spatial resolution of 10 m × 10 m. The geo-temperature index represents the relative temperature variation with elevation (EL; m), considering the temperature lapse rate (TLR; °C/m), base elevation (BaEL; m), and base temperature (BaT; °C) (Equation (2)). In South Korea, a temperature lapse rate of 0.006 °C/m is applied [45].

$$gT=BaT-(EL-BaEL)\times TLR$$

(2)

To calculate the geo-temperature index, monthly temperature data over the past 10 years (2013–2023) were averaged from eight automatic weather stations (AWSs; Korea Meteorological Administration; Figure 1)—the only three stations within Muju County (Muju, Deogyusan, and Seolcheonbong) and five stations located closest to and surrounding the outskirts of Muju County (Gagok, Daedeok, Buksang, Donghyang, and Jinan-Jucheon) (

Table 2. Average temperatures for temperature factors over the last 10 years (2014–2023) obtained from AWS.

Temperature Factor	Average Temperatures by AWS Locations (°C)
	Muju (212 y)	Deogyusan (660)	Seolcheonbong (1515)	Gagok (118)	Daedeok (205)	Buksang (324)	Donghyang (320)	Jinan-Jucheon (269)
GST z	17.6	15.5	11.1	17.9	17.4	17.2	16.6	16.6
AGPT	20.8	18.6	14.1	21.2	20.5	20.3	19.9	20.0
ST	29.9	26.5	21.6	30.3	29.3	29.0	29.2	29.9
WT	−13.9	−14.8	−20.6	−13.7	−12.3	−12.5	−15.6	−15.4

^z GST = Growing season temperature (April–November), AGPT = Active growth period temperature (June–October), ST = Summer temperature (June–August), and WT = Winter temperature (December–February). ^y Numbers in parentheses indicate the elevations (m) of the AWS installation locations.

). Since Muju County is larger than the coverage area of the three internal stations, five additional external stations surrounding the county were used to estimate temperatures in areas farther from the internal stations. These AWSs are considered to currently provide precise geographic coverage of Muju County in South Korea. The temperatures at each AWS location were converted to 0 m elevation temperatures using Equation (2). These data were then spatially interpolated using the spherical semi-variogram function and ordinary kriging method [46,47], followed by the generation of thematic maps in the raster format, with a spatial resolution of 10 m × 10 m, for temperature factors across Muju County using the DEM and Equation (2).

2.4. Generation and Analysis of Composite Suitability Map and Final Suitability Map

The composite suitability map is a thematic map reclassified into four grades from the initial weighted sum map, which was created by applying a weighted sum to the thematic maps of environmental factors to explore suitable cultivation sites for GE. To generate a composite suitability map, thematic maps for each factor were created as raster data with a spatial resolution of 10 m × 10 m. The attribute values of the raster data were reclassified into ranks corresponding to the criteria for each factor. Using the weighted sum function, an initial weighted sum map was created by assigning weights to the ranks of 11 thematic maps, excluding the extreme-temperature maps (summer and winter temperatures). The initial weighted sum map was reclassified using various classification methods to generate a composite suitability map. The suitability grades for GE cultivation were reclassified into four grades: suitable site (SS), possibly suitable site (PSS), probably unsuitable site (PUS), and unsuitable site (US). The classification methods used for reclassification were the Natural Breaks (Jenks) (NB), Quantile (Qu), Equal Interval (EI), and Geometric Interval (GI) methods [48].

The NB groups had similar values and maximized the differences between grades. This method is useful when there are no meaningful intervals in the data, and an optimal grouping is sought [49,50,51]. Qu is a useful method for understanding the ranking of each value within a dataset. EI is recommended when intervals are meaningful and a specific number of grades is expected. Finally, the GI is useful for finding natural groupings in the data while keeping the range of grades approximately equal [49,51]. The reclassification results of the various methods were compared to analyze their differences.

A final suitability map was generated by removing areas corresponding to extreme temperatures from the composite suitability map. Using this map, the impact of applying extreme temperatures on the area proportions of each grade according to different classification methods was analyzed.

