the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 05 May 2025
Bakaano-Hydro (v1.1). A distributed hydrology-guided deep learning model for streamflow prediction
Abstract. Reliable streamflow prediction is fundamental to hydrological forecasting, water resources planning, and climate adaptation. However, existing data-driven approaches often lack physical interpretability and struggle to incorporate spatial heterogeneity and hydrological connectivity. Conversely, traditional process-based models are limited by high calibration demands and structural uncertainty, especially in data-scarce regions. These challenges underscore the need for hybrid frameworks that combine the strengths of physically based modeling with the predictive capacity of machine learning. Here, I present Bakaano-Hydro, a distributed hydrology-guided deep learning model for streamflow prediction. The model integrates a gridded runoff generation method, a topographic flow routing scheme, and a temporal convolutional network to capture both spatial and temporal hydrological dynamics. This architecture enables incorporation of spatial heterogeneity and explicitly represents hydrological connectivity, while using neural networks to learn streamflow dynamics and enhance predictive performance. Bakaano-Hydro’s performance is evaluated across six river basins spanning four continents, encompassing diverse climate zones, land-use patterns, and hydrological regimes. Results indicate that Bakaano-Hydro demonstrates robust performance in humid and snow-fed basins where saturation-excess runoff dominates, while revealing key limitations in arid and semi-arid regions characterized by infiltration-excess processes. Bakaano-Hydro advances the state of the art in data-driven hydrological modeling by integrating physical realism with deep learning. Its modular and fully automated pipeline enables rapid deployment in data-scarce regions, while maintaining high reliability and interpretability. These features make Bakaano-Hydro a promising tool for operational forecasting, climate risk assessment, and adaptation planning across diverse hydrological and socio-environmental contexts. The model code is publicly available at https://github.com/confidence-duku/bakaano-hydro to facilitate reproducibility and community-driven development.
- Preprint
(1814 KB) - Metadata XML
- BibTeX
- EndNote
Status: final response (author comments only)
-
RC1: 'Comment on egusphere-2025-1633', Anonymous Referee #1, 30 Jun 2025
-
AC1: 'Reply on RC1', Confidence Duku, 23 Jan 2026
I thank the reviewer for their careful reading of my manuscript and for their constructive comments and suggestions. In the following, I respond to each comment point by point. Reviewer comments are reproduced in full (RC), followed by my author responses (AC). All revisions indicated will be incorporated in the revised manuscript.
RC: The chief shortcoming is the lack of an empirical benchmark against the data‑driven approaches that motivate the study. At minimum the authors should compare against (i) a lumped LSTM trained on catchment‑aggregated forcings and (ii) ideally a Conv‑LSTM fed with the same gridded inputs; a physics‑only baseline (VegET + routing) would further contextualise gains. Without these, neither the added predictive value nor computational overhead of the proposed architecture can be quantified.
AC: I welcome the suggestion by the reviewer to include benchmarking against other data-driven approaches. In the revised version, I intend to benchmark performance against widely recognized hydrological models including a lumped data-driven model and a physicaly based model.
RC: Abstract: state training (1989–2016) and evaluation (1982–1988) periods and, once baselines are added, give a headline improvement (e.g. median ΔKGE vs lumped LSTM).
AC: I will revise the abstract accordingly
RC: Baseline experiment: implement and report at least one well‑tuned baseline (lumped LSTM, Conv‑LSTM or both) on the same split; a small table of KGE/NSE and wall‑time for 2–3 basins suffices.
AC: I thank the reviewer for the comments. In the revised version, I will benchmark performance against lumped data-driven models and include wall-time.
RC: Runoff generation (Sect 2.1): clarify whether VegET parameters are default or calibrated and explain beforehand why saturation‑excess may fail in arid basins.
AC: VegET parameters are not calibrated. The runoff generated from VegET are not calibrated and used directly. Additionally, the flow-routed runoff are also not calibrated and used directly. As stated in page 6 line 140 – 146, It is important to note that this initial routing implementation does not explicitly account for hydrological processes such as transmission losses, overbank flow, or in-channel travel time delays, which can influence both the magnitude and timing of streamflow. The primary objective at this stage is to simulate maximum potential daily flow at each hydrological point. These routed runoff time-series serve as inputs to the subsequent deep learning stage, where temporal dependencies and additional hydrological dynamics are learned directly from data.
