Predictive modeling for the distribution of plant communities of the order Quercetalia pubescenti-petraeae Klika 1933

Abstract

© 2019 Tomsk State University. All rights reserved. The Braun-Blanquet method has been widely used by Russian researchers during recent years for the purposes of vegetation classification. In this context, some regions, such as the Republic of Bashkortostan, have been investigated in detail and their complete prodromes are available. However, for plant communities of other territories, including the Republic of Tatarstan, this work has just begun. The following two major questions are commonly faced by researchers when studying the syntaxonomic diversity of the territory: "Which of the already identified syntaxa are present in the study area?" and "Where can they be found?". In this paper, we showed and discussed one of the possible ways to answer the above questions. Our research focused on xeromesophytic oak forests of the order Quercetalia pubescenti-petraeae Klika 1933 that belong to the class Quercetea pubescentis Doing-Kraft ex Scamoni et Passarge 1959 and occupy an area within the forest-steppe zone of Eastern Europe (40-60°N; 15-60°E). The obtained results add considerably to what is already known about the plant communities in the territory of the Republic of Tatarstan (54-57°N; 47-55°E). We based our model for potential distribution of xeromesophytic oak forests on the following strategy: "first collect, then predict". According to this strategy, field data are first classified, arranged, or aggregated to further create objects or attributes at the community level. The initial data for model construction were the locations provided in the relevés for plant communities assigned to the alliances of Aceri tatarici-Quercion Zólyomi 1957 and Lathyro pisiformis-Quercion roboris Solomeshch et Grigoriev in Willner et al. 2016 within the territories of Slovakia, Romania, Ukraine, and Russia (Kursk region, Bryansk region, Belgorod region, Republic of Tatarstan, Republic of Bashkortostan, and Orenburg region). A total of 227 relevés were analyzed (Fig. 1). The data of the WorldClim 2.0 and SoilGrids models were used as the abiotic predictors. To build a model of the easternmost part of the study area (Tatarstan and Bashkortostan), we employed the data on the forest cover from the Global Forest Change project. The spatial data were compiled with the help of the QGIS 2.14 package. The Maxent method was applied as a modeling algorithm. The main probability models using the climate and soil data as the predictors were combined in different ways, calibrated using 75% of the data, and assessed by area under the ROC curve (AUC) using the remaining 25%. All models using the climate and soil model data as the predictors yielded highly accurate results-AUC=0.861-0.817. On the contrary, when the data on forest presence were used as the predictor, the accuracy of the models turned out to be significantly lower, up to 0.582. Furthermore, at some threshold values the models based on the forest cover data behaved worse than a random choice. At the stage of visualization, all models demonstrated that oak communities belonging to the order Quercetalia pubescentipetraeae are most likely to be found in such areas of Russia as Bashkiria, Tatarstan, Chuvashia, Samara region, Ulyanovsk region, Belgorod region, Bryansk region, Kursk region, Belgorod region, Stavropol region (in the foothills of the Caucasus), as well as in Ukraine, Romania, and Moldova, which is in good agreement with the published data (Fig. 2). The models with the best AUC values were used to assess the suitable habitats of Quercetalia pubescenti-petraeae communities in Tatarstan (Fig. 3). We carried out 85 relevés at the localities where the probability of presence of the modeled communities in the territory of Tatarstan was the highest. The preliminary presence of the localities with the high probability of finding communities of a certain type allowed to optimize the procedure for detection of these phytocenoses and to investigate an area about 30 000 km2 during a single field season. Further phytocoenotic processing of the obtained data confirmed that these plant communities belong to the order Quercetalia pubescenti-petraeae, the alliance Lathyro pisiformis-Quercion roboris. The analysis of the statistical characteristics of the obtained results revealed that the model using the data on climatic characteristics (WorldClim) as the predictors shows better results than the one using the soil data (SoilGrids) as the predictors (See Table 1). At the same time, combining these two models (WorldClim and SoilGrids) as the predictors showed better results than the model using only WorldClim. The very form of combination is unimportant, because no statistically significant differences were observed between the models (See Table 2). Therefore, it was concluded that using the mechanism for modeling communities of a certain syntaxonomic rank enables us to either register this type of communities within the borders of a particular area or to suggest with high probability that they do not occur here. A similar approach can be helpful when studying vast and poorly known, in terms of syntaxonomic composition, territories for optimization of field research. In the case of xeromesophytic oak forests, the models based on the combination of climate and soil data as the predictors proved to be the most appropriate ones.