Finally, to evaluate the similarity between the classification methods, a confusion matrix based on pixel counts was generated for each classification method as a reference. The overall accuracy (Equation (3)), which represents the proportion of agreement between the classification methods, and kappa ($\widehat{K}$) analysis (Equation (4)), which measures classification accuracy while considering the probability of agreement by chance, were performed. A $\widehat{K}$ value of 0.8 or above indicates a high level of agreement, 0.4 to 0.8 indicates a moderate level of agreement, and below 0.4 indicates a low level of agreement [52]. Additionally, the reclassification accuracy for the same grade among the different classification methods was evaluated.

$$Overall accuracy (\%)={\displaystyle \frac{\sum _{i=1}^k{x}_{ii}}{N}}\times 100$$

(3)

where N is the total number of observations, k is the number of columns in the matrix, and x_ii is the number of observations in column i and row i.

$$\widehat{K}={\displaystyle \frac{N\sum _{i=1}^k{x}_{ii}-\sum _{i=1}^k(x_{i+}\times x_{+i})}{N^2-\sum _{i=1}^k(x_{i+}\times x_{+i})}}$$

(4)

where N represents the total number of observations, k represents the number of columns in the matrix, x_ii represents the number of observations in column i and row i, and x₁₊ and x₊₁ represent the sums of column i and row i, respectively.

Meanwhile, the spatial resolution for the initial weighted sum map, composite suitability map, and final suitability map was set to 10 m × 10 m. The GIS software used for analysis in this study included ArcGIS Pro 3.3.0 and ArcMap 10.3.1 (ESRI Inc., Redlands, CA, USA).

3. Results and Discussion

3.1. Analysis of Composite Suitability Map by Classification Method

To identify suitable cultivation sites for GE, an initial weighted sum map was created using the weighted sum function with 11 environmental factors, excluding the two extreme-temperature factors (Figure 2). This map was then reclassified using four classification methods to compare the composite suitability maps.

Among the classification methods, Qu reclassified the SS into the largest area. NB and EI reclassified the PUS into the largest areas within their respective methods. In the case of EI, even though the weighted score range was the same for each grade, there were significant differences in the areas of the grades. The GI reclassified the US as the largest area (

Table 3. Area proportions and weighted score ranges for the four-grade reclassification of composite suitability map by classification method.

Classification Method	Grade	Area (ha, %)	Weighted Score Range
NB z	SS y	4425.24 (9.4)	7.97–9.96 (1.99) x
	PSS	11,870.49 (25.2)	7.48–7.97 (0.49)
	PUS	26,565.21 (56.3)	6.80–7.48 (0.69)
	US	4308.26 (9.1)	5.48–6.80 (1.32)
	Total	47,169.20 (100.0)	5.48–9.96 (4.48)
Qu	SS	12,154.09 (25.8)	7.78–9.96 (2.18)
	PSS	11,162.89 (23.7)	7.31–7.78 (0.47)
	PUS	11,508.54 (24.4)	6.97–7.31 (0.33)
	US	12,343.68 (26.2)	5.48–6.97 (1.49)
	Total	47,169.20 (100.0)	5.48–9.96 (4.48)
EI	SS	129.41 (0.3)	8.84–9.96 (1.12)
	PSS	12,024.68 (25.5)	7.72–8.84 (1.12)
	PUS	32,800.01 (69.5)	6.60–7.72 (1.12)
	US	2215.10 (4.7)	5.48–6.60 (1.12)
	Total	47,169.20 (100.0)	5.48–9.96 (4.48)
GI	SS	2826.45 (6.0)	8.15–9.96 (1.81)
	PSS	9327.64 (19.8)	7.72–8.15 (0.43)
	PUS	17,609.01 (37.3)	7.29–7.72 (0.43)
	US	17,406.10 (36.9)	5.48–7.29 (1.81)
	Total	47,169.20 (100.0)	5.48–9.96 (4.48)

^z NB = Natural breaks (Jenks), Qu = Quantile, EI = Equal interval, and GI = Geometric interval. ^y SS = Suitable site, PSS = Possibly suitable site, PUS = Probably unsuitable site, and US = Unsuitable site. ^x Parentheses indicate the width of the weighted score range.