In the revised version, I will explicitly state that runoff generation by VegET and flow routing are uncalibrated. I will discuss why saturation-excess may fail in arid basins as suggested by the reviewer.
RC: Neural network (Sect 2.3): list trainable parameters for both variants and typical wall‑time per basin on CPU/GPU (batch size, optimiser). Justify the 365‑day look‑back or summarise sensitivity tests (this mirrors Kratzert 2018 and provides a natural benchmark).
AC: Figure 2 illustrates the neural network architecture used in Bakaano-Hydro. In the revised manuscript, I will explicitly list the trainable parameters and report typical computational requirements, including wall-clock training time per basin on CPU and GPU, batch size, and optimizer settings, as requested. The look-back window is a user-defined hyperparameter rather than a fixed or universally optimal choice. In the current study, a 365-day look-back was selected for the case studies to capture seasonal to interannual hydrological memory and to enable comparison with commonly used configurations in the literature (e.g., Kratzert et al., 2018). Bakaano-Hydro is designed such that this parameter can be varied depending on basin characteristics and application needs. I will clarify this explicitly in Sect. 2.3 and either justify the chosen value more clearly or summarize sensitivity tests to demonstrate its effect on model performance.
RC: Data split & basin stats (Sect 4.1): justify the 1989 cut‑off and supply a table with initial vs final stations, mean record length, missing‑data threshold. Diagnostics (Sect 4.3): add an extreme‑flow metric (Peak Flow Bias, FHV/FLV) and fix the truncated Figure 5 caption. Figures/layout: ensure ≥ 300 dpi and greyscale‑safe colour palettes, consistent across plots.
AC: I thank the reviewer for these useful suggestions. I will revise the paper accordingly and include these suggestions.
RC: Complete Figure 5 caption and define all metrics. Standardise “VegET” capitalisation; use km², mm day⁻¹, etc. Line 200: “relavant” → “relevant”. Define β‑KGE and α‑NSE at first mention. Format all DOIs with https://doi.org/…
AC: I thank the reviewer for pointing out these. I will revise the paper accordingly.
Citation: https://doi.org/10.5194/egusphere-2025-1633-AC1
-
AC1: 'Reply on RC1', Confidence Duku, 23 Jan 2026
-
RC2: 'Comment on egusphere-2025-1633', Anonymous Referee #2, 30 Dec 2025
General Assessment
This manuscript presents Bakaano-Hydro, a hybrid modeling framework that combines a gridded process-based runoff generation scheme (VegET), topographic flow routing, and a deep learning architecture (TCN with attention and FiLM conditioning) to simulate distributed streamflow. The paper is ambitious in scope, technically detailed, and addresses an important and timely problem in hydrology: how to reconcile physical realism, spatial heterogeneity, and predictive skill within data-driven modeling frameworks.
The manuscript is generally well written, clearly structured, and accompanied by open-source code, which is a strong asset. The author demonstrates a deep understanding of both hydrological theory and modern machine learning architectures. The evaluation across six large basins spanning multiple hydroclimatic regimes is a notable strength, as is the explicit diagnostic discussion of where and why the model fails.
However, despite these strengths, I have substantial concerns regarding the framing of novelty, the strength of some claims (especially concerning data-scarce regions), the choice and rigidity of the runoff generation mechanism, and the lack of comparison against relevant baselines. In its current form, the manuscript would benefit from significant revision to better position Bakaano-Hydro within the rapidly evolving literature on physics-guided and hybrid hydrological machine learning, and to more carefully delimit the conditions under which the model is genuinely advantageous.
Overall, I believe this work has clear potential for publication, but major revisions are required before it can be considered for acceptance.
Major Comments
I. Novelty and Positioning Relative to Existing Hybrid and Physics-Guided ML Models
The manuscript repeatedly emphasizes that Bakaano-Hydro addresses limitations of “state-of-the-art data-driven hydrological models” by incorporating spatial heterogeneity and hydrological connectivity. While this motivation is valid, the literature review and framing do not sufficiently acknowledge how much progress has already been made in this direction.
