Research Assistant Professor
Global Water Futures
Hydrology Research Group
Dept. Civil and Environmental Engineering
University of Waterloo
Email:juliane.mai[at]uwaterloo.ca
Phone:+1 (519) 888-4567 ext. 30016
Office:PHY-3016
Address:200 University Ave West
Waterloo, ON N2L 3G1
ORCID:ORCID iD iconorcid.org/0000-0002-1132-2342

Conferences

Talks
Mai, J., BA Tolson, KC Kornelsen, P Coulibaly, D Schaefer, N Gasset, V Fortin, D Bouhemhem, M Leahy, F Anctil, B Hall
The Canadian Surface Prediction Archive of ECCC's Numerical Weather Predictions (and the GWF Cuizinart)
Workshop on Flood Forecasting, Vancouver, BC, Canada. 20-21/02/2019
Mai, J., j Lin, Z Li, H Kheyrollah Pour, M Gauch, Y Wang, L Tan, Y Li, A Pietroniro
The Global Water Futrures Data Slicer-and-Dicer: The Cuizinart
GWF Operations Committee Meeting, Saskatoon, SK, Canada. 22-23/01/2019
Mai, J., BA Tolson, KC Kornelsen, D Schaefer, N Gasset, V Fortin, D Bouhemhem, M Leahy, P Coulibaly, F Anctil, B Hall
The Canadian Surface Prediction Archive of ECCC's Numerical Weather Predictions
52nd Ontario Climate Advisory Committee Meeting, Ontario Ministry of the Environment and Climate Change, Toronto, ON, Canada. 29/11/2018
Mai, J., BA Tolson, KC Kornelsen, D Schaefer, N Gasset, V Fortin,D Bouhemhem, M Leahy, P Coulibaly, F Anctil, B Hall
The Canadian Surface Prediction Archive of ECCC's Numerical Weather Predictions
52nd Ontario Climate Advisory Committee Meeting, Ontario Ministry of the Environment and Climate Change, Toronto, ON, Canada. 29/11/2018
Mai, J., BA Tolson, KC Kornelsen, D Schaefer, N Gasset, V Fortin, M Leahy, P Coulibaly, F Anctil, B Hall
The Canadian Surface Prediction Archive of ECCC's Numerical Weather Predictions: Five Reasons Why You Should Know it
Environment and Climate Change Canada. Dorval, QC, Canada. 09/10/2018
Mai, J., KC Kornelsen, BA Tolson, D Schaefer, N Gasset, V Fortin, M Leahy, P Coulibaly, F Anctil, B Hall
Five Reasons Why You Should Know the Canadian Surface Prediction Archive CaSPAr
Floodnet Annual General Meeting. Quebec City, QC, Canada. 18-20/06/2018
Mai, J., BA Tolson, KC Kornelsen, D Schaefer, N Gasset, V Fortin, M Leahy, and P Coulibaly
Five Reasons Why You Should Know the Canadian Surface Prediction Archive CaSPAr
CGU Annual Meeting. Niagara Falls, ON, Canada. 10-14/06/2018
Mai, J., BA Tolson, H Shen, E Gaborit, N Gasset, V Fortin, et al.
Status report on the Great Lakes Runoff Inter-comparison Project for Lake Erie (GRIP-E)
GWF Science Meeting. Hamilton, ON, Canada. 04-06/06/2018
Mai, J., BA Tolson.
Model Variable Augmentation MVA for Online Diagnostic Assessment of Sensitivity Analysis Results
EGU General Assembly. Vienna, Austria. 09-13/04/2018
Mai, J., KC Kornelsen, BA Tolson, P Coulibaly, F Anctil, V Fortin, M Leahy, B Hall.
The Canadian Surface Prediction Archive (CaSPAr): A Platform to Enhance Environmental Modelling in Canada and Globally
AGU Fall Meeting. New Orleans, USA. 11-15/12/2017
Kornelsen, KC, J Mai, BA Tolson, P Coulibaly, F Anctil, V Fortin, M Leahy, B Hall.
The Canadian Surface Prediction Archive (CaSPAr): A Platform to Enhance Environmental Modelling in Canada and Globally
GIS in Education and Research Conference 2017, Toronto, Canada. 11/10/2017
Mai, J., KC Kornelsen, B Tolson, P Coulibaly, F Anctil, V Fortin, M Leahy, B Hall.
The Canadian Surface Prediction Archive CaSPAr
Floodnet Annual General Meeting, Montreal, Canada. 27-28/06/2017
Mai, J., M. Cuntz, M. Zink, S. Attinger, L. Samaniego.
The mesoscale Hydrologic Model mHM – Concept, Case studies & Model analyses –
University of Adelaide, Australia. 24/08/2016
Mai, J., M. Shafii, M. Cuntz, BA Tolson.
Multi-objective vs. single objective calibration of a hydrologic model exploring the benefit of hydrologic signatures.
CMWR, Toronto, Canada. 20-24/06/2016
Mai, J., M. Cuntz, L. Samaniego, D. Schaefer.
Computationally inexpensive identification of noninformative model parameters.
REKLIM workshop, Merseburg, Germany. 27-29/04/2016
Mai, J., M. Cuntz, M. Shafii, M. Zink, D. Schaefer, S. Thober, L. Samaniego, and B. Tolson (2016).
Single-objective vs. multi-objective calibration of a hydrologic model using single- and multi-objective screening.
EGU General Assembly. Vienna, Austria. 17-22/04/2016
Mai, J., M. Cuntz, L. Samaniego, D. Schäfer (2014).
Parameter sensitivity estimation: The different information in derivative- and variance-based methods.
EGU General Assembly. Vienna, Austria. 28/04-02/05/2014
Mai, J., S. Trump, R. Ali, G. Hager, T. Hanke, I. Lehmann, S. Attinger (2011).
Modeling spatio-temporal dynamics within living cells.
SETAC Europe 21st Annual Meeting, Milano, Italy. 16/05/2011.
Mai, J., S. Trump, R. Ali, T. Hanke, I. Lehmann, S. Attinger (2011).
Imaging in Systems Biology.
Contaminants In The Environment CITE, Leipzig, Germany. 05/04/2011
Mai, J., G. Wittum, S. Attinger (2008).
The BaP-AhR-pathway: Spatio-temporal dynamics of the receptor-ligand-complex.
Helmholtz Alliance on Systems Biology Status Seminar, Potsdam, Germany. 24/06/2008
Conference Papers
Mai, J., S. Trump, S. Attinger (2009).
Analysis of spatio-temporal dynamics by atificial and real FRAP data.
Sixth International Workshop on Computational Systems Biology - WCSB 2009 - Aarhus, Denmark. 10-12/06/2009
Download pdf
Mai, J., S. Attinger (2008).
Modeling Motion of Contaminant BaP in Cytoplasm.
Fifth International Workshop on Computational Systems Biology - WCSB 2008 - Leipzig, Germany. 11-13/06/2008
Download pdf
Poster

