Predicting environmental variables using vegetation composition

In this study, the efficiency of the two nearest neighbor (KNN) and weighted average (WA) methods was evaluated for indirect estimation of environmental variables in plant communities. For this purpose, vegetation composition data of 324 relevés with an area 400 m2 of the Hyrcanian yew forests database were used. Then, environmental variables in each relevés were indirectly estimated by using KNN and WA methods based on two kind of vegetation data (incidence based and abundance based of floristic data) as well as the initial values of that environmental variables.Validation of the models were evaluated using determinant coefficient of linear regression analysis, which done based on the initial values and followed by estimated one of each environmental variables as the predictor and response variables. Results showed that using KNN method based on abundance data due to having the highest determination coefficient value has the priority in comparison to another three algorithms. The main reason of the differences between KNN and WA was influenced by different approaches of interpolation (KNN) and extrapolation (WA) in the process of environmental variables point estimation. The better performance of the KNN compared with WA in the point estimating of environmental variables is due to using the environmental data of the only adjacent plot data with the most similarly floristically features to each points in the KNN, while the results of the WA are globally affected by the range of each environmental variables at the whole of the dataset.