Milan Dotlić
Milan Dotlić has a total of 24 co-authors. Here are the top co-authors by number of shared publications:
- Ana Dodig
University of Belgrade
3 shared publications B. Pokorni
3 shared publications- Luka Vinokić
Institute for Artificial Intelligence R&D of Serbia, Fruškogorska 1, 21000 Novi Sad, Serbia
3 shared publications - Milan Dimkić
Jaroslav Černi Institute Jaroslava Černog 80 11226 Pinosava Belgrade Serbia
3 shared publications - Milan Stojković
Institute for Artificial Intelligence R&D of Serbia, Fruškogorska 1, 21000 Novi Sad, Serbia
3 shared publications - Milenko Pusic
Academy of Engineering Sciences of Serbia, Belgrade
3 shared publications
Latest Publications
- Open Access
- 9 September 2026
- Published by Elsevier in SSRN Electronic Journal
Weather-Informed Streamflow Forecasting with Temporal Convolution Enhanced Graph Transformer Networks
In recent years, machine learning has emerged as a powerful tool for high-accuracy streamflow forecasting. However, most existing approaches do not fully exploit the inherent directional and hierarchical graph structure of hydrological systems. This work introduces Temporal Convolution Graph Transformer Network (TC-GTN), a novel hybrid model for streamflow forecasting that specifically integrates temporal convolution for temporal pattern extraction with a graph transformer for advanced relational reasoning. It utilizes structured graph representation of river network with accompanying meteorological stations where the transformer's attention mechanism is critical for better understanding of interactions between different nodes/stations and for capturing self-dependencies within each station. To benchmark performance, TC-GTN is compared against two strong baselines: a Graph Transformer without a temporal encoder and a Graph Transformer with a long short-term memory encoder. Experiments on the Drina - Lim River Basin dataset demonstrate that the TC-GTN achieves high performance, with an average coefficient of determination of 0.88 across all hydrological stations, outperforming baselines in 75% of evaluated cases. High-flow event analysis (streamflows above the 75th percentile) further shows that TC-GTN maintains strong accuracy during extreme conditions, achieving average coefficient of determination above 0.75 at major hydropower and flood monitoring stations. This robustness is further strengthened by the quantile-loss formulation, which provides balanced and stable performance for both overall and high-flow conditions, essential for effective flood risk mitigation and sustainable hydropower management. Beyond its hydrological application, TC-GTN offers a general design pattern for spatio-temporal prediction on directed heterogeneous graphs, contributing to the broader literature on spatio-temporal representation learning.
- 9 September 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessWeather-Informed Streamflow Forecasting with Temporal Convolution Enhanced Graph Transformer Networks
In recent years, machine learning has emerged as a powerful tool for high-accuracy streamflow forecasting. However, most existing approaches do not fully exploit the inherent directional and hierarchical graph structure of hydrological systems. This work introduces Temporal Convolution Graph Transformer Network (TC-GTN), a novel hybrid model for streamflow forecasting that specifically integrates temporal convolution for temporal pattern extraction with a graph transformer for advanced relational reasoning. It utilizes structured graph representation of river network with accompanying meteorological stations where the transformer's attention mechanism is critical for better understanding of interactions between different nodes/stations and for capturing self-dependencies within each station. To benchmark performance, TC-GTN is compared against two strong baselines: a Graph Transformer without a temporal encoder and a Graph Transformer with a long short-term memory encoder. Experiments on the Drina - Lim River Basin dataset demonstrate that the TC-GTN achieves high performance, with an average coefficient of determination of 0.88 across all hydrological stations, outperforming baselines in 75% of evaluated cases. High-flow event analysis (streamflows above the 75th percentile) further shows that TC-GTN maintains strong accuracy during extreme conditions, achieving average coefficient of determination above 0.75 at major hydropower and flood monitoring stations. This robustness is further strengthened by the quantile-loss formulation, which provides balanced and stable performance for both overall and high-flow conditions, essential for effective flood risk mitigation and sustainable hydropower management. Beyond its hydrological application, TC-GTN offers a general design pattern for spatio-temporal prediction on directed heterogeneous graphs, contributing to the broader literature on spatio-temporal representation learning.