AGU 2018: The Great Lakes Runoff Inter-comparison Project for Lake Erie (GRIP-E)

Juliane Mai, Bryan A. Tolson, Hongren Shen, Etienne Gaborit, Vincent Fortin, Milena Dimitrijevic, Nicolas Gasset, Dorothy Durnford, Young Lan Shin, Tricia A. Stadnyk, Lauren M. Fry, Tim Hunter, Andrew D. Gronewold, Joeseph P. Smith, Lacey Mason, Laura Read, Katelyn FitzGerald, Kevin M. Sampson, Alan F. Hamlet, Frank Seglenieks, Shervan Gharari, Saman Razavi, Amin Haghnegahdar, Daniel G. Princz, and Alain Pietroniro
image

Abstract: The Great Lakes Runoff Inter-comparison Project (GRIP) includes a wide range of lumped and distributed models that are used operationally and/or for research purposes across Canada and the United States. Participating models are GEM- Hydro, WRF-Hydro, MESH, VIC, WATFLOOD, HYPE, and LBRM. As part of the Integrated Modelling Program for Canada (IMPC) under the Global Water Futures (GWF) program, the project is aiming to run all these models over several regions in Canada with Lake Erie as the initial domain (GRIP-E). One of the main contributions of the project is to identify a standard, consistent dataset for model building that all participants in the inter-comparison project can access and then process to generate their model-specific required inputs. This presentation will give an update on the design of the inter-comparison and will report on preliminary comparative results. The results of this project will not only serve as a classical framework for model inter-comparison demonstrating the differences in model capabilities, but will also develop strategies to handle cross-border inconsistencies of available data and develop unifying approaches. During development, the project team will share all codes and data on a private GitHub repository. In the final stage of the project all scripts and datasets will be transferred to a public GitHub repository. The inter-comparison project will furthermore test the operational applicability of participating models and identify respective model strengths, i.e., learning which models perform best under certain conditions. The generated multi-model ensembles will help to quantify the uncertainty of hydrologic processes including states and fluxes. Model outputs will be archived3 to enable benchmarking of some of the models which will continue to be developed and improved into the future. Download pdf

Integrated Modelling Program for Canada (IMPC) First Annual Meeting (July 2018): Automatic subsetting of WRF derived climate change scenario forcings

Juliane Mai and Zhenhua Li
image

Abstract: We will present a semi-automatized approach to subset WRF derived climate change forcings. The approach will help GWF scientists to easier access the data and request data according to their individual needs. The data amount handled can hence be reduced significantly. The data is in CF-1.6 conform NetCDF format. The exact domain of interest, the variables and the time period can be requested. The data and the processing are hosted on the Graham super-computer facility. Download pdf

AGU 2017: Method-independent, Computationally Frugal Convergence Testing for Sensitivity Analysis Techniques