).

Examining the classification methods, the reclassification results using NB showed that out of the total 47,169.20 ha, 4425.24 ha (9.4%) were classified as an SS, with the PUS accounting for the highest proportion (56.3%) (Table 3). Despite having the widest weighted score range (1.99), the area of the SS was reclassified as the second smallest. This reflects the characteristic of NB to maximize the differences between grades [49].

The reclassification results using Qu showed that the SS accounted for 12,154.09 ha, which is 25.8% of the total area, making it the highest proportion for the SS. In Qu, the weighted score range was the widest for the SS at 2.18, and the narrowest was for the PUS. However, there were no significant differences in the area proportions for each grade (Table 3).

The reclassification results using EI showed that the weighted score range was the same for all grades at 1.12. However, the SS accounted for only 129.41 ha or 0.3% of the total area. In contrast, the PUS accounted for 69.5% of the total area, and when combined with the US, they accounted for 74.2% of the total area (Table 3). This indicates that despite the EI reclassification with the same weighted score range [49], the initial weighted sum map had a relatively high distribution of low scores.

The reclassification results using GI showed that the SS accounted for 2826.45 ha or 6.0% of the total area, making it the smallest area. The combined area of the SS and PSS was 25.8% of the total area, which was lower than that of the PUS (37.3%) and US (36.9%) (Table 3). Although the GI classification method is useful for finding natural groupings in data while keeping the range of grades approximately equal [49], it appears that this characteristic is not well reflected in the reclassification into four grades for GE cultivation suitability.

3.2. Analysis of Final Suitability Map Excluding Extreme-Temperature Regions

The final suitability map, created by removing the extreme-temperature regions from the composite suitability map, was analyzed. Among the environmental factors for analyzing suitable cultivation sites for GE, the extreme-temperature regions, where summer temperatures exceed 30 °C and winter temperatures are below −15 °C, covered an area of 32,055.06 ha. Regions with summer temperatures above 30 °C were primarily located in the northwestern part of Muju County in areas with relatively low elevations. In contrast, regions with winter temperatures below −15 °C were widely distributed in the southeastern part of Muju County. This area includes Deogyusan Mountain (altitude 1614 m), which has a relatively high elevation (Figure 3).

Extreme-temperature regions were removed from the four composite suitability maps generated by the different classification methods, and the final suitability maps were created and the areas analyzed by grade (Figure 4).

Compared to the composite suitability map, the most notable changes were the common increases in the area proportions of the SS and PSS by 0.1–4.2%p and 2.2–4.1%p, respectively, while the area proportion of the US decreased by 3.3–7.7%p. Additionally, the area proportion of the PUS decreased by 0.8%p and 0.9%p in Qu and EI, respectively (

Table 4. Final area proportions and changes after excluding extreme-temperature regions for the four-grade reclassification of the composite suitability map.

Classification Method	Grade	Area (ha, %)	Change
NB z	SS y	1667.18 (11.0)	1.6 x
	PSS	4134.60 (27.4)	2.2
	PUS	8561.26 (56.6)	0.3
	US	751.10 (5.0)	−4.2
	Total	15,114.14 (100.0)	–
Qu	SS	4530.94 (30.0)	4.2
	PSS	4139.02 (27.4)	3.7
	PUS	3562.12 (23.6)	−0.8
	US	2882.06 (19.1)	−7.1
	Total	15,114.14 (100.0)	–
EI	SS	58.72 (0.4)	0.1
	PSS	4472.22 (29.6)	4.1
	PUS	10,368.05 (68.6)	−0.9
	US	215.15 (1.4)	−3.3
	Total	15,114.14 (100.0)	–
GI	SS	1071.69 (7.1)	1.1
	PSS	3459.25 (22.9)	3.1
	PUS	6169.69 (40.8)	3.5
	US	4413.51 (29.2)	−7.7
	Total	15,114.14 (100.0)	–

^z NB = Natural breaks (Jenks), Qu = Quantile, EI = Equal interval, and GI = Geometric interval. ^y SS = Suitable site, PSS = Possibly suitable site, PUS = Probably unsuitable site, and US = Unsuitable site. ^x Changes (%p) represent the difference in area proportions before and after excluding the extreme-temperature regions.

