Nonlinear analysis and forecasting of a brackish karstic spring

Nonlinear methods and artificial neural network techniques are applied to t he study of the regime and the possibility of short-term forecasting of dis charges of the spring of Almyros, Iraklion, Crete. Questions regarding the nonlinearity and chaotic characteristics of the system necessitate the exam ination of dynamical properties. Toward this objective the time series of d aily average discharges is analyzed in detail. First, the dimensionality of the dynamics in the reconstructed phase space is found to be quite low, si milar to 3-4. Then several tests are applied to examine the nonlinearity an d the presence of noise in the data. Using the surrogate time series test, a high degree of nonlinearity and a deterministic nature are revealed, whil e the differentiation test showed that the presence of high-frequency noise in the series of the discharge is not dynamically important. These suggest that an attempt to forecast the short-term future behavior of this time se ries may turn out to be quite successful. Nonlinear methods, such as Farmer 's algorithm and artificial neural networks, were employed and found to exh ibit a very satisfactory predictive ability, with neural networks achieving a slightly better performance.