scholarly journals Streamflow prediction in ungauged basins: benchmarking the efficiency of deep learning

2020 ◽  
Vol 163 ◽  
pp. 01001
Author(s):  
Georgy Ayzel ◽  
Liubov Kurochkina ◽  
Eduard Kazakov ◽  
Sergei Zhuravlev
Keyword(s):  
Deep Learning ◽  
Hydrological Model ◽  
State Of The Art ◽  
Flash Flood ◽  
Early Warning Systems ◽  
Model Parameters ◽  
Ungauged Basins ◽  
Streamflow Prediction ◽  
Basin Scale ◽  
The Impact

Streamflow prediction is a vital public service that helps to establish flash-flood early warning systems or assess the impact of projected climate change on water management. However, the availability of streamflow observations limits the utilization of the state-of-the-art streamflow prediction techniques to the basins where hydrometric gauging stations exist. Since the most river basins in the world are ungauged, the development of the specialized techniques for the reliable streamflow prediction in ungauged basins (PUB) is of crucial importance. In recent years, the emerging field of deep learning provides a myriad of new models that can breathe new life into the stagnating PUB methods. In the presented study, we benchmark the streamflow prediction efficiency of Long Short-Term Memory (LSTM) networks against the standard technique of GR4J hydrological model parameters regionalization (HMREG) at 200 basins in Northwest Russia. Results show that the LSTM-based regional hydrological model significantly outperforms the HMREG scheme in terms of median Nash-Sutcliffe efficiency (NSE), which is 0.73 and 0.61 for LSTM and HMREG, respectively. Moreover, LSTM demonstrates the comparable median NSE with that for basin-scale calibration of GR4J (0.75). Therefore, this study underlines the high utilization potential of deep learning for the PUB by demonstrating the new state-of-the-art performance in this field.

scholarly journals Assessing the Impact of the Loss Function, Architecture and Image Type for Deep Learning-Based Wildfire Segmentation

2021 ◽  
Vol 11 (15) ◽  
pp. 7046
Author(s):  
Jorge Francisco Ciprián-Sánchez ◽  
Gilberto Ochoa-Ruiz ◽  
Lucile Rossi ◽  
Frédéric Morandini
Keyword(s):  
Deep Learning ◽  
Loss Function ◽  
State Of The Art ◽  
Fire Detection ◽  
Loss Functions ◽  
Wildfire Spread ◽  
Combine Information ◽  
The Impact ◽  
Image Type ◽  
Segmentation Models

Wildfires stand as one of the most relevant natural disasters worldwide, particularly more so due to the effect of climate change and its impact on various societal and environmental levels. In this regard, a significant amount of research has been done in order to address this issue, deploying a wide variety of technologies and following a multi-disciplinary approach. Notably, computer vision has played a fundamental role in this regard. It can be used to extract and combine information from several imaging modalities in regard to fire detection, characterization and wildfire spread forecasting. In recent years, there has been work pertaining to Deep Learning (DL)-based fire segmentation, showing very promising results. However, it is currently unclear whether the architecture of a model, its loss function, or the image type employed (visible, infrared, or fused) has the most impact on the fire segmentation results. In the present work, we evaluate different combinations of state-of-the-art (SOTA) DL architectures, loss functions, and types of images to identify the parameters most relevant to improve the segmentation results. We benchmark them to identify the top-performing ones and compare them to traditional fire segmentation techniques. Finally, we evaluate if the addition of attention modules on the best performing architecture can further improve the segmentation results. To the best of our knowledge, this is the first work that evaluates the impact of the architecture, loss function, and image type in the performance of DL-based wildfire segmentation models.

scholarly journals Investigating the Impact of the Bit Depth of Fluorescence-Stained Images on the Performance of Deep Learning-Based Nuclei Instance Segmentation

Diagnostics ◽  
2021 ◽  
Vol 11 (6) ◽  
pp. 967
Author(s):  
Amirreza Mahbod ◽  
Gerald Schaefer ◽  
Christine Löw ◽  
Georg Dorffner ◽  
Rupert Ecker ◽  
...  
Keyword(s):  
Deep Learning ◽  
State Of The Art ◽  
Computer Assisted ◽  
Important Criterion ◽  
Histological Image ◽  
Computer Mediated ◽  
Image Patches ◽  
Processing Techniques ◽  
The Impact ◽  
Instance Segmentation