In recent years, numerous studies have: Incorporated physical constraints directly into neural networks; Used distributed or graph-based representations of river networks; Combined conceptual or process-based runoff modules with ML-based routing or correction layers; Explicitly targeted spatial generalization and ungauged basins. As a result, statements implying that most existing data-driven models are lumped and physically uninterpretable are overly broad and somewhat outdated.
The manuscript would benefit from: 1. A clearer comparison between Bakaano-Hydro and other physics-guided or hybrid frameworks, not only lumped LSTM baselines. 2. Explicit discussion of why Bakaano-Hydro’s serial hybridization (VegET → routing → TCN) offers conceptual or practical advantages over alternative coupling strategies. 3. Without this, the novelty risks being perceived as incremental rather than transformative, especially given that the runoff generation and routing components are externally imposed rather than learned or dynamically coupled.
II. Strong Claims About Applicability in Data-Scarce Regions Are Not Fully Supported
The manuscript repeatedly argues that Bakaano-Hydro is well suited for data-scarce regions and may outperform traditional process-based models under such conditions. However, the presented results do not convincingly support this claim.
Key concerns: 1. All experiments rely on GRDC stations with at least three years of data, many of them in well-monitored basins. 2. The neural network is trained on multi-station, multi-year observations, which are precisely what data-scarce regions lack. 3. Performance degrades substantially in arid and semi-arid basins (e.g., Orange), which are often the most data-scarce and management-critical regions.
In fact, the results suggest the opposite: where hydrology is complex, threshold-driven, or poorly aligned with the assumed runoff mechanism, the model struggles.
I strongly recommend that the author: 1. Substantially tone down claims regarding superiority or robustness in data-scarce regions, or 2. Provide explicit experiments demonstrating performance under reduced training data availability (e.g., leave-one-basin-out, reduced station density, or short-record training).
As written, the claim that Bakaano-Hydro is “particularly well-suited” for data-scarce regions is not adequately justified.
III Structural Dependence on Saturation-Excess Runoff Is a Major Limitation
The manuscript commendably provides an honest diagnosis of model weaknesses, especially the reliance on VegET’s saturation-excess runoff formulation. However, this limitation is not merely a secondary detail—it is structural and foundational.
Key issues: 1. VegET does not represent infiltration-excess (Hortonian) runoff, transmission losses, or event-scale runoff generation. 2. Routing ignores travel time, channel storage, and attenuation. 3. The neural network can only correct patterns that are present in the routed runoff signal; it cannot invent missing hydrological processes.
As a result: 1. The model systematically underperforms in arid, semi-arid, and regulated basins. 2. Peak flows and flash responses are poorly captured. 3. The claim of “hydrology-guided” modeling becomes ambiguous when the guiding physics are incomplete for many real-world systems.
I encourage the author to: 1. More explicitly acknowledge that Bakaano-Hydro is not process-agnostic, but rather strongly tailored to humid, saturation-dominated systems. 2. Reframe the contribution as a successful prototype architecture, rather than a broadly applicable solution. 3. Discuss whether alternative runoff generators (or multiple regimes) could be modularly integrated, and what that would imply for training and interpretability.
IV. Lack of Benchmark Comparisons Limits Interpretability of Performance
While the manuscript presents extensive diagnostic metrics (NSE, KGE, log-transformed variants, decomposition terms), there is no direct comparison to meaningful baselines such as: 1. A lumped LSTM or TCN, 2. A conceptual hydrological model (e.g., VIC model), 3. A routing-only ML model using the same inputs.
Without such benchmarks, it is difficult to assess: 1. How much predictive skill comes from the neural network versus the runoff generator, 2. Whether spatial routing materially improves performance, 3. Whether the additional complexity is justified.
Even a limited comparison in one or two basins would significantly strengthen the manuscript and help readers contextualize the reported skill levels.