- Open Access
- 9 May 2026
- Published by Elsevier in SSRN Electronic Journal
Exploring uncertainties across the modelling chain in ML-based streamflow forecasting
Operational streamflow forecasts underpin flood preparedness and reservoir operations, yet their utility is often constrained by poorly characterized and attributed predictive uncertainty. In machine-learning-based forecasting, uncertainty is frequently omitted or reported as a single aggregate output, leaving it unclear which parts of the end-to-end forecasting chain drive overconfidence and forecast degradation, particularly with increasing lead time. In this work, we develop an end-to-end uncertainty decomposition framework for operational streamflow forecasting that attributes predictive uncertainty across meteorological forcing choice, feature design, model architecture, hyperparameter optimization, and training variability, evaluated across multi-day horizons. The decomposition reveals a systematic, horizon-dependent shift in dominant uncertainty sources, with forcing-related contributions increasing with lead time while model-structure and feature choices remain influential at shorter horizons. During high-flow events, predictive intervals remain essential because pipeline heterogeneity can bias the central estimate even when ensemble dispersion widens appropriately. Tuning contributes little to the uncertainty budget but strongly affects compute–skill trade-offs, with Bayesian optimization delivering the most favorable cost–benefit performance under the tested constraints. Together, these results provide actionable guidance for operational freshwater management, showing where investment yields the largest reliability gains: model design at short lead times and forcing quality at longer lead times. This guidance can reduce the risk of costly or unsafe decisions in flood preparedness, reservoir operation, and other critical decision-making contexts in water management.
- 9 May 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessExploring uncertainties across the modelling chain in ML-based streamflow forecasting
Operational streamflow forecasts underpin flood preparedness and reservoir operations, yet their utility is often constrained by poorly characterized and attributed predictive uncertainty. In machine-learning-based forecasting, uncertainty is frequently omitted or reported as a single aggregate output, leaving it unclear which parts of the end-to-end forecasting chain drive overconfidence and forecast degradation, particularly with increasing lead time. In this work, we develop an end-to-end uncertainty decomposition framework for operational streamflow forecasting that attributes predictive uncertainty across meteorological forcing choice, feature design, model architecture, hyperparameter optimization, and training variability, evaluated across multi-day horizons. The decomposition reveals a systematic, horizon-dependent shift in dominant uncertainty sources, with forcing-related contributions increasing with lead time while model-structure and feature choices remain influential at shorter horizons. During high-flow events, predictive intervals remain essential because pipeline heterogeneity can bias the central estimate even when ensemble dispersion widens appropriately. Tuning contributes little to the uncertainty budget but strongly affects compute–skill trade-offs, with Bayesian optimization delivering the most favorable cost–benefit performance under the tested constraints. Together, these results provide actionable guidance for operational freshwater management, showing where investment yields the largest reliability gains: model design at short lead times and forcing quality at longer lead times. This guidance can reduce the risk of costly or unsafe decisions in flood preparedness, reservoir operation, and other critical decision-making contexts in water management.
- Open Access
- 15 April 2026
- Published by Elsevier in SSRN Electronic Journal
Improved Spatio-Temporal Streamflow Forecasting with Temporal Encoder Enhanced Graph Transformer Network
In recent years, machine learning has emerged as a powerful tool for achieving high-accuracy hydrological streamflow forecasting. However, most existing approaches do not fully exploit the inherent directional and hierarchical graph structure of hydrological systems. This work introduces TC-GTN (Temporal Convolution Graph Transformer Network), a novel hybrid model for streamflow forecasting that integrates temporal convolution for temporal pattern extraction with a graph transformer for advanced relational reasoning. It utilizes structured graph representation of river network with accompanying meteorological stations where the transformer's attention mechanism are critical for better understanding of interactions between different nodes/stations and for capturing self-dependencies within each station. To benchmark performance, TC-GTN is compared against two strong baselines: a Graph Transformer without a temporal encoder (GTN) and a Graph Transformer with a long short-term memory encoder (LS-GTN). Comprehensive experiments on the Drina-Lim River Basin dataset demonstrate that the TC-GTN achieves high performance, with an average R2 of 0.88 across all hydrological stations, outperforming two strong baselines in 75% of all evaluated cases. High-flow event analysis (flows above the 75th percentile) further shows that TC-GTN maintains strong accuracy during extreme conditions, achieving average R2 values above 0.75 at major hydropower and flood monitoring stations. This robustness is further strenghtened by the quantile-loss formulation, which provides the most balanced and stable performance for both overall and high-flow conditions, an essential feature for effective flood risk mitigation and sustainable hydropower management.
