Neural networks are increasingly popular in geosciences due to big progress in neural network modelling techniques and imperative demands in geosciences. Without assuming parametric relationship, artificial neural networks have the ability to learn patterns of relationships in data from being shown a given set of inputs including combinations of descriptive and quantitative data, generalize or abstract results from imperfect data, and be insensitive to minor variations in input such as noise in the data, missing data, or a few incorrect values. When the neural network is trained appropriately, it generalizes therelationship so that it can be applied to other new data sets. The theoretical basis of this technique is the universal approximation theorem, which states that a multi-layer feed-forward neural network, such as the radial-basis function or perceptron neural network, is capable of performing any nonlinear input-output mapping.
    In this paper, the applications of ANNs in various branches of geosciences have been examined here. It is found that ANNs are robust tools and alternative approaches for modeling many of the nonlinear processes in geosciences. After appropriate training, they are able to generate satisfactory results for solving many problems such as prediction, classification, pattern recognition and optimization. By counting the 349 ANNs-based papers in geosciences during the 1997—2000, three predominant subjects are hydrology, geology and atmospheric science, andprediction is major modelling purpose.
    After reviewing the state of the art of geoscience's ANN modelling, author thinks that integrated ANNs modelling framework must be developed in order to dealwith more complex nonlinear processes. Such integrated framework consists of non-parametric techniques such as neural network, fuzzy logic, genetic algorithm,simulated annealing algorithm, fractal theory, Cellular automata and wavelet transform etc. It is helpful for selecting appropriate input and output neurons and designing more efficient networks to profoundly understand the linear or nonlinear processes being modeled in geosciences. Important aspects such as physical interpretation of ANN architecture, optimal training data set, adaptive learning, and generalization must be explored further. The merits and limitations of ANN applications have been discussed, and potential research avenues have been explored briefly.