ATTRIBUTES OF NEURAL NETWORKS FOR EXTRACTING CONTINUOUS VEGETATION VARIABLES FROM OPTICAL AND RADAR MEASUREMENTS

Efficient algorithms that incorporate different types of spectral data and ancillary data are being developed to extract continuous vegetati on variables. Inferring continuous variables implies that functional r elationships must be found among the predicted variable(s), the remote ly sensed data and the ancillary data. Neural networks have attributes which facilitate the extraction of vegetation variables. The advantag es and power of neural networks for extracting continuous vegetation v ariables using optical and/or radar data and ancillary data are discus sed and compared to traditional techniques. Studies that have made adv ances in this research area are reviewed and discussed. Neural network s can provide accurate initial models for extracting vegetation variab les when an adequate amount of data is available. Networks provide a p erformance standard for evaluating existing physically based models. M any practical problems occur when inverting physically based models us ing traditional techniques and neural network techniques can provide a solution to these problems. Networks can be used as a tool to find a set of variables relevant to the desired variables to be inferred for measurement and modelling studies. Neural networks adapt to incorporat e new data sources that would be difficult or impossible to use with c onventional techniques. Neural networks employ a more powerful and ada ptive nonlinear equation form as compared to traditional linear and si mple nonlinear analyses. This power and flexibility is gained by repea ting nonlinear activation functions in a network structure. This uniqu e structure allows the neural network to learn complex functional rela tionships between the input and output data that cannot be envisioned by a researcher.