Tsz Ping Chan · KU ScholarWorks (The University of Kansas) · 2008
Two novel approaches are developed for direction-of-arrival (DOA) estimation and functional brain imaging estimation, which are denoted as ReIterative Super-Resolution (RISR) and Source AFFine Image REconstruction (SAFFIRE), respectively. Both recursive approaches are based on a minimum mean-square error (MMSE) framework. The RISR estimator recursively determines an optimal filter bank by updating an estimate of the spatial power distribution at each successive stage. Unlike previous non-parametric covariance-based approaches, which require numerous time snapshots of data, RISR is a parametric approach thus enabling operation on as few as one time snapshot, thereby yielding very high temporal resolution and robustness to the deleterious effects of temporal correlation. RISR has been found to resolve distinct spatial sources several times better than that afforded by the nominal array resolution even under conditions of temporally correlated sources and spatially colored noise. The SAFFIRE algorithm localizes the underlying neural activity in the brain based on the response of a patient under sensory stimuli, such as an auditory tone. The estimator processes electroencephalography (EEG) or magnetoencephalography (MEG) data simulated for sensors outside the patient's head in a recursive manner converging closer to the true solution at each consecutive stage. The algorithm requires a minimal number of time samples to localize active neural sources, thereby enabling the observation of the neural activity as it progresses over time. SAFFIRE has been applied to simulated MEG data and has shown to achieve unprecedented spatial and temporal resolution. The estimation approach has also demonstrated the capability to precisely isolate the primary and secondary auditory cortex responses, a challenging problem in the brain MEG imaging community.
Cited by 0Open access
Lindrith Cordell · Geophysics · 1973
Abstract In the analysis of gravity data over thick sedimentary basins, density contrast can sometimes be approximated by a continuous function decreasing exponentially with depth. Typical values of the exponential decrement are in the order of 0.3 to 1.5 km (super -1) . The gravity effect of an infinite (Bouguer) slab in this system tends to a finite limit as the slab becomes infinitely thick, leading to quantitative, and sometimes stringent, limitations on the resolving power of gravity analysis over deep structures. The exponential density term acts as an integrating factor in deriving the gravity effect of a prismatic building element, leading to a simple expression involving both the gravity field and its vertical derivative. A recursive computer algorithm giving the inverse solution automatically in terms of assigned density parameters is applied to a gravity profile over a graben structure at San Jacinto, California. The computed basin configuration is consistent with the seismically determined basement depth of 2.4 km and the estimated density layering of the graben fill.
Cited by 183
A. F. Gualtieri, A. Viani, G. Banchio et al. · Research Padua Archive (University of Padua) · 2001
One of the limits of quantitative phase analysis of natural clays using the Rietveld method is the lack of an applicable structure model for the highly disordered phases such as montmorillonite. The description of disorder of montmorillonite, the main dioctrahedral smectite, was accomplished using the DIFFaX program [1] which allows the calculation of simulated powder patterns. We followed the strategy successfully used for kaolinite [2] in which simulated powder patterns calculated using different models of disorder are compared to the observed one. Since DIFFaX is bases upon a recursive method and produces a statistical crystal, a super-cell composed of n layers describing the average disordered structure was re-defined to be used in the codes which apply the deterministic Rietveld algorithm. Therefore, a working structure composed of 4 layers (an ideal layer, a b/3 shifted layer, a -b/3 shifted layer, and a -a/3 shifted layer) was described and used in GSAS [3] to fit a natural international standard of Ca-montmorillonite.