V. Conceptual Ambiguity Between “Physical Guidance” and “Preprocessing”
The paper repeatedly describes Bakaano-Hydro as “physically guided” or “physics-informed.” However, in practice: 1. The physics-based components operate entirely upstream of the neural network, 2. There is no physical constraint enforced during learning, 3. The network does not feed back into runoff generation or routing.
This raises a conceptual question: Is Bakaano-Hydro a physics-guided model, or a data-driven model with physically motivated preprocessing?
This distinction matters, especially for readers comparing this framework to: 1. Physics-informed neural networks (PINNs), 2. Differentiable hydrological models, 3. Hybrid models with joint optimization.
Clarifying this distinction would improve conceptual clarity and prevent overinterpretation of the model’s physical grounding.
Minor Comments:
The introduction occasionally overgeneralizes shortcomings of process-based models (e.g., calibration demands, structural uncertainty). These statements would benefit from more nuanced phrasing.
The rationale for choosing a 365-day lookback window is not fully justified. Is this optimal across all basins?
The choice of Hargreaves PET should be briefly justified, given its known limitations in humid or energy-limited environments.
The description of the three-branch vs. two-branch architectures could be streamlined; the practical implications of choosing one over the other remain somewhat unclear.
Figures are generally clear, but some captions (e.g., Fig. 5) are overly dense and could be simplified.
Technical and Editorial Suggestions:
Consider adding a concise table summarizing model assumptions, including runoff generation, routing, and learning components.
Explicitly state computational requirements (runtime, memory) for basin-scale applications.
Minor grammatical issues are present but do not impede understanding.
Ensure consistent use of terminology (e.g., “hydrology-guided,” “physics-based,” “process-based”).
Citation: https://doi.org/10.5194/egusphere-2025-1633-RC2 -
AC2: 'Reply on RC2', Confidence Duku, 23 Jan 2026
RC: This manuscript presents Bakaano-Hydro, a hybrid modeling framework that combines a gridded process-based runoff generation scheme (VegET), topographic flow routing, and a deep learning architecture (TCN with attention and FiLM conditioning) to simulate distributed streamflow. The paper is ambitious in scope, technically detailed, and addresses an important and timely problem in hydrology: how to reconcile physical realism, spatial heterogeneity, and predictive skill within data-driven modeling frameworks.
The manuscript is generally well written, clearly structured, and accompanied by open-source code, which is a strong asset. The author demonstrates a deep understanding of both hydrological theory and modern machine learning architectures. The evaluation across six large basins spanning multiple hydroclimatic regimes is a notable strength, as is the explicit diagnostic discussion of where and why the model fails.
However, despite these strengths, I have substantial concerns regarding the framing of novelty, the strength of some claims (especially concerning data-scarce regions), the choice and rigidity of the runoff generation mechanism, and the lack of comparison against relevant baselines. In its current form, the manuscript would benefit from significant revision to better position Bakaano-Hydro within the rapidly evolving literature on physics-guided and hybrid hydrological machine learning, and to more carefully delimit the conditions under which the model is genuinely advantageous.
Overall, I believe this work has clear potential for publication, but major revisions are required before it can be considered for acceptance.
AC: I thank the reviewer for their careful reading of my manuscript and for their constructive comments and suggestions. In the following, I respond to each comment point by point. All revisions indicated will be incorporated in the revised manuscript.
RC: The manuscript repeatedly emphasizes that Bakaano-Hydro addresses limitations of “state-of-the-art data-driven hydrological models” by incorporating spatial heterogeneity and hydrological connectivity. While this motivation is valid, the literature review and framing do not sufficiently acknowledge how much progress has already been made in this direction.
In recent years, numerous studies have: Incorporated physical constraints directly into neural networks; Used distributed or graph-based representations of river networks; Combined conceptual or process-based runoff modules with ML-based routing or correction layers; Explicitly targeted spatial generalization and ungauged basins. As a result, statements implying that most existing data-driven models are lumped and physically uninterpretable are overly broad and somewhat outdated.
