MHMarvin Höge

Marvin Höge

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article

Previous Organizations / Institutes

The organizations / institutes from the scholar's other articles in the past ten years


Marvin Höge has a total of 71 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Towards improved Water Quality Modelling using Neural ODE models

    by Marvin Höge, Florian Wenk, Andreas Scheidegger, Carlo Albert, Andreas Frömelt

    Neural Ordinary Differential Equations (ODEs) fuse neural networks with a mechanistic equation framework. This hybrid structure offers both traceability of model states and processes, as it is typical for physics-based models, and the ability of machine learning to encode new functional relations. Neural ODE models have demonstrated high potential in hydrologic predictions and scientific investigation of the related process in the hydrologic cycle, i.e. tasks of water quantity estimation (Höge et al., 2022).This explicit representation of state variables is key to water quality modelling. There, we typically have several interrelated state variables like nitrate, nitrite, phosphorous, organic matter,… Traditionally, these states are modelled based on mechanistic kinetic rate expressions that are often only rough approximations of the underlying dynamics. At the same time, this domain of water research suffers from data scarcity and therefore solely data-driven methods struggle to provide accurate predictions reliably. We show how to improve predictions of state dynamics and to foster knowledge gain about the processes in such interrelated systems with multiple states using Neural ODEs. Höge, M., Scheidegger, A., Baity-Jesi, M., Albert, C., & Fenicia, F.: Improving hydrologic models for predictions and process understanding using Neural ODEs. Hydrol. Earth Syst. Sci., 26, 5085-5102, https://hess.copernicus.org/articles/26/5085/2022/

  • Open Access

    EStreams: Building an integrated dataset of streamflow, hydro-climatic variables and landscape attributes for catchments in Europe

    by Thiago V. M. Do Nascimento, Julia Rudlang, Marvin Höge, Ruud van der Ent, Jan Seibert, Markus Hrachowitz, Fabrizio Fenicia

    High-quality datasets are essential to hydrological analysis​1​. Although many such datasets exist, their accessibility is typically time-consuming and often challenging. Recently, there has been a significant spread of large-sample hydrology (LSH) datasets. Many of these datasets are referred to as Catchment Attributes and MEteorology for Large-sample Studies (CAMELS) or derivations1–4, covering hydro-climatic and landscape static attributes and time series data. These data have collectively been made available5 including first extensionsbased on daily time series such as the Global Runoff Data Base (https://www.bafg.de/GRDC)6. Additionally, there have been collection efforts for global streamflow data indices and signatures7–9. However, such globally accessible dataset represent only a small fraction of what is currently available. Here we present EStreams, a new dataset and data-access catalogue of streamflow, hydro-climatic variables and landscape descriptors for over 15,000 catchments in 39 European countries, set to be released in 2024. The data spans up to 100 years of streamflow data and includes several open-source catchment aggregated landscape attributes on topography, soil, lithology, vegetation, and land cover, as well as climatic forcing and streamflow time-series, hydro-climatic signatures and a catalogue of streamflow providers (“European streamflow data and where to find them”). EStreams offers both an extensive and extensible data collection along with codes for data retrieval, aggregation and processing. Our goal is to extend current large-sample datasets and take a step towards integrating hydro-climatic and landscape data across Europe. References1. Addor, N., Newman, A. J., Mizukami, N. & Clark, M. P. The CAMELS data set: Catchment attributes and meteorology for large-sample studies. Hydrol Earth Syst Sci 21, 5293–5313 (2017).2. Coxon, G. et al. CAMELS-GB: hydrometeorological time series and landscape attributes for 671 catchments in Great Britain. Earth Syst Sci Data 12, 2459–2483 (2020).3. Höge, M. et al. CAMELS-CH: hydro-meteorological time series and landscape attributes for 331 catchments in hydrologic Switzerland. Earth Syst Sci Data 15, 5755–5784 (2023).4. Klingler, C., Schulz, K. & Herrnegger, M. LamaH-CE: LArge-SaMple DAta for Hydrology and Environmental Sciences for Central Europe. Earth Syst Sci Data 13, 4529–4565 (2021).5. Kratzert, F. et al. Caravan - A global community dataset for large-sample hydrology. Scientific Data 2023 10:1 10, 1–11 (2023).6. Färber, C. et al. GRDC-Caravan: extending the original dataset with data from the Global Runoff Data Centre (0.1) [Data set]. Zenodo https://zenodo.org/records/8425587 (2023) doi:10.5281/ZENODO.8425587.7. Do, H. X., Gudmundsson, L., Leonard, M. & Westra, S. The Global Streamflow Indices and Metadata Archive (GSIM)-Part 1: The production of a daily streamflow archive and metadata. Earth Syst Sci Data 10, 765–785 (2018).8. Gudmundsson, L., Do, H. X., Leonard, M. & Westra, S. The Global Streamflow Indices and Metadata Archive (GSIM)-Part 2: Quality control, time-series indices and homogeneity assessment. Earth Syst Sci Data 10, 787–804 (2018).9. Chen, X., Jiang, L., Luo, Y. & Liu, J. A global streamflow indices time series dataset for large-sample hydrological analyses on streamflow regime (until 2022). Earth Syst Sci Data 15, 4463–4479 (2023).