Nuclei instance segmentation can be considered as a key point in the computer-mediated analysis of histological fluorescence-stained (FS) images. Many computer-assisted approaches have been proposed for this task, and among them, supervised deep learning (DL) methods deliver the best performances. An important criterion that can affect the DL-based nuclei instance segmentation performance of FS images is the utilised image bit depth, but to our knowledge, no study has been conducted so far to investigate this impact. In this work, we released a fully annotated FS histological image dataset of nuclei at different image magnifications and from five different mouse organs. Moreover, by different pre-processing techniques and using one of the state-of-the-art DL-based methods, we investigated the impact of image bit depth (i.e., eight bits vs. sixteen bits) on the nuclei instance segmentation performance. The results obtained from our dataset and another publicly available dataset showed very competitive nuclei instance segmentation performances for the models trained with 8 bit and 16 bit images. This suggested that processing 8 bit images is sufficient for nuclei instance segmentation of FS images in most cases. The dataset including the raw image patches, as well as the corresponding segmentation masks is publicly available in the published GitHub repository.

scholarly journals Climate-dependent propagation of precipitation uncertainty into the water cycle

2020 ◽  
Author(s):  
Ali Fallah ◽  
Sungmin O ◽  
Rene Orth
Keyword(s):  
Hydrological Model ◽  
Water Cycle ◽  
State Of The Art ◽  
Rain Gauge ◽  
Multiple Sources ◽  
Current State ◽  
Wide Range ◽  
The Impact ◽  
Conceptual Hydrological Model

Abstract. Precipitation is a crucial variable for hydro-meteorological applications. Unfortunately, rain gauge measurements are sparse and unevenly distributed, which substantially hampers the use of in-situ precipitation data in many regions of the world. The increasing availability of high-resolution gridded precipitation products presents a valuable alternative, especially over gauge-sparse regions. Nevertheless, uncertainties and corresponding differences across products can limit the applicability of these data. This study examines the usefulness of current state-of-the-art precipitation datasets in hydrological modelling. For this purpose, we force a conceptual hydrological model with multiple precipitation datasets in > 200 European catchments. We consider a wide range of precipitation products, which are generated via (1) interpolation of gauge measurements (E-OBS and GPCC V.2018), (2) combination of multiple sources (MSWEP V2) and (3) data assimilation into reanalysis models (ERA-Interim, ERA5, and CFSR). For each catchment, runoff and evapotranspiration simulations are obtained by forcing the model with the various precipitation products. Evaluation is done at the monthly time scale during the period of 1984–2007. We find that simulated runoff values are highly dependent on the accuracy of precipitation inputs, and thus show significant differences between the simulations. By contrast, simulated evapotranspiration is generally much less influenced. The results are further analysed with respect to different hydro-climatic regimes. We find that the impact of precipitation uncertainty on simulated runoff increases towards wetter regions, while the opposite is observed in the case of evapotranspiration. Finally, we perform an indirect performance evaluation of the precipitation datasets by comparing the runoff simulations with streamflow observations. Thereby, E-OBS yields the best agreement, while furthermore ERA5, GPCC V.2018 and MSWEP V2 show good performance. In summary, our findings highlight a climate-dependent propagation of precipitation uncertainty through the water cycle; while runoff is strongly impacted in comparatively wet regions such as Central Europe, there are increasing implications on evapotranspiration towards drier regions.

scholarly journals A physically-based parsimonious hydrological model for flash floods in Mediterranean catchments

2011 ◽  
Vol 11 (9) ◽  
pp. 2567-2582 ◽  
Author(s):  
H. Roux ◽  
D. Labat ◽  
P.-A. Garambois ◽  
M.-M. Maubourguet ◽  
J. Chorda ◽  
...  
Keyword(s):  
Sensitivity Analysis ◽  
Soil Properties ◽  
Overland Flow ◽  
Hydrological Model ◽  
Extreme Event ◽  
Model Parameters ◽  
Peak Time ◽  
Efficiency Coefficient ◽  
Soil Saturation ◽  
The Impact

