Martin Gauch
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Martin Gauch has a total of 145 co-authors. Here are the top co-authors by number of shared publications:
- Frederik Kratzert
Google research
77 shared publications - Daniel Klotz
Interdisciplinary Transformation University Austria, Linz, Austria
70 shared publications Grey Nearing
37 shared publicationsOren Gilon
Google Research, Electra Tower, Yigal Alon St 98, Tel Aviv-Yafo, 6789141, Israel
33 shared publicationsSepp Hochreiter
ELLIS Unit Linz and LIT AI Lab, Institute for Machine Learning, Johannes Kepler University Linz, Linz, Austria
18 shared publicationsGuy Shalev
Google DeepMind, Electra Tower, Yigal Alon St 98, Tel Aviv-Yafo, 6789141, Israel
16 shared publications
Latest Publications
- Open Access
- 18 September 2026
- Published by California Digital Library (CDL) in EarthArXiv
A Global Event-Based Benchmark for Tide and Surge Models Beyond the Tide Gauge
Coastal flood hindcasting and forecasting form the foundation of effective risk management in coastal regions. They drive emergency response efforts and inform the design of coastal defenses, adaptation strategies, and evidence-based policy. Global coastal sea level models are traditionally validated using time-series regression metrics (i.e., Root Mean Square Error [RMSE], Mean Absolute Error [MAE], Pearson correlation) at tide gauge sites. However, this approach has two main drawbacks: (1) gauge coverage is geographically sparse outside North America, Northern Europe, and limited parts of Asia, and (2) spatially aggregated metrics obscure prediction errors during extreme sea level events. Here we introduce a novel event-based framework that benchmarks models against a global database of news-reported flood events captured in the Groundsource database. Our curated benchmark of ~33,000 global coastal flood records across 10 years was constructed using a Large Language Model (LLM) calibrated with tide-gauge data to identify coastal floods. Evaluating the performance of two global storm tide reanalyses, namely Global Tide and Surge Model (GTSM) simulations and Environment and Climate Change Canada's (ECCC) hindcast, against our curated flood database, reveals detection rates of only 39% (GTSM) and 43% (ECCC), with nearly half of all flood reportings missed by the two models. Further, we show that spatial aggregation, such as averaging RMSE across the gauges of a country, can mask flood impacts, whereas per-gauge extreme-quantile errors align more closely with performance measured on records of actual floods. This framework offers an essential complementary lens for assessing model quality by providing broad global coverage beyond gauge locations.
- 18 September 2026
- Published by California Digital Library (CDL) in EarthArXiv
Open AccessA Global Event-Based Benchmark for Tide and Surge Models Beyond the Tide Gauge
Coastal flood hindcasting and forecasting form the foundation of effective risk management in coastal regions. They drive emergency response efforts and inform the design of coastal defenses, adaptation strategies, and evidence-based policy. Global coastal sea level models are traditionally validated using time-series regression metrics (i.e., Root Mean Square Error [RMSE], Mean Absolute Error [MAE], Pearson correlation) at tide gauge sites. However, this approach has two main drawbacks: (1) gauge coverage is geographically sparse outside North America, Northern Europe, and limited parts of Asia, and (2) spatially aggregated metrics obscure prediction errors during extreme sea level events. Here we introduce a novel event-based framework that benchmarks models against a global database of news-reported flood events captured in the Groundsource database. Our curated benchmark of ~33,000 global coastal flood records across 10 years was constructed using a Large Language Model (LLM) calibrated with tide-gauge data to identify coastal floods. Evaluating the performance of two global storm tide reanalyses, namely Global Tide and Surge Model (GTSM) simulations and Environment and Climate Change Canada's (ECCC) hindcast, against our curated flood database, reveals detection rates of only 39% (GTSM) and 43% (ECCC), with nearly half of all flood reportings missed by the two models. Further, we show that spatial aggregation, such as averaging RMSE across the gauges of a country, can mask flood impacts, whereas per-gauge extreme-quantile errors align more closely with performance measured on records of actual floods. This framework offers an essential complementary lens for assessing model quality by providing broad global coverage beyond gauge locations.