  • Open Access

    Learning Catchment Features with Autoencoders

    by Alberto Bassi, Antonietta Mira, Marvin Höge, Fabrizio Fenicia, Carlo Albert

    By employing Machine Learning techniques on the US-CAMELS dataset, we discern a minimal number of streamflow features. Together with meteorological forcing, these features enable an approximate reconstruction of the entire streamflow time-series. This task is achieved through the application of an explicit noise conditional autoencoder, wherein the meteorological forcing is inputted to the decoder to encourage the encoder to learn streamflow features exclusively related to landscape properties. The optimal number of encoded features is determined with an intrinsic dimension estimator. The accuracy of reconstruction is then compared with models that take a subset of static catchment attributes (both climate and landscape attributes) in addition to meteorological forcing variables. Our findings suggest that attributes gathered by experts encompass nearly all pertinent information regarding the input/output relationship. This information can be succinctly summarized with merely three independent streamflow features. These features exhibit a strong correlation with the baseflow index and aridity indicators, aligning with the observation that predicting streamflow in dry catchments or with a high baseflow index is more challenging. Furthermore, correlation analysis underscores the significance of soil-related and vegetation attributes. These learned features can also be associated with parameters in conceptual hydrological models such as the GR model family.

  • Open Access

    Learning landscape features from streamflow with autoencoders

    by Alberto Bassi, Marvin Höge, Antonietta Mira, Fabrizio Fenicia, Carlo Albert

    Recent successes with machine learning (ML) models in catchment hydrology have highlighted their ability to extract crucial information from catchment properties pertinent to the rainfall–runoff relationship. In this study, we aim to identify a minimal set of catchment signatures in streamflow that, when combined with meteorological drivers, enable an accurate reconstruction of the entire streamflow time series. To achieve this, we utilize an explicit noise-conditional autoencoder (ENCA), which, assuming an optimal architecture, separates the influences of meteorological drivers and catchment properties on streamflow. The ENCA architecture feeds meteorological forcing and climate attributes into the decoder in order to incentivize the encoder to only learn features that are related to landscape properties minimally related to climate. By isolating the effect of meteorology, these hydrological features can thus be interpreted as landscape fingerprints. The optimal number of features is found by means of an intrinsic dimension estimator. We train our model on the hydro-meteorological time series data of 568 catchments of the continental United States from the Catchment Attributes and Meteorology for Large-sample Studies (CAMELS) dataset. We compare the reconstruction accuracy with models that take as input a subset of static catchment attributes (both climate and landscape attributes) along with meteorological forcing variables. Our results suggest that available landscape attributes can be summarized by only two relevant learnt features (or signatures), while at least a third one is needed for about a dozen difficult-to-predict catchments in the central United States, which is mainly characterized by a high aridity index. The principal components of the learnt features strongly correlate with the baseflow index and aridity indicators, which is consistent with the idea that these indicators capture the variability of catchment hydrological responses. The correlation analysis further indicates that soil-related and vegetation attributes are of importance.