Abstract. A spatially distributed hydrological model, dedicated to flood simulation, is developed on the basis of physical process representation (infiltration, overland flow, channel routing). Estimation of model parameters requires data concerning topography, soil properties, vegetation and land use. Four parameters are calibrated for the entire catchment using one flood event. Model sensitivity to individual parameters is assessed using Monte-Carlo simulations. Results of this sensitivity analysis with a criterion based on the Nash efficiency coefficient and the error of peak time and runoff are used to calibrate the model. This procedure is tested on the Gardon d'Anduze catchment, located in the Mediterranean zone of southern France. A first validation is conducted using three flood events with different hydrometeorological characteristics. This sensitivity analysis along with validation tests illustrates the predictive capability of the model and points out the possible improvements on the model's structure and parameterization for flash flood forecasting, especially in ungauged basins. Concerning the model structure, results show that water transfer through the subsurface zone also contributes to the hydrograph response to an extreme event, especially during the recession period. Maps of soil saturation emphasize the impact of rainfall and soil properties variability on these dynamics. Adding a subsurface flow component in the simulation also greatly impacts the spatial distribution of soil saturation and shows the importance of the drainage network. Measures of such distributed variables would help discriminating between different possible model structures.

scholarly journals Calibration of physically based models: back to basics?

2003 ◽  
Vol 5 (4) ◽  
pp. 233-244 ◽  
Author(s):  
Vincent Guinot ◽  
Philippe Gourbesville
Keyword(s):  
Hydrological Model ◽  
Flash Flood ◽  
Model Parameters ◽  
Model Structure ◽  
Ungauged Catchments ◽  
Model Response ◽  
Physically Based ◽  
Physically Based Models ◽  
Extreme Hydrological Events ◽  
Main Factor

The modelling of extreme hydrological events often suffers from a lack of available data. Physically based models are the best available modelling option in such situations, as they can in principle provide answers about the behaviour of ungauged catchments provided that the geometry and the forcings are known with sufficient accuracy. The need for calibration is therefore limited. In some situations, calibration (seen as adjusting the model parameters so that they fit the calculation as closely to the measurements as possible) is impossible. This paper presents such a situation. The MIKE SHE physically based hydrological model is used to model a flash flood over a medium-sized catchment of the Mediterranean Alps (2820 km2). An examination of a number of modelling alternatives shows that the main factor of uncertainty in the model response is the model structure (what are the dominant processes). The second most important factor is the accuracy with which the catchment geometry is represented in the model. The model results exhibit very little sensitivity to the model parameters, and therefore calibration of these parameters is found to be useless.

scholarly journals Improving streamflow predictions at ungauged locations with real-time updating: application of an EnKF-based state-parameter estimation strategy

2014 ◽  
Vol 18 (10) ◽  
pp. 3923-3936 ◽  
Author(s):  
X. Xie ◽  
S. Meng ◽  
S. Liang ◽  
Y. Yao
Keyword(s):  
Real Time ◽  
Land Surface ◽  
Hydrological Model ◽  
Similarity Measures ◽  
State Parameter ◽  
Model Parameters ◽  
Prediction Errors ◽  
Distributed Hydrological Model ◽  
Streamflow Prediction ◽  
Short Period

Abstract. The challenge of streamflow predictions at ungauged locations is primarily attributed to various uncertainties in hydrological modelling. Many studies have been devoted to addressing this issue. The similarity regionalization approach, a commonly used strategy, is usually limited by subjective selection of similarity measures. This paper presents an application of a partitioned update scheme based on the ensemble Kalman filter (EnKF) to reduce the prediction uncertainties. This scheme performs real-time updating for states and parameters of a distributed hydrological model by assimilating gauged streamflow. The streamflow predictions are constrained by the physical rainfall-runoff processes defined in the distributed hydrological model and by the correlation information transferred from gauged to ungauged basins. This scheme is successfully demonstrated in a nested basin with real-world hydrological data where the subbasins have immediate upstream and downstream neighbours. The results suggest that the assimilated observed data from downstream neighbours have more important roles in reducing the streamflow prediction errors at ungauged locations. The real-time updated model parameters remain stable with reasonable spreads after short-period assimilation, while their estimation trajectories have slow variations, which may be attributable to climate and land surface changes. Although this real-time updating scheme is intended for streamflow predictions in nested basins, it can be a valuable tool in separate basins to improve hydrological predictions by assimilating multi-source data sets, including ground-based and remote-sensing observations.

