Next Article in Journal
Fungi Inhabiting Stem Wounds of Quercus robur following Bark Stripping by Deer Animals
Previous Article in Journal
Full-Length Transcriptome Sequencing and Identification of Genes Related to Terpenoid Biosynthesis in Cinnamomum migao H. W. Li
 
 
Search for Articles:
Title / Keyword
Author / Affiliation / Email
Journal
Article Type
 
 
Advanced Search
 
Section
Special Issue
Volume
Issue
Number
Page
 
Journals
Forests
Volume 14
Issue 10
10.3390/f14102076
forests-logo
Submit to this Journal Review for this Journal Propose a Special Issue
► Article Menu

Article Menu

share Share announcement Help format_quote Cite question_answer Discuss in SciProfiles thumb_up
...
Endorse textsms
...
Comment
first_page
settings
Order Article Reprints
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

Potential Distribution and Suitable Habitat for Chestnut (Castanea sativa)

by
Vasil Metreveli
1,2,3,*,
Holger Kreft
2,
Ilia Akobia
4,
Zurab Janiashvili
5,
Zaza Nonashvili
5,
Lasha Dzadzamia
1,
Zurab Javakhishvili
1 and
Alexander Gavashelishvili
6
1
School of Natural Sciences and Medicine, Ilia State University, Cholokashvili Ave. 3/5, 0162 Tbilisi, Georgia
2
Department of Biodiversity, Macroecology and Biogeography, University of Göttingen, 37077 Göttingen, Germany
3
Department of Geobotany, Institute of Botany, Ilia State University, Botanical Str. 1, 0105 Tbilisi, Georgia
4
Department of Geoinformatics, Institute of Botany, Ilia State University, Botanical Str. 1, 0105 Tbilisi, Georgia
5
Department of Biodiversity and Forestry, Ministry of Environmental Protection and Agriculture of Georgia, Marshal Archil Gelovani Ave. 34, 0159 Tbilisi, Georgia
6
Center of Biodiversity Studies, Institute of Ecology, Ilia State University, Cholokashvili Str. 5, 0162 Tbilisi, Georgia
*
Author to whom correspondence should be addressed.
Forests 2023, 14(10), 2076; https://doi.org/10.3390/f14102076
Submission received: 28 June 2023 / Revised: 12 August 2023 / Accepted: 19 August 2023 / Published: 17 October 2023
(This article belongs to the Section Forest Ecology and Management)
Browse Figures
Review Reports Versions Notes

Abstract

:
Chestnut, Castanea sativa Miller (Fagales: Fagaceae), is an ecologically and economically important tree species of the forest ecosystem in Southern Europe, North-Western Europe, Western Asia, North Africa, and the Caucasus. The distributional range of chestnut in Europe has been highly modified by humans since ancient times. Biotic and abiotic factors have dramatically changed its distribution. Historic anthropogenic range expansion makes it difficult to identify habitat requirements for natural stands of chestnut. In the Caucasus, natural stands of chestnut survived in glacial forest refugia and landscapes that have been difficult for humans to colonize. To identify the species reliable habitat requirements, we estimated the relationship between climatic variables and 620 occurrence locations of natural chestnut stands from the Caucasus and validated the model using GBIF data from outside the Caucasus. We found that our best model is reasonably accurate and the data from the Caucasus characterize chestnut stands throughout the species range well.

1. Introduction

Chestnut, Castanea sativa Miller (Fagales: Fagaceae), is one of the relict deciduous tree species and is the only native species from the genus Castanea in Europe and the Caucasus. It grows 20–35 m in height and is an important part of the forest ecosystem [1,2]. It occurs either as nearly monospecific stands or mixed with beech, Fagus orientalis L. (Fagales: Fagaceae), hornbeam, Carpinus betulus L. (Fagales: Betulaceae), and different oak species, mostly presented as a subdominant and submissive species [3]. Chestnut is a light-demanding species. The tree can grow in shade, but needs light in the early stages of growth [4].
Chestnut occurs in Southern Europe (Iberian Peninsula, Italy, Balkans, including larger Mediterranean Islands), North Africa (Morocco), North-Western Europe (England and Belgium), Western Asia (NE Türkiye and Syria), and the Caucasus (Georgia, Azerbaijan, Armenia, Russia, Türkiye, and Iran). According to evolutionary theories, different European forest woody species including chestnut spread to Europe from the Caucasus. Palynological studies support the presence of chestnut in Europe around 9000 years ago [5].
As a result of human activities and climate change, the distribution area of the chestnut has been highly modified and enlarged to the present period [6]. There were mainly two rapid expansions of chestnut—approximately 5000 years ago and approximately 2000 years ago, during the Roman empire [5]. However, its occurrence in some places is historically controversial. For instance, the species is considered to be cultivated in Great Britain; however, as the climate gets warmer due to climate change, the chestnut species is expanding further north [7,8]. At the same time, rising summer droughts impede its expansion in some areas [9].
Chestnuts prefers medium to high precipitation. The precipitation range is between 500 mm and 3500 mm/year [3]. Chestnuts occur at a mean annual temperature between 8° and 15 °C [4]. They have been reported to require an average temperature for the coldest month that is higher than −3.5 °C, whereas they can resist short periods of frost from −18 °C to −25 °C [2,10]. However, in the northern part of the Caucasus, it can survive under −35 °C because of the insulator effect of snow coverage [11]. Castanea sativa commonly occurs from 100 to 1800 m above sea level [3,4].
The main soil texture for chestnut represents sand, sandy loams, loams, and clay loams. It typically grows on poor sandy to loamy soil on slopes [12]. Castanea sativa avoids calcareous soils. On the other hand, it can be found on limestone substratum in several places in the Caucasus [3]. Deep soil is important to preserve the water capacity to help trees during the dry hot summer period [13], as chestnut is sensitive to summer droughts [9]. Water and heat stress affects the photosynthetic productivity and favor insects and diseases [14].
The increased human activities in the forest, anthropogenic habitat alteration, climate change, fungi as Phytophthora cambivora (Petri) Buisman (Peronosporales: Peronosporaceae), and emerging diseases such as Cryphonectria parasitica (Murrill) M.E.Barr (Diaporthales: Cryphonectriaceae) have a strong negative impact on chestnut stands [15,16]. Cryphonectria parasitica has become a big problem for chestnut trees in the last hundred years all over the species range. The native range of the disease is China and Japan. It was introduced in Europe in the 1930s from the USA and it caused relevant problems in the whole distribution area of the chestnut as it resulted in the death of many chestnut trees [17,18].
Historically, chestnut is a very important economical species for many people in Europe and the Caucasus. Because of its economic value, chestnuts have been introduced in many parts of the world. Fruits and wood have a crucial role in European and Caucasian countries’ economies. People have actively been using it for silvicultural and agricultural purposes, and for building houses with chestnut wood. The fruits have been used to make flour and they can be eaten in fresh, boiled, or fried. Chestnut is famous for its coppice utilization, which was the traditional method of stand regeneration to produce the wood rapidly. In the Caucasus, a lot of chestnut stands were used as coppice stands to produce firewood [2,13]. The demand for chestnut in Europe and the Caucasus is still high, which has resulted in more intensive management activities to increase nut and wood production [19].
The period of industrialization changed the species composition and shape of the forests in the world, particularly in parts of Europe. Primeval woodlands were removed and most of the forests in the continent were planted. At the same time, during reforestation, local species were not prioritized. Because of their great commercial value, different species were introduced from different continents. Many of these introduced species that exhibited good adaptations to the new conditions became naturalized and widespread [20]. That has made it difficult to delineate natural and non-natural habitats of chestnuts [21]. On the other hand, the forests in the Caucasus have been less impacted. Even though extensive industrial cutting operations were conducted in the 20th century (1930 to 1950), following this time (between 1950 and 1990), forests were mostly managed with protective and recreational goals [22]. In addition, many woodland areas in the Caucasus are still regarded as virgin because of the difficult terrain of the region [23]. Therefore, the natural stands of chestnuts can still be seen and assessed [24]. The Caucasus is an area where natural populations of C. sativa with a long evolutionary history can still be found [25]. Therefore, the analysis of the conditions of the presence of chestnuts in the Caucasus provides a good opportunity to find a suitable natural habitat throughout the species range.
Understanding the niche characteristics and the overall environmental requirements of a species can be effectively achieved through modeling [26]. This will be beneficial for bridging the knowledge gap about an important tree species such as C. sativa. On the other hand, modeling such widespread species as chestnut also has certain difficulties [27]. In addition, the selection of the special extent, climate datasets, and environmental variables to illustrate species distributions may result in different distribution maps for the same species [28,29]. Several chestnut distribution models have been developed during the past years for the Caucasus and Europe [4,25]. Unfortunately, these models do not always address the reliable habitat suitability and distribution boundaries of the chestnut trees due to the lack of representative presence data from its natural range [4]. However, since Georgia’s first national forest inventory (2019–2021) was completed, the representativeness of occurrence data in natural areas of taxa have improved [30,31]. This provides a good opportunity to model the natural distribution range of chestnuts.
The main objective of this study is as follows: (1) To model the suitable habitats and potential distribution of chestnuts in the Caucasus. (2) To see whether this distribution model is reliable in explaining the species global range. We anticipate that the model created using the data from the Caucasus will identify the species entire range. (3) To select the best climate dataset for our distribution model from the CHELSA climatology data (CHELSA version 2.1) [32] and the WorldClim climate data (WorldClim 2 version 2.1) [33]. Both of these datasets are actively used in the different species distribution and niche models. However, not a lot of studies have compared the modeling results evaluated by different climate datasets [29,32]. Furthermore, this is the first study that illustrates the comparison of the climate datasets for chestnut distribution.