- 15 April 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessImproved Spatio-Temporal Streamflow Forecasting with Temporal Encoder Enhanced Graph Transformer Network
In recent years, machine learning has emerged as a powerful tool for achieving high-accuracy hydrological streamflow forecasting. However, most existing approaches do not fully exploit the inherent directional and hierarchical graph structure of hydrological systems. This work introduces TC-GTN (Temporal Convolution Graph Transformer Network), a novel hybrid model for streamflow forecasting that integrates temporal convolution for temporal pattern extraction with a graph transformer for advanced relational reasoning. It utilizes structured graph representation of river network with accompanying meteorological stations where the transformer's attention mechanism are critical for better understanding of interactions between different nodes/stations and for capturing self-dependencies within each station. To benchmark performance, TC-GTN is compared against two strong baselines: a Graph Transformer without a temporal encoder (GTN) and a Graph Transformer with a long short-term memory encoder (LS-GTN). Comprehensive experiments on the Drina-Lim River Basin dataset demonstrate that the TC-GTN achieves high performance, with an average R2 of 0.88 across all hydrological stations, outperforming two strong baselines in 75% of all evaluated cases. High-flow event analysis (flows above the 75th percentile) further shows that TC-GTN maintains strong accuracy during extreme conditions, achieving average R2 values above 0.75 at major hydropower and flood monitoring stations. This robustness is further strenghtened by the quantile-loss formulation, which provides the most balanced and stable performance for both overall and high-flow conditions, an essential feature for effective flood risk mitigation and sustainable hydropower management.
- Open Access
- 1 January 2018
- Published by National Library of Serbia in Filomat
- Vol. 32 (20) , 6857-6867
Non-linear multi-point flux approximation in the near-well region
We consider non-linear multi-point flux approximations (MPFA) scheme for flow simulations in a model of anisotropic porous medium that includes wells. The hydraulic head varies logarithmically and its gradient changes rapidly in the well vicinity. Due to this strong non-linearity of the near-well flow, use of the MPFA scheme in the near well region results in a completely wrong total well flux and an inaccurate hydraulic head distribution. In this article we propose correction of the MPFA scheme. The outcome is a scheme that is second-order accurate even in the well vicinity for anisotropic medium. Solution obtained with this scheme respects minimum and maximum principle, and also, it is non-oscillating.
- 1 January 2018
- Published by National Library of Serbia in Filomat
- Vol. 32 (20) , 6857-6867
Open AccessNon-linear multi-point flux approximation in the near-well region
We consider non-linear multi-point flux approximations (MPFA) scheme for flow simulations in a model of anisotropic porous medium that includes wells. The hydraulic head varies logarithmically and its gradient changes rapidly in the well vicinity. Due to this strong non-linearity of the near-well flow, use of the MPFA scheme in the near well region results in a completely wrong total well flux and an inaccurate hydraulic head distribution. In this article we propose correction of the MPFA scheme. The outcome is a scheme that is second-order accurate even in the well vicinity for anisotropic medium. Solution obtained with this scheme respects minimum and maximum principle, and also, it is non-oscillating.
- Open Access
- 1 January 2017
- Published by National Library of Serbia in Geoloski Anali Balkanskoga Poluostrva
- No. 78,p. 47-59
Modeling of sorption and degradation of selected pharmaceuticals: Case study of Belgrade groundwater source
The application of a mathematical model that analyzes the transport of selected pharmaceuticals from the Sava River to a corresponding radial collector well at Belgrade?s groundwater source is assessed. The occurrence of the selected pharmaceuticals in surface water and the corresponding well was monitored from 2009 to 2015. The pharmaceuticals selected for the present study are carbamazepine, trimethoprim, and metamizole metabolites 4-AAA and 4-FAA. Transport is analyzed based on experimental data (sorption isotherms) and a field tracer experiment that includes injection of the selected pharmaceuticals. The analysis shows that sorption of carbamazepine is relatively low and that this pharmaceutical does not degrade under the studied conditions, so it is not possible to accurately determine the degradation half-life. Trimethoprim is detected in the Sava River with an average concentration 8.5 ng/L, but there is no positive detection in well Rb-16. The average concentration of 4-AAA in the surface water is 34 ng/L and of 4-FAA 13 ng/L. The average concentrations of 4-FAAand 4-AAA in the groundwater are in the range from 1 and 1.85 ng/L. The objective of the research is to use an existing hydrogeologic model and apply a transport model to determine the minimum degradation half-life of the investigated pharmaceuticals.