AC: The statements highlighted by the reviewer are motivated by the observation that a substantial portion of data-driven approaches to streamflow prediction still rely on lumped basin representations, particularly in large-scale and operational settings. Nonetheless, I fully acknowledge that important advances have been made in recent years toward incorporating spatial heterogeneity, hydrological connectivity, and physical constraints into machine-learning-based hydrological models. At lines 67–81, I synthesize this progress by summarizing existing approaches that explicitly address spatial structure and connectivity, while also highlighting remaining gaps that Bakaano-Hydro is designed to help close. In response to the reviewer’s comment, I will revise the introduction to more explicitly reflect these advances and to refine the framing so that it avoids overly broad generalizations, while clearly positioning Bakaano-Hydro within the evolving landscape of hybrid and distributed hydrological modeling.
RC : The manuscript would benefit from: 1. A clearer comparison between Bakaano-Hydro and other physics-guided or hybrid frameworks, not only lumped LSTM baselines. 2. Explicit discussion of why Bakaano-Hydro’s serial hybridization (VegET → routing → TCN) offers conceptual or practical advantages over alternative coupling strategies. 3. Without this, the novelty risks being perceived as incremental rather than transformative, especially given that the runoff generation and routing components are externally imposed rather than learned or dynamically coupled.
AC: I agree that positioning Bakaano-Hydro relative to other modeling frameworks is important. In the revised version, I will include a benchmarking strategy designed to compare Bakaano-Hydro against widely used, open-access large-scale models for which data, assumptions, and evaluation protocols are transparent and reproducible and which represent relevant and practical reference points for continental-scale flood forecasting and streamflow simulation in data-scarce regions.
At the same time, it is not my intention to exhaustively compare Bakaano-Hydro against the full spectrum of existing data-driven model variants, given the substantial heterogeneity in architectures, data requirements, and training protocols across the literature. Instead, I emphasize rigorous validation and cross-validation experiments using a diverse set of widely adopted hydrological performance metrics across multiple hydroclimatic regimes. Such evaluation strategies are commonly used to assess reliability, generalization, and operational usefulness, particularly when the focus is on scalability and real-world applicability rather than architectural optimization.
In response to the reviewer’s second point, I will include explicit discussion of the conceptual and practical rationale behind the serial hybridization adopted in Bakaano-Hydro (VegET → routing → TCN). Specifically, I will clarify how this structure enables physically interpretable runoff generation, explicit hydrological connectivity through routing, and learning by neural networks,
RC: The manuscript repeatedly argues that Bakaano-Hydro is well suited for data-scarce regions and may outperform traditional process-based models under such conditions. However, the presented results do not convincingly support this claim.
Key concerns: 1. All experiments rely on GRDC stations with at least three years of data, many of them in well-monitored basins. 2. The neural network is trained on multi-station, multi-year observations, which are precisely what data-scarce regions lack. 3. Performance degrades substantially in arid and semi-arid basins (e.g., Orange), which are often the most data-scarce and management-critical regions. In fact, the results suggest the opposite: where hydrology is complex, threshold-driven, or poorly aligned with the assumed runoff mechanism, the model struggles. I strongly recommend that the author: 1. Substantially tone down claims regarding superiority or robustness in data-scarce regions, or 2. Provide explicit experiments demonstrating performance under reduced training data availability (e.g., leave-one-basin-out, reduced station density, or short-record training). As written, the claim that Bakaano-Hydro is “particularly well-suited” for data-scarce regions is not adequately justified.
AC: I acknowledge the reviewer’s point that, in the current manuscript, the claim that Bakaano-Hydro is particularly well suited for data-scarce regions is implied rather than explicitly supported by targeted experimental evidence. I also agree that this claim requires stronger empirical justification. In the revised version, I will therefore include additional cross-validation experiments explicitly designed to assess model performance under reduced data availability, as well as expanded benchmarking against widely used open-access large-scale models. These additions are intended to directly test the robustness and generalization behavior of Bakaano-Hydro under data-scarce conditions.
I would, however, like to clarify two points regarding the interpretation of the current results. First, the assertion that arid and semi-arid basins are necessarily the most data-scarce is not fully accurate in the context of the GRDC database used in this study. Several semi-arid to arid basins included in our analysis—such as the Orange, Zambezi, and Okavango—are relatively well monitored and contain long, continuous gauge records. In contrast, some of the most data-scarce regions represented in the GRDC archive are semi-humid to humid basins in West, Central, and parts of Eastern Africa (e.g., Volta, Congo), where station density and record continuity are substantially lower.