- Open Access
- 3 September 2026
- Published by Copernicus GmbH in EGUsphere
- Vol. 2026, 1-21
Everything everywhere all at once: A single-cell LSTM network unifying multi-frequency, missing data, and discharge assimilation for robust operational flood forecasting
Long short-term memory (LSTM) networks have demonstrated state-of-the-art capabilities in operational flood forecasting, with recent missing-data handling strategies further improving pipeline robustness. Building on these advancements, we introduce the multi-frequency masked-forecasting LSTM (MF2LSTM), a model architecture that combines missing-data workflows with multi-frequency approaches to generate hourly streamflow forecasts. Additionally, our framework features a flexible data assimilation strategy to include real-time discharge information. This approach enhances forecast performance when real-time observations are available, and allows the model to keep operating when the discharge signal is absent, either by gauge failure or for prediction in ungauged basins. We benchmarked the MF2LSTM against the Large Area Runoff Simulation (LARSIM) model, the current operational flood-forecasting model used by several European countries, forcing both models with ICON-D2 meteorological products. Our results indicate that the MF2LSTM yields higher predictive accuracy than LARSIM. Overall, this framework presents a robust operational pipeline that demonstrates the viability of deep learning for hourly forecasting. Furthermore, by relying on a standard single-cell LSTM architecture, the approach highlights that structurally simple deep learning architectures can achieve high operation performance.
- 3 September 2026
- Published by Copernicus GmbH in EGUsphere
- Vol. 2026, 1-21
Open AccessEverything everywhere all at once: A single-cell LSTM network unifying multi-frequency, missing data, and discharge assimilation for robust operational flood forecasting
Long short-term memory (LSTM) networks have demonstrated state-of-the-art capabilities in operational flood forecasting, with recent missing-data handling strategies further improving pipeline robustness. Building on these advancements, we introduce the multi-frequency masked-forecasting LSTM (MF2LSTM), a model architecture that combines missing-data workflows with multi-frequency approaches to generate hourly streamflow forecasts. Additionally, our framework features a flexible data assimilation strategy to include real-time discharge information. This approach enhances forecast performance when real-time observations are available, and allows the model to keep operating when the discharge signal is absent, either by gauge failure or for prediction in ungauged basins. We benchmarked the MF2LSTM against the Large Area Runoff Simulation (LARSIM) model, the current operational flood-forecasting model used by several European countries, forcing both models with ICON-D2 meteorological products. Our results indicate that the MF2LSTM yields higher predictive accuracy than LARSIM. Overall, this framework presents a robust operational pipeline that demonstrates the viability of deep learning for hourly forecasting. Furthermore, by relying on a standard single-cell LSTM architecture, the approach highlights that structurally simple deep learning architectures can achieve high operation performance.
- Open Access
- 27 July 2026
- Published by Copernicus GmbH in Earth System Science Data
- Vol. 18 (7) , 5485-5504
EARLS: a runoff reconstruction dataset for Europe
Data drives our understanding of hydrological processes, supports model development, and enables anticipatory water management. This contribution introduces EARLS: European Aggregated Reconstructions for Large-sample Studies. EARLS offers daily streamflow reconstructions for more than 10 000 basins in Europe including uncertainty estimates, covering the period from 1953 to 2023. The reconstruction is derived from a single Long Short-Term Memory (LSTM) based rainfall–runoff model trained on more than 5000 basins. LSTMs represent the state of the art in rainfall–runoff modeling and are well suited to provide predictions in ungauged basins. We evaluate the quality of the reconstruction through quantitative evaluation on two held-out sets of basins and by conducting a qualitative assessment that compares EARLS-based peak flows and flood timing to previous large-scale hydrological studies. EARLS represents a new generation of datasets that harness the capabilities of Deep Learning to obtain accurate and high-resolution data. EARLS is available at https://doi.org/10.5281/zenodo.13864842 (Klotz et al., 2026b).