2. Materials and Methods

2.1. Study Area

The study area represents the distribution range of the chestnut, located between −16° and 72° E longitude and 26° and 77° N latitude. It covers the whole distribution range (Figure 1).
Modeling the species global range using the data from its natural habitat is a common tactic used for invasive species [34]. However, given that it is challenging to distinguish between natural and anthropogenic chestnut stands in Europe [4], we decided to sample the Caucasus ecoregion rather than the entire study zone because it is a less impacted area compared with the other territories of the chestnut distribution range [22]. By employing this technique, we hope to eliminate the bias in the data caused by human intervention.
The ecoregion is located between 36° and 51° E longitude and 36° and 47° N latitude and is located between the Black Sea (oceanic nature) and the Caspian Sea (Figure 2); it represents the important Pleistocene refugia [7,35]. It covers an area of 580,000 km² and is composed of several prominent elements, including the North Caucasus Plain (below sea level); the Greater Caucasus Range, which is 1300 km in length from northwest to southeast between the Black Sea and Caspian Sea (highest peak 5642 m); the Lesser Caucasus 100 km to the south (from the Black Sea coast and Colchic lowlands to 4500 m); and the Highland of South Caucasus, which covers parts of Asia Minor (highest peak 5165 m) [35,36] (Figure 2).
The Caucasus is one of the 35 biodiversity hotspots in the world [37] and it has a very rich flora with over 6500 species of vascular plants, many of which are endemic species [38]. Forests are spread mostly in mesic climates. The biggest part of the forested areas is located close to the southern and eastern Black Sea coast, northwest of the Greater Caucasus, and along the southern coast of the Caspian Sea. The Caucasus region has rich native tree flora with 153 tree species [39]. The most dominant tree species are beech, hornbeam, chestnut, spruce, and fir [40].

2.2. Chestnut Occurrence Data

The fieldwork was conducted from 2018 to 2022. The stratified random sampling method was used for planning 166 sample plots in Türkiye and Georgia. The geographical locations were predetermined by the random stratified sampling method, using forest inventory maps (Forest Management Plans from 1990 to the present period) [41]. The chestnut stands were polygonised from the maps and sampled using the pre-GPS points (sample plots) [42]. The locations in the field were found and recorded using a GPS Garmin Montana 64s device. The coordinates of the single chestnut tree in the stand were also considered occurrences and were used in the analyses.
The Ministry of Environmental Protection and Agriculture of Georgia provided 647 presence sample plots, sampled using the random systematic method [43], from the first national forest inventory data in the country (collected in 2019–2021). The sample plots that were close to 1 km apart from the other occurrence spots were removed. In the end, 330 of the 647 locations were applied. An additional 30 occurrences were derived from herbarium specimens deposited in the Institute of Botany at Ilia State University, taken from 1960 until the present day. Four coordinates from Azerbaijan were provided by our colleague Berika Beridze in 2018. More than 30 randomly sampled records were provided by the United Nations Development Program (UNDP), among the project “Expansion and Improved Management Effectiveness of the Adjara Region’s Protected Areas (00088000) /2015”. Another 60 randomly sampled coordinates were provided by the Caucasus Nature Fund (CNF), collected from the project—“Technical assistance to conduct monitoring of the state of Sweet Chestnut and Chestnut Blight in Mtirala and Kintrishi Protected Areas in Georgia” (CNF/2021/TAGA-GEO-141).
In the end, 620 chestnut presence samples were collected in Georgia, Türkiye, and Azerbaijan. To reduce spatial autocorrelation in the species distribution modeling, we used locations with a minimum of 1 × 1 km distance between them, because we applied the climate raster layers with a spatial resolution of 1 km cells [44,45] (Figure 2).
We also used GBIF (Global Biodiversity Information Facility) data for Europe (downloaded from: https://doi.org/10.15468/dl.k6d6hn accessed on 25 February 2023) to check how the data from the Caucasus illustrates the favorable climate conditions for C. sativa in Europe [46]. The occurrences in parks, museums, botanical gardens, herbaria, and urban areas were removed [47]. For cleaning the coordinates from the urban areas, we used a land cover map from MODIS (https://search.earthdata.nasa.gov/search accessed on 9 April 2018) [48]. The data were also cleaned of duplicate records and places with inaccurate geo-references [49]. Finally, 37954 GBIF chestnut presence records were used for validation of the chestnut distribution in Europe (Figure S1 and Table S1).

2.3. Climatic Data

We used 1 km resolution present climate scenario raster layers of 19 bioclimatic variables from the CHELSA climatology data (CHELSA version 2.1, 30 s~1 km2, 1981–2010, downloaded from https://chelsa-climate.org, accessed on 3 February 2023) [32] and WorldClim climate data (World-Clim 2 version 2.1, 30 s~1 km2, 1970–2000, downloaded from http://www.worldclim.org/, accessed on 3 February 2023) [33] to compare the two different climate datasets (Table 1). The temperature variables of the CHELSA climatology data are multiplied by 10 (°C × 10) [32]. The raster files from the climatic sources were cropped to the extent of the study area using the packages “rgdal” and “Raster” in R and were prepared for modeling [50,51].

2.4. Chestnut Suitable Habitats Modeling

To examine the pairwise correlations between predictor variables and to identify a source of error, the correlation matrices were built for both CHELSA and WorldClim climate variables using the “Hmisc” package in R (Figures S2 and S3) [52]. We used the predictors based on Pearson’s correlation coefficient r < 0.9 [53]. Using the “gtools” package in R, 119 uncorrelated variable combinations were created using both climate datasets [54].
In addition to the uncorrelated variable combinations, we selected four additional groups of factors using CHELSA and WorldClim climate datasets (Table 2)—two variants using all of the variables and two others using the mechanistic predictor selection approach. In contrast with the correlative models, mechanistic species distribution models take into account how the environment affects the physiological performance of the species [55]. The survival of chestnut is connected to a combination of temperature and precipitation factors. For instance, chestnut trees are very sensitive to summer droughts and a lack of precipitation during the hot months is vital to their survival [9]. Simultaneously, several studies concur that annual mean precipitation, as well as winter temperatures and precipitation, impact chestnut distribution [2,56]. Therefore, we created variable groups for both climate datasets based on this information.
To model the suitable habitats of chestnuts using the current climate, we used MaxEnt 3.4.1. It is a widely used algorithm for species distribution and ecological niche modeling, especially when only presence data are available [57]. MaxEnt is well suited for presence-only data [58]. In practice, it is not always easy to find the true absence coordinates because, in addition to human activities, herbivores and competition between tree species significantly alter species distribution [59]. As a result, determining the threshold between true absence and absence data is difficult [60]. Therefore, we sampled 10,000 background points (i.e., pseudo-absence points) inside the extent of the Caucasus ecoregion boundaries to study the distribution of environmental conditions in the study region [57,61]. A total of 48 models based on CHELSA climate variables and 75 models based on WorldClim variables were created. The models were validated using 10-fold cross-validation and the mean area under the curve (AUC) of the receiver operating characteristic (ROC) curve [62]. The AUC (the area under the ROC curve) is believed to be one of the best options for estimating the appropriateness of a species environment, especially when background data are used [63].
Models were run with a maximum of 2500 iterations, quadratic and hinge features only, and default settings for convergence thresholds and regularization [57]. Using the testing AUC, we selected 2 models out of 123 models, one from the CHELSA climate data and the other with WorldClim data. In addition, for each dataset, one model with all variables and one using a mechanistic approach was also selected. Finally, six models were chosen for the final inspection (Table 2).