Understanding factors influencing the wetland parameters of a monthly rainfall-runoff model in the Upper Congo River basin

2020 ◽  
Author(s):  
Pierre Kabuya ◽  
Denis Hughes ◽  
Raphael Tshimanga ◽  
Mark Trigg
Keyword(s):  
Hydrological Model ◽  
Return Flow ◽  
Model Parameters ◽  
Morphometric Characteristics ◽  
Practical Advantage ◽  
Flow Parameters ◽  
Basin Scale ◽  
Storage Volume ◽  
Wetland Processes ◽  
Clockwise Hysteresis

<p>Wetland processes considerably influence the flow regime of the downstream river channel, and are important to consider for a better representation of runoff generation within a basin scale hydrological model. The need to understand these processes lead to the development of a wetland sub-model for the monthly time step Pitman hydrological model. However, previous studies highlighted the need to provide guidance to explicitly estimate the wetland parameters rather than using a trial and error calibration approach. In this study, a 2D hydrodynamic river-wetland model (LISFLOOD-FP) is used to explicitly represent the inundation process exchanges between river channels and wetland systems and thereby inform the choice of Pitman wetland model parameters. The hysteretic patterns of these river-wetland processes are quantified through the use of hysteresis indices. Additionally, the hysteretic patterns are connected with the spill and return flow parameters of the wetland sub-model and eventually with the wetland morphometric characteristics. The results show that there is a consistent connection between the degree of hysteresis found in the channel-wetland exchange processes and the Pitman wetland parameters which are also explicitly linked to the wetland morphometric characteristics. The channel capacity to spill (Qcap) is consistently correlated with the hysteresis found between the channel inflow and the wetland storage volume. This anti-clockwise hysteresis represents the time delay between the inundation and drainage processes. The channel spill factor (QSF), in addition to the inundation processes, is also connected with the drainage processes represented by the wetland storage volume and channel outflow anti-clockwise hysteresis. On the other hand, the parameters of the return flow equation have shown a strong consistent relationship with the channel inflow-wetland storage hysteresis. It has also been observed that the wetland average surface slope and the proportion of the wetland storage below the channel banks are the morphometric characteristics that influence the spill and the return flow parameters of the Pitman wetland sub-model. This understanding has a practical advantage for the estimation of the Pitman wetland parameters in the many areas where it is not possible to run complex hydrodynamic models.</p>

scholarly journals Attentional Factorization Machines: Learning the Weight of Feature Interactions via Attention Networks

2017 ◽  
Author(s):  
Jun Xiao ◽  
Hao Ye ◽  
Xiangnan He ◽  
Hanwang Zhang ◽  
Fei Wu ◽  
...  
Keyword(s):  
Deep Learning ◽  
State Of The Art ◽  
Feature Interaction ◽  
Model Parameters ◽  
Learning Approach ◽  
Attention Networks ◽  
Feature Interactions ◽  
Factorization Machine ◽  
Real World Datasets ◽  
Novel Model

Factorization Machines (FMs) are a supervised learning approach that enhances the linear regression model by incorporating the second-order feature interactions. Despite effectiveness, FM can be hindered by its modelling of all feature interactions with the same weight, as not all feature interactions are equally useful and predictive. For example, the interactions with useless features may even introduce noises and adversely degrade the performance. In this work, we improve FM by discriminating the importance of different feature interactions. We propose a novel model named Attentional Factorization Machine (AFM), which learns the importance of each feature interaction from data via a neural attention network. Extensive experiments on two real-world datasets demonstrate the effectiveness of AFM. Empirically, it is shown on regression task AFM betters FM with a 8.6% relative improvement, and consistently outperforms the state-of-the-art deep learning methods Wide&Deep [Cheng et al., 2016] and DeepCross [Shan et al., 2016] with a much simpler structure and fewer model parameters. Our implementation of AFM is publicly available at: https://github.com/hexiangnan/attentional_factorization_machine

scholarly journals An evaluation of the potential of Sentinel 1 for improving flash flood predictions via soil moisture–data assimilation

2017 ◽  
Vol 44 ◽  
pp. 89-100 ◽  
Author(s):  
Luca Cenci ◽  
Luca Pulvirenti ◽  
Giorgio Boni ◽  
Marco Chini ◽  
Patrick Matgen ◽  
...  
Keyword(s):  
Soil Moisture ◽  
Data Assimilation ◽  
Spatial Resolution ◽  
Temporal Resolution ◽  
Hydrological Model ◽  
High Spatial Resolution ◽  
Flash Flood ◽  
Soil Moisture Data ◽  
Time Period ◽  
The Impact