Cited by 0
Gennadi Malaschonok, Andrii Ivaskevych · NaUKMA Research Papers Computer Science · 2020
This paper investigates the block-recursive parallel algorithm for Cholesky decomposition for a super-computer with distributed memory. For ease of scaling, the number of cores is a degree of the number two. We investigate the resistance of the algorithm to the accumulation of computational errors and the scaling efficiency of the static block-recursive algorithm.Note that the problem does not have an exact solution in rational numbers (for example, LU-decompo-sition has an exact solution), so it is necessary to use approximate calculations and there are no other approaches than calculations with some accuracy. We have not been able to find systematic studies on the accumulation of error in the Cholesky algorithm, so we conduct such studies in this paper. elements are integers with a fixed number of binary digits and a uniform distribution.We have shown that for dense matrices, starting with matrix size 64, the use of double precision (standard floating-point arithmetic and 64-bit machine word) does not allow to obtain an error less than 1, even in simple cases. As the size of the coefficients increases, the error only increases, so the use of such arithmetic for matrices of even larger size loses its meaning. To solve this problem, we used BigDecimal arithmetic for large matrices. It allows you to programmatically specify the accuracy, which is specified as the number of decimal places. First, we determine the required number of characters for BigDecimal so that the calculation error of this matrix does not exceed one, and then we experiment with different numbers of cores for such a matrix. We give recommendations for dense matrices, the elements of which were given randomly with a uniform distribution. For such matrices, based on their size and the number of decimal places in their elements, we recommend the choice of accuracy for machine arithmetic and the number of cores for calculations.Manuscript received 09.06.2020
Cited by 0Open access
Martı́n Farach-Colton, Paolo Ferragina, S. Muthukrishnan · Journal of the ACM · 2000
The suffix tree of a string is the fundamental data structure of combinatorial pattern matching. We present a recursive technique for building suffix trees that yields optimal algorithms in different computational models. Sorting is an inherent bottleneck in building suffix trees and our algorithms match the sorting lower bound. Specifically, we present the following results. (1) Weiner [1973], who introduced the data structure, gave an optimal 0 ( n )-time algorithm for building the suffix tree of an n -character string drawn from a constant-size alphabet. In the comparison model, there is a trivial Ο( n log n )-time lower bound based on sorting, and Weiner's algorithm matches this bound. For integer alphabets, the fastest known algorithm is the O ( n log n )time comparison-based algorithm, but no super-linear lower bound is known. Closing this gap is the main open question in stringology. We settle this open problem by giving a linear time reduction to sorting for building suffix trees. Since sorting is a lower-bound for building suffix trees, this algorithm is time-optimal in every alphabet mode. In particular, for an alphabet consisting of integers in a polynomial range we get the first known linear-time algorithm. (2) All previously known algorithms for building suffix trees exhibit a marked absence of locality of reference, and thus they tend to elicit many page faults (I/Os) when indexing very long strings. They are therefore unsuitable for building suffix trees in secondary storage devices, where I/Os dominate the overall computational cost. We give a linear-I/O reduction to sorting for suffix tree construction. Since sorting is a trivial I/O-lower bound for building suffix trees, our algorithm is I/O-optimal.
Cited by 210Open access
Suzuki, Yudai, Tiang, Bi Hong, Son, Jeongrak et al. · arXiv (Cornell University) · 2025
Quantum signal processing (QSP), a framework for implementing matrix-valued polynomials, is a fundamental primitive in various quantum algorithms. Despite its versatility, a potentially underappreciated challenge is that all systematic protocols for implementing QSP rely on post-selection. This can impose prohibitive costs for tasks when amplitude amplification cannot sufficiently improve the success probability. For example, in the context of ground-state preparation, this occurs when using a too poor initial state. In this work, we introduce a new formula for implementing QSP transformations of Hermitian matrices, which requires neither auxiliary qubits nor post-selection. Rather, using approximation to the exact unitary synthesis, we leverage the theory of the double-bracket quantum algorithms to provide a new quantum algorithm for QSP, termed Double-Bracket QSP (DB-QSP). The algorithm requires the energy and energetic variance of the state to be measured at each step and has a recursive structure, which leads to circuit depths that can grow super exponentially with the degree of the polynomial. With these strengths and caveats in mind, DB-QSP should be viewed as complementing the established QSP toolkit. In particular, DB-QSP can deterministically implement low-degree polynomials to "warm start" QSP methods involving post-selection.