Juliane Mai and Bryan Tolson
image

Abstract: The increasing complexity and runtime of environmental models lead to the current situation that the calibration of all model parameters or the estimation of all of their uncertainty is often computationally infeasible. Hence, techniques to determine the sensitivity of model parameters are used to identify most important parameters. All subsequent model calibrations or uncertainty estimation procedures focus then only on these subsets of parameters and are hence less computational demanding. While the examination of the convergence of calibration and uncertainty methods is state-of-the-art, the convergence of the sensitivity methods is usually not checked. If any, bootstrapping of the sensitivity results is used to determine the reliability of the estimated indexes. Bootstrapping, however, might as well become computationally expensive in case of large model outputs and a high number of bootstraps. We, therefore, present a Model Variable Augmentation (MVA) approach to check the convergence of sensitivity indexes without performing any additional model run. This technique is method- and model-independent. It can be applied either during the sensitivity analysis (SA) or afterwards. The latter case enables the checking of already processed sensitivity indexes. To demonstrate the method’s independency of the convergence testing method, we applied it to two widely used, global SA methods: the screening method known as Morris method or Elementary Effects (Morris 1991) and the variance-based Sobol’ method (Solbol’ 1993). The new convergence testing method is first scrutinized using 12 analytical benchmark functions (Cuntz & Mai et al. 2015) where the true indexes of aforementioned three methods are known. This proof of principle shows that the method reliably determines the uncertainty of the SA results when different budgets are used for the SA. The results show that the new frugal method is able to test the convergence and therefore the reliability of SA results in an efficient way. The appealing feature of this new technique is the necessity of no further model evaluation and therefore enables checking of already processed sensitivity results. This is one step towards reliable and transferable, published sensitivity results. Download pdf

EGU 2017: Method-independent, Computationally Frugal Convergence Testing for Sensitivity Analysis Techniques

Juliane Mai and Bryan Tolson
image

Abstract: The increasing complexity and runtime of environmental models lead to the current situation that the calibration of all model parameters or the estimation of all of their uncertainty is often computationally infeasible. Hence, techniques to determine the sensitivity of model parameters are used to identify most important parameters or model processes. All subsequent model calibrations or uncertainty estimation procedures focus then only on these subsets of parameters and are hence less computational demanding. While the examination of the convergence of calibration and uncertainty methods is state-of-the-art, the convergence of the sensitivity methods is usually not checked. If any, bootstrapping of the sensitivity results is used to determine the reliability of the estimated indexes. Bootstrapping, however, might as well become computationally expensive in case of large model outputs and a high number of bootstraps. We, therefore, present a Model Variable Augmentation (MVA) approach to check the convergence of sensitivity indexes without performing any additional model run. This technique is method- and model-independent. It can be applied either during the sensitivity analysis (SA) or afterwards. The latter case enables the checking of already processed sensitivity indexes. To demonstrate the method independency of the convergence testing method, we applied it to three widely used, global SA methods: the screening method known as Morris method or Elementary Effects (Morris 1991, Campolongo et al., 2000), the variance-based Sobol’ method (Solbol’ 1993, Saltelli et al. 2010) and a derivative-based method known as Parameter Importance index (Goehler et al. 2013). The new convergence testing method is first scrutinized using 12 analytical benchmark functions (Cuntz & Mai et al. 2015) where the true indexes of aforementioned three methods are known. This proof of principle shows that the method reliably determines the uncertainty of the SA results when different budgets are used for the SA. Subsequently, we focus on the model-independency by testing the frugal method using the hydrologic model mHM (www.ufz.de/mhm) with about 50 model parameters. The results show that the new frugal method is able to test the convergence and therefore the reliability of SA results in an efficient way. The appealing feature of this new technique is the necessity of no further model evaluation and therefore enables checking of already processed (and published) sensitivity results. This is one step towards reliable and transferable, published sensitivity results. Download pdf

AGU 2016: Multi-objective calibration of a hydrologic model using multi-objective screening

Juliane Mai, Matthias Cuntz, Stephan Thober, Luis Samaniego, and Bryan Tolson
image

Abstract: Hydrologic models are traditionally calibrated against observed streamflow. Recent studies have shown however, that only a few global model parameters are constrained using this integral signal. They can be identified using prior screening techniques. Since different objectives might constrain different parameters, it is advisable to use multiple information to calibrate those models and apply a multi-objective (MO) optimization algorithm. In this study, we address the question if MO algorithms benefit from a prior parameter screening and how much calibration results deteriorate when reducing the model complexity to only the screened set of parameters. The model employed in this study is a distributed hydrologic model mHM with 52 parameters. The MO calibrations are achieved using the Pareto Archived Dynamically Dimensioned Search algorithm using various budgets. We used two objectives here, i.e., the Nash Sutcliffe Efficiencies of the logarithmic low and high streamflows. The screening is based on the sequential single-objective parameter screening introduced by Cuntz et al. (2015). Different options to perform MO screenings are introduced and tested. One is state-of-the- art applying SO screenings to MO. Others are extensions, porting the ideas of Cuntz et al. to MO. It can be shown that the different options identify different number of parameters. The most sensitive parameters, however, are detected with all options. The multi-objective calibrations are then applied using the different sets of screened parameters. Two major results of this coupled screening-calibration experiment can be reported. (1) The performance loss of the calibrated parameter sets can directly be related to the reduction of parameter space through screening. The reduction error is determined by the loss in model variability induced by the reduction of the parameter domain. (2) The MO calibration algorithm however converges faster when the reduced parameter set is used. In general the budget can be reduced by x% when the parameter space is reduced by the same amount. Download pdf

AGU 2016: Impact of the hard-coded parameters on the hydrologic and atmospheric fluxes of the land surface model Noah-MP

