Large-scale Modeling Applications Research Articles

Multiscale model reduction of finite-difference frequency-domain wave modelling in acoustic media

SUMMARY Frequency-domain wave modelling can easily describe the visco-acoustic behaviour of wave propagation using frequency-dependent velocities. Conventional finite-difference (FD) modelling in the frequency domain is computationally prohibitive for solving the acoustic Helmholtz equation in complicated and large geological models. To reduce the computational cost of traditional FD Helmholtz solvers, we develop a multiscale FD frequency-domain method that uses multiscale basis functions to significantly reduce the dimension of system matrices associated with the Helmholtz equation. Due to the insufficient accuracy of the first-order multiscale basis functions in the case of strongly heterogeneous models, we introduce the multinode coarse-element scheme into the scalar Helmholtz equation, a scheme previously developed in the extended multiscale finite-element method for vector problems. This multinode scheme enables multiscale basis functions to capture accurate fine-scale medium property variations. We use one homogeneous model and two heterogeneous models to validate our multiscale method for accuracy and computational cost. Numerical results demonstrate that our new approach can significantly reduce the time and memory costs of acoustic wave modelling while maintaining accuracy, indicating the great potential of our multiscale method in large-scale modelling applications.

Incorporating parameter variability into Monod models of nutrient-limited growth of non-diazotrophic and diazotrophic cyanobacteria.

Models are widely used tools in aquatic science to understand the mechanism of phytoplankton growth and anticipate the occurrence of harmful algal blooms. However, model parameterization remains challenging and issues that may introduce prediction uncertainty exist. Many models use the Monod equation to predict cyanobacteria growth rate based on ambient nutrient concentrations. The half-saturation concentrations in the Monod equation varies greatly among different studies and depends on environmental conditions. In this study, we estimated the growth rate due to nutrient limitations for two cyanobacteria species (Microcystis aeruginosa and Dolichospermum flos-aquae) using a modified Monod model which allows the half-saturation concentration to vary according to initial nitrogen (N) conditions. The model is calibrated against observations from laboratory experiment where cyanobacteria growth and ambient nutrient concentrations were measured simultaneously, which is rarely done in the literature. Our results show this modified model produce better predictions on growth rate and biomass, indicating many commonly used mechanistic models may need improvement regarding phytoplankton growth representation. Furthermore, our study quantifies the flexibility in cyanobacteria growth parameter across a wide range of environmental N in eutrophic lakes thus provides important information for large-scale modelling applications.

Estimation of spatial distribution of irrigated crop areas in Europe for large-scale modelling applications

Regional-to-global studies of hydrology, water availability and water use, and their interaction with agricultural production and food security require accurate information on the location and extent of irrigated croplands. In this study, we derive the European Irrigation Map for the year 2010 (EIM2010) - underpinned by the agricultural census conducted in the European Union in 2010. The map contains 14 irrigated crop classes as well as total irrigable and irrigated areas centred on the year 2010. The thematic resolution of the map is tailored specifically to European agriculture, and contains most relevant staple crops (e.g. maize, rice, potato, other cereals) as well as major permanent crops (e.g. irrigated olive orchards and vineyards). The map was created using a consistent methodology and data gathered across the European Union Member States and the UK, and therefore provides comparable and representative information across the continental domain. The input data used to construct the map consisted of farm-level records on irrigated areas and total crop areas, collected during the 2010 agricultural census and aggregated to 10 km x 10 km grid, as well as regional-level statistics (at NUTS2 level) on irrigated areas by crop type. A custom-developed disaggregation algorithm was used to distribute the regional-level statistics into the 10 km x 10 km grid. Evaluation of the map against independent reference geospatial data indicated satisfactory agreement for both large-scale spatial patterns as well as specific grid cell values for the main crop production regions. The EIM2010 is publicly available and provides valuable inputs for large scale modelling applications involving irrigated agriculture.