  • Open Access

    Metamorphic testing of machine learning and conceptual hydrologic models

    by Peter Reichert, Kai Ma, Marvin Höge, Fabrizio Fenicia, Marco Baity-Jesi, Dapeng Feng, Chaopeng Shen

    Predicting the response of hydrologic systems to modified driving forces beyond patterns that have occurred in the past is of high importance for estimating climate change impacts or the effect of management measures. This kind of prediction requires a model, but the impossibility of testing such predictions against observed data makes it difficult to estimate their reliability. Metamorphic testing offers a methodology for assessing models beyond validation with real data. It consists of defining input changes for which the expected responses are assumed to be known, at least qualitatively, and testing model behavior for consistency with these expectations. To increase the gain of information and reduce the subjectivity of this approach, we extend this methodology to a multi-model approach and include a sensitivity analysis of the predictions to training or calibration options. This allows us to quantitatively analyze differences in predictions between different model structures and calibration options in addition to the qualitative test of the expectations. In our case study, we apply this approach to selected conceptual and machine learning hydrological models calibrated for basins from the CAMELS data set. Our results confirm the superiority of the machine learning models over the conceptual hydrologic models regarding the quality of fit during calibration and validation periods. However, we also find that the response of machine learning models to modified inputs can deviate from the expectations and the magnitude, and even the sign of the response can depend on the training data. In addition, even in cases in which all models passed the metamorphic test, there are cases in which the quantitative response is different for different model structures. This demonstrates the importance of this kind of testing beyond and in addition to the usual calibration–validation analysis to identify potential problems and stimulate the development of improved models.


  • Open Access

    Quantifying Glacier Melt Contribution to Streamflow in Switzerland with CAMELS-CH using Deep Learning

    by Corinna Frank, Marvin Höge

    Meltwater from glaciers is an essential resource of water in Alpine catchments that is however subject to large inter-annual fluctuations due to varying meteorological conditions. To optimize energy production and support water resource management, it is crucial to understand the expected range of arriving glacier melt volumes during the ablation season. Glacier contribution is generally estimated through dynamic glacier models, in-situ isotope measurements or a data-driven approach using hydrological modeling. We present a simplified data-driven approach to indicate the potential of Deep Learning models for the determination of streamflow contributions. A regional LSTM was set up on the recently released CAMELS-CH dataset that includes over 20 catchments with substantial glacier cover (>10% catchment area coverage according to the Swiss glacier inventory GLAMOS). The LSTM was trained on all river catchments with dynamic meteorological forcings and static catchment attributes, including the glacier cover percentage, as inputs. Thereby reaching a median KGE of 0.83 over the test period. In a second test run, the glacier cover percentage was set to zero. Glacier melt time series were then interpreted as the difference in simulated streamflow with respect to the test simulation with the actual glacier cover percentage. The resulting discharge time series exhibits the expected seasonal pattern for glacier-influenced catchments. Removing the glacier cover reduces the simulated streamflow during summer months, with the largest reduction observed around August/September. Simultaneously, an increase in streamflow is observed during spring months, which might be associated with the melt from excess snow that is not assimilated to the glacier during winter. We find relative glacier melt contributions to streamflow of 0% to 60% for the period of 2006-2020, with an average of 5% over all catchments, which are generally higher than contributions reported in literature determined using estimates from commonly used conceptual hydrologic models. This hints at the potential of DL models as an additional method to quantify glacier melt volumes and streamflow contributions. Further, this enables us to investigate glacier melt and to study glacier dynamics in a data-driven way with a minimum set of assumptions in our modeling approach.

  • Open Access

    Learning Landscape Features from Streamflow with Autoencoders

    by Alberto Bassi, Marvin Höge, Antonietta Mira, Fabrizio Fenicia, Carlo Albert

    Understanding the number and types of signatures that best describe streamflow time series is a crucial objective in hydrological science, serving applications such as catchment classification, hydrological model development and calibration. With the main objective of learning a minimal number of streamflow features, we employ an explicit noise conditional autoencoder (ENCA), which, together with meteorological forcings, allows for an accurate reconstruction of the whole streamflow time series. The ENCA architecture feeds the meteorological forcing to the decoder in order to incentivize the encoder to only learn features that are related to landscape properties. By isolating the effect of meteorology, these features can thus be interpreted as landscape fingerprints. The optimal number of features is found by means of an intrinsic dimension estimator. We train our model on the hydro-meteorologic time series data of 568 catchments of the continental United States from the CAMELS dataset. We compare the reconstruction accuracy with state-of-the-art models that take as input a subset of static catchment attributes (both climate and landscape attributes) along with the meteorological forcing variables. Our results suggest that available static catchment attributes compiled by experts account for almost all the relevant information about the rainfall-runoff relationship. Yet, these catchment attributes can be summarized by only two relevant learnt features (or signatures), while a third one is needed for about a dozen difficult catchments in the central US, mainly characterized by high aridity index and intermittent flow. The principal components of the learnt features strongly correlate with the baseflow index and aridity indicators, which is consistent with the idea that these indicators capture the variability of catchment hydrological response and relate to needed model complexity. The correlation analysis further indicates that soil-related and vegetation attributes are of high importance. Finally, in the attempt to interpret the learnt catchment features, we relate them to typical hydrological model components, with specific reference to the parameters of the GR4J model and their function on the hydrograph.