Matthias Cuntz, Juliane Mai, Luis Samaniego, Martyn Clark, Volker Wulfmeyer, Sabine Attinger, and Stephan Thober
image

Abstract: Land surface models incorporate a large number of processes, described by physical, chemical and empirical equations. The agility of the models to react to different meteorological conditions is artificially constrained by having hard-coded parameters in their equations. Here we searched for hard-coded parameters in the computer code of the land surface model Noah with multiple process options (Noah-MP) to assess the model’s agility during parameter estimation. We found 139 hard-coded values in all Noah-MP process options in addition to the 71 standard parameters. We performed a Sobol’ global sensitivity analysis to variations of the standard and hard-coded parameters. The sensitivities of the hydrologic output fluxes latent heat and total runoff, their component fluxes, as well as photosynthesis and sensible heat were evaluated at twelve catchments of the Eastern United States with very different hydro-meteorological regimes. Noah-MP’s output fluxes are sensitive to two thirds of its standard parameters. The most sensitive parameter is, however, a hard-coded value in the formulation of soil surface resistance for evaporation, which proved to be oversensitive in other land surface models as well. Latent heat and total runoff show very similar sensitivities towards standard and hard-coded parameters. They are sensitive to both soil and plant parameters, which means that model calibrations of hydrologic or land surface models should take both soil and plant parameters into account. Sensible and latent heat exhibit almost the same sensitivities so that calibration or sensitivity analysis can be performed with either of the two. Photosynthesis has almost the same sensitivities as transpiration, which are different from the sensitivities of latent heat. Including photosynthesis and latent heat in model calibration might therefore be beneficial. Surface runoff is sensitive to almost all hard-coded snow parameters. These sensitivities get, however, diminished in total runoff. It is thus recommended to include the most sensitive hard-coded model parameters that were exposed in this study when calibrating Noah-MP. Download pdf

AGU 2015: Multi-objective vs. single-objective calibration of a hydrologic model

Juliane Mai, Matthias Cuntz, Matthias Zink, David Schaefer, Stephan Thober, Luis Samaniego, Mahyar Shafii, and Bryan Tolson
image

Abstract: Hydrologic models are traditionally calibrated against discharge. Recent studies have shown however, that only a few global model parameters are constrained using the integral discharge measurements. It is therefore advisable to use additional information to calibrate those models. Snow pack data, for example, could improve the parametrization of snow-related processes, which might be underrepresented when using only discharge. One common approach is to combine these multiple objectives into one single objective function and allow the use of a single-objective algorithm. Another strategy is to consider the different objectives separately and apply a Pareto-optimizing algorithm. Both methods are challenging in the choice of appropriate multiple objectives with either conflicting interests or the focus on different model processes. A first aim of this study is to compare the two approaches employing the mesoscale Hydrologic Model mHM at several distinct river basins over Europe and North America. This comparison will allow the identification of the single-objective solution on the Pareto front. It is elucidated if this position is determined by the weighting and scaling of the multiple objectives when combing them to the single objective. The principal second aim is to guide the selection of proper objectives employing sensitivity analyses. These analyses are used to determine if an additional information would help to constrain additional model parameters. The additional information are either multiple data sources or multiple signatures of one measurement. It is evaluated if specific discharge signatures can inform different parts of the hydrologic model. The results show that an appropriate selection of discharge signatures increased the number of constrained parameters by more than 50\% compared to using only NSE of the discharge time series. It is further assessed if the use of these signatures impose conflicting objectives on the hydrologic model. The usage of signatures is furthermore contrasted to the use of additional observations such as soil moisture or snow height. The gain of using an auxiliary dataset is determined using the parametric sensitivity on the respective modeled variable. Download pdf

AGU 2015: Towards constraining hydrologic models using satellite retrieved soil moisture

Matthias Zink, Juliane Mai, Martin Schrön, Oldrich Rakovec, David Schäfer, and Luis Samaniego
image

Abstract: Hydrological models are usually calibrated against observed discharge at the catch- ment outlet and thus are conditioned by an integral catchment information. This pro- cedure ensures the fulfillment of the catchment’s water balance but can lead to high predictive uncertainties in model internal states, like soil moisture, or a lack in spatial representativeness of the model. However, some hydrologic applications, as e.g. soil drought monitoring and prediction, rely on this information. Within this study we propose a framework in which the mesoscale Hydrologic Model (mHM) is calibrated with soil moisture data from various sources. The aim is to con- dition the model on soil moisture (SM), while preserving good performance in discharge estimation. We identify the most appropriate objective functions by conducting synthetic experiments. The best objective function is determined based on i) deviation of synthetic and simulated soil moisture, ii) nonparametric comparison of SM fields (e.g copulas), and iii) by euclidian distance of model parameters. Those objective functions performing best are used to calibrate mHM against satellite soil moisture products (e.g. ESA-CCI) and local in situ observations. This procedure is tested in 3 distinct European basins ranging from snow domination to semi arid climatic conditions. The results of the synthetic experiment indicate that objective functions focusing on the temporal dynamics of SM are preferable to objective functions aiming on spatial patterns or catchment averages. The best performance in the sense of parameter distance is achieved using temporal correlation or the sum of squared distances from soil moisture anomalies of observed and estimated soil moisture. By comparing the copulas of the different objective functions no significant differences between the methods is observed. Employing satellite data, in a consecutive step, the calibrated model is able to catch soil moisture dynamics but deteriorates the discharge signal. Download pdf