2.5. Comparison of the Climate Datasets

To determine the best habitat suitability model of chestnut for the Caucasus and Europe and to compare the performance of the WorldClim 2 and the CHELSA climate datasets, the presence coordinates of the GBIF (Global Biodiversity Information Facility) in Europe were used. Because of the large special bias of the GBIF data [64], the Habitat Electivity Indices (Jacob’s Modified Electivity) were also calculated [65] in order to eliminate the influence of the tendentious coordinates during validation.
The habitat electivity index (HEI) is a regularly used technique for assessing species habitat preferences. It is used to compute the likelihood of the existence of a given organism in a specific habitat type based on the occurrences in that environment. The habitat selection of numerous species, including mammals, reptiles, amphibians, and birds, is often assessed using HEI. The index compares the preferences of various species for distinct environments by measuring the ratio of observed relative to anticipated abundances of a species in a given habitat. Furthermore, by quantifying the degree of selectivity for that environment, the index may be used to estimate the appropriateness and the quality of a habitat for a certain species. The index (HEI) is generated by subtracting two values: a species preference for a certain habitat and the likelihood of finding that habitat type in the environment.
The HEI formula is given as follows:
(P1 − P2)/[(P1 + P2) − (2 × P1 × P2)]
where P1 represents the abundance of the species in a specific habitat type (proportion of habitat used, the proportion of chestnut occurrence points in the presence polygon) and P2 represents the availability of the habitat type in the environment (proportion of habitat available and proportion of the chestnut presence polygon). The numerator (P1 − P2) reflects how favorable the environment is for the species, the denominator (P1 + P2) represents the entire amount of habitat accessible in the environment, and (2 × P1 × P2) represents the degree of connection between the P1 and P2 values [65]. A higher degree of association would imply a reduced overall availability of habitat, resulting in a smaller denominator and a higher index score. If −1 < HEI < 0, then the site is unsuitable. If 0 < HEI < 1, then the location is valid. If HEI = 0, then the place is neither bad nor good [66].
Two predictions (one from CHELSA and one from Wordclim 2) were chosen out of the six based on how well they fit the GBIF presence data from throughout Europe (with the highest HEI). To choose the best predictive model, the resulting models were mapped and the validity of the explanation of the ecological variables was determined through a comparison of the response curves. In addition, the binary maps were intersected to highlight the geographic disparity between the two climatic datasets. The binary maps were created using Maximum training sensitivity plus specificity threshold in Maxent [67]. To explain the discrepancies between the models, the variables were also intersected through cell-by-cell subtraction of CHELSA minus WorldClim [68].

3. Results

3.1. Comparison of the Models

The AUCs of the 10-fold cross-validation for models using both climate datasets were greater than 0.95 and barely differed between the WorldClim and CHELSA models. Following validation with GBIF data, the majority of WorldClim models had a higher Habitat Electivity Index (HEI) than CHELSA. On the other hand, the mechanistic models for both datasets showed the highest HEI (more than 0.93) (Table 3).
The models with BIO10 (mean temperature of warmest quarter), BIO11 (mean temperature of coldest quarter), BIO12 (annual precipitation), BIO18 (precipitation of warmest quarter), and BIO19 (precipitation of coldest quarter) variables for both climate data showed the highest performance. Nevertheless, the distribution pattern of chestnut still highly differed between these two models and the HEI was not very different.
The cell-by-cell subtraction of CHELSA minus WorldClim indicates that the temperature and precipitation variable distributions were also different in these two models. For instance, the mean temperature of the warmest quarter (BIO 10) of CHELSA was nearly higher than the mean temperature of the warmest quarter of WorldClim everywhere. For CHELSA, the mean precipitation during the hottest quarter (BIO 18) was also greater in northern Europe and Hyrcanian forests, but lower in northern Africa. WorldClim in northern Europe had lower BIO 11 values than CHELSA, as well as in North Africa and the Hyrcanian region. The coldest quarter precipitation (BIO 19) was higher for WorldClim in northern Europe and North Africa, but lower in the Hyrcanian regions. As for BIO 12, CHELSA was higher than WorldClim in the Hyrcanian mountains. It was, however, lower in northern Europe and Africa (Figure S4).
The results show that the WorldClim model predicted a wither distribution probability in the northwest of the European continent. The Faroe Islands, the southeast coast of Iceland, and north Norway are among the suitable locations for chestnut, according to the WorldClim model. On the other hand, it did not show occurrence areas in the Hyrcanian forest of Azerbaijan and Iran, unlike CHELSA, which has a significant distribution in Hyrcanian woodlands, but a narrow distribution in northern Europe. Additionally, the CHELSA model did not illustrate any distribution across the continent of Africa, while the Wordclim model exhibited it (Figure S5).
Considering temperature-related factors (mean temperature of warmest quarter, mean temperature of coldest quarter) the WorldClim model was more difficult to explain. The BIO 11 curve of Wordclim did not display any distributional limits for high temperatures (higher than 10°) in contrast with CHELSA. Furthermore, the WorldClim BIO 10 curve did not indicate the presence of any lower distributional limits (Figure S6). As a result, we found a high suitability in locations where the forests could not exist (Figure S8). Therefore, for showing the final results, we selected the mechanistic model of CHELSA as it better revealed reliable suitable habitats for chestnuts in the Caucasus and Europe.

3.2. The Significance and Performance of the Variables

Jackknife of regularization training gain of the model shows the importance of variables. During the analyses, each variable was removed one at a time and a model was built using the variables that remained. Then, the test made a model with one variable at each stage. In addition, a model was also generated with all of the variables. The changes in the evaluation metric were recorded each time (AUC). Finally, we obtained the graph of individual environmental variable importance in the development of the MaxEnt model relative to all environmental variables (red bars), for each predictor variable alone (blue bars), and the drop in training gain when the variable was removed from the full model (green bars) [57]. According to the Jackknife test, the warmest quarter precipitation (BIO18), the mean temperature of the warmest quarter (BIO10), and the annual precipitation (BIO12) were most significant for the model of suitable habitats of chestnuts (Figure 3).
The minimal annual precipitation (BIO 12) was 500 mm for the chestnut habitats, according to the habitat suitability model. As the precipitation increased from 500 mm to 1000 mm and eventually reached the plateau, the suitability of habitats rapidly increased. The likelihood of an occurrence likewise increased when the warmest quarter’s precipitation (BIO 18) grew from 0 to 100. After that, it similarly became a plateau and then gradually declined. Simultaneously, the average temperature of the warmest quarter (BIO 10) for optimal suitability ranged from 15 to 30 degrees.
The extremely low winter temperature for the chestnut was −18 degrees Celsius, while the optimal mean temperature of the coldest quarter (BIO 11) ranged between −15 and 5 degrees. The chestnut optimal range of precipitation in the coldest quarter (BIO 19) was 100 mm to 300 mm, after which it reached a plateau. Meanwhile, the extreme minimum winter precipitation for chestnut was 30 mm (Figure 4).