  • Top Cited Papers
    Open Access

    CAMELS-CH: hydro-meteorological time series and landscape attributes for 331 catchments in hydrologic Switzerland

    by Marvin Höge, Martina Kauzlaric, Rosi Siber, Ursula Schönenberger, Pascal Horton, Jan Schwanbeck, Marius Günter Floriancic, Daniel Viviroli, Sibylle Wilhelm, Anna E. Sikorska-Senoner, et al.

    We present CAMELS-CH (Catchment Attributes and MEteorology for Large-sample Studies – Switzerland), a large-sample hydro-meteorological data set for hydrologic Switzerland in central Europe. This domain covers 331 basins within Switzerland and neighboring countries. About one-third of the catchments are located in Austria, France, Germany and Italy. As an Alpine country, Switzerland covers a vast diversity of landscapes, including mountainous environments, karstic regions, and several strongly cultivated regions, along with a wide range of hydrological regimes, i.e., catchments that are glacier-, snow- or rain dominated. Similar to existing data sets, CAMELS-CH comprises dynamic hydro-meteorological variables and static catchment attributes. CAMELS-CH (Höge et al., 2023; available at https://doi.org/10.5281/zenodo.7784632) encompasses 40 years of data between 1 January 1981 and 31 December 2020, including daily time series of stream flow and water levels, and of meteorological data such as precipitation and air temperature. It also includes daily snow water equivalent data for each catchment starting from 2 September 1998. Additionally, we provide annual time series of land cover change and glacier evolution per catchment. The static catchment attributes cover location and topography, climate, hydrology, soil, hydrogeology, geology, land use, human impact and glaciers. This Swiss data set complements comparable publicly accessible data sets, providing data from the “water tower of Europe”.


  • Open Access

    Supplementary material to "Metamorphic Testing of Machine Learning and Conceptual Hydrologic Models"

    by Peter Reichert, Kai Ma, Marvin Höge, Fabrizio Fenicia, Marco Baity-Jesi, Dapeng Feng, Chaopeng Shen


  • Open Access

    Metamorphic Testing of Machine Learning and Conceptual Hydrologic Models

    by Peter Reichert, Kai Ma, Marvin Höge, Fabrizio Fenicia, Marco Baity-Jesi, Dapeng Feng, Chaopeng Shen

    Predicting the response of hydrologic systems to modified driving forces, beyond patterns that have occurred in the past, is of high importance for estimating climate change impacts or the effect of management measures. This kind of predictions requires a model, but the impossibility of testing such predictions against observed data makes it still difficult to estimate their reliability. Metamorphic testing offers a methodology for assessing models beyond validation with real data. It consists of defining input changes for which the expected responses are assumed to be known at least qualitatively, and to test model behavior for consistency with these expectations. To increase the gain of information and reduce the subjectivity of this approach, we extend this methodology to a multi-model approach and include a sensitivity analysis of the predictions to training or calibration options. This allows us to quantitatively analyse differences in predictions between different model structures and calibration options in addition to the qualitative test to the expectations. In our case study, we apply this approach to selected conceptual and machine learning hydrological models calibrated to basins from the CAMELS data set. Our results confirm the superiority of the machine learning models over the conceptual hydrologic models regarding the quality of fit during calibration and validation periods. However, we also find that the response of machine learning models to modified inputs can deviate from the expectations and the magnitude and even the sign of the response can depend on the training data. In addition, even in cases in which all models passed the metamorphic test, there are cases in which the quantitative response is different for different model structures. This demonstrates the importance of this kind of testing beyond the usual calibration-validation analysis to identify potential problems and stimulate the development of improved models.