- 27 July 2026
- Published by Copernicus GmbH in Earth System Science Data
- Vol. 18 (7) , 5485-5504
Open AccessEARLS: a runoff reconstruction dataset for Europe
Data drives our understanding of hydrological processes, supports model development, and enables anticipatory water management. This contribution introduces EARLS: European Aggregated Reconstructions for Large-sample Studies. EARLS offers daily streamflow reconstructions for more than 10 000 basins in Europe including uncertainty estimates, covering the period from 1953 to 2023. The reconstruction is derived from a single Long Short-Term Memory (LSTM) based rainfall–runoff model trained on more than 5000 basins. LSTMs represent the state of the art in rainfall–runoff modeling and are well suited to provide predictions in ungauged basins. We evaluate the quality of the reconstruction through quantitative evaluation on two held-out sets of basins and by conducting a qualitative assessment that compares EARLS-based peak flows and flood timing to previous large-scale hydrological studies. EARLS represents a new generation of datasets that harness the capabilities of Deep Learning to obtain accurate and high-resolution data. EARLS is available at https://doi.org/10.5281/zenodo.13864842 (Klotz et al., 2026b).
- Open Access
- 8 June 2026
- Published by Copernicus GmbH
- Vol. 2026, 1-31
Comment on egusphere-2026-2283 - No compliance with the policy of the journal
This paper evaluates an updated flood forecasting system that significantly extends reliable lead times. We evaluated this updated model (v2) against the prior system (v1) and established third-party benchmarks across 1,223 global test basins. The primary finding is that the v2 system extends the reliable predictive horizon by 6 days in gauged basins and 1 day in ungauged basins relative to the v1 nowcast, as measured by the Nash Sutcliffe Efficiency. Along with this paper, we release an open-source codebase for training both the v1 and v2 forecast models with the open-source Caravan dataset.
- 8 June 2026
- Published by Copernicus GmbH
- Vol. 2026, 1-31
Open AccessComment on egusphere-2026-2283 - No compliance with the policy of the journal
This paper evaluates an updated flood forecasting system that significantly extends reliable lead times. We evaluated this updated model (v2) against the prior system (v1) and established third-party benchmarks across 1,223 global test basins. The primary finding is that the v2 system extends the reliable predictive horizon by 6 days in gauged basins and 1 day in ungauged basins relative to the v1 nowcast, as measured by the Nash Sutcliffe Efficiency. Along with this paper, we release an open-source codebase for training both the v1 and v2 forecast models with the open-source Caravan dataset.
- Open Access
- 28 May 2026
- Published by Copernicus GmbH
- Vol. 2026, 1-31
Comment on egusphere-2026-2283
This paper evaluates an updated flood forecasting system that significantly extends reliable lead times. We evaluated this updated model (v2) against the prior system (v1) and established third-party benchmarks across 1,223 global test basins. The primary finding is that the v2 system extends the reliable predictive horizon by 6 days in gauged basins and 1 day in ungauged basins relative to the v1 nowcast, as measured by the Nash Sutcliffe Efficiency. Along with this paper, we release an open-source codebase for training both the v1 and v2 forecast models with the open-source Caravan dataset.
- 28 May 2026
- Published by Copernicus GmbH
- Vol. 2026, 1-31
Open AccessComment on egusphere-2026-2283
This paper evaluates an updated flood forecasting system that significantly extends reliable lead times. We evaluated this updated model (v2) against the prior system (v1) and established third-party benchmarks across 1,223 global test basins. The primary finding is that the v2 system extends the reliable predictive horizon by 6 days in gauged basins and 1 day in ungauged basins relative to the v1 nowcast, as measured by the Nash Sutcliffe Efficiency. Along with this paper, we release an open-source codebase for training both the v1 and v2 forecast models with the open-source Caravan dataset.