Abstract. The assimilation of satellite-derived soil moisture estimates (soil moisture–data assimilation, SM–DA) into hydrological models has the potential to reduce the uncertainty of streamflow simulations. The improved capacity to monitor the closeness to saturation of small catchments, such as those characterizing the Mediterranean region, can be exploited to enhance flash flood predictions. When compared to other microwave sensors that have been exploited for SM–DA in recent years (e.g. the Advanced SCATterometer – ASCAT), characterized by low spatial/high temporal resolution, the Sentinel 1 (S1) mission provides an excellent opportunity to monitor systematically soil moisture (SM) at high spatial resolution and moderate temporal resolution. The aim of this research was thus to evaluate the impact of S1-based SM–DA for enhancing flash flood predictions of a hydrological model (Continuum) that is currently exploited for civil protection applications in Italy. The analysis was carried out in a representative Mediterranean catchment prone to flash floods, located in north-western Italy, during the time period October 2014–February 2015. It provided some important findings: (i) revealing the potential provided by S1-based SM–DA for improving discharge predictions, especially for higher flows; (ii) suggesting a more appropriate pre-processing technique to be applied to S1 data before the assimilation; and (iii) highlighting that even though high spatial resolution does provide an important contribution in a SM–DA system, the temporal resolution has the most crucial role. S1-derived SM maps are still a relatively new product and, to our knowledge, this is the first work published in an international journal dealing with their assimilation within a hydrological model to improve continuous streamflow simulations and flash flood predictions. Even though the reported results were obtained by analysing a relatively short time period, and thus should be supported by further research activities, we believe this research is timely in order to enhance our understanding of the potential contribution of the S1 data within the SM–DA framework for flash flood risk mitigation.

Model adequacy tests for improving predictions in ungauged basins

2020 ◽  
Author(s):  
Cristina Prieto ◽  
Nataliya Le Vine ◽  
Dmitri Kavetski ◽  
César Álvarez ◽  
Raúl Medina
Keyword(s):  
Time Series ◽  
Hydrological Model ◽  
Flow Time ◽  
Flow Index ◽  
Model Parameters ◽  
Ungauged Basins ◽  
Ungauged Catchments ◽  
Subsequent Work ◽  
Study Results

<p>Flow prediction in ungauged catchments is a major unresolved challenge in scientific and engineering hydrology. Meeting this challenge is made difficult by the uncertainty in the “regionalization” model used to transpose hydrological data (e.g., flow indices) from gauged to ungauged basins, and by the uncertainty in the hydrological model used to predict streamflow in the ungauged basin. This study combines recent advances in flow index selection, regionalization via machine learning methods, and a Bayesian inference framework. In addition, it proposes two new statistical metrics, “DistanceTest” and “InfoTest”, to assess the adequacy of a model before estimating its parameters. “DistanceTest” quantifies whether a model (hydrological or regionalization) is likely to reproduce the available hydrological information in a catchment. “InfoTest” is based on Bayes Factors and quantifies the information added by a model (hydrological or regionalization) over prior knowledge about the available hydrological information in a catchment). The proposed adequacy tests can be seen as a prerequisite for a model (hydrological or regionalization) being considered capable of providing meaningful and high quality flow time series predictions in ungauged catchments. If a model is found inadequate a priori and rejected, the modeler is spared the effort in estimating the model parameters, which can be a substantial saving.</p><p>The proposed regionalization approach is applied to 92 northern Spain catchments, with 16 catchments treated as ungauged. It is found that (1) a small number of PCs capture approximately 87% of variability in the flow indices, and (2) adequacy tests with respect to regionalized information are indicative of (but do not guarantee) the ability of a hydrological model to predict flow time series. The adequacy tests identify the regionalization of flow index PCs as adequate in 12 of 16 catchments but the hydrological model as adequate in only 1 of 16 catchments. In addition, the case study results suggest that the hydrological model is the main source of uncertainty in comparison to the regionalization model, and hence should receive the main priority in subsequent work at the case study catchments.</p>

Sign in / Sign up

Export Citation Format

Share Document