Second, we respectfully push back on the interpretation that reduced performance in certain basins implies that the model fundamentally struggles “where hydrology is complex or poorly aligned with the assumed runoff mechanism.” The runoff generation phase of Bakaano-Hydro, like most process based hydrological models, cannot capture the full range of hydrological processes or threshold-driven dynamics in all environments. Rather, it serves as a physically constrained, parsimonious runoff approximation that provides a consistent baseline for routing and subsequent learning by the neural network. Performance degradation in specific settings therefore reflects the known limitations of simplified runoff representations, rather than a failure of the hybrid modeling concept itself. I should emphasize here also that the modularity of Bakaano-hydro architecture also allows the VegET runoff generation to be replaced by another runoff generation process without affecting learning. Obviously these are features that can be included in future versions.
In response to the reviewer’s recommendation, I will refine the wording of the manuscript to avoid overgeneralized claims of superiority in data-scarce regions and instead ground such statements in the outcomes of the newly added experiments. This revision will ensure that claims regarding applicability and robustness are directly supported by evidence presented in the paper.
RC: The manuscript commendably provides an honest diagnosis of model weaknesses, especially the reliance on VegET’s saturation-excess runoff formulation. However, this limitation is not merely a secondary detail—it is structural and foundational.
Key issues: 1. VegET does not represent infiltration-excess (Hortonian) runoff, transmission losses, or event-scale runoff generation. 2. Routing ignores travel time, channel storage, and attenuation. 3. The neural network can only correct patterns that are present in the routed runoff signal; it cannot invent missing hydrological processes.
As a result: 1. The model systematically underperforms in arid, semi-arid, and regulated basins. 2. Peak flows and flash responses are poorly captured. 3. The claim of “hydrology-guided” modeling becomes ambiguous when the guiding physics are incomplete for many real-world systems.
I encourage the author to: 1. More explicitly acknowledge that Bakaano-Hydro is not process-agnostic, but rather strongly tailored to humid, saturation-dominated systems. 2. Reframe the contribution as a successful prototype architecture, rather than a broadly applicable solution. 3. Discuss whether alternative runoff generators (or multiple regimes) could be modularly integrated, and what that would imply for training and interpretability.`
AC: I thank the reviewer for the detailed assessment of the VegET-based runoff formulation. However, I disagree with the interpretation that the identified limitations are structural in a way that constrains the broader applicability or conceptual validity of Bakaano-Hydro. Bakaano-Hydro employs a serial hybridization strategy in which runoff generation, routing, and learning are explicitly separated into modular components. This design is intentional and enables modularity and substitutability of individual components without altering the overall framework. In particular, the runoff-generation module is not fixed to VegET: it serves as one physically constrained instantiation that can be replaced with alternative, regime-specific formulations (e.g., infiltration-excess–dominated, transmission-loss–aware, or regulation-aware runoff schemes) where hydrological context or application demands it. The current implementation reflects a pragmatic choice aligned with large-scale, daily simulations and available data, rather than an assumption of universal process dominance. Importantly, the serial hybrid architecture of Bakaano-Hydro ensures that the learning component operates on routed runoff signals while remaining agnostic to the specific runoff generator employed, thereby preserving interpretability and physical consistency even when alternative formulations are substituted.
I will revise the manuscript to clarify these points.
RC: While the manuscript presents extensive diagnostic metrics (NSE, KGE, log-transformed variants, decomposition terms), there is no direct comparison to meaningful baselines such as: 1. A lumped LSTM or TCN, 2. A conceptual hydrological model (e.g., VIC model), 3. A routing-only ML model using the same inputs.
Without such benchmarks, it is difficult to assess: 1. How much predictive skill comes from the neural network versus the runoff generator, 2. Whether spatial routing materially improves performance, 3. Whether the additional complexity is justified.
Even a limited comparison in one or two basins would significantly strengthen the manuscript and help readers contextualize the reported skill levels.