Developing pedotransfer functions for predicting soil bulk density in Campania

Oven-dry soil bulk density (BD) is a key soil parameter in biophysical models. Yet, its direct determination for large-scale modeling applications is limited by excessive efforts required for labor-demanding, time-consuming, expensive field campaigns and laboratory-based measurements. To circumvent these shortcomings, BD can be estimated using pedotransfer functions (PTFs) that, however, offer their optimal prediction capability if calibrated and validated within the area of interest. In this study, we exploited the availability of a dataset comprising 3,316 soil samples collected in the farmlands of Campania (a region of southern Italy) to develop regional PTFs for predicting BD using the Random Forest (RF) algorithm. RF was executed considering different combinations of seven soil and three terrain attributes with a 10-fold cross-validation approach to avoid performance overestimation. In light of the RF-based results, we further developed two new PTFs based on multiple linear regression equations. The first regression-based PTF was multiparametric and employed eight features (i.e., six soil properties and two terrain features as environmental covariates), whereas the second PTF was parsimonious and based on three easily available soil predictors. Both regression-based PTFs consistently outperformed 62 existing published PTFs. We also enhanced PTF prediction capabilities by employing regionalization through a clustering approach by grouping soil samples in ten land system classes. Finally, transferability of our models was tested using an external large independent dataset of 12,019 soil samples extracted from the European EU-HYDI database. The parsimonious PTF proved satisfactory prediction performance by corroborating results found in the Campania dataset.

Identification of uncertainty sources in quasi-global discharge and inundation simulations using satellite-based precipitation products

Predicting river discharge and inundation is crucial for water resources management and flood hazard reduction; however, it is still unclear to what extent their variabilities can be captured on global scale. This study evaluates uncertainty sources in the quasi-global river discharge and inundation simulations using the Variable Infiltration Capacity (VIC) macroscale hydrologic model and the Catchment-based Macroscale Floodplain (CaMa-Flood) hydrodynamic model, forced with five high-resolution satellite precipitation datasets. The simulated discharge is first evaluated against more than 2852 sites selected from the Global Streamflow Indices and Metadata Archive (GSIM) dataset, and then the simulated inundation is compared with complementary multiple satellite observations. Globally, about 38% – 43% of the stations produce reasonable discharge simulations with positive Kling-Gupta Efficiency (KGE) on monthly time scale. The simulations show good agreement for flood fractions with mean correlations ranging from 0.47 to 0.62 for satellite detected events. The potential uncertainties sources of discharge and inundation simulation related to physics setting and forcing datasets, such as precipitation, land surface model, routing model, and observation from site and satellite are discussed, as well as future directions for improving large-scale model applications. By using default model settings, we hope our study can offer valuable insights into the applicability of flood simulations and provide guides for model development.

Investigating regionalization techniques for large-scale hydrological modelling

This work investigates regionalization techniques for large-scale model applications in the frame of a pan-European assessment of water resources covering approx. 740,000 km2 in Western Europe. Using the SWAT platform, four variants of the similarity based regionalization approach were compared. The first two involved unsupervised clustering to define hydrological regions before performing hydrological model calibration, whereas the last two involved supervised clustering after performing calibration. Similarity is defined using Partial Least Squares Regression (PLSR) analysis that identifies watershed physiographic characteristics that are most relevant for the selected hydrological response indices.The PLSR results indicate that typically available watershed characteristics such as geomorphology, land-use, climate, and soil properties describe reasonably well the average hydrological conditions but poorly the extreme events. Regionalization variants considering unsupervised clustering and supervised clustering performed similarly well when using all available information. However, results indicate that supervised clustering uses data more efficiently and may be more suitable when data are scarce. It is demonstrated that parsimonious use of available data can be achieved using both regionalization techniques. Finally, model performance consistently becomes acceptable by calibrating watersheds covering only 10% of the model domain, thus, making the calibration task affordable in terms of time and computational resources required.

Toward continental hydrologic–hydrodynamic modeling in South America

Abstract. Providing reliable estimates of streamflow and hydrological fluxes is a major challenge for water resources management over national and transnational basins in South America. Global hydrological models and land surface models are a possible solution to simulate the terrestrial water cycle at the continental scale, but issues about parameterization and limitations in representing lowland river systems can place constraints on these models to meet local needs. In an attempt to overcome such limitations, we extended a regional, fully coupled hydrologic–hydrodynamic model (MGB; Modelo hidrológico de Grandes Bacias) to the continental domain of South America and assessed its performance using daily river discharge, water levels from independent sources (in situ, satellite altimetry), estimates of terrestrial water storage (TWS) and evapotranspiration (ET) from remote sensing and other available global datasets. In addition, river discharge was compared with outputs from global models acquired through the eartH2Observe project (HTESSEL/CaMa-Flood, LISFLOOD and WaterGAP3), providing the first cross-scale assessment (regional/continental × global models) that makes use of spatially distributed, daily discharge data. A satisfactory representation of discharge and water levels was obtained (Nash–Sutcliffe efficiency, NSE > 0.6 in 55 % of the cases) and the continental model was able to capture patterns of seasonality and magnitude of TWS and ET, especially over the largest basins of South America. After the comparison with global models, we found that it is possible to obtain considerable improvement on daily river discharge, even by using current global forcing data, just by combining parameterization and better routing physics based on regional experience. Issues about the potential sources of errors related to both global- and continental-scale modeling are discussed, as well as future directions for improving large-scale model applications in this continent. We hope that our study provides important insights to reduce the gap between global and regional hydrological modeling communities.