- Open Access
- 25 May 2026
- Published by Copernicus GmbH
- Vol. 2026, 1-31
Comment on egusphere-2026-2283
This paper evaluates an updated flood forecasting system that significantly extends reliable lead times. We evaluated this updated model (v2) against the prior system (v1) and established third-party benchmarks across 1,223 global test basins. The primary finding is that the v2 system extends the reliable predictive horizon by 6 days in gauged basins and 1 day in ungauged basins relative to the v1 nowcast, as measured by the Nash Sutcliffe Efficiency. Along with this paper, we release an open-source codebase for training both the v1 and v2 forecast models with the open-source Caravan dataset.
- 25 May 2026
- Published by Copernicus GmbH
- Vol. 2026, 1-31
Open AccessComment on egusphere-2026-2283
This paper evaluates an updated flood forecasting system that significantly extends reliable lead times. We evaluated this updated model (v2) against the prior system (v1) and established third-party benchmarks across 1,223 global test basins. The primary finding is that the v2 system extends the reliable predictive horizon by 6 days in gauged basins and 1 day in ungauged basins relative to the v1 nowcast, as measured by the Nash Sutcliffe Efficiency. Along with this paper, we release an open-source codebase for training both the v1 and v2 forecast models with the open-source Caravan dataset.
- Open Access
- 29 April 2026
- Published by Copernicus GmbH in EGUsphere
- Vol. 2026, 1-31
Extending Medium-Range Global Flood Forecasts: The Google Global Flood Forecasting Model Version 2
This paper evaluates an updated flood forecasting system that significantly extends reliable lead times. We evaluated this updated model (v2) against the prior system (v1) and established third-party benchmarks across 1,223 global test basins. The primary finding is that the v2 system extends the reliable predictive horizon by 6 days in gauged basins and 1 day in ungauged basins relative to the v1 nowcast, as measured by the Nash Sutcliffe Efficiency. Along with this paper, we release an open-source codebase for training both the v1 and v2 forecast models with the open-source Caravan dataset.
- 29 April 2026
- Published by Copernicus GmbH in EGUsphere
- Vol. 2026, 1-31
Open AccessExtending Medium-Range Global Flood Forecasts: The Google Global Flood Forecasting Model Version 2
This paper evaluates an updated flood forecasting system that significantly extends reliable lead times. We evaluated this updated model (v2) against the prior system (v1) and established third-party benchmarks across 1,223 global test basins. The primary finding is that the v2 system extends the reliable predictive horizon by 6 days in gauged basins and 1 day in ungauged basins relative to the v1 nowcast, as measured by the Nash Sutcliffe Efficiency. Along with this paper, we release an open-source codebase for training both the v1 and v2 forecast models with the open-source Caravan dataset.
- Open Access
- 14 March 2026
- Published by Copernicus GmbH
Towards Deep Learning Models for Global Coastal Sea Level Prediction
Accurate sea level prediction is crucial for coastal communities and infrastructure. Unfortunately, classical approaches can be computationally expensive and inaccurate, especially in data-sparse regions. Recently, scientists have started to explore to which degree deep learning models could help address these problems. Combined with long historical tidal gauge records, these data-driven approaches offer the potential to improve the accuracy and spatial resolution compared to classical modeling schemes, while at the same time being computationally far cheaper to operate.Yet, deep learning for sea level modeling is still in its infancy: existing studies typically struggle to accurately predict extreme events, differ in their strategies of (pre-)processing input and output data, and focus on individual gauges or small regions. In this contribution, we take inspiration from recent successes in global riverine hydrologic modeling, where deep learning models greatly improved the accuracy of predictions in ungauged regions. Translating these findings to a coastal floods context, we present our work towards globally applicable deep learning models of sea level and storm surge. Specifically, we focus on these models’ ability to make predictions in places that lack historical gauge measurements.Our framework utilizes state-of-the-art deep learning architectures to capture complex spatial dependencies between atmospheric drivers and the ocean state. To ensure robust performance, we integrate a diverse set of input features, including ERA5 atmospheric reanalysis (wind and pressure), FES tidal predictions, and high-resolution static geospatial data such as bathymetry and land-sea masks. Furthermore, we explore the utility of pre-trained geospatial embedding data to encode local station properties.We compare the data-driven predictions with established hydrodynamic model baselines, such as the ocean model run by Environment and Climate Change Canada (ECCC) and the Global Tide and Surge Model (GTSM). Our findings indicate that deep learning approaches can exceed the performance of physics-based models across standard metrics. Furthermore, validation against real-world extreme events confirms the model's superior ability to identify high-impact storm surges.