AC. I agree with the reviewer that explicit benchmarking against meaningful baselines would substantially strengthen the manuscript and improve the interpretability of the reported performance. In the revised version, I will therefore include additional benchmark experiments to more clearly contextualize the predictive skill of Bakaano-Hydro.
RC: The paper repeatedly describes Bakaano-Hydro as “physically guided” or “physics-informed.” However, in practice: 1. The physics-based components operate entirely upstream of the neural network, 2. There is no physical constraint enforced during learning, 3. The network does not feed back into runoff generation or routing. This raises a conceptual question: Is Bakaano-Hydro a physics-guided model, or a data-driven model with physically motivated preprocessing? This distinction matters, especially for readers comparing this framework to: 1. Physics-informed neural networks (PINNs), 2. Differentiable hydrological models, 3. Hybrid models with joint optimization.
Clarifying this distinction would improve conceptual clarity and prevent overinterpretation of the model’s physical grounding.
AC: I thank the reviewer for raising this important point of conceptual clarity. I would like to clarify that nowhere in the manuscript is Bakaano-Hydro described as a “physics-informed” or “physically guided” neural network in the sense commonly associated with PINNs, differentiable hydrological models, or joint physics–learning optimization frameworks. Instead, I consistently refer to Bakaano-Hydro as a distributed hydrology-guided neural network approach, and I agree that making this distinction more explicit will improve clarity.
The key difference lies in where and how hydrological structure is introduced. In Bakaano-Hydro, physical hydrological reasoning is embedded upstream of the neural network through explicit runoff generation and routing, which transform dynamic meteorological and static biophysical inputs into spatially connected hydrological signals. The neural network then operates on these routed signals for learning. This is conceptually different from physics-informed learning, where physical laws are enforced as constraints during optimization, or from jointly optimized hybrid models where physical and learned components are trained end-to-end.
In simple terms, while lumped data-driven models typically feed catchment-aggregated meteorological and static descriptors directly into a sequential neural network, Bakaano-Hydro first passes these inputs through a distributed runoff generation and routing scheme to explicitly represent hydrological connectivity and spatial structure before learning. The model is therefore hydrology-guided in its architecture and data flow, rather than physics-informed in the optimization sense.
I agree with the reviewer that this distinction matters, particularly for readers comparing Bakaano-Hydro to PINNs, differentiable hydrological models, or hybrid frameworks with joint optimization. In the revised manuscript, I will therefore (i) explicitly clarify this terminology, (ii) state unambiguously that Bakaano-Hydro does not enforce physical constraints during learning nor perform feedback from the neural network to the runoff or routing components, and (iii) position the framework as a distributed, hydrology-guided preprocessing and learning pipeline rather than a physics-informed or fully coupled hybrid model.
RC: The introduction occasionally overgeneralizes shortcomings of process-based models (e.g., calibration demands, structural uncertainty). These statements would benefit from more nuanced phrasing.
AC: I will revise the introduction accordingly.
RC: The rationale for choosing a 365-day lookback window is not fully justified. Is this optimal across all basins? The choice of Hargreaves PET should be briefly justified, given its known limitations in humid or energy-limited environments.
AC: The length of the look-back window is a user-defined hyperparameter in Bakaano-Hydro rather than a fixed or universally optimal choice. In the current study, a 365-day look-back window was selected for the case studies to capture seasonal and interannual hydrological memory and to enable comparison with commonly used configurations in the literature (e.g., Kratzert et al., 2018). diverse hydroclimatic regimes.
I agree that the choice of the Hargreaves formulation for potential evapotranspiration (PET) should be more explicitly justified. In the revised manuscript, I will briefly clarify that Hargreaves PET was selected due to its low data requirements and robustness in data-scarce settings, which aligns with the large-scale and multi-regional scope of the study. I will also acknowledge its known limitations in humid or energy-limited environments and clarify that this choice represents a pragmatic trade-off rather than an assumption of optimality across all hydroclimatic regimes.
RC: The description of the three-branch vs. two-branch architectures could be streamlined; the practical implications of choosing one over the other remain somewhat unclear.
AC: I will revise the methodologically description accordingly and provide clarity.
