Scotland's Rural College Global Research Alliance ... · Global Research Alliance Modelling Platform (GRAMP): An open web platform for modelling greenhouse gas emissions from agro-ecosystems.

Scotland's Rural College

Global Research Alliance Modelling Platform (GRAMP): An open web platform formodelling greenhouse gas emissions from agro-ecosystemsYeluripati, JB; del Prado, A; Sanz-Cobena, A; Rees, RM; Li, C; Chadwick, D; Tilston, E; Topp,CFE; Cardenas, LM; Ingraham, P; Gilhespy, S; Anthony, S; Vetter, SH; Misselbrook, T;Salas, W; Smith, PPublished in:Computers and Electronics in Agriculture

DOI:10.1016/j.compag.2014.11.016

Print publication: 01/01/2015

Document VersionPeer reviewed version

Link to publication

Citation for pulished version (APA):Yeluripati, JB., del Prado, A., Sanz-Cobena, A., Rees, RM., Li, C., Chadwick, D., Tilston, E., Topp, CFE.,Cardenas, LM., Ingraham, P., Gilhespy, S., Anthony, S., Vetter, SH., Misselbrook, T., Salas, W., & Smith, P.(2015). Global Research Alliance Modelling Platform (GRAMP): An open web platform for modelling greenhousegas emissions from agro-ecosystems. Computers and Electronics in Agriculture, 111, 112 - 120.https://doi.org/10.1016/j.compag.2014.11.016

General rightsCopyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright ownersand it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights.

• Users may download and print one copy of any publication from the public portal for the purpose of private study or research. • You may not further distribute the material or use it for any profit-making activity or commercial gain • You may freely distribute the URL identifying the publication in the public portal ?

Take down policyIf you believe that this document breaches copyright please contact us providing details, and we will remove access to the work immediatelyand investigate your claim.

Download date: 23. Dec. 2020

Global Research Alliance Modelling Platform (GRAMP): An 1

open web platform for modelling greenhouse gas emissions from 2

agro-ecosystems. 3

Jagadeesh B. Yeluripati1,9*, Agustin del Prado2, Alberto Sanz-Cobeña10, Robert Rees3, 4 Changsheng Li4, Dave Chadwick5, Emma Tilston3, Cairistiona F. E. Topp3, Laura Cardenas6, 5 Pete Ingraham7, Sarah Gilhespy6, Steven Anthony8, Sylvia H. Vetter9, Tom Misselbrook6, 6 William Salas7, Pete Smith9 7

1The James Hutton Institute, Craigiebuckler, Aberdeen, AB15 8QH, UK; Email: 8 [email protected] 9

2BC3 Basque Centre for Climate Change, Alameda Urquijo 4, 4a 48008 Bilbao Bizkaia, 10 Spain. 11

3SRUC, King’s Buildings, West Mains Road, Edinburgh EH9 3JG, UK. 12

4Institute for the Study of Earth, Oceans, and Space, University of New Hampshire, Durham, 13 NH 03824, USA. 14

5School of Environment, Natural Resources and Geography, Environment Centre Wales, 15 Deiniol Road, Bangor University, Bangor, LL57 2UW, UK. 16

6Rothamsted Research, North Wyke, Okehampton, Devon, EX20 2SB, UK. 17

7Applied Geosolutions, LLC, 87 Packers Falls Road, Durham, NH 03824, USA. 18

8ADAS Group Ltd, HQ Pendeford House, Pendeford Business Park, Wolverhampton, WV9 19 5AP, UK. 20

9Institute of Biological and Environmental Sciences, University of Aberdeen, 23 St Machar 21 Drive, Aberdeen, AB24 3UU, UK 22

10Technical University of Madrid. School of Agriculture. Av Complutense s/n. 28040 23 Madrid. Spain 24

*Corresponding author 25 26

Abstract: 27

Carbon (C) and nitrogen (N) process-based models are important tools for estimating and 28

reporting greenhouse gas emissions and changes in soil C stocks. There is a need for 29

continuous evaluation, development and adaptation of these models to improve scientific 30

understanding, national inventories and assessment of mitigation options across the world. To 31

1

date, much of the information needed to describe different processes like transpiration, 32

photosynthesis, plant growth and maintenance, above and below ground carbon dynamics, 33

decomposition and nitrogen mineralization etc., 34

in ecosystem models remains inaccessible to the wider community, being stored within 35

model computer source code, or held internally by modelling teams. Here we describe the 36

Global Research Alliance Modelling Platform (GRAMP), a web-based modelling platform to 37

link researchers with appropriate datasets, models and training material. It will provide access 38

to model source code and an interactive platform for researchers to form a consensus on 39

existing methods, and to synthesize new ideas, which will help to advance progress in this 40

area. The platform will eventually support a variety of models, but to trial the platform and 41

test the architecture and functionality, it was piloted with variants of the DNDC model. The 42

intention is to form a worldwide collaborative network (a virtual laboratory) via an 43

interactive website with access to models and best practice guidelines; appropriate datasets 44

for testing, calibrating and evaluating models; on-line tutorials and links to modelling and 45

data provider research groups, and their associated publications. A graphical user interface 46

has been designed to view the model development tree and access all of the above functions. 47