EGU 2015: Sensitivity Analysis of the Land Surface Model Noah-MP for Different Output Fluxes in 12 Distinct Catchments

Juliane Mai, Stephan Thober, Luis Samaniego, Oliver Branch, Volker Wulfmeyer, Martyn Clark, Sabine Attinger, Rohini Kumar, Matthias Cuntz
image

Abstract: Land Surface Models (LSMs) use a plenitude of process descriptions to represent the carbon, energy and water cycles. They are highly complex and computationally expensive. Practitioners, however, are often only interested in specific parts of the model such as latent heat or surface runoff. In model applications like parameter estimation, the most important parameters are then chosen by experience or expert knowledge. Hydrologists interested in surface runoff therefore chose mostly soil parameters while biogeochemists focus on carbon fluxes and hence on vegetation parameters. However, this might lead to the omission of parameters that are important, for example, through strong interactions with the parameters chosen. It also happens during model development that some process descriptions contain fixed values, which are supposedly unimportant parameters. However, these hidden parameters remain normally undetected although they might be highly relevant during model calibration. Sensitivity analyses are used to identify informative model parameters for a specific model output. Standard methods for sensitivity analysis such as Sobol indexes require large amounts of model evaluations, specifically in case of many model parameters. We hence propose to first use a recently developed inexpensive sequential screening method based on Elementary Effects that has proven to identify the relevant informative parameters. This reduces the number parameters and therefore model evaluations for subsequent analyses such as sensitivity analysis or model calibration. In this study, we quantify parametric sensitivities of the land surface model Noah-MP that is a state-of-the-art LSM and used at regional scale as the land surface scheme of the atmospheric Weather Research and Forecasting Model (WRF). NOAH-MP contains multiple process parameterizations yielding a considerable amount of parameters (~100). Sensitivities for the three model outputs (a) surface runoff, (b) underground runoff and (c) latent heat are calculated on twelve Model Parameter Estimation Experiment (MOPEX) catchments ranging in size from 1020 to 4421 km2. This allows investigation of parametric sensitivities for distinct hydro-climatic characteristics, emphasizing different land-surface processes. The sequential screening identifies the most informative parameters of Noah-MP for different model output variables. The number of parameters is reduced substantially for all of the three model outputs to approximately 25. The subsequent Sobol method quantifies the sensitivities of these informative parameters. The study demonstrates the existence of sensitive, important parameters in almost all parts of the model for the three output variables. This contrasts to the choice of only soil parameters in hydrological studies and only plant parameters when using Eddy covariance data. The sequential screening identified several important hidden parameters that carry large sensitivities and have hence to be included during model calibration. Download pdf

AGU 2014: Computationally inexpensive identification of non-informative model parameters

Juliane Mai, Matthias Cuntz, Stephan Thober, Matthias Zink, Rohini Kumar, Oldrich Rakovec, Sabine Attinger, John Craven, Giovanni Dalmasso, Ben Langenberg, Jude Musuuza, Vladyslav Prykhodko, David Schaefer, Martin Schrön, Diana Spieler, and Luis Samaniego
image

Abstract: Sensitivity analysis is used, for example, to improve the efficiency of model calibration by discarding insensitive parameters or to detect weaknesses in model formulation and structure. A variety of methods are available for global sensitivity analysis of computational models. They can broadly be divided into derivative-based and variance-based approaches. Derivative-based sensitivity methods highlight key processes for output fluxes and are therefore used to identify major components of a model or detect model weaknesses. Variance-based techniques on the other hand emphasize uncertain processes or process descriptions in the model. Hence, these methods are used to identify parameters which induce the largest variability in the model output and are thus important during calibration. They unfortunately require a large number of model evaluations, but evaluations of complex environmental models are expensive. Therefore, we developed an inexpensive method which applies the Elementary Effects method sequentially and automatically determines informative and non-informative parameters. This computational inexpensive method for identification of parameters important during calibration is applied to a distributed hydrologic model at the mesoscale (mHM) with 52 parameters. The model is applied in three distinct catchments of different hydrological characteristics over Europe, i.e. Neckar (Germany), Sava (Slovenia), and Guadalquivir (Spain). The method identifies the same informative parameters as the standard Sobol method but with less than 1% of model runs. In Germany and Slovenia 22 of 52 parameters are informative mostly in the formulations of evapotranspiration, interflow and percolation. In Spain 19 of 52 parameters are informative with an increased importance of soil parameters. The proposed screening method required only 433, 413, and 406 model runs while the standard Sobol method required 75 600, 43 200, and 70 200 in Germany, Slovenia, and Spain, respectively. We show further that the Sobol indexes calculated only for the subset of informative parameters are practically the same as the Sobol indexes before the screening but the number of model runs is reduced by more than 50%. The model mHM was then calibrated twice in the three test catchments. First all 52 parameters are taken into account and then only the informative parameters are calibrated while all others are kept fixed at their initial value. In the German catchment the Nash-Sutcliff efficiencies (NSE) were 0.87 and 0.83, in the Slovene basin 0.89 and 0.88, and in the Spanish catchment they were 0.86 and 0.85, respectively.This minor loss of at most 4% in model performance comes along with a substantial decrease of at least 65% in model evaluations, i.e. iterations of the optimization algorithm. The model performs equally well during validation periods in all three catchments, which proves that the proposed method retained all informative model parameters. In summary, we propose an efficient screening method to identify non-informative model parameters that can be discarded during further applications. We have shown that sensitivity analysis keeps its information content even after screening. We could further demonstrate no loss of model performance after screening and subsequent calibration. However, model evaluations could be reduced by at least 50% in both applications. Download pdf