Cited by 0Open access
Elaine Levey, Thomas Rothvoß · arXiv (Cornell University) · 2015
In a classical problem in scheduling, one has $n$ unit size jobs with a\nprecedence order and the goal is to find a schedule of those jobs on $m$\nidentical machines as to minimize the makespan. It is one of the remaining four\nopen problems from the book of Garey & Johnson whether or not this problem is\n$\\mathbf{NP}$-hard for $m=3$.\n We prove that for any fixed $\\varepsilon$ and $m$, an LP-hierarchy lift of\nthe time-indexed LP with a slightly super poly-logarithmic number of $r =\n(\\log(n))^{\\Theta(\\log \\log n)}$ rounds provides a $(1 +\n\\varepsilon)$-approximation. For example Sherali-Adams suffices as hierarchy.\nThis implies an algorithm that yields a $(1+\\varepsilon)$-approximation in time\n$n^{O(r)}$. The previously best approximation algorithms guarantee a $2 -\n\\frac{7}{3m+1}$-approximation in polynomial time for $m \\geq 4$ and\n$\\frac{4}{3}$ for $m=3$. Our algorithm is based on a recursive scheduling\napproach where in each step we reduce the correlation in form of long chains.\nOur method adds to the rather short list of examples where hierarchies are\nactually useful to obtain better approximation algorithms.\n
Cited by 0Open access
Yudai Suzuki, Bi Hong Tiang, Jeongrak Son et al. · Quantum · 2025
Quantum Signal Processing (QSP), a framework for implementing matrix-valued polynomials, is a fundamental primitive in various quantum algorithms. Despite its versatility, a potentially underappreciated challenge is that all systematic protocols for implementing QSP rely on post-selection. This can impose prohibitive costs for tasks when amplitude amplification cannot sufficiently improve the success probability. For example, in the context of ground-state preparation, this occurs when using a too poor initial state. In this work, we introduce a new formula for implementing QSP transformations of Hermitian matrices, which requires neither auxiliary qubits nor post-selection. Rather, using approximation to the exact unitary synthesis, we leverage the theory of the double-bracket quantum algorithms to provide a new quantum algorithm for QSP, termed Double-Bracket QSP (DB-QSP). The algorithm requires the energy and energetic variance of the state to be measured at each step and has a recursive structure, which leads to circuit depths that can grow super exponentially with the degree of the polynomial. With these strengths and caveats in mind, DB-QSP should be viewed as complementing the established QSP toolkit. In particular, DB-QSP can deterministically implement low-degree polynomials to ``warm start" QSP methods involving post-selection.
Cited by 0Open access
Foad Fereidoony, Ali Jishi, Maziar Hedayati et al. · IEEE Sensors Journal · 2021
Multipath signals are known to be the main source of error in localization procedures; they are generated by the reflection from different objects through non-line of sight (NLOS) paths and overlap with the line of sight (LOS) signal. The time delay and strength of the LOS signal are critical parameters for ranging which are both affected by the overlapped NLOS components. The situation is worse when the frequency bandwidth (BW) is limited that inherently leads to decrease in the resolution of system obeying the Fourier limit ( <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$3^\ast 10^{\wedge }8$ </tex-math></inline-formula> /BW). Leveraging the classical super-resolution techniques such as Maximum Likelihood (ML), MUSIC and compressive sensing (CS), the Fourier limit can be overcome but at the cost of having higher complexity and more computational time. In this paper, the Magnitude-Delay Least Mean Squares Equalization (MD-LMSE) is proposed uniquely for super-resolution time delay estimation. The proposed estimation algorithm is based on the recursive time domain Least Mean Squares (LMS) minimization, which is established through an equalizer structure. The MD-LMSE approach provides high accuracy, super resolution and low computational time that makes it suitable for real time implementation. The proposed MDLMSE algorithms is used through an experimental setup consisting one LOS signal and one NLOS signal to improve the range resolution of a narrowband system 95% with low error of 6.66%.The obtained results verify that the proposed approach is a good candidate for accurate ranging, especially in low frequency applications (with limited bandwidth) by enabling higher penetration depths.
Cited by 14
Junqiu Zhang, Ying Rao, Yanghua Wang · IEEE Transactions on Geoscience and Remote Sensing · 2024
In seismic full waveform inversion (FWI), the shot-encoding technique can reduce computational costs. It randomly encodes all individual shot gathers and combines them into a super shot gather. This allows the number of wavefield simulations to be reduced. To mitigate crosstalk noise caused by interference between different shots, shot-encoded FWI requires the shot encoding to be sufficiently random. However, this may destabilize the standard limited memory BFGS (L-BFGS) algorithm and lead to instability and nonconvergence in the interactive inversion of FWI. In this article, we propose the improved L-BFSG (iL-BFGS) method, which uses the formula of the outer product of the Hessian matrix to calculate the gradient difference for the recursive L-BFSG computation. We also redesigned the workflow for shot-encoded FWI to increase the randomness within the shot encoding. We have shown that the proposed method improves the image quality of shot-encoded FWI, especially for intermediate and deep formations, accelerates the convergence of iterative inversion, and improves the ability of FWI to suppress crosstalk effects.