3.3. Suitable Habitats Model in the Caucasus

According to the model, suitable habitats for chestnuts cover a significant part of the Caucasus land. In glaciers and high mountain locations, as well as semi-desert and desert areas of Georgia, Azerbaijan, Dagestan, Armenia, Türkiye, and Iran, the distribution model predicted a zero probability of chestnut occurrence. The mountains near the Lesser Caucasus’ Black Sea shoreline were predicted to have the highest occurrence rate (>93%), as well as in the northwest part of Georgia (Svaneti and Abkhazia) of the Greater Caucasus (>93%). We obtained an interesting image in central Georgia (in the Kartli and Mtskheta-Mtianeti areas), where the likelihood of occurrence was lower (50%) than in the eastern Caucasus (>78%).
The likelihood of finding suitable C. sativa habitats was very high (87%) in the northwest Caucasus (Krasnodar region, Adygea, Karachaevo-Cherkess and Kabardino-Balkaria), where the habitats span a large area. A probability of 67% was found for the central part of the North Greater Caucasus (Ingushetia, North Ossetia). The possible distribution also covered forests in the eastern Greater Caucasus and the ecoregion southeast (Hyrcanian woods in Iran and Azerbaijan) with a likelihood of 74%–90%. The model performed poorly in the area of Azerbaijan’s southwest and Armenia’s southeast, as well as in the northern section of Armenia and the southern half of Georgia (Figure 5).

3.4. Suitable Habitats Model in Europe

According to the model, the current potentially suitable areas of chestnuts were estimated to be 3,758,621 km2 in our study area. The suitable habitats of chestnut were found to be distributed in Southern Europe (Portugal, Spain, Greece, Italy, and Türkiye), Central and Eastern Europe (Albania, Bosnia and Herzegovina, Croatia, Czech Republic, Hungary, Poland, Serbia, Slovakia, Slovenia, North Macedonia, Bulgaria, Georgia, Lithuania, Moldova, Romania, Ukraine, Armenia, Azerbaijan, and the Russian Federation), Northern Europe (Lithuania, Latvia, Estonia, Norway, and Denmark), Western Europe (France, Liechtenstein, Monaco, Switzerland, Belgium, Germany, Luxembourg, Netherlands, Austria, Montenegro, San Marino, and the United Kingdom), and Western Asia (Türkiye). However, Greece had the highest probability (>93%).
Furthermore, potentially suitable areas were detected in Central Asia (Kyrgyzstan) and Southern Asia (Iran, Afghanistan, and Pakistan). On the other hand, Latvia, Estonia, Denmark, Liechtenstein, Monaco, Belgium, Luxembourg, the Netherlands, the Czech Republic, Hungary, the United Kingdom, Poland, San Marino, Slovakia, Norway, Lithuania, Moldova, Armenia, and Kyrgyzstan had a relatively low probability (<50%) of chestnut occurrence (Figure 6).

4. Discussion

The evaluation of the model performance using various climatic datasets has not frequently been addressed in ecological niche modeling so far [68]. In this work, climatic data from CHELSA and WorldClim are compared for the first time in the Caucasus to estimate the possible distribution of chestnuts.
The research demonstrated numerous interesting findings and approaches. (1) Evaluating and choosing the top climatic datasets first time for the Caucasus for developing a chestnut distribution model. (2) Developing a valid suitable habitat model on a local and global scale using data from the species natural distribution range. (3) Validation of the model in the global range using Jacob’s Modified Electivity Index [65], as the model evaluation using only AUC values based on presence-only GBIF data may show a contradictory result [64].
The special bias of the GBIF data can be substantially eliminated using habitat electivity indices (HEI), which are determined by the proportions of habitat used and the proportion of chestnut occurrence points in the presence polygon [65]. However, neither AUC nor other quality evaluation methods can always indicate model performances that are closely related to ecological-based knowledge [69].

4.1. Comparison of the Models

The mechanistic models with BIO 10 (mean temperature of the warmest quarter), BIO 11 (mean temperature of the coldest quarter), BIO 12 (annual precipitation), BIO 18 (precipitation of the warmest quarter), and BIO 19 (precipitation of the coldest quarter) variables showed the highest performance for both climate data. This demonstrates that mechanistic models are often more successful as the physiological characteristics of the species are taken into account [55].
Even though the best WorldClim and CHELSA models’ habitat electivity indices (HEI) computed for validation with GBIF data were not very different, the intersection of their binary maps demonstrates the geographic discrepancy between them. The WorldClim model predicts a wither distribution probability in northern Europe. It does not, however, depict occurrence areas in the Hyrcanian woodland (Azerbaijan and Iran), unlike CHELSA, which has a wide range in the Hyrcanian forests, but a small range in northern Europe. Furthermore, the CHELSA model does not show any suitability throughout the African continent, but the Wordclim model does (Figure S5). This should be due to differences in the data of the same variables in Wordclim and CHELSA [68]. According to the cell-by-cell subtraction, Chelsea shows a dry and hot summer in North Africa compared with WorldClim. On the other hand, Hyrkan is less hot and wetter in summer and has a milder winter. At the same time, according to WorldClim, northern Europe has more annual precipitation but lower summer temperatures than CHELSA. Therefore, it is relatively dry according to CHELSA (Figure S4).
In addition, the Jackknife graph for the CHELSA climate datasets shows that precipitation of the warmest quarter (BIO 18), mean temperature of the warmest quarter (BIO 10), and annual precipitation (BIO 12) contribute the most to chestnut distribution (Figure 4), in contrast with WorldClim where the temperature-related variables are ranked as the most important (Figure S7).
Based on the temperature-related factors (mean temperature of the warmest quarter and mean temperature of the coldest quarter), the WorldClim model is less reliable. Unlike CHELSA, the curve of the mean temperature of the coldest quarter of Wordclim does not depict any restrictions for distribution during high temperatures. In addition, the BIO 10 graph shows no lower limitations for the mean temperature of the warmest quarter for chestnut distribution (Figure S6). As a result, the WorldClim indicates highly suitable areas in western Britain, northern Norway, the Faroe Islands, and Iceland (Figure S8), which are historically mostly covered by grasses and where only a few tree species can survive [70].
As a consequence, we chose the CHELSEA mechanistic model as it indicates more reliable habitats for chestnuts in the Caucasus and Europe. This is most likely because CHELSA better explains the orographic effects, which improves habitat models, particularly in mountain locations [32]. Many researchers, however, continue to use WorldClim data for species distribution modeling in the Caucasus and other mountain regions [71], with various approaches, species, time situations, locations, and datasets [68]. As a result, before finalizing the models, data from various climatic sources should be checked in each scenario, as two different climate datasets allowed us to find the best possible habitat suitability model for chestnut.

4.2. The Significance and Performance of Variables

The model predictors adequately describe the habitat requirements of the species. Our research shows that the minimal rainfall required for chestnuts is 500 mm. This is in agreement with the study of Nakhutsrisvili and Conedera (minimum 500–600 mm of rainfall) [3,4]. The increase in the species habitat suitability with winter precipitation can be explained by snow coverage, which is also crucial for chestnuts, in addition to the temperature. This means that when the temperature is low, snow coverage is necessary for the survival of chestnut seedlings and saplings.
The fact that snow cover is also important for chestnuts, in addition to temperature, could explain why the habitat appropriateness of the species increases with winter precipitation. This means that snow cover is essential for chestnut seedlings and saplings in order to survive in cold climates [56].
Furthermore, the figures reveal that the mean temperature of the hottest and coldest quarters, as well as winter and summer precipitation, have an important effect on chestnut distribution. In general, we can tell that chestnut prefers moderate climates. According to our model, a mean temperature of more than 30 degrees in the hottest quarter is stressful for chestnuts, most likely because it is an anisohydric species [14].