RapidRMSD: rapid determination of RMSDs corresponding to motions of flexible molecules.

The root mean square deviation (RMSD) is one of the most used similarity criteria in structural biology and bioinformatics. Standard computation of the RMSD has a linear complexity with respect to the number of atoms in a molecule, making RMSD calculations time-consuming for the large-scale modeling applications, such as assessment of molecular docking predictions or clustering of spatially proximate molecular conformations. Previously, we introduced the RigidRMSD algorithm to compute the RMSD corresponding to the rigid-body motion of a molecule. In this study, we go beyond the limits of the rigid-body approximation by taking into account conformational flexibility of the molecule. We model the flexibility with a reduced set of collective motions computed with e.g. normal modes or principal component analysis. The initialization of our algorithm is linear in the number of atoms and all the subsequent evaluations of RMSD values between flexible molecular conformations depend only on the number of collective motions that are selected to model the flexibility. Therefore, our algorithm is much faster compared to the standard RMSD computation for large-scale modeling applications. We demonstrate the efficiency of our method on several clustering examples, including clustering of flexible docking results and molecular dynamics (MD) trajectories. We also demonstrate how to use the presented formalism to generate pseudo-random constant-RMSD structural molecular ensembles and how to use these in cross-docking. We provide the algorithm written in C++ as the open-source RapidRMSD library governed by the BSD-compatible license, which is available at http://team.inria.fr/nano-d/software/RapidRMSD/. The constant-RMSD structural ensemble application and clustering of MD trajectories is available at http://team.inria.fr/nano-d/software/nolb-normal-modes/. Supplementary data are available at Bioinformatics online.

A thinning routine for large-scale biogeochemical mechanistic ecosystem models

Biogeochemical mechanistic models (BGC models) are used to model the carbon balance of forest ecosystems. Since European forests are managed intensively, a crucial part of carbon modelling is integrating management and thinning routines in the modelling process. In this study, forest inventory data are used to derive information concerning forest management practice. Based on this, a harvesting model is calibrated for simulating the ‘business as usual’ management that can be used in large-scale BGC models. Our approach is based on data from the Austrian National Forest Inventory. The model comprises two sub-models: (1) a logistic model to assess the probability of an inventory point to be thinned and (2) a non-parametric model based on empirical probability density maps to assess the thinning intensity. Since BGC models operate on the stand level, only stand level parameters are integrated in the model such as standing timber carbon, site quality, cover type, elevation and age. A comparison of the predicted and observed proportion of thinned points and the thinning intensity suggests that the model is able to correctly mimic the management regime derived from the inventory data. No systematic trends in the results are evident. Using this thinning model in combination with a mechanistic model will enable assessment of the overall carbon stored in managed forest ecosystems, especially in large-scale modelling applications.

Impact of input data resolution and extent of harvested areas on crop yield estimates in large-scale agricultural modeling for maize in the USA