Cited by 2
K. Deepthi, B. Mounika, Y. Chalapathi Rao · 2025
This paper presents a hardware implementation of 32-bit matrix multiplication using Strassen's algorithm, targeting improvements in speed, power efficiency, and chip area. Strassen's method reduces the number of required multiplications by dividing matrices recursively into smaller blocks, achieving a computational complexity of approximately <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathrm{O}\left(\mathrm{n}^{2}. 81\right)$</tex> compared to the conventional <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathrm{O}\left(\mathrm{n}^{3}\right)$</tex> method. The design is implemented in Verilog and synthesized using 90 nm SAED technology. Comparative evaluation shows a 30 % reduction in area, 20 % savings in power, and 42 % improvement in timing over traditional matrix multiplication. These improvements construct the design fit for super-computing applications including steganography, artificial intelligence, and display adapter, where large-scale matrix operations are general. The work shows that a recursive, hardware-aware approach can critically accelerate matrix computation while remaining are agreeable with 32-bit architectures.
Cited by 0
Lutz Oettershagen, Nils M. Kriege, Petra Mutzel · 2023
The H-index of a node in a static network is the maximum value h such that at least h of its neighbors have a degree of at least h. Recently, a generalized version, the n-th order H-index, was introduced, allowing to relate degree centrality, H-index, and the k-core of a node. We extend the n-th order H-index to temporal networks and define corresponding temporal centrality measures and temporal core decompositions. Our n-th order temporal H-index respects the reachability in temporal networks leading to node rankings, which reflect the importance of nodes in spreading processes. We derive natural decompositions of temporal networks into subgraphs with strong temporal coherence. We analyze a recursive computation scheme and develop a highly scalable streaming algorithm. Our experimental evaluation demonstrates the efficiency of our algorithms and the conceptional validity of our approach. Specifically, we show that the n-th order temporal H-index is a strong heuristic for identifying possible super-spreaders in evolving social networks and detects temporally well-connected components.
Cited by 6Open access
Kiyotaka Kohno, Yasushi Inouye, Mitsuru Kawamoto · IEEE Transactions on Circuits and Systems I Regular Papers · 2008
In the simplest case, the matrix inversion Lemma gives an explicit formula of the inverse of a positive-definite matrix A added to a rank-one matrix bb <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">H</sup> as follows:(A + bb <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">H</sup> ) <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">-1</sup> = A <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">-1</sup> -A <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">-1</sup> b(1 + b <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">H</sup> A <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">-1</sup> b) <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">-1</sup> b <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">H</sup> A <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">-1</sup> . It is well known in the literature that this formula is very useful to develop a recursive least-squares algorithm for the recursive identification of linear systems or the design of adaptive filters. We extend this result to the case when the matrix A is singular and present a matrix pseudo-inversion lemma along with some illustrative examples. Such a singular case may occur in a situation where a given problem is overdeter-mined in the sense that it has more equations than unknowns. This lemma is important in its own right, but in order to show the usefulness of the lemma, we apply it to develop an adaptive super-exponential algorithm for the blind deconvolution of multi-input multi-output systems.
Cited by 18
Badis Ouahab, Farès Boudjema, Mohamed Amine Alouane et al. · Journal of Vibration and Control · 2026
This paper presents a prescribed performance-based hierarchical fast terminal sliding mode control (PP-HR-FTSMC) approach for stabilizing a rotary inverted pendulum (RIP). The proposed controller ensures finite-time convergence while enforcing a prescribed performance bound on the system tracking error. A hierarchical recursive control structure is designed to handle the highly nonlinear and underactuated dynamics of the RIP, enhancing robustness against uncertainties and disturbances. By utilizing an error transformation function, the constrained control problem is converted into an unconstrained one, ensuring compliance with the preset performance of tracking errors. The fast terminal sliding mode (FTSM) framework guarantees rapid error convergence while mitigating chattering effects. To optimize the controller gains, a multi-objective golden eagle optimization (MO-GEO) algorithm is employed, balancing key performance metrics such as convergence speed, control effort, and robustness. The proposed control strategy is validated through high-fidelity simulations in Simscape, followed by real-time experiments on a custom-built RIP platform. The results demonstrate superior tracking accuracy, robustness, and finite-time stabilization compared to linear quadratic Gaussian (LQG) controller, adaptive super-twisting sliding mode controller (ASTSMC), and proportional integral derivative (PID) controller.