AGU 2014: Sensitivity Analysis for the Land Surface Model NOAH-MP for Different Model Fluxes

Stephan Thober, Juliane Mai, Luis Samaniego, Martyn Clark, Pablo Mendoza, Volker Wulfmeyer, Oliver Branch, Sabine Attinger, Rohini Kumar, Matthias Cuntz
image

Abstract: The land-atmosphere fluxes of water, energy and carbon, as computed by the Land Surface Model (LSM), are a critical component of Earth System Models and Numerical Weather Prediction models. Processes and parameters of LSMs are validated mostly against point measurements, for example from Eddy-covariance towers, with much attention given to biophysical processes and vegetation parameters. River discharge on the other hand is not considered very often although it provides an integrated signal of the hydrologic cycle over a catchment. Sensitivity analyses of hydrologic models have shown that soil parameters have then the largest impact on modeled river discharge. In this study, we quantify parametric sensitivities of the land surface model NOAH-MP simultaneously for model outputs at different spatial resolutions. NOAH-MP is a state-of-the-art LSM, which is used at regional scale as the land surface scheme of the atmospheric Weather Research and Forecasting Model (WRF). NOAH-MP contains multiple process parameterizations (hence MP), yielding a considerable amount of parameters (> 500). Standard methods for sensitivity analysis such as Sobol indexes require too many model evaluations in case of many parameters. We therefore use first a recently developed inexpensive screening method based on Elementary Effects that has proven to identify the same informative parameters as the Sobol method but requires only 1% of model evaluations. This reduces the number of parameters to a feasible amount for a thorough sensitivity analysis. The study is conducted on twelve Model Parameter Estimation Experiment (MOPEX) catchments. This allows investigation of parametric sensitivities for distinct hydro-climatic characteristics, emphasizing different land-surface processes. The river basins range in size from 1020 to 4421 km², allowing fast model evaluation. The screening and sensitivity analysis identifies the most informative parameters of NOAH-MP for different model output variables. These parameters can subsequently be used in model calibration and adaptation for better representation of land-atmosphere fluxes at different scales. The long-term objective is to establish flux preservation at multiple resolutions in NOAH-MP to allow assimilation of observations at their representative scale. Download pdf

AGU 2014: Multi-criteria hydrologic parameterization over European river basins

Oldrich Rakovec, Rohini Kumar, Luis Samaniego, Juliane Mai, Stephan Thober and Matthias Cuntz
image

Abstract: Understanding hydrologic model uncertainty and meaningful representation of hydrological processes leads to more reliable hydrologic forecasts, which can be in particular critical under extreme hydrometeorologic conditions. Therefore, hydrologic model development and evaluation should not only focus on the simulated streamflow (model output), but also on other key land surface variables. However, scale discrepancy between available observations and modeling resolution is often neglected. In this study we introduce multiscale and multivariable parameter estimation approach to bridge the scaling gap. The basic components of this framework include the mesoscale hydrologic model (mHM 5.1, http://www.ufz.de/mhm) and the multiscale parameter regionalization (MPR) technique. This framework enables assimilation of various sources of information at their native spatial scales. Additionally, it allows to scrutinize model simultaneously at multiple spatial scales. The application of this framework is demonstrated over 7 large European basins and evaluated for 350 smaller basins. Besides traditional calibration of hydrologic model against observed discharge, model parametrization is further constrained by a range of other variables available at different spatial scales: GRACE terrestrial water storage content (1° × 1° resolution), eddy flux data (≈500 m footprint) and ESA soil moisture product (0.25° × 0.25° resolution). Initial results shows that model parameterization constrained by streamflow only can not be outperformed. However, while parameterization based on complementary data sets leads to slight deterioration in streamflow performance, this marginal loss is balanced by improved simulation of other model states and fluxes. This becomes beneficial especially during the forecast applications, for which correct model initialization is crucial. Download pdf