Large-scale modeling applications are associated with various assumptions and spatial resolutions. In this study, the GIS-based Environmental Policy Integrated Climate (GEPIC) model was used to examine the effects of resampling input data from a resolution of 5arcmin to 10 and 30arcmin on simulated crop grain yields. Maize cultivation in the USA was used as a case study. The biggest impact was found to be the resampling of land use datasets. Rain-fed and irrigated areas are simulated separately and the yields are subsequently weighted according to irrigated and rain-fed fractions in each grid cell. The aggregation causes some grid cells to become rain-fed and irrigated at coarser resolutions after being only rain-fed or only irrigated at 5arcmin. The estimated yield can increase or decrease largely in the affected grid cells due to the fact that irrigated areas generally have much higher yields than rain-fed agriculture in dry regions. The resampling of the climate data has a low impact on crop yields. However, changes in yield can still be large in regions with rain-fed agriculture and low precipitation. The aggregation of soil data by selecting the major soil type within neighboring grid cells has the lowest impact on crop yields. The impact of resampling input data from 5 to 10 and 30arcmin is not significant when modeled maize yields were aggregated on the US national scale, and still quite comparable at the state level. Therefore, for estimating agricultural productivity over large areas, the coarser resolutions can be considered sufficient. Fine resolutions can be important when the goal is to make spatially detailed conclusions. Model performance shows that large deviations between simulated and reported yields can occur at all spatial resolutions in grid cells with harvested areas below 5% of the total cell area. This indicates that a sampling approach that uses representative pixels is preferable to a wall to wall approach using all grid cells.

Modeling Spatial Variation in Microbial Degradation of Pesticides in Soil

Currently, no general guidance is available on suitable approaches for dealing with spatial variation in the first-order pesticide degradation rate constant k even though it is a very sensitive parameter and often highly variable at the field, catchment, and regional scales. Supported by some mechanistic reasoning, we propose a simple general modeling approach to predict k from the sorption constant, which reflects bioavailability, and easily measurable surrogate variables for microbial biomass/activity (organic carbon and clay contents). The soil depth was also explicitly included as an additional predictor variable. This approach was tested in a meta-analysis of available literature data using bootstrapped partial least-squares regression. It explained 73% of the variation in k for the 19 pesticide-study combinations (n = 212) in the database. When 4 of the 19 pesticide-study combinations were excluded (n = 169), the approach explained 80% of the variation in the degradation rate constant. We conclude that the approach shows promise as an effective way to account for the effects of bioavailability and microbial activity on microbial pesticide degradation in large-scale model applications.

Measurements and modeling of pesticide persistence in soil at the catchment scale

Degradation of pesticides in soils is both spatially variable and also one of the most sensitive factors determining losses to surface water and groundwater. To date, no general guidance is available on suitable approaches for dealing with spatial variation in pesticide degradation in catchment or regional scale modeling applications. The purpose of the study was therefore to study the influence of various soil physical, chemical and microbiological characteristics on pesticide persistence in the contrasting cultivated soils found in a small (13 km 2) agricultural catchment in Sweden and to develop and test a simple model approach that could support catchment scale modeling. Persistence of bentazone, glyphosate and isoproturon was investigated in laboratory incubation experiments. Degradation rate constants were highly variable with coefficients of variation ranging between 42 and 64% for the three herbicides. Multiple linear regression analysis and Mallows Cp statistic were employed to select the best set of independent parameters accounting for the variation in degradation. Soil pH and the proportion of active microorganisms (r) together explained 69% of the variation in the bentazone degradation rate constant; the Freundlich sorption co-efficient (K f) and soil laccase activity together explained 88% of the variation in degradation rate of glyphosate, while soil pH was a significant predictor (p < 0.05) for isoproturon persistence. However, correlations between many potential predictor variables made clear interpretations of the statistical analysis difficult. Multiplicative models based on two predictors chosen ‘a priori’, one accounting for microbial activity (e.g. microbial respiration, laccase activity or the surrogate variable soil organic carbon, SOC) and one accounting for the effects of sorption on bioavailability, showed promise to support predictions of degradation for large-scale modeling applications, explaining up to 50% of the variation in herbicide persistence.

Mass residuals in implicit finite volume models for overland and groundwater flow

Summary A primary advantage in using the finite volume method for simulating groundwater flow and overland flow is the conservation property or the ability to conserve mass. However, when implicit finite volume methods are used with large time steps, small cell areas, or parameters with extreme value ranges, the conservation of mass equation becomes slightly unbalanced with a residual. Problems with large mass residuals can be predicted using the condition number of the solution matrix, and the convergence criterion used in the sparse matrix solver. The amount of practical guidance available on how to manage the magnitude of the mass residual or the matrix condition number is limited. To address this need, the current paper shows the usefulness of the mesh ratio. The mesh ratio is a dimensionless number that is a function of the mesh resolution and the temporal resolution. It is directly related to the condition number of the matrix, which in turn affects the mass residual and the model run time. During the current study, several numerical experiments are carried out to determine how the mesh ratio and the water level are related to the condition number, how the critical mesh ratio is related to the number of cells, how the run time is related to the mesh ratio, and how the mass residual is related to the mesh ratio. The results are useful in creating guidelines for mesh design during large-scale model applications. These guidelines can be applied to reducing the mass residual and the run time. The usefulness of the mesh ratio is illustrated using a Regional Simulation Model (RSM) (Lal, A.M.W., Van Zee, Randy, Belnap, Mark, 2005. Case study: model to simulate regional flow in South Florida. Journal of Hydraulic Engineering 131 (4), 247–258) application in south Florida.