). This is due to the inclusion of a large area classified as a US in the extreme-temperature regions, resulting in a relative increase in the area proportion of the SS or PSS.

When examining the classification methods, in the final suitability map reclassified by NB, the SS was 1667.18 ha, which was 11.0% of 15,114.14 ha. This represents a 1.6%p increase compared with the composite suitability map. The proportion of the PSS area increased by 2.2%p, showing the highest increase in NB. In contrast, the proportion of the US population decreased by 4.2%p (Table 4).

In the final suitability map reclassified by Qu, the area of the SS was 4530.94 ha, which is 30.0% of the total area, making it the largest SS area among the four classification methods. Additionally, the SS area increased by 4.2%p compared to the composite suitability map, which was also the largest increase among the four classification methods. Furthermore, the area proportion of USs decreased by 7.1%p, and the area proportion of the PUS decreased by 0.8%p (Table 4). In the composite suitability map reclassified by Qu, the differences in area proportions between grades were small, ranging from 0.7 to 2.1%p (see Table 3). However, in the final suitability map, the differences in the area proportions between grades were relatively large, ranging from 2.6 to 4.5%p (Table 4).

In the final suitability map, the area of the SS reclassified by EI was 58.72 ha, which was 0.4% of the total area, making it the smallest SS area among the four classification methods. The proportion of the SS increased by 0.1%p compared to the composite suitability map, but this difference was not considered significant. In contrast, the area proportion of USs decreased by 3.3%p, and the area proportion of the PUS also decreased by 0.9%p. However, even with the removal of extreme-temperature regions, the EI classification method still resulted in the combined area of the PUS and US, accounting for 70% of the total area (Table 4). This was likely due to the influence of the initial weighted sum map, which had a relatively high distribution of low scores (see Figure 2).

In the final suitability map reclassified by GI, the SS was 1071.69 ha, which was 7.1% of the total area. Compared with the composite suitability map, the area proportions of the SS, PSS, and PUS increased by 1.1%p, 3.1%p, and 3.5%p, respectively, whereas the area proportion of the US decreased by 7.7%p. This represents the largest decrease in the area proportion among the four classification methods. This is likely because the GI reclassified the extreme-temperature regions, which had a high distribution of low scores in the initial weighted sum map, to a high proportion in the US. Additionally, similar to EI, the combined areas of the PUS and US accounted for 70% of the total area of GI (Table 4).

In the final suitability map applying extreme-temperature factors, NB exhibited characteristics that maximized the differences between grades, as it effectively characterized the distribution of each factor [53]. In groundwater potential studies [54] and disaster-related fields such as landslide susceptibility studies [55] and flood vulnerability assessment studies [51], NB was also found to be the most suitable reclassification method. Qu reclassified the area of the SS at a relatively high rate, which can be considered a reclassification method for securing large areas for cultivation. In contrast, EI reclassified the SS and US areas at the lowest rate when compared with the other classification methods, making it a useful classification method when the strict selection of the SS is desired. The GI most accurately reflected areas with extreme temperatures, such as the US, because it tends to underestimate the weighted score when compared with other classification methods [51]. The application of extreme temperatures resulted in distinct differences in the distribution of area proportions by grade on the composite suitability map. This emphasizes the significant impact of extreme-temperature factors on the spatial distribution of the cultivation suitability grades. In addition, extreme temperatures can be a critical factor in effectively identifying suitable growth areas for GE and maximizing its viability.

3.3. Evaluation of Similarities Among Classification Methods

3.3.1. Similarity Evaluation Based on NB

A confusion matrix was generated based on the pixel count for the reclassification results by grading the final suitability map using NB, and the results were compared with those from other classification methods. Based on this, the overall accuracy, kappa coefficient, and reclassification accuracy for the same grade were compared and analyzed. The overall accuracy and kappa coefficient between NB and EI were 77.40% and 0.57, respectively, indicating the highest similarity among the classification methods. However, the average value of the reclassification accuracy for the same grade between the different classification methods was the highest (72.7%) between NB and GI. This means that based on NB, GI reclassified the same areas (locations) more similarly for all four grades compared to Qu or EI. When the primary objective was to explore the SS, the reclassification accuracy of the SS with EI was very low at 3.5%. In contrast, the reclassification accuracy for the SS was 100.0% for Qu and 64.3% for GI (

Table 5. Overall accuracy and kappa coefficient for the reclassification methods based on NB using pixel counts.

