Wavelet Analysis, Artificial Neural Networks And Their Applications In Ecological Research

The Use of the Mexican Hat and the Morlet wavelets for Detection ofEcological PatternsIn this study, we apply methods of statistical significance test to wavelet analysis for pattern analysis by using Monte Carlo assessments. In order to understand the inherent strength and weakness of the Morelet and the Mexican Hat wavelets, we also investigate and compare the properties of two frequently used wavelets by testing with field data and four artificial transects of different typical patterns which is often encountered in ecological research. It is shown that the Mexican Hat provides better detection and localization of patch and gap events over the Morlet, whereas the Morlet offers improved detection and localization of scale over the Mexican Hat. There is always a trade-off between the detection and localization of scale versus patch and gap events. Therefore, the best composite analysis is the combination of their advantages. The properties of wavelet in dealing with ecological data may be affected by characteristics intrinsic to wavelet itself. The peaks of different scales in isograms of wavelet power spectrum from the Mexican Hat may overlap with each other. Alternatively, these peaks of different scales in isograms of wavelet power spectrum may combine with each other unless the size of the analyzed scales is significantly different. These overlapping or combining lead to combining of peaks for different scales, or the masking of trough between peaks of different scales in the scalogram. Ecologists should combine all the information in scalogram and isograms of wavelet coefficient and wavelet power spectrum from different wavelets, which can provide us a broader view and precise pattern information.Analyzing Regeneration Pattern of Quercus liaotungensis in TemperateForest with Two-Dimensional Wavelet AnalysisThis study introduces two-dimensional wavelet analysis as a general interrogative technique for the detection of spatial structure in lattice data. No only is it able to detect constituent components of hierarchical structure, but it can also display the locational information of the components. Patches and gaps of different spatial scales in graphical presentation of wavelet coefficient can be directly linked to the local ecological processes that determine patterns at stand or landscape scales. Derived from 2D wavelet transform, the calculation of wavelet variance can reduce the four-dimensional data of wavelet coefficient to two-dimensional wavelet variance function, and quantify the contribution of the given scale to the overall pattern. We illustrate the use of 2D wavelet analysis and investigate the properties of Mexican Hat wavelet and Halo wavelet by analyzing two simulated patterns. The Halo wavelet can offer higher resolution than Mexican Hat wavelet because of their intrinsic characteristics. The 2D wavelet analysis is also applied to identify the regeneration pattern of Quercus liaotungensis in warm temperate forest. Northern China. The results of analysis indicate that the recruitment of Q. liaotungensis occurs in the overlapping area between of the patch of adult and canopy gap.Testing the Generalization of Artificial Neural Networks with Cross Validation and Independent-Testing Validation in Modelling RiceTillering DynamicsNeural networks rely on the inner structure of the available data sets rather than on the comprehension of the modeled processes, therefore they were generally taken as highly empirical models that are not able to extrapolate, and were always thought to certainly fail to make accurate predictions outside of the range of the training and validation data sets. By comparing the performance of the cross-validated neural networks versus the independent-testing-validated neural networks, the generalization ability of neural networks in predicting rice tillering dynamics was tested in this study. Several techniquesinducing generalization of neural network were also compared. Neural networks in predicting tillering dynamics were proved to have ability to extrapolate t...