An important step in vegetation distribution modeling is selection of an appropriate formulation for the statistical model. The formulation should consider both the shapes of vegetation response curves along environmental gradients, and adaptability of the selected model to considering new environmental drivers in the modeling. Formulations in commonly used statistical models (e.g., GLM) or non-parametric statistical methods (e.g., GAM, CART, etc.) have their own limits or disadvantages in satisfying requirements of the curve shapes and model adaptability. Here, we propose a new formulation based on a joint probability method. Our new model framework use the response curves directly fitting to vegetation survey data instead of making any assumption of the curve shapes, therefore different curve shapes (e.g., skewed) can be applied to the modeling. Moreover, our model framework had a high adaptability, allowing new environmental parameters to be added to the model while avoiding re-formulation and re-calibration. This joint-probability based model framework was compared with a multinomial regression based Bayesian model framework by using vegetation survey data of freshwater marsh communities in the Everglades, Florida. The environmental drivers included hydroperiod, water depth, and soil phosphorus and carbon contents.
Results/Conclusions
Our results showed that: (1) the joint-probability based model framework could simulate vegetation responses to spatial changes in the environmental drives, and (2) difference in the modeling accuracies between the two model frameworks was small. Therefore, this joint-probability based model framework has a great potential to estimate the vegetation responses to changes of important environmental drivers, thereby contributing to more informed conservation management.