Classification Method	Grade	NB					Overall Accuracy (%)	Kappa Coefficient
		SS y	PSS	PUS	US	Total
Qu z	SS	166,718	286,376	0	0	453,094	47.98	0.31
	PSS	0	127,084	286,818	0	413,902
	PUS	0	0	356,212	0	356,212
	US	0	0	213,096	75,110	288,206
	Total	166,718 (100.0) x	413,460 (30.7)	856,126 (41.6)	75,110 (100.0)	1,511,414 (68.1) w
EI	SS	5872	0	0	0	5872	77.40	0.57
	PSS	160,846	286,376	0	0	447,222
	PUS	0	127,084	856,126	53,595	1,036,805
	US	0	0	0	21,515	21,515
	Total	166,718 (3.5)	413,460 (69.3)	856,126 (100.0)	75,110 (28.6)	1,511,414 (50.4)
GI	SS	107,169	0	0	0	107,169	63.42	0.47
	PSS	59,549	286,376	0	0	345,925
	PUS	0	127,084	489,885	0	616,969
	US	0	0	366,241	75,110	441,351
	Total	166,718 (64.3)	413,460 (69.3)	856,126 (57.2)	75,110 (100.0)	1,511,414 (72.7)

^z NB = Natural breaks (Jenks), Qu = Quantile, EI = Equal interval, and GI = Geometric interval. ^y SS = Suitable site, PSS = Possibly suitable site, PUS = Probably unsuitable site, and US = Unsuitable site. ^x Parentheses indicate the percentage (%) of the number of pixels in the grade of the comparative classification method relative to the number of pixels in the grade of the reference classification method. ^w Parentheses indicate the arithmetic mean value (%) of reclassification accuracy for the same grade between the different classification methods.

).

3.3.2. Similarity Evaluation Based on Qu

Based on Qu, the overall accuracy with NB was 47.98%, and the kappa coefficient was 0.31, indicating the highest similarity among the compared classification methods. The reclassification accuracy of the SS with NB was 36.8%, which was the highest among the compared classification methods. Additionally, the average value of reclassification accuracy for the same grade between the different classification methods was the highest at 48.4% between Qu and NB. Notably, both EI and GI reclassified the PSS from Qu as a PUS (

Table 6. Overall accuracy and kappa coefficient for the reclassification methods based on Qu using pixel counts.

Classification Method	Grade	Qu					Overall Accuracy (%)	Kappa Coefficient
		SS y	PSS	PUS	US	Total
NB z	SS	166,718	0	0	0	166,718	47.98	0.31
	PSS	286,376	127,084	0	0	413,460
	PUS	0	286,818	356,212	213,096	856,126
	US	0	0	0	75,110	75,110
	Total	453,094 (36.8) x	413,902 (30.7)	356,212 (100.0)	288,206 (26.1)	1,511,414 (48.4) w
EI	SS	5872	0	0	0	5872	25.38	0.01
	PSS	447,222	0	0	0	447,222
	PUS	0	413,902	356,212	266,691	1,036,805
	US	0	0	0	21,515	21,515
	Total	453,094 (1.3)	413,902 (0.0)	356,212 (100.0)	288,206 (7.5)	1,511,414 (27.2)
GI	SS	107,169	0	0	0	107,169	39.59	0.21
	PSS	345,925	0	0	0	345,925
	PUS	0	413,902	203,067	0	616,969
	US	0	0	153,145	288,206	441,351
	Total	453,094 (23.7)	413,902 (0.0)	356,212 (57.0)	288,206 (100.0)	1,511,414 (45.2)

^z NB = Natural breaks (Jenks), Qu = Quantile, EI = Equal interval, and GI = Geometric interval. ^y SS = Suitable site, PSS = Possibly suitable site, PUS = Probably unsuitable site, and US = Unsuitable site. ^x Parentheses indicate the percentage (%) of the number of pixels in the grade of the comparative classification method relative to the number of pixels in the grade of the reference classification method. ^w Parentheses indicate the arithmetic mean value (%) of reclassification accuracy for the same grade between the different classification methods.