EGU 2014: Parameterization of PET approaches for distributed hydrologic modeling

Matthias Zink, Juliane Mai, Luis Samaniego, and Matthias Cuntz
image

Abstract: Reliable soil moisture products are needed for the estimation of plant available water or agricultural droughts. For the simulation of hydrological states, e.g. soil moisture, the estimation of evapotranspiration is crucial since it has the largest contribution to the water balance besides precipitation. In hydrological modeling the evapotranspiration is usually estimated based on potential evapotranspiration (PET). The common approaches for PET estimation and their parameterization are sufficient at the point or field scale for which they have been developed. But for spatially distributed estimations on the mesoscale, e.g. 4 km, their robust parameterization is still a challenge in current research. The aim of this study is to find scale and location independent parameters for three different potential evapotranspiration formulations, which are applied in the mesoscale Hydrologic Model (mHM). PET is estimated using the 1) Hargreaves-Samani, 2) Priestley-Taylor, and 3) Penman-Monteith equations. The Hargreaves-Samani method is a temperature driven approach, whereas the other two methods are based on radiation. For estimating the parameters of the above mentioned PET formulations, the Multiscale Parameter Regionalization technique is used. This technique accounts for subgrid variabilities by connecting morphological terrain properties, which are available in a higher resolution than the model resolution, with the parameters for the particular PET approach. The parameters, which needed to be estimated, are the coefficient of the Hargreaves-Samani equation, the Priestley-Taylor coefficient, and the aerodynamic and bulk surface resistance for the Penman-Monteith equation. The Hargreaves-Samani coefficient is regionalized based on the aspect of the terrain. The Priestley-Taylor coefficient as well as the aerodynamic and bulk surface resistance have been estimated using static land cover information combined with leaf area index (LAI) development curves and thus an approximation for vegetation information. This new parameterized PET approaches are evaluated in six different German river basins ranging from 6,000 km2 to 38,000 km2 including a spatial variety from catchments in the northern German lowlands to alpine catchments in the south. The comparison of the results is focusing on evapotranspiration, soil moisture and discharge. Whereas only slight changes in the discharge hydrograph have been observed in the comparison of the three PET equations, the impact on soil moisture is significant. Especially during the summer period the soil moisture is lower for the Priestley-Taylor and Penman-Monteith formulation compared to the Hargreaves-Samani equation. This effect is due to higher estimates in PET for those two methods. Furthermore a validation against eddy covariance measurements showed that the dynamics of evapotranspiration is captured well by the three methods. Download pdf

AGU 2013: Calibration of a hydrological model using patterns of satellite derived land surface temperature

Matthias Zink, Luis Samaniego, Juliane Mai, and Matthias Cuntz
image

Abstract: Hydrological models are usually calibrated against discharge and thus rely on gauging station information for estimating hydrological fluxes and state variables within the catchment of interest. But gauging data are rarely available and do only deliver information about spatially integrated signals of water fluxes within a catchment. In contrast, satellite data which include spatial distributed information are broadly available. Since satellite derived soil moisture and evapotranspiration data are prone to large uncertainty, more certain land surface temperature data are used in this study. The objective of this study is to assess the predictive skill of satellite derived land surface temperature Ts regarding discharge by calibrating the distributed hydrological model mHM using Ts. Since mHM was not capable for estimating Ts, an additional module using the energy balance has been developed. By providing net radiation (Rn) as an input, using evapotranspiration (ET) as residual of the water balance and assuming that soil heat flux and storage terms are negligible, the remaining term of the energy balance is sensible heat (H). By rearranging the sensible heat formulation and substituting H with the before mentioned remainder (Rn-ET), Ts can be estimated. Since comparison of ground measurements with satellite Ts shows an inherent bias (~4.1K), classical quantitative distance measures like mean squared error would yield unreliable calibration results. Though, it is assumed that the spatial distribution of Ts is trustworthy. Hence, a non-parametric, local pattern matching technique has been developed and used for the calibration of the model. This methodology has been tested in six different German river basins ranging from 6,000 km2 to 38,000 km2 including a spatial variety from catchments in the northern German lowlands to alpine influenced catchments in the south. The calibration of the six catchments against Ts leads to Nash Sutcliffe efficiencies (NSEs) of the estimated discharge between 0.3 and 0.6 (medians of 20 optimization runs per basin). While high flows are matched satisfactorily, the prediction of low flows is weak. Conducting cross location experiments which means the application of optimal parameter sets of the five catchments within the sixth yield stable NSEs ranging between 0.3 and 0.6. Furthermore, a change of the spatial distribution of ET within the catchments is observed compared to calibration against discharge. For a relatively homogeneous catchment (regarding soils and land cover) in the north of Germany, the spatial variability of ET decreases compared to results of optimization against discharge. Further, model parameters related to ET are better constrained. In summary, the proposed method shows that satellite derived land surface temperature does not only have an impact on the patterns of ET but also has a predictive skill regarding discharge. Hence, if no discharge data are available, this approach can be used to get an estimate of discharge especially when focusing on high flows. Download pdf

SBHD 2012: A FRAP simulator to quantify influences of experimental setup and model simplifications