4.3. Suitable Habitats Model

The other meaningful result of our study is that we developed a model of suitable habitats for chestnuts over all of the distribution ranges using data from the natural distribution areas of the species. Using this approach, we obtained a global model avoiding the spatial bias of the global data [72], as, from the Last Glacial Maximum, the forests in the Caucasus have not changed as dramatically as in Europe [2].
Even though the predictive factors contributed differently to the model, our findings revealed that the species response to environmental conditions are complicated, and it is impossible to describe habitats using single particular variables [72]. Therefore, we have to discuss the temperatures and precipitations of the season together. We can observe from the analysis that precipitation factors are more relevant. However, temperature significantly restricts its dispersal in both the north and south.
Suitable habitats for chestnuts cover significant areas of the Caucasus and Europe. In the Caucasus, they are mostly found in Georgia, Azerbaijan, Armenia, the Russian Federation (Adyghe, Kabardino-Balkaria, Ingushetia, and North Ossetia), Iran, and Türkiye. Historically, the majority of these locations were appropriate for chestnut cultivation. The genetic variety of the groups demonstrates that chestnut stands naturally occurred in these places [25], especially in the Hyrcanian highlands, which served as a south Caspian refugium [73]. Chestnuts are now extinct in central Armenia and southern Türkiye [25]. Our model suggests a low suitability in those areas.
The distribution model correctly predicted the occurrence of chestnuts as having zero probability in glaciers and high mountain areas, as well as in semi-desert and desert areas. Our model also provides a reliable picture of distribution in Europe, where the suitable habitats of chestnuts occur in Southern Europe, Central Europe, Eastern Europe, Northern Europe, Western Europe, and Türkiye. On the other hand, we found that the habitat suitability for chestnut is very low in Northern Europe. This confirms the hypothesis that chestnut was introduced there [21]. However, with the climate change scenarios, Great Britain and the northern parts of Europe are expected to become more suitable for chestnuts [8].
According to our model, suitable habitats for chestnuts were also detected in the Lesser Himalayas of Pakistan, Afghanistan, and Kyrgyzstan. Some studies confirm the occurrence of chestnuts in Pakistan. However, it is difficult to discuss whether it is natural or not in this part [74]. Our model also shows that there is very low habitat suitability for chestnuts in northern Africa. Even though they exist in the northern parts of the continent, they are cultivated [75]. One of the main factors, together with the climate for the distribution of the trees, is competition [76]. Therefore, with the highest probability, the whole suitable habitats of chestnut are not covered by the chestnut trees, because of the high density of competitors. Shade tolerance species as beech, spruce, and fir occupy large parts of the suitable habitats for chestnuts [2]. The other factor is fruit availability, as they need to be dispersed [76]. Due to these and other factors, the actual distribution area is comparably small.

5. Conclusions

Finally, we conclude that the suitable habitats model by CHELSA are less diffuse compared with WorldClim. It is in closer correspondence to the existing knowledge about chestnut habitat suitability. The model predictions match the actual existing distribution range of chestnut and correctly explain most of the natural global range of chestnuts [4].
The weakness of our model is that we did not insert the soil parameters and proximity to glacial refugia into the model, whose inclusion substantially improves the current species distribution models [35]. This is because sampling representativeness and the spatial resolution of the soil models are poor [77] and the human-facilitated spread of chestnut downplays the effect of proximity to the glacial refugia [7].
However, we were still able to model the suitable reliable habitats in the Caucasus and Europe thanks to representative samples of chestnut distribution from the part of the definite species natural range. In addition, testing climate datasets helped us to select the best climate datasets and improve our model. Therefore, we can say that the chestnut habitat suitability modeled from the data collected in the Caucasus is transferable and answers our research questions. This approach can be used to find suitable habitats for other chestnut species, as well as taxa of similar biogeographic histories.

Supplementary Materials

The following supporting information can be downloaded at: https://www.mdpi.com/article/10.3390/f14102076/s1, Figure S1: The map of the GBIF data distribution; Figure S2: The figure represents the correlation matrix of CHELSA climate data variables; Figure S3: The figure represents the correlation matrix of WorldClim climate data variables; Figure S4: The figure represents the cell-by-cell subtraction of CHELSA minus WorldClim, showing the differences between the temperature and precipitation variables of these two climate models; Figure S5: The figure shows the differences between the Chestnut distribution models made by CHELSA and WorldClim climate data; Figure S6: The response curves of the WorldClim climate variables; Figure S7: Contribution graph of the environmental variables in the Mechanistic model done by Worldclim Climate Data; Figure S8: Representation of the Maxent model for habitat suitability of chestnut in the species distribution range done by WorldClim 2 data; Table S1: GBIF_data.

Author Contributions

Conceptualization, V.M., H.K., A.G. and Z.J. (Zurab Javakhishvili); methodology, V.M., A.G. and H.K.; software, V.M. and A.G.; validation, V.M., A.G. and H.K.; formal analysis, V.M., A.G., I.A. and Z.J. (Zurab Janisahvili); investigation, V.M. and A.G.; resources, V.M., H.K., L.D. and Z.J. (Zurab Javakhishvili); data curation, V.M., Z.N., I.A., L.D. and Z.J. (Zurab Janisahvili); writing—original draft preparation, V.M.; writing—review and editing, V.M., A.G., H.K., Z.N., Z.J. (Zurab Javakhishvili) and I.A.; visualization, V.M., H.K., I.A. and Z.J. (Zurab Janisahvili); supervision, A.G. and H.K.; project administration, A.G. and H.K.; funding acquisition, A.G. and H.K. All authors have read and agreed to the published version of the manuscript.

Funding

This research was funded by Volkswagen Stiftung in the framework of the joint project “Structured Education Quality Assurance Freedom to Think” and Shota Rustaveli National Science Foundation of Georgia. V.M. was supported by Carl Friedrich Lehman-Haupt International Doctoral Program.

Data Availability Statement

The data presented in this study are available on request from the corresponding author.

Acknowledgments

We are grateful to Marco Conedera, Frank Bohlander, Lars Drossler, Nikoloz Lachashvili, Joachim Puhe and Tamar Beridze for their suggestions and recommendations. Furthermore, we are grateful to Berika Beridze, the Ministry of Environmental Protection and Agriculture of Georgia, the Institute of Botany of Ilia State University, The United Nations Development Program (UNDP), and Caucasus Nature Fund (CNF) for providing the field data.

Conflicts of Interest

The authors declare no conflict of interest.