). This is likely because EI and GI have wider weighted score ranges for the PUS than for Qu.

3.3.3. Similarity Evaluation Based on EI

For EI, the overall accuracy with NB was 77.40% and the kappa coefficient was 0.57, indicating the highest similarity among the compared classification methods. Additionally, the reclassification accuracy of the SS was 100.0% for NB, Qu, and GI. Notably, the reclassification accuracy of the US was 100.0% across all three classification methods (

Table 7. Overall accuracy and kappa coefficient for the reclassification methods based on EI using pixel counts.

Classification Method	Grade	EI					Overall Accuracy (%)	Kappa Coefficient
		SS y	PSS	PUS	US	Total
NB z	SS	5872	160,846	0	0	166,718	77.40	0.57
	PSS	0	286,376	127,084	0	413,460
	PUS	0	0	856,126	0	856,126
	US	0	0	53,595	21,515	75,110
	Total	5872 (100.0) x	447,222 (64.0)	1,036,805 (82.6)	21,515 (100.0)	1,511,414 (86.7) w
Qu	SS	5872	447,222	0	0	453,094	25.38	0.01
	PSS	0	0	413,902	0	413,902
	PUS	0	0	356,212	0	356,212
	US	0	0	266,691	21,515	288,206
	Total	5872 (100.0)	447,222 (0.0)	1,036,805 (34.4)	21,515 (100.0)	1,511,414 (58.6)
GI	SS	5872	101,297	0	0	107,169	65.52	0.47
	PSS	0	345,925	0	0	345,925
	PUS	0	0	616,969	0	616,969
	US	0	0	419,836	21,515	441,351
	Total	5872 (100.0)	447,222 (77.3)	1,036,805 (59.5)	21,515 (100.0)	1,511,414 (84.2)

^z NB = Natural breaks (Jenks), Qu = Quantile, EI = Equal interval, and GI = Geometric interval. ^y SS = Suitable site, PSS = Possibly suitable site, PUS = Probably unsuitable site, and US = Unsuitable site. ^x Parentheses indicate the percentage (%) of the number of pixels in the grade of the comparative classification method relative to the number of pixels in the grade of the reference classification method. ^w Parentheses indicate the arithmetic mean value (%) of reclassification accuracy for the same grade between the different classification methods.

). This is likely because the areas of the SS and US in the final suitability map by EI were reclassified at very low rates of 0.4% and 1.4% of the total area, respectively, thus encompassing the reclassification results of the SS and US using the comparison classification methods. Meanwhile, the average value of the reclassification accuracy for the same grade between the different classification methods was the highest, at 86.7%, between EI and NB (Table 7).

3.3.4. Similarity Evaluation Based on GI

Based on the GI, the overall accuracy with EI was 65.52%, and the kappa coefficient was 0.47, indicating the highest similarity among the compared classification methods. However, the reclassification accuracy of the SS was very low, with an EI of 5.5%, but it was 100.0% for both NB and Qu. The average reclassification accuracy for the same grade between the different classification methods was the highest, at 69.8%, between GI and NB (

Table 8. Overall accuracy and kappa coefficient for the reclassification methods based on GI using pixel counts.