Cited by 0
Zhenjun Ma, Shengwei Wang · Research Online (University of Wollongong) · 2010
This paper presents an optimization strategy for optimal control and operation of building central chilling systems in order to minimize their energy consumption and provide improved control performance. The strategy is formulated using a systematic approach, in which the characteristics and interactions among the components and subsystems in the central chilling system are considered. The simplified models of major components are used in the strategy as the performance predictors to estimate the system energy performance and responses to the changes of control settings and working conditions. To ensure the models to provide reliable estimates when the working condition changes, the model parameters are updated online using the recursive least squares (RLS) estimation technique. A genetic algorithm (GA) is used to solve the optimization problem and search for globally optimal control settings on the basis of a cost function estimator defined. The performance of this strategy was tested and evaluated in a simulated virtual system representing the complex central chilling system in a super high-rise building under various working conditions. The model parameter identification and performance validation as well as the performance evaluation of the optimization strategy are presented.
Cited by 2Open access
Pengfei Liang, Qingyun Di, Changmin Fu et al. · Radio Science · 2023
Abstract Electromagnetic (EM) fields excited by a vertical electric dipole (VED) at the super‐low frequency and the extremely low frequency have been widely used in different studies such as space weather, earthquake predictions, geophysical investigations, and communications. Accurate and efficient algorithms are necessary for quick estimations of worldwide EM fields. This study proposes a recursive algorithm to calculate EM fields excited by a VED embedded in a layered spherical model. A set of local/general reflection and transmission coefficients is defined to describe propagations of EM fields in each spherical layer and the outgoing‐type waves and the standing‐type waves of EM fields are expressed by normalized spherical Bessel/Hankel functions. The normalized spherical Bessel/Hankel functions are not only useful to overcome the overflow problem in calculations of spherical Bessel/Hankel functions but also helpful to speed up the convergence of summation expressions of EM fields, because the normalized spherical Bessel/Hankel functions mitigate small oscillations terms. Calculating the primary fields and the secondary fields of the total EM fields, respectively, is also helpful to reach convergent summations, because their convergence radiuses are different.
Cited by 0Open access
Liu , Zongying · University of Malaya Students Repository · 2019
Multi-ahead time series prediction model is essential to human activities in the financial domain, electrical load, and natural disaster analysis/forecasting. Recent years, Machine correlation and potential non stationary of the data can be automatically analyzed. However, the problems with traditional offline and online learning algorithms in machine learning algorithms are usually faced with parameter dependency, concept drift handling problem, connectionless of neural net and unfixed reservoir. In this study, the proposed models consist of the off-line and on-line algorithm for time series prediction: Recurrent Kernel Extreme Reservoir Machine with Quantum Particle Swarm Optimization (RKERM-QPSO) and Meta-cognitive Recurrent Recursive Kernel On-line Sequential Extreme Learning Machine with the Drift Detection Machine (Meta-RRKOS-ELM-DDM) to solve above mentioned problems and improved overall prediction accuracy. In the forecasting process, the restriction of the prediction horizon is solved by the recurrent multi-steps-ahead algorithm for both off-line and on-line prediction models. The parameter dependency problem of the off-line and on-line model is handled with QPSO and new meta-cognitive learning strategy, respectively. Moreover, reservoir computing is applied to generate a fixed reservoir with high dimension information in the off-line prediction model in order to improve the forecasting accuracy. In the on-line learning model, the recursive kernel is also successfully used to generate a fixed reservoir with optimized information by dissipating and overwriting the information from the coming data in the learning part of the prediction model, which is helpful for improvement of forecasting performance. Besides, concept drift problem in on-line learning model is solved by Drift Detection Machine (DDM). In this study presents the theoretical insights which supporting the proposed arguments. The experiment is carried out by analyzed with benchmark data and a real-world example. In the proposed online learning model, the SMAPE result of 1-18 step periods in real-world data sets are 3.39% for S&P 500, 3.45% for Shanghai, and 5.48% for Ozone, respectively. It shown the super prediction ability compared to other state-of-the-art time-series-specific.