References

  1. Konstantinidis, P.; Tsiourlis, G.; Xofis, P.; Buckley, G.P. Taxonomy and Ecology of Castanea sativa Mill. Forests in Greece. Plant Ecol. 2008, 195, 235–256. [Google Scholar] [CrossRef]
  2. Dolukhanov, A. Forest Vegetation of Georgia; Universal: Tbilisi, GA, USA, 2010. [Google Scholar]
  3. Nakhutsrishvili, G. The Vegetation of Georgia (South Caucasus); Springer: Berlin/Heidelberg, Germany, 2013; ISBN 978-3-642-29914-8. [Google Scholar]
  4. Conedera, M.; Tinner, W.; Krebs, P.; de Rigo, D.; Caudullo, G. Castanea sativa in Europe: Distribution, Habitat, Usage and Threats; European Atlas of Forest Tree Species. EU, Luxembourg. 2016, pp. 78–79. Available online: chrome-extension://efaidnbmnnnibpcajpcglclefindmkaj/https://ies-ows.jrc.ec.europa.eu/efdac/download/Atlas/pdf/Castanea_sativa.pdf (accessed on 2 June 2023).
  5. Mattioni, C.; Cherubini, M.; Micheli, E.; Villani, F.; Bucci, G. Role of Domestication in Shaping Castanea sativa Genetic Variation in Europe. Tree Genet. Genomes 2008, 4, 563–574. [Google Scholar] [CrossRef]
  6. Kvavadze, E. Identification of Anthropological Landscapes and Human Activity in Georgia in Correlation with Holocene Black Sea Level Fluctuations. Earth 2015, 4, 120. [Google Scholar] [CrossRef]
  7. Krebs, P.; Conedera, M.; Pradella, M.; Torriani, D.; Felber, M.; Tinner, W. Quaternary Refugia of the Sweet Chestnut (Castanea sativa Mill.): An Extended Palynological Approach. Veget. Hist. Archaeobot. 2004, 13, 145–160. [Google Scholar] [CrossRef]
  8. Broadmeadow, M.S.J.; Ray, D.; Samuel, C.J.A. Climate Change and the Future for Broadleaved Tree Species in Britain. For. Int. J. For. Res. 2005, 78, 145–161. [Google Scholar] [CrossRef]
  9. Conedera, M.; Krebs, P.; Gehring, E.; Wunder, J.; Hülsmann, L.; Abegg, M.; Maringer, J. How Future-Proof Is Sweet Chestnut (Castanea sativa) in a Global Change Context. For. Ecoly Manag. 2021, 494, 119320. [Google Scholar] [CrossRef]
  10. Gulisashvili, V. Chestnut in the Caucasus; Nature: Tbilisi, Georgia, 1967. [Google Scholar]
  11. Tugushi, L. Chestnut forests of Abkhazia and ways to improve them, Forest Institute of Tbilisi. Ph.D. Thesis, Ochamchire, 1965; pp. 33–37. [Google Scholar]
  12. Velizarova, E. Physico-Chemical and Morphological Properties of Soils in Chestnut (Castanea sativa Mill.) Habitats of Belasitsa Mountain. Silva Balc. 2015, 16, 60–70. [Google Scholar]
  13. Badenes, M.L.; Byrne, D.H. (Eds.) Fruit Breeding; Springer US: Boston, MA, USA, 2012; ISBN 978-1-4419-0762-2. [Google Scholar]
  14. Gomes-Laranjo, J.; Dinis, L.-T.; Martins, L.; Portela, E.; Pinto, T.; Ciordia, M.; Feito, I.; Majada, J.; Peixoto, F.; Pereira, S.; et al. Characterization of Chestnut Behavior with Photosynthetic Traits. In Applied Photosynthesis; Najafpour, M., Ed.; InTech: London, UK, 2012; ISBN 978-953-51-0061-4. [Google Scholar]
  15. Vettraino, A.M.; Morel, O.; Perlerou, C.; Robin, C.; Diamandis, S.; Vannini, A. Occurrence and Distribution of Phytophthora Species in European Chestnut Stands, and Their Association with Ink Disease and Crown Decline. Eur. J. Plant Pathol. 2005, 111, 169–180. [Google Scholar] [CrossRef]
  16. Beridze, B.; Dering, M. Problems and Threats to the Caucasus Forest Ecosystems on the Example of Castanea sativa. Kosmos 2021, 70, 19–26. [Google Scholar] [CrossRef]
  17. Heiniger, U.; Rigling, D. Biological Control of Chestnut Blight in Europe. Annu. Rev. Phytopathol. 1994, 32, 581–599. [Google Scholar] [CrossRef]
  18. Anagnostakis, S.L. American Chestnut Sprout Survival with Biological Control of the Chestnut-Blight Fungus Population. For. Ecol. Manag. 2001, 152, 225–233. [Google Scholar] [CrossRef]
  19. Venanzi, R.; Picchio, R.; Piovesan, G. Silvicultural and Logging Impact on Soil Characteristics in Chestnut (Castanea sativa Mill.) Mediterranean Coppice. Ecol. Eng. 2016, 92, 82–89. [Google Scholar] [CrossRef]
  20. Newbigin, M.I. Man and the Forest in Europe: The Pre-Industrial Period. Emp. For. J. 1928, 7, 209–224. [Google Scholar]
  21. Conedera, M.; Manetti, M.C.; Giudici, F.; Amorini, E. Distribution and economic potential of the Sweet chestnut (Castanea sativa Mill.) in Europe. Ecmed 2004, 30, 179–193. [Google Scholar] [CrossRef]
  22. Akhalkatsi, M. Forest Habitat Restoratio in Georgia, Caucaus Ecoregion; Mtsigobari: Tbilisi, GA, USA, 2015; ISBN 978-9941-450-68-6. [Google Scholar]
  23. Quinn, J. Gatekhili Mountains, Gatekhili State: Fractured Alpine Forest Governance and Post-Soviet Development in the Republic of Georgia. Rga 2017, 105, 1–14. [Google Scholar] [CrossRef]
  24. Prospero, S.; Lutz, A.; Tavadze, B.; Supatashvili, A.; Rigling, D. Discovery of a New Gene Pool and a High Genetic Diversity of the Chestnut Blight Fungus Cryphonectria Parasitica in Caucasian Georgia. Infect. Genet. Evol. 2013, 20, 131–139. [Google Scholar] [CrossRef]
  25. Beridze, B.; Sękiewicz, K.; Walas, Ł.; Thomas, P.A.; Danelia, I.; Fazaliyev, V.; Kvartskhava, G.; Sós, J.; Dering, M. Biodiversity Protection against Anthropogenic Climate Change: Conservation Prioritization of Castanea sativa in the South Caucasus Based on Genetic and Ecological Metrics. Ecol. Evol. 2023, 13, e10068. [Google Scholar] [CrossRef] [PubMed]
  26. Franklin, J.; Miller, J.A. Mapping Species Distributions: Spatial Inference and Prediction; Ecology, biodiversity and conservation; Cambridge University Press: Cambridge, UK; New York, NY, USA, 2009; ISBN 978-0-521-87635-3. [Google Scholar]
  27. Araújo, M.B.; Guisan, A. Five (or so) Challenges for Species Distribution Modelling. J. Biogeogr. 2006, 33, 1677–1688. [Google Scholar] [CrossRef]
  28. Luoto, M.; Virkkala, R.; Heikkinen, R.K. The Role of Land Cover in Bioclimatic Models Depends on Spatial Resolution. Glob. Ecol. Biogeogr. 2006, 16, 34–42. [Google Scholar] [CrossRef]
  29. Soria-Auza, R.W.; Kessler, M.; Bach, K.; Barajas-Barbosa, P.M.; Lehnert, M.; Herzog, S.K.; Böhner, J. Impact of the Quality of Climate Models for Modelling Species Occurrences in Countries with Poor Climatic Documentation: A Case Study from Bolivia. Ecol. Model. 2010, 221, 1221–1229. [Google Scholar] [CrossRef]
  30. Bakuradze, G. Assessment of Above-Ground Carbon Stocks in the Forest Ecosystem of Georgia; Ilia State University: Tbilisi, GA, USA, 2023. [Google Scholar]
  31. Kurdadze, T. Influence of Competition on Beech Growth in Mtskheta-Mtianeti Region Based on National Forest Inventory Data; Ilia State University: Tbilisi, GA, USA, 2020. [Google Scholar]
  32. Karger, D.N.; Conrad, O.; Böhner, J.; Kawohl, T.; Kreft, H.; Soria-Auza, R.W.; Zimmermann, N.E.; Linder, H.P.; Kessler, M. Climatologies at High Resolution for the Earth’s Land Surface Areas. Sci. Data 2017, 4, 170122. [Google Scholar] [CrossRef]
  33. Fick, S.E.; Hijmans, R.J. WorldClim 2: New 1-km Spatial Resolution Climate Surfaces for Global Land Areas. Int. J. Climatol. 2017, 37, 4302–4315. [Google Scholar] [CrossRef]
  34. Thuiller, W.; Richardson, D.M.; Pyšek, P.; Midgley, G.F.; Hughes, G.O.; Rouget, M. Niche-based Modelling as a Tool for Predicting the Risk of Alien Plant Invasions at a Global Scale. Glob. Chang. Biol. 2005, 11, 2234–2250. [Google Scholar] [CrossRef] [PubMed]
  35. Tarkhnishvili, D.; Gavashelishvili, A.; Mumladze, L. Palaeoclimatic Models Help to Understand Current Distribution of Caucasian Forest Species: Modeling West Asian Forest Refugia. Biol. J. Linn. Soc. 2012, 105, 231–248. [Google Scholar] [CrossRef]