Classification Method	Grade	GI					Overall Accuracy (%)	Kappa Coefficient
		SS y	PSS	PUS	US	Total
NB z	SS	107,169	59,549	0	0	166,718	63.42	0.47
	PSS	0	286,376	127,084	0	413,460
	PUS	0	0	489,885	366,241	856,126
	US	0	0	0	75,110	75,110
	Total	107,169 (100.0) x	345,925 (82.8)	616,969 (79.4)	441,351 (17.0)	1,511,414 (69.8) w
Qu	SS	107,169	345,925	0	0	453,094	39.59	0.21
	PSS	0	0	413,902	0	413,902
	PUS	0	0	203,067	153,145	356,212
	US	0	0	0	288,206	288,206
	Total	107,169 (100.0)	345,925 (0.0)	616,969 (32.9)	441,351 (65.3)	1,511,414 (49.6)
EI	SS	5872	0	0	0	5872	65.52	0.47
	PSS	101,297	345,925	0	0	447,222
	PUS	0	0	616,969	419,836	1,036,805
	US	0	0	0	21,515	21,515
	Total	107,169 (5.5)	345,925 (100.0)	616,969 (100.0)	441,351 (4.9)	1,511,414 (52.6)

^z NB = Natural breaks (Jenks), Qu = Quantile, EI = Equal interval, and GI = Geometric interval. ^y SS = Suitable site, PSS = Possibly suitable site, PUS = Probably unsuitable site, and US = Unsuitable site. ^x Parentheses indicate the percentage (%) of the number of pixels in the grade of the comparative classification method relative to the number of pixels in the grade of the reference classification method. ^w Parentheses indicate the arithmetic mean value (%) of reclassification accuracy for the same grade between the different classification methods.

).

When evaluating the similarity between the classification methods by comparing the overall accuracy and kappa coefficient, NB and EI showed the highest similarity. However, because EI reclassified the SS and US at relatively low rates, it is difficult to conclude whether its similarity is the highest. Therefore, considering the overall accuracy, the kappa coefficient, peculiarities of EI reclassification results, and the average value of the reclassification accuracy for the same grade between different classification methods, NB and GI were judged to be the classification methods that provided the most similar results in the analysis of GE cultivation suitability. Meanwhile, in the landslide susceptibility study, NB and Qu exhibited very similar spatial patterns [55], unlike the reclassification results for the exploration of suitable cultivation sites for GE in this study.

4. Conclusions

The use of GIS allowed the analysis of GE cultivation suitability in mountainous areas, clearly identifying the characteristics and limitations of various classification methods. Among the factors crucial to crop growth conditions, temperature conditions, especially the application of extreme temperatures, can provide significant scientific benefits, such as the accurate identification of suitable cultivation areas for GE, the reduction of cultivation failure risks, sustainable cultivation management, the efficient use of cultivation areas, environmental monitoring, and ecosystem maintenance. Therefore, the application of extreme temperatures is essential for analyzing GE cultivation suitability, acting as a crucial methodological result for the successful cultivation of GE and the practice of sustainable agriculture and forestry.

Additionally, this study evaluated the characteristics and similarities of various classification methods in the exploration of GE cultivation suitability, confirming the applicability of GIS-based GE cultivation suitability exploration techniques and providing essential foundational data for selecting appropriate classification methods. However, empirical research on whether farmers have identified and cultivated GE in suitable areas in mountainous regions or whether they are producing GE stably remains a task that needs to be addressed. Therefore, further research is needed to investigate GE cultivation areas in mountainous regions through field surveys and to compare the accuracy of these areas with the results of various classification methods through overlay analysis. Additionally, by investigating the native habitats of GE in mountainous regions and applying correlative algorithms such as MaxEnt, Random Forest, and ENFA, and comparing these results, the completeness of the GE cultivation suitability analysis can be enhanced, providing a means to maximize the practical benefits for farmers. Future research should include additional environmental factors and climate change scenarios [56] that affect GE growth to predict future GE cultivation suitability, thereby improving the sustainability of GE cultivation and production.

Furthermore, we acknowledge that the limited number of AWSs used in this study may affect the accuracy of the estimated temperatures. Future research should aim to incorporate a greater number of AWSs to improve the precision of temperature estimates, thereby enhancing the reliability of GE cultivation suitability analyses. Additionally, research will be needed to elucidate the relationship between micro-climate conditions and GE growth, as these factors are not yet detailed in national cultivation manuals. Incorporating these findings into cultivation strategies could lead to the more precise identification of suitable cultivation sites, maximizing the practical benefits for farmers and contributing to the advancement of sustainable agriculture practices.