- 1 January 2017
- Published by National Library of Serbia in Geoloski Anali Balkanskoga Poluostrva
- No. 78,p. 47-59
Open AccessModeling of sorption and degradation of selected pharmaceuticals: Case study of Belgrade groundwater source
The application of a mathematical model that analyzes the transport of selected pharmaceuticals from the Sava River to a corresponding radial collector well at Belgrade?s groundwater source is assessed. The occurrence of the selected pharmaceuticals in surface water and the corresponding well was monitored from 2009 to 2015. The pharmaceuticals selected for the present study are carbamazepine, trimethoprim, and metamizole metabolites 4-AAA and 4-FAA. Transport is analyzed based on experimental data (sorption isotherms) and a field tracer experiment that includes injection of the selected pharmaceuticals. The analysis shows that sorption of carbamazepine is relatively low and that this pharmaceutical does not degrade under the studied conditions, so it is not possible to accurately determine the degradation half-life. Trimethoprim is detected in the Sava River with an average concentration 8.5 ng/L, but there is no positive detection in well Rb-16. The average concentration of 4-AAA in the surface water is 34 ng/L and of 4-FAA 13 ng/L. The average concentrations of 4-FAAand 4-AAA in the groundwater are in the range from 1 and 1.85 ng/L. The objective of the research is to use an existing hydrogeologic model and apply a transport model to determine the minimum degradation half-life of the investigated pharmaceuticals.
- Open Access
Second-order accurate finite volume method for well-driven flows
We consider a finite volume method for a well-driven fluid flow in a porous medium. Due to the singularity of the well, modeling in the near-well region with standard numerical schemes results in a completely wrong total well flux and an inaccurate hydraulic head. Local grid refinement can help, but it comes at computational cost. In this article we propose two methods to address well singularity. In the first method the flux through well faces is corrected using a logarithmic function, in a way related to the Peaceman correction. Coupling this correction with a second-order accurate two-point scheme gives a greatly improved total well flux, but the resulting scheme is still not even first order accurate on coarse grids. In the second method fluxes in the near-well region are corrected by representing the hydraulic head as a sum of a logarithmic and a linear function. This scheme is second-order accurate.
Open AccessSecond-order accurate finite volume method for well-driven flows
We consider a finite volume method for a well-driven fluid flow in a porous medium. Due to the singularity of the well, modeling in the near-well region with standard numerical schemes results in a completely wrong total well flux and an inaccurate hydraulic head. Local grid refinement can help, but it comes at computational cost. In this article we propose two methods to address well singularity. In the first method the flux through well faces is corrected using a logarithmic function, in a way related to the Peaceman correction. Coupling this correction with a second-order accurate two-point scheme gives a greatly improved total well flux, but the resulting scheme is still not even first order accurate on coarse grids. In the second method fluxes in the near-well region are corrected by representing the hydraulic head as a sum of a logarithmic and a linear function. This scheme is second-order accurate.
- 1 February 2016
- Published by Elsevier in Journal of Computational Physics
- Vol. 307, 460-475
Second-order accurate finite volume method for well-driven flows
- 1 February 2016
- Published by Elsevier in Journal of Computational Physics
- Vol. 307, 460-475
Second-order accurate finite volume method for well-driven flows
- 1 February 2015
- Published by Elsevier in Journal of Computational Physics
- Vol. 282, 227-237
Piecewise linear transformation in diffusive flux discretization
- 1 February 2015
- Published by Elsevier in Journal of Computational Physics
- Vol. 282, 227-237
Piecewise linear transformation in diffusive flux discretization
- 1 October 2014
- Published by Walter de Gruyter GmbH in Computational Methods in Applied Mathematics
- Vol. 14 (4) , 473-483
Finite Volume Methods for Well-Driven Flows in Anisotropic Porous Media
We consider a finite volume method for flow simulations in an anisotropic porous medium in the presence of a well. The hydraulic head varies logarithmically and its gradient changes rapidly in the well vicinity. Thus, the use of standard numerical schemes results in a completely wrong total well flux and an inaccurate hydraulic head. In this article we propose two finite volume methods to model the well singularity in an anisotropic medium. The first method significantly reduces the total well flux error, but the hydraulic head is still not even first-order accurate. The second method gives a second-order accurate hydraulic head and at least first-order accurate total well flux.
- 1 October 2014
- Published by Walter de Gruyter GmbH in Computational Methods in Applied Mathematics
- Vol. 14 (4) , 473-483
Finite Volume Methods for Well-Driven Flows in Anisotropic Porous Media
We consider a finite volume method for flow simulations in an anisotropic porous medium in the presence of a well. The hydraulic head varies logarithmically and its gradient changes rapidly in the well vicinity. Thus, the use of standard numerical schemes results in a completely wrong total well flux and an inaccurate hydraulic head. In this article we propose two finite volume methods to model the well singularity in an anisotropic medium. The first method significantly reduces the total well flux error, but the hydraulic head is still not even first-order accurate. The second method gives a second-order accurate hydraulic head and at least first-order accurate total well flux.
- 16 May 2014
- Published by Springer Nature
- p. 723-730
Piecewise Linear Transformation in Diffusive Flux Discretizations
- 16 May 2014
- Published by Springer Nature
- p. 723-730
Piecewise Linear Transformation in Diffusive Flux Discretizations