RC: Figures are generally clear, but some captions (e.g., Fig. 5) are overly dense and could be simplified. Consider adding a concise table summarizing model assumptions, including runoff generation, routing, and learning components. Explicitly state computational requirements (runtime, memory) for basin-scale applications. Minor grammatical issues are present but do not impede understanding. Ensure consistent use of terminology (e.g., “hydrology-guided,” “physics-based,” “process-based”).
AC: I thank the reviewer for these constructive suggestions. I agree that some figure captions (e.g., Fig. 5) are overly dense and will revise them to improve clarity and readability. In the revised manuscript, I will also add a concise summary table outlining key model assumptions, including runoff generation, routing, and learning components. In addition, I will explicitly report computational requirements (runtime and memory) for representative basin-scale applications to better inform readers about practical applicability.
Citation: https://doi.org/10.5194/egusphere-2025-1633-AC2
-
AC2: 'Reply on RC2', Confidence Duku, 23 Jan 2026
-
EC1: 'Comment on egusphere-2025-1633 (Anonymous Referee #3)', Charles Onyutha, 18 Jan 2026
Dear Author,
An anonymous Referee #3 submitted the below set of comments that you should consider.
1. The manuscript proposes Bakaano-Hydro, a fully distributed hybrid framework for streamflow prediction that incorporates hydrology guidance into deep learning.
2. However, the current manuscript does not include adequate comparisons with (i) purely machine-learning approaches and (ii) process-based hydrologic models. This missing benchmarking makes it difficult to quantify the added value of the proposed hybrid framework and to demonstrate its significance relative to established alternatives.
Citation: https://doi.org/10.5194/egusphere-2025-1633-EC1 -
AC3: 'Reply on EC1', Confidence Duku, 23 Jan 2026
RC: The manuscript proposes Bakaano-Hydro, a fully distributed hybrid framework for streamflow prediction that incorporates hydrology guidance into deep learning. However, the current manuscript does not include adequate comparisons with (i) purely machine-learning approaches and (ii) process-based hydrologic models. This missing benchmarking makes it difficult to quantify the added value of the proposed hybrid framework and to demonstrate its significance relative to established alternatives.
AC: I agree that positioning Bakaano-Hydro relative to other models is important. In the revised version, I will include a benchmarking strategy designed to compare Bakaano-Hydro against widely used, open-access models for which data, assumptions, and evaluation protocols are transparent and reproducible and which represent relevant and practical reference points.
At the same time, it is not my intention to exhaustively compare Bakaano-Hydro against the full spectrum of existing data-driven and physically-based model variants, given the substantial heterogeneity in architectures, data requirements, and training protocols across the literature. Instead, I emphasize rigorous validation and cross-validation experiments using a diverse set of widely adopted hydrological performance metrics across multiple hydroclimatic regimes. Such evaluation strategies are commonly used to assess reliability, generalization, and operational usefulness, particularly when the focus is on scalability and real-world applicability rather than architectural optimization.
Citation: https://doi.org/10.5194/egusphere-2025-1633-AC3
-
AC3: 'Reply on EC1', Confidence Duku, 23 Jan 2026
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 2,014 | 149 | 24 | 2,187 | 31 | 43 |
- HTML: 2,014
- PDF: 149
- XML: 24
- Total: 2,187
- BibTeX: 31
- EndNote: 43
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
General comments
The pre‑print presents Bakaano‑Hydro (v1.1), a fully distributed hybrid framework combining VegET‑based grid‑cell runoff generation, MFD routing and a Temporal Convolutional Network (TCN) with attention + FiLM conditioning. The code is open, the design modular and the evaluation spans six hydro‑climatic basins.
The chief shortcoming is the lack of an empirical benchmark against the data‑driven approaches that motivate the study. At minimum the authors should compare against (i) a lumped LSTM trained on catchment‑aggregated forcings and (ii) ideally a Conv‑LSTM fed with the same gridded inputs; a physics‑only baseline (VegET + routing) would further contextualise gains. Without these, neither the added predictive value nor computational overhead of the proposed architecture can be quantified.
Specific comments
Technical corrections