Search

Deep learning excels at managing spatial and temporal time series with variable patterns for streamflow forecasting, but traditional machine learning algorithms may struggle with complicated data, including non-linear and multidimensional complexity. Empirical heterogeneity within watersheds and limitations inherent to each estimation methodology pose challenges in effectively measuring and appraising hydrological statistical frameworks of spatial and temporal variables. This study emphasizes streamflow forecasting in the region of Johor, a coastal state in Peninsular Malaysia, utilizing a 28-year streamflow-pattern dataset from Malaysia’s Department of Irrigation and Drainage for the Johor River and its tropical rainforest environment. For this dataset, wavelet transformation significantly improves the resolution of lag noise when historical streamflow data are used as lagged input variables, producing a 6% reduction in the root-mean-square error. A comparative analysis of convolutional neural networks and artificial neural networks reveals these models’ distinct behavioral patterns. Convolutional neural networks exhibit lower stochasticity than artificial neural networks when dealing with complex time series data and with data transformed into a format suitable for modeling. However, convolutional neural networks may suffer from overfitting, particularly in cases in which the structure of the time series is overly simplified. Using Bayesian neural networks, we modeled network weights and biases as probability distributions to assess aleatoric and epistemic variability, employing Markov chain Monte Carlo and bootstrap resampling techniques. This modeling allowed us to quantify uncertainty, providing confidence intervals and metrics for a robust quantitative assessment of model prediction variability.

Learning automata (LA), which intellectually explores its optimal state by interacting with an external environment continuously, is encountered widely in artificial intelligence. In the evaluation of LA, it has always been a key issue how to tradeoff between “accuracy” and “speed,” which substantially touches on parameter tuning. A latest issue in the design of LA methodology involves bearing a parameter-free property, thus removing the tremendous expenses brought by parameter tuning. Nevertheless, the currently existing parameter-free LA schemes generally maintain a Monte-Carlo technique, which helps avoid the tuning process at the cost of more computations. This paper examines a new measurement of parameter-free LA schemes based on statistics which overcome the difficulties found in other counterparts. Specifically, it has innovatively disengaged from the dependance on Monte-Carlo methods. Of greater significance, the learning mechanisms operating in the common stationary environments are likewise extended to the nonstationary environments. Simulations confirm the effectiveness and efficiency of the proposed algorithm, especially its low computation consumption as well as the strong tracking capability to abrupt environmental changes. [ABSTRACT FROM AUTHOR]

A method based on random field theory and Gibbs-Markov random fields equivalency within Bayesian statistical framework is used to derive 3-D surface motion maps from sparse global positioning system (GPS) measurements and differential interferometric synthetic aperture radar (DInSAR) interferogram in the southern California region. The minimization of the Gibbs energy function is performed analytically, which is possible in the case when neighboring pixels are considered independent. The problem is well posed and the solution is unique and stable and not biased by the continuity condition. The technique produces a 3-D field containing estimates of surface motion on the spatial scale of the DInSAR image, over a given time period, complete with error estimates. Significant improvement in the accuracy of the vertical component and moderate improvement in the accuracy of the horizontal components of velocity are achieved in comparison with the GPS data alone. The method can be expanded to account for other available data sets, such as additional interferograms, lidar, or leveling data, in order to achieve even higher accuracy. [ABSTRACT FROM AUTHOR]

A revised method for derivation of three-dimensional surface motions maps from sparse global positioning system (GPS) measurements and two differential interferometric synthetic aperture radar (DInSAR) interferograms based on a random field theory and Gibbs-Markov random fields equivalency within Bayesian statistical framework is proposed. It is shown that the Gibbs energy function can be optimized analytically in the absence of a neighboring relationship between sites of a regular lattice. Because the problem is well posed, its solution is unique and stable, and additional regularization in the form of smoothness is not required. The proposed algorithm is simple in realization, does not require extensive computer power, and is very quick in execution. The results of inverse computer modeling are presented and show a drastic improvement of accuracy when both GPS and DInSAR data are used. [ABSTRACT FROM PUBLISHER]