Root-Water-Uptake Based upon a New Water Stress Reduction and an Asymptotic Root Distribution Function

Abstract A water stress–compensating root-water-uptake module was developed based upon a newly proposed water stress reduction function and an asymptotic root distribution function. The water stress reduction function takes into account both soil water pressure head and soil resistance to water flow. It requires only physically based parameters that eliminate the need for empirical calibration. The root-water-uptake module, incorporated into a complete Soil–Vegetation–Atmosphere Transfer (SVAT) simulation model, was tested for a variety of soil, crop, and climatic conditions across Canada. Both the proposed water stress reduction and the asymptotic root distribution function performed similarly to existing ones, with the maximum difference in normalized root-mean-square error (NRMSE) between the new and existing water stress reduction function being 0.6%, and between the asymptotic and an exponential root distribution function being 1.2%. The entire root-water-uptake module worked as well as, or better than, published ones. Because the new module uses fewer empirical parameters, it becomes particularly useful in large-scale modeling applications of land surface, hydrology, and terrestrial ecosystems where such parameters are usually not readily available.

The Effects of Scaling and Model Complexity in Simulating the Transport of Inorganic Micropollutants in a Lowland River Reach

Abstract A hypothesis stating that more complex descriptions of processes in models simulate reality better (less error) but with more unreliable predictability (more sensitivity) is tested using a river water quality model. This hypothesis was extended stating that applying the model on a domain of smaller scale requires greater complexity to capture the same accuracy as in large-scale model applications which, however, leads to increased model sensitivity. The sediment and pollutant transport model TOXI, a module in the WASP5 package, was applied to two case studies of different scale: a 90-km course of the 5th order (sensu Strahler 1952) lower Saale river, Germany (large scale), and the lock-and-weir system at Calbe (small scale) situated on the same river course. A sensitivity analysis of several parameters relating to the physical and chemical transport processes of suspended solids, chloride, arsenic, iron and zinc shows that the coefficient, which partitions the total heavy metal mass into its dissolved and sorbed fraction, is a very sensitive parameter. Hence, the complexity of the sorptive process was varied to test the hypotheses.

Stream flow simulation within UK ungauged catchments using a daily rainfall-runoff model

One of the MOPEX science goals is to develop improved a priori model parameter estimation techniques for both large scale modelling applications and ungauged catchments. The access to daily stream flow data within ungauged catchments is also a central component of many aspects of water resource and water quality management.This paper explores the problem of regionalisation a rainfall-runoff model within the United Kingdom for water resources applications. The results from a regionalisation study using a large dataset of 260 catchments within the United Kingdom are presented.Two approaches to the regionalisation approach are developed and evaluated; a regression-based approach relating model parameters to catchment characteristics and a nearest neighbour based approach. The value of both approaches is evaluated in the context of the quality of the simulations obtained with regionalised parameter sets. The paper identifies that the regression-based approach yields the best predictive results and that these results are encouraging in the context of operational water management. However, the ultimate limitations of the approach are recognised and recommendations are made for further research on nearest neighbour approaches. The paper concludes by raising the question whether the regionalisation of the parameters for prior calibrated model is the most appropriate way to approach the regionalisation question and suggests an alternative modelling framework based on hydrological response units.

Screening Sensitivity Theory

A comprehensive rigorous theory is developed for screening sensitivity coefficients in largescale modeling applications. The theory uses Bayesian inference and group theory to establish a probabilistic framework for solving an underdetermined system of linear equations. The underdetermined problem is directly related to statistical screening sensitivity theory as developed in recent years. Several examples of the new approach to screening are worked out in detail and comparisons are made with statistical approaches to the problem. The drawbacks of these latter methods are discussed at some length.