Cited by 0
Dongsu Bak, Hyunsoo Min, Soo-Jong Rey · Physical review. D. Particles, fields, gravitation, and cosmology/Physical review. D, Particles, fields, gravitation, and cosmology · 2010
We study issues concerning perturbative integrability of $\mathcal{N}=6$ Chern-Simons theory at planar and weak `t Hooft coupling regime. By Feynman diagrammatics, we derive so-called maximal-ranged interactions in the quantum dilatation generator, originating from homogeneous and inhomogeneous diagrams. The dilatation operator requires proper regularization of ultraviolet and infrared divergences and also bears scheme dependence depending on operator-mixing or two-point function methods adopted. We first consider the standard operator-mixing method. We show that homogeneous diagrams are obtainable by recursive method to all orders. The method, however, is not easily extendable to inhomogeneous diagrams. We thus consider two-point function method and study both operator contents and the spectrum of the quantum dilatation generator up to six-loop orders. Within this scheme, we show that, of two possible classes of operators, only one linear combination actually contributes. Curiously, this is exactly the same combination as in $\mathcal{N}=4$ super Yang-Mills theory. From these operators, we extract a spectrum of anomalous dimension up to six loops. We find that the spectrum agrees perfectly with the prediction based on quantum integrability. In evaluating the six-loop diagrams, we utilized a remarkable integer-relation algorithm developed by Ferguson, Baily, and Arno.
Cited by 19Open access
N. Denizcan Vanli, Kaan Gökcesu, Muhammed O. Sayin et al. · IEEE Transactions on Signal Processing · 2016
We study sequential compound decision problems in the context of sequential prediction of real valued sequences. In particular, we consider finite state (FS) predictors that are constructed based on a hierarchical structure, such as the order preserving patterns of the sequence history. We define hierarchical equivalence classes by tying certain models at a hierarchy level in a recursive manner in order to mitigate undertraining problems. These equivalence classes defined on a hierarchical structure are then used to construct a super exponential number of sequential FS predictors based on their combinations and permutations. We then introduce truly sequential algorithms with computational complexity only linear in the pattern length that 1) asymptotically achieve the performance of the best FS predictor or the best linear combination of all the FS predictors in an individual sequence manner without any stochastic assumptions over any data length n under a wide range of loss functions; 2) achieve the mean square error of the best linear combination of all FS filters or predictors in the steady-state for certain nonstationary models. We illustrate the superior convergence and tracking capabilities of our algorithm with respect to several state-of-the-art methods in the literature through simulations over synthetic and real benchmark data.
Cited by 10
Eyob Betru Wegayehu, Fiseha Behulu Muluneh · Heliyon · 2023
Traditional data-driven streamflow predictions usually apply a single model with inconsistent performance in different variability conditions. These days model ensembles or merging the benefits of different models without losing the general character of the data are becoming a trend in hydrology. This study compared three super ensemble learners with eight base models. Twelve years of monthly rolled daily time series data in three river catchments of Ethiopia (Borkena watershed: Awash river basin), (Gummera watershed: Abay river basin), and (Sore watershed: Baro Akobo river basin) is used for single-step daily streamflow simulation using previous thirty-day input timesteps. Five input scenarios are applied: three vegetation indices, three remote sensing-based precipitation products, ground-gauged rainfall, all fused inputs, and selected inputs with the Recursive Feature Elimination (RFE) algorithm. The time series is then divided into training and testing datasets with a ratio of 80:20. The performance of the proposed models was evaluated using the Root Mean Squared Error (RMSE), coefficient of determination (R<sup>2</sup>), Mean Absolute Error (MAE), and Median Absolute Error (MEDAE). Finally, the result is presented with the corresponding five input scenarios. The catchment's and input scenario's average performance indicated that the three super ensemble learners outperformed the eight base models with relatively stable performance. The top-ranked WASE model exceeded the linear regression baseline by 13.3%. XGB, CNN-GRU, and LSTM proved the highest performance of the base models. This study also revealed that LSTM's key downside is its performance drop in the absence of feature selection criteria. In comparison, XGB showed its superior performance after controlling redundant inputs internally. Moreover, this study uniquely highlights the potential of remote sensing-based vegetation indices in the science of data-driven streamflow modelling for non-gauged catchments with no meteorological time series.
Cited by 14Open access