  36. Tielidze, L.G.; Wheate, R.D. The Greater Caucasus Glacier Inventory (Russia, Georgia and Azerbaijan). Cryosphere 2018, 12, 81–94. [Google Scholar] [CrossRef]
  37. Mittermeier, R.A.; Turner, W.R.; Larsen, F.W.; Brooks, T.M.; Gascon, C. Global Biodiversity Conservation: The Critical Role of Hotspots. In Biodiversity Hotspots; Zachos, F.E., Habel, J.C., Eds.; Springer: Berlin/Heidelberg, Germany, 2011; pp. 3–22. ISBN 978-3-642-20991-8. [Google Scholar]
  38. Zazanashvili, N.; Mallon, D. Status and Protection of Globally Threatened Species in the Caucasus; CEPF, WWF: Tbilisi, GA, USA, 2009; Volume 232, ISBN 978-9941-0-2203-6. [Google Scholar]
  39. Akhalkatsi, M. Pine Forest on Tree-Line Ecotone in the Mountain Kazbegi in the Georgia (South Caucasus). Agri. Res. Technol. Open Access J. 2019, 21, 556149. [Google Scholar] [CrossRef]
  40. Tarkhnishvili, D.N. Historical Biogeography of the Caucasus; Wildlife protection, destruction and extinction; Nova Publishers: New York, NY, USA, 2014; ISBN 978-1-63321-910-6. [Google Scholar]
  41. Aoyama, H. A Study of the Stratified Random Sampling. Ann. Inst. Stat. Math. 1954, 6, 1–36. [Google Scholar] [CrossRef]
  42. Hayek, L.-A.C.; Buzas, M.A. Surveying Natural Populations: Quantitative Tools for Assessing Biodiversity; Columbia University Press: New York, NY, USA, 2010; ISBN 978-0-231-14620-3. [Google Scholar]
  43. Iachan, R. Systematic Sampling: A Critical Review. Int. Stat. Rev. Rev. Int. Stat. 1982, 50, 293. [Google Scholar] [CrossRef]
  44. Dormann, C.F.; McPherson, J.M.; Araújo, M.B.; Bivand, R.; Bolliger, J.; Carl, G.; Davies, R.G.; Hirzel, A.; Jetz, W.; Daniel Kissling, W.; et al. Methods to Account for Spatial Autocorrelation in the Analysis of Species Distributional Data: A Review. Ecography 2007, 30, 609–628. [Google Scholar] [CrossRef]
  45. Veloz, S.D. Spatially Autocorrelated Sampling Falsely Inflates Measures of Accuracy for Presence-Only Niche Models. J. Biogeogr. 2009, 36, 2290–2299. [Google Scholar] [CrossRef]
  46. Occurrence data of Castanea sativa, Global Biodiversity Information Facility (GBIF), 2023. Available online: https://doi.org/10.15468/dl.k6d6hn (accessed on 25 February 2023).
  47. Zizka, A.; Silvestro, D.; Andermann, T.; Azevedo, J.; Duarte Ritter, C.; Edler, D.; Farooq, H.; Herdean, A.; Ariza, M.; Scharn, R.; et al. CoordinateCleaner: Standardized Cleaning of Occurrence Records from Biological Collection Databases. Methods Ecol. Evol. 2019, 10, 744–751. [Google Scholar] [CrossRef]
  48. Friedl, M.; Sulla-Menashe, D. MODIS/Terra+Aqua Land Cover Type Yearly L3 Global 500m SIN Grid MCD12Q1. NASA EOSDIS Land Processes Distributed Active Archive Center, South Dakota, USA, 2019. Available online: https://search.earthdata.nasa.gov/search (accessed on 20 May 2023).
  49. Maldonado, C.; Molina, C.I.; Zizka, A.; Persson, C.; Taylor, C.M.; Albán, J.; Chilquillo, E.; Rønsted, N.; Antonelli, A. Estimating Species Diversity and Distribution in the Era of B Ig D Ata: To What Extent Can We Trust Public Databases? Glob. Ecol. Biogeogr. 2015, 24, 973–984. [Google Scholar] [CrossRef] [PubMed]
  50. Bivand, R.; Keitt, T.; Rowlingson, B.; Pebesma, E.; Sumner, M.; Hijmans, R.; Rouault, E.; Maintainer, R.B. Package ‘rgdal’. Bindings for the Geospatial Data Abstraction Library. 2022, p. 172. Available online: https://cran.r-project.org/web/packages/rgdal/index.html (accessed on 15 May 2022).
  51. Hijmans, R. Raster package in R, The Comprehensive R Archive Network, 2023. p. 172. Available online: https://rspatial.org/raster (accessed on 8 February 2022).
  52. Frank, E.H., Jr. Package ‘Hmisc’, The Comprehensive R Archive Network, 2023. Available online: https://hbiostat.org/R/Hmisc/ (accessed on 10 May 2022).
  53. Schober, P.; Boer, C.; Schwarte, L.A. Correlation Coefficients: Appropriate Use and Interpretation. Anesth. Analg. 2018, 126, 1763–1768. [Google Scholar] [CrossRef]
  54. Warnes, G.R.; Bolker, B.; Lumley, T.; Warnes, M.G.R. Package‘Gtools’, The Comprehensive R Archive Network, R Package version, 3(1), 2021. Available online: https://github.com/r-gregmisc/gtools (accessed on 5 April 2022).
  55. Evans, T.G.; Diamond, S.E.; Kelly, M.W. Mechanistic Species Distribution Modelling as a Link between Physiology and Conservation. Conserv. Physiol. 2015, 3, cov056. [Google Scholar] [CrossRef] [PubMed]
  56. Gurney, K.M.; Schaberg, P.G.; Hawley, G.J.; Shane, J.B. Inadequate Cold Tolerance as a Possible Limitation to American Chestnut Restoration in the Northeastern United States. Restor. Ecol. 2011, 19, 55–63. [Google Scholar] [CrossRef]
  57. Phillips, S.J.; Dudík, M. Modeling of Species Distributions with Maxent: New Extensions and a Comprehensive Evaluation. Ecography 2008, 31, 161–175. [Google Scholar] [CrossRef]
  58. Elith, J.; Graham, C.H.; Anderson, R.P.; Dudík, M.; Ferrier, S.; Guisan, A.; Hijmans, R.J.; Huettmann, F.; Leathwick, J.R.; Lehmann, A.; et al. Novel Methods Improve Prediction of Species’ Distributions from Occurrence Data. Ecography 2006, 29, 129–151. [Google Scholar] [CrossRef]
  59. Vera, F.W.M.; Bakker, E.S.; Olff, H. Large Herbivores: Missing Partners of Western European Light-demanding Tree and Shrub Species? In Large Herbivore Ecology, Ecosystem Dynamics and Conservation; Danell, K., Bergström, R., Duncan, P., Pastor, J., Eds.; Cambridge University Press: Cambridge, UK, 2006; pp. 203–231. ISBN 978-0-521-83005-8. [Google Scholar]
  60. Meier, E.S.; Kienast, F.; Pearman, P.B.; Svenning, J.-C.; Thuiller, W.; Araújo, M.B.; Guisan, A.; Zimmermann, N.E. Biotic and Abiotic Variables Show Little Redundancy in Explaining Tree Species Distributions. Ecography 2010, 33, 1038–1048. [Google Scholar] [CrossRef]
  61. VanDerWal, J.; Shoo, L.P.; Graham, C.; Williams, S.E. Selecting Pseudo-Absence Data for Presence-Only Distribution Modeling: How Far Should You Stray from What You Know? Ecol. Model. 2009, 220, 589–594. [Google Scholar] [CrossRef]
  62. Bleyhl, B.; Arakelyan, M.; Askerov, E.; Bluhm, H.; Gavashelishvili, A.; Ghasabian, M.; Ghoddousi, A.; Heidelberg, A.; Khorozyan, I.; Malkhasyan, A.; et al. Assessing Niche Overlap between Domestic and Threatened Wild Sheep to Identify Conservation Priority Areas. Divers. Distrib. 2019, 25, 129–141. [Google Scholar] [CrossRef]
  63. Merow, C.; Smith, M.J.; Silander, J.A. A Practical Guide to MaxEnt for Modeling Species’ Distributions: What It Does, and Why Inputs and Settings Matter. Ecography 2013, 36, 1058–1069. [Google Scholar] [CrossRef]
  64. Beck, J.; Böller, M.; Erhardt, A.; Schwanghart, W. Spatial Bias in the GBIF Database and Its Effect on Modeling Species’ Geographic Distributions. Ecol. Inform. 2014, 19, 10–15. [Google Scholar] [CrossRef]
  65. Jacobs, J. Quantitative Measurement of Food Selection: A Modification of the Forage Ratio and Ivlev’s Electivity Index. Oecologia 1974, 14, 413–417. [Google Scholar] [CrossRef] [PubMed]
  66. Lechowicz, M.J. The Sampling Characteristics of Electivity Indices. Oecologia 1982, 52, 22–30. [Google Scholar] [CrossRef]
  67. Hu, J.; Jiang, Z. Predicting the Potential Distribution of the Endangered Przewalski’s Gazelle. J. Zool. 2010, 282, 54–63. [Google Scholar] [CrossRef]