Juliane Mai, Saskia Trump, Irina Lehmann and Sabine Attinger
image

Abstract: The interest in molecular interaction and dynamics in living cells is steadily increasing. Several experimental approaches based on fluorescent microscopy have been developed over the years. One of the most widespread techniques to investigate molecular mobility in living cells, is fluorescence recovery after photobleaching (FRAP). To evaluate the effects of a bleaching experiment in a cellular environment, we established a new approach allowing simulations of spatio-temporal dynamics in real cell geometries. These simulations were used to quanitatively determine the effects of several experimental and analytical aspects. The computational feasibility of FRAP analysis requires different assumptions which usually not hold in the environment of a living cell. The compartment bleached in a FRAP experiment has a finite extension as opposed to the commonly used hypothesis of infinity. Another simplification often considered when analyzing FRAP data is complete destruction of all fluorescent particles in the ROI immediately after bleaching. In many cases, this is an oversimplification of the experimental observation. It is easy to imagine that by these model assumptions only approximate model functions describing the time course of the fluorescence recovery within the ROI can be deduced, which will lead to inaccurate parameter values (i.e. diffusion coefficients and reactions rates). We investigated the quantitative influence of the most commonly used model assumptions (i.e. the initial distribution within the ROI and boundary conditions relating membrane features) as well as the influence of the experimental setup (i.e. the simplification of 3D to 2D processes, the position of the bleaching spot, and the geometry of the bleached compartment). The results of such an analysis enables a guidance for the experimentalists to use optimal experimental setups as well as to determine the quality of deduced FRAP parameter sets. Download pdf

ICSB 2011: Analysis of spatio-temporal dynamics by FRAP data

Juliane Mai, Saskia Trump, Rizwan Ali, Gordon Hager, Thomas Hanke, Irina Lehmann and Sabine Attinger
image

Abstract: Exposure to environmental contaminants can lead to a complex cellular response, including toxic effects that might even be involved in carcinogenesis and immunosuppression. Unraveling the underlying mechanisms is essential not only for a comprehensive understanding of such processes but might help to develop new strategies for therapy and prevention. Since biological experiments are often expensive and the number of different conditions and time points is limited, simulations of such intracellular processes constitute an essential tool. Such simulations not only allow the prediction of a response e.g. at an arbitrary time point, but enable the identification of the primary factors that determine the cellular response to contamination. Download pdf

JSMB-JCBC 2010: Analysis of spatio-temporal dynamics by FRAP data

Juliane Mai, Saskia Trump, Rizwan Ali, Gordon Hager, Thomas Hanke, Irina Lehmann and Sabine Attinger
image

Abstract: Polycyclic aromatic hydrocarbons (PAHs) are a large group of environmental contaminants. Many of them, such as Benzo[a]pyrene (B[a]P) are carcinogens and are formed as products of incomplete combustion of fossil fuels. Exposure to B[a]P results in rapid uptake, intracellular distribution and binding to the aryl hydrocarbon receptor (AhR). Within the framework of the systems biology project ”From contaminant molecules to cellular response” we focus on the uptake and distribution of AhR inside cytoplasm and nucleus using Fluorescence Recovery After Photobleaching (FRAP) experiments. We introduce a novel approach for the analysis of FRAP data. By using a (semi-) analytical solution for reaction diffusion equations, allowing for multiple diffusion, we start from the assumption that all involved molecular fractions, whether bound or unbound, could be mobile with different diffusion coefficients. The Laplace transformed equation of the analytical solution is found and inverted numerically using the Stehfest algorithm. For fitting purposes the Simulated Annealing strategy proves to be a better alternative to the conventionally used Levenberg-Marquardt algorithm. We asses performance of our model by fitting different analytical solutions to artificial FRAP data as well as by applying our approach to FRAP data on yellow protein labelled aryl hydrocarbon receptor (transiently transfected into mouse hepatoma cells) comparing the results to previously introduced models. Subsequently we test the capability of our fitting algorithm for identifying the characteristics of binding and diffusion (i.e. number of binding partners, percentage of bound and unbound fraction, binding and diffusion constants). Our new approach shows that intracellular molecular mobility can only be described adequately by allowing for multiple diffusion processes which need to be considered when modelling such data. Download pdf

WCSB 2009: Analysis of spatio-temporal dynamics by artificial and real FRAP data

Juliane Mai, Saskia Trump, Rizwan Ali, Gordon Hager, Thomas Hanke, Irina Lehmann and Sabine Attinger
image

Abstract: The response of cells to contaminants is of great importance for human health. The project focus is to model the motion of a contaminants. Therefore, we use the well- known experiment Fluorescence Recovery After Photobleaching FRAP, which enables us to analyse binding and diffusion of fluorescent particles. Already published analytical solutions which describe the FRAP recovery for several cases only deal with diffusion of unbounded particles. First, we derived the laplace transformed solution for diffusion of all particles. Second, we fit different solutions to artificial data and try to identify both the strategy of acting and the parameters of binding and diffusion. Third, to give an example of use we fit the analytical solutions to real FRAP measurements. Download pdf

WCSB 2008: Modeling Motion of Contaminant BaP in Cytoplasm

Juliane Mai and Sabine Attinger
image

Abstract: The response of cells to contaminant stressors like nicotine is of great importance for human health. The focus of the project is to model the way of the contaminant until the entrance of the nucleus. Therefore, in the first step the cell culture surrounded by the fluorescent contaminant is imaged by a laser microscope. Filters and contour extracting algorithms are used to extract the cell geometry. Finally the movement of the contaminant is modeled using reaction-diffusion-equations and random-walk-processes. The long term goal of the project is to understand the influence of contaminant molecules on biological cell functions. Download pdf