- 14 March 2026
- Published by Copernicus GmbH
Open AccessTowards Deep Learning Models for Global Coastal Sea Level Prediction
Accurate sea level prediction is crucial for coastal communities and infrastructure. Unfortunately, classical approaches can be computationally expensive and inaccurate, especially in data-sparse regions. Recently, scientists have started to explore to which degree deep learning models could help address these problems. Combined with long historical tidal gauge records, these data-driven approaches offer the potential to improve the accuracy and spatial resolution compared to classical modeling schemes, while at the same time being computationally far cheaper to operate.Yet, deep learning for sea level modeling is still in its infancy: existing studies typically struggle to accurately predict extreme events, differ in their strategies of (pre-)processing input and output data, and focus on individual gauges or small regions. In this contribution, we take inspiration from recent successes in global riverine hydrologic modeling, where deep learning models greatly improved the accuracy of predictions in ungauged regions. Translating these findings to a coastal floods context, we present our work towards globally applicable deep learning models of sea level and storm surge. Specifically, we focus on these models’ ability to make predictions in places that lack historical gauge measurements.Our framework utilizes state-of-the-art deep learning architectures to capture complex spatial dependencies between atmospheric drivers and the ocean state. To ensure robust performance, we integrate a diverse set of input features, including ERA5 atmospheric reanalysis (wind and pressure), FES tidal predictions, and high-resolution static geospatial data such as bathymetry and land-sea masks. Furthermore, we explore the utility of pre-trained geospatial embedding data to encode local station properties.We compare the data-driven predictions with established hydrodynamic model baselines, such as the ocean model run by Environment and Climate Change Canada (ECCC) and the Global Tide and Surge Model (GTSM). Our findings indicate that deep learning approaches can exceed the performance of physics-based models across standard metrics. Furthermore, validation against real-world extreme events confirms the model's superior ability to identify high-impact storm surges.
- Open Access
- 14 March 2026
- Published by Copernicus GmbH
How to deal w___ missing input data
Hydrologic deep learning models have made their way from research to applications. More and more national hydrometeorological agencies, hydro power operators, and consulting companies are building Long Short-Term Memory (LSTM) models for operational use cases. However, all of these efforts are confronted with similar sets of challenges—issues that are different from those in controlled scientific studies. One common issue is the question: how to deal with missing input data? Operational systems depend on the real-time availability of various data products—most notably, meteorological forcings. Additional forcings generally improve the model performance, but at the same time, every new dependency increases the likelihood of an outage in one of the input data products. In a recent study, we evaluated different solutions to generate predictions even when some of the meteorological input data do not arrive in time, or not arrive at all (Gauch et al., 2025). In this presentation, we will introduce these methods and discuss how they can help (1) operational forecasters to run reliable real-time flood forecasting systems, and (2) researchers and modelers to build accurate models that leverage as much data as possible. Gauch, M., Kratzert, F., Klotz, D., Nearing, G., Cohen, D., and Gilon, O.: How to deal w___ missing input data, Hydrol. Earth Syst. Sci., 29, 6221–6235, 2025.