  68. Bobrowski, M.; Schickhoff, U. Why Input Matters: Selection of Climate Data Sets for Modelling the Potential Distribution of a Treeline Species in the Himalayan Region. Ecol. Model. 2017, 359, 92–102. [Google Scholar] [CrossRef]
  69. Hijmans, R.J. Cross-Validation of Species Distribution Models: Removing Spatial Sorting Bias and Calibration with a Null Model. Ecology 2012, 93, 679–688. [Google Scholar] [CrossRef]
  70. Johansen, J. Pollen Diagrams From the Shetland and Faroe Islands. New Phytol. 1975, 75, 369–387. [Google Scholar] [CrossRef]
  71. Akobia, I.; Janiashvili, Z.; Metreveli, V.; Zazanashvili, N.; Batsatsashvili, K.; Ugrekhelidze, K. Modelling the Potential Distribution of Subalpine Birches (Betula Spp.) in the Caucasus. Community Ecol. 2022, 23, 209–218. [Google Scholar] [CrossRef]
  72. Bowler, D.E.; Callaghan, C.T.; Bhandari, N.; Henle, K.; Benjamin Barth, M.; Koppitz, C.; Klenke, R.; Winter, M.; Jansen, F.; Bruelheide, H.; et al. Temporal Trends in the Spatial Bias of Species Occurrence Records. Ecography 2022, 2022, e06219. [Google Scholar] [CrossRef]
  73. Gavashelishvili, A.; Tarkhnishvili, D. Biomes and Human Distribution during the Last Ice Age: Biomes and Humans during the Ice Age. Glob. Ecol. Biogeogr. 2016, 25, 563–574. [Google Scholar] [CrossRef]
  74. Rahman, I.U.; Afzal, A.; Iqbal, Z.; Abd_Allah, E.F.; Alqarawi, A.A.; Calixto, E.S.; Ali, N.; Ijaz, F.; Kausar, R.; Alsubeie, M.S.; et al. Role of Multivariate Approaches in Floristic Diversity of Manoor Valley (Himalayan Region), Pakistan. Appl. Ecol. Env. Res. 2019, 17, 1475–1498. [Google Scholar] [CrossRef]
  75. Urbisz, A.; Urbisz, A. European Chestnut (Castanea sativa Mill.)-a Tree Naturalized on the Baltic Sea Coast? Pol. J. Ecol. 2007, 55, 175–179. [Google Scholar]
  76. Howe, H.F.; Estabrook, G.F. On Intraspecific Competition for Avian Dispersers in Tropical Trees. Am. Nat. 1977, 111, 817–832. [Google Scholar] [CrossRef]
  77. Hengl, T.; De Jesus, J.M.; MacMillan, R.A.; Batjes, N.H.; Heuvelink, G.B.M.; Ribeiro, E.; Samuel-Rosa, A.; Kempen, B.; Leenaars, J.G.B.; Walsh, M.G.; et al. SoilGrids1km—Global Soil Information Based on Automated Mapping. PLoS ONE 2014, 9, e105992. [Google Scholar] [CrossRef]
Forests 14 02076 g001
Figure 1. Map of the study area. The green color represents the forest cover from the Moderate Resolution Imaging Spectroradiometer (MODIS).
Figure 1. Map of the study area. The green color represents the forest cover from the Moderate Resolution Imaging Spectroradiometer (MODIS).
Forests 14 02076 g001
Forests 14 02076 g002
Figure 2. The map of the Caucasus ecoregion. The green color represents the forest cover of the Moderate Resolution Imaging Spectroradiometer (MODIS). The black points are the presence locations of C. sativa (620 locations).
Figure 2. The map of the Caucasus ecoregion. The green color represents the forest cover of the Moderate Resolution Imaging Spectroradiometer (MODIS). The black points are the presence locations of C. sativa (620 locations).
Forests 14 02076 g002
Forests 14 02076 g003
Figure 3. Importance of the environmental variables in the final (second) model.
Figure 3. Importance of the environmental variables in the final (second) model.
Forests 14 02076 g003
Forests 14 02076 g004
Figure 4. The response curves of the CHELSA climate variables affected by Maxent predictions.
Figure 4. The response curves of the CHELSA climate variables affected by Maxent predictions.
Forests 14 02076 g004
Forests 14 02076 g005
Figure 5. Representation of the Maxent model for suitable habitats for chestnuts in the Caucasus ecoregion. Darker colors show areas with more habitat suitability.
Figure 5. Representation of the Maxent model for suitable habitats for chestnuts in the Caucasus ecoregion. Darker colors show areas with more habitat suitability.
Forests 14 02076 g005
Forests 14 02076 g006
Figure 6. Representation of the Maxent model for habitat suitability of chestnut in the species distribution range. Darker colors show areas with more habitat suitability.
Figure 6. Representation of the Maxent model for habitat suitability of chestnut in the species distribution range. Darker colors show areas with more habitat suitability.
Forests 14 02076 g006
Table 1. Nineteen climate variables were used in the chestnut distribution model.
Table 1. Nineteen climate variables were used in the chestnut distribution model.
Coded Bioclimatic VariablesBioclimatic Variables
BIO1Annual mean temperature
BIO2Mean diurnal range (mean of monthly (maximum temperature − inimum temperature))
BIO3Isothermally (BIO2/BIO7) (×100)
BIO4Temperature seasonality (standard deviation ×100)
BIO5Maximum temperature of warmest month
BIO6Minimum temperature of coldest month
BIO7Temperature annual range (BIO5–BIO6)
BIO8Mean temperature of wettest quarter
BIO9Mean temperature of driest quarter
BIO10Mean temperature of warmest quarter
BIO11Mean temperature of coldest quarter
BIO12Annual precipitation
BIO13Precipitation of wettest month
BIO14Precipitation of driest month
BIO15Precipitation seasonality (coefficient of variation)
BIO16Precipitation of wettest quarter
BIO17Precipitation of driest quarter
BIO18Precipitation of warmest quarter
BIO19Precipitation of coldest quarter
Table 2. The models for the final inspection.
Table 2. The models for the final inspection.
IDClimate DataModelsVariables
1CHELSA V2.1The model with all variablesAll 19 BIO
2CHELSA V2.1Mechanistic modelBIO_19, BIO_18, BIO_12, BIO_11, BIO_10
3CHELSA V2.1The correlative model with the best AUC performancesBIO_19, BIO_18, BIO_16, BIO_15, BIO_11, BIO_10, BIO_09, BIO_08, BIO_07, BIO_06, BIO_04, BIO_03, BIO_02, BIO_01
4Wordclim 2 V2.1The model with all variablesAll 19 BIO
5Wordclim 2 V2.1Mechanistic modelBIO_ 19, BIO_18, BIO_12, BIO_11, BIO_10
6Wordclim 2 V2.1The correlative model with the best AUC performances BIO_19, BIO_18, BIO_17, BIO_15, BIO_13, BIO_09, BIO_08, BIO_06, BIO_04, BIO_03, BIO_02
Table 3. HEI and AUC indices of the models.
Table 3. HEI and AUC indices of the models.
IDClimate DataModelsAUCHEI
1CHELSA V2.1The model with all variables0.9634−0.616084571
2CHELSA V2.1Mechanistic model0.95350.932336488
3CHELSA V2.1The correlative model with the best AUC performances0.9625−0.382977772
4Wordclim 2 V2.1The model with all variables0.96250.956727365
5Wordclim 2 V2.1Mechanistic model0.95590.969826464
6Wordclim 2 V2.1The correlative model with the best AUC performances 0.96240.92322457
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Share and Cite

MDPI and ACS Style

Metreveli, V.; Kreft, H.; Akobia, I.; Janiashvili, Z.; Nonashvili, Z.; Dzadzamia, L.; Javakhishvili, Z.; Gavashelishvili, A. Potential Distribution and Suitable Habitat for Chestnut (Castanea sativa). Forests 2023, 14, 2076. https://doi.org/10.3390/f14102076

AMA Style

Metreveli V, Kreft H, Akobia I, Janiashvili Z, Nonashvili Z, Dzadzamia L, Javakhishvili Z, Gavashelishvili A. Potential Distribution and Suitable Habitat for Chestnut (Castanea sativa). Forests. 2023; 14(10):2076. https://doi.org/10.3390/f14102076

Chicago/Turabian Style

Metreveli, Vasil, Holger Kreft, Ilia Akobia, Zurab Janiashvili, Zaza Nonashvili, Lasha Dzadzamia, Zurab Javakhishvili, and Alexander Gavashelishvili. 2023. "Potential Distribution and Suitable Habitat for Chestnut (Castanea sativa)" Forests 14, no. 10: 2076. https://doi.org/10.3390/f14102076

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Back to TopTop