| Summary: | 碩士 === 國立臺灣大學 === 生物環境系統工程學研究所 === 105 === Riverine ecosystems are usually under the risk of anthropogenic contamination and climate change. With an eye to improving this situation, it is imperative to formulate corresponding policies. To provide the government with better policy-making standard, understanding of the future development of the ecosystems is quite necessary.
Species distribution models (SDMs) is a good tool to calculate habitat suitability index and understand the relationship between environmental variables and species occurrence. There are various species distribution models that have been developed and used in stream ecology. However, there is no consensus on the selection of different models and the relationship between hydrosphere and biosphere is highly complex. What is worse, inappropriate model selection not only expand the uncertainty but also cause the prediction error.
This study collects river channel data and fish data (Sicyopterus Japonic and Onychostoma barbatulum) in the Datuan stream. We use River2D as an efficient tool for simulating the two-dimensional flow condition of a stream segment. Then we combine six SDMs with the outputs of River2D and quantify the relationship between environmental variables and species occurrence by using six SDMs, which are generalized linear model (GLM), generalized additive model (GAM), random forest model (RF), support vector machine (SVM), artificial neural network model (ANN), and ensemble model (the average of other five SDMs). We randomly split the fish data to train(70%) and validate(30%), and each model repeats this step for 1000 times. Additionally, Through Correlation Coefficient, Akaike information criterion, root-mean-square error and Kullback–Leibler divergence, we can know which model has better performance. Moreover, we calculate the information entropy to quantify the spatial heterogeneity of flow conditions and habitat suitability index.
The results demonstrated the accuracy of River2D is affected by measurement deeply. Sicyopterus Japonic likes areas where the water velocity is rapid, and Onychostoma barbatulum likes areas where the velocity and the depth value is small. Moreover, through the result of Correlation Coefficient, root-mean-square error and Kullback–Leibler divergenc, we can know that the performance of the ensemble model is better than others. Therefore, next time we can use the ensemble model when we have not identified which model to use. Moreover, we consider this model for integrating hydraulic and species distribution models can become a criterion when government planning of Ecological Engineering in the future.
|
|---|