- 14 March 2026
- Published by Copernicus GmbH
Open AccessHow to deal w___ missing input data
Hydrologic deep learning models have made their way from research to applications. More and more national hydrometeorological agencies, hydro power operators, and consulting companies are building Long Short-Term Memory (LSTM) models for operational use cases. However, all of these efforts are confronted with similar sets of challenges—issues that are different from those in controlled scientific studies. One common issue is the question: how to deal with missing input data? Operational systems depend on the real-time availability of various data products—most notably, meteorological forcings. Additional forcings generally improve the model performance, but at the same time, every new dependency increases the likelihood of an outage in one of the input data products. In a recent study, we evaluated different solutions to generate predictions even when some of the meteorological input data do not arrive in time, or not arrive at all (Gauch et al., 2025). In this presentation, we will introduce these methods and discuss how they can help (1) operational forecasters to run reliable real-time flood forecasting systems, and (2) researchers and modelers to build accurate models that leverage as much data as possible. Gauch, M., Kratzert, F., Klotz, D., Nearing, G., Cohen, D., and Gilon, O.: How to deal w___ missing input data, Hydrol. Earth Syst. Sci., 29, 6221–6235, 2025.
- Open Access
- 14 March 2026
- Published by Copernicus GmbH
Mapping riverine heatwave trends across Europe using reconstructed river temperature time-series
Riverine heatwaves, that is, periods of anomalously high water temperature, have become more frequent and intense over the past decades across Europe. To improve our understanding of their spatial distribution and to analyze potential shifts in the causes of such events, we require long time-series of river temperature over a large spatial domain. However, records of measurement stations are often of limited length or, in some regions of Europe, not available at all. To fill these temporal and spatial data gaps, we train a deep learning Long Short-Term Memory (LSTM) model to reconstruct historic time-series (1985-2020) of daily river temperature from meteorological records and catchment characteristics. To train and evaluate the model for the simulation of riverine heatwaves, we use the new TempER (Temperature of European Rivers) dataset that contains over 4000 temperature measurement stations in Europe. TempER covers 26 European countries and contains daily water temperature records of 1 to 72 years length. 48% of the stations additionally provide streamflow observations that we integrate into the model to enhance the fidelity of the reconstructed data. In regions not covered by TempER we use streamflow records from EStreams[1] to guide the temperature simulation. We analyze the reconstructed river temperatures with respect to trends in riverine heatwave characteristics (frequency, duration, intensity) and their spatial distribution across Europe. With our findings we intend to improve the understanding of the hydro-meteorological processes that drive riverine heatwaves and provide a continuous dataset allowing further analysis of river temperatures in Europe. [1] do Nascimento, T.V.M., Rudlang, J., Höge, M. et al. EStreams: An integrated dataset and catalogue of streamflow, hydro-climatic and landscape variables for Europe. Sci Data 11, 879 (2024).
- 14 March 2026
- Published by Copernicus GmbH
Open AccessMapping riverine heatwave trends across Europe using reconstructed river temperature time-series
Riverine heatwaves, that is, periods of anomalously high water temperature, have become more frequent and intense over the past decades across Europe. To improve our understanding of their spatial distribution and to analyze potential shifts in the causes of such events, we require long time-series of river temperature over a large spatial domain. However, records of measurement stations are often of limited length or, in some regions of Europe, not available at all. To fill these temporal and spatial data gaps, we train a deep learning Long Short-Term Memory (LSTM) model to reconstruct historic time-series (1985-2020) of daily river temperature from meteorological records and catchment characteristics. To train and evaluate the model for the simulation of riverine heatwaves, we use the new TempER (Temperature of European Rivers) dataset that contains over 4000 temperature measurement stations in Europe. TempER covers 26 European countries and contains daily water temperature records of 1 to 72 years length. 48% of the stations additionally provide streamflow observations that we integrate into the model to enhance the fidelity of the reconstructed data. In regions not covered by TempER we use streamflow records from EStreams[1] to guide the temperature simulation. We analyze the reconstructed river temperatures with respect to trends in riverine heatwave characteristics (frequency, duration, intensity) and their spatial distribution across Europe. With our findings we intend to improve the understanding of the hydro-meteorological processes that drive riverine heatwaves and provide a continuous dataset allowing further analysis of river temperatures in Europe. [1] do Nascimento, T.V.M., Rudlang, J., Höge, M. et al. EStreams: An integrated dataset and catalogue of streamflow, hydro-climatic and landscape variables for Europe. Sci Data 11, 879 (2024).