Keywords: Biogeochemical modelling, Integrated modelling platform, Unified modelling 48

approach, Climate change, ecosystems, greenhouse gas emissions. 49

50

2

1 Introduction 51

Agriculture plays a vital role in food security, poverty reduction, rural employment and 52

sustainable development (Foresight, 2009). There is a need to produce more food with fewer 53

resources, while safeguarding the environment and reinvigorating rural economies to feed a 54

growing population (Smith, 2013). The agriculture sector is particularly vulnerable to the 55

impacts of climate change and faces significant challenges in meeting a dramatic increase in 56

global food demand, while reducing its contribution to greenhouse gas emissions (GHG) 57

(Smith and Gregory, 2013). The agricultural sector contributes ~14% of world’s annual direct 58

anthropogenic GHG emissions (Smith et al., 2008), and these emissions are expected to rise 59

by 30-40% above 2005 levels, in line with the projected increase in food production by 2050 60

if current trends continue (Godfray et al., 2011). Farmers need new strategies to produce 61

goods with anticipated changes in climate and agro-ecological conditions. Modelling can be 62

used to support decision making that introduces new management practices to reduce GHG 63

emissions and maintain productivity. 64

Recently, many models (Del Grosso et al., 2009; Giltrap et al., 2010.; Smith et al., 2010) have 65

been developed and are in use to address the challenges of sustainable agricultural 66

development (Shepherd et al., 2011). The active use of simulation modelling techniques is 67

one of the few means to enable us to verify hypotheses about the operating principles in agro-68

ecosystems and their subsystems. Carbon (C) and nitrogen (N) process-based models are an 69

important tool in the quantification, prediction and reporting of GHG emissions from 70

different ecosystems. We need to evaluate, develop and adapt models that can be used to 71

improve national inventories of GHG emissions by meeting Tier 2 and Tier 3 reporting 72

requirements, as countries upgrade from Tier 1. If models are accessible enough, they can act 73

as a medium for wider participation in environmental management. However, using, testing, 74

3

calibrating and evaluating these models are far from straightforward. There are already 75

several models that can address the questions related to C and N cycling and GHG emissions 76

from soils (Del Grosso et al., 2009; Li, 2007) and there are about 4000 more general 77

mathematical models in the field of ecology and environmental sciences (Jørgensen, 1999; 78

Rivington and Koo, 2010) . These models represent a large collection of scientific knowledge 79

and experience about structure, function and behaviour of ecosystems. 80

There have been many contributions to a more profound understanding of ecosystems in the 81

past two decades (Smith et al., 2012) unifying approaches and identifying and removing 82

artefacts contributes to the development of more comprehensive ecosystem theory. There are 83

major benefits that can be delivered by the consolidation of existing models and theories in 84

order to address the challenges of representing different spatial and temporal scales, avoiding 85

the redundancy in model development (Rotmans, 2009). There has been debate about the 86

different approaches used (e.g. empirical vs. process-based, simple vs. complex, importance 87

of different processes) in ecosystem modelling. Most of the scientific knowledge associated 88

with these models is heterogeneous and dispersed and, therefore, not directly available to the 89

scientific and user community. Furthermore, there is limited information available on the 90

mechanistic hypotheses used in most of the existing models. Lack of adequate model 91

documentation has been described in previous studies (Russell and Layton, 1992). Because 92

of this, there is often a gap in understanding model structure, or expectations and certainty of 93

measured and modelled results between model developers and model users. There is a need 94

for a resource that unifies thoughts, ideas and observations to achieve the state-of-the-art in 95

ecosystem modelling. As of now, much of the critical information needed to describe 96

different processes in ecosystem models can only be found with individual model developers 97

and the “comment statements” found in their computer codes, hence it is often largely 98

4

inaccessible by the broader community. In addition, experimental conditions influence the 99

choice of model parameterization which can lead to differences in simulations. Hence, in 100

addition to detailed documentation of the models themselves, the experimental conditions and 101

choices made by modellers on how to set different parameters must also be fully documented. 102

This information is very important for scientific understanding of different ecosystem 103

processes and of model performance. 104

Acknowledging these challenges, and in an attempt to improve the communication and 105

understanding, an open web-platform, GRAMP, has been developed (www.gramp.org.uk). 106

This paper describes how the GRAMP web platform was initially developed, and 107

demonstrates several uses in scientific projects, and for policy formulation. We also present 108

the initial case study using the DNDC model, to illustrate its functionality and utility. Section 109

3 discusses the future development of GRAMP and the ways in which it can help with 110

unifying environmental modelling and assessment. 111

2 GRAMP 112

113

2.1 Aim and scope of GRAMP 114

1) To create an open web-platform with existing data and prior knowledge, in collaboration 115

with end-users, with every stage open to critical review and revision, to improve the 116

predictions of soil C & N cycling in agro-ecosystems in the context of climate change. This 117

will involve classifying the various models according to their capabilities and specificities. 118

2) Establish a vibrant network of specialist researchers, model developers and users who can 119

work together, to examine strategically what the various models currently available can 120

deliver in accounting for the effect of ecosystem management on GHG emissions, to identify 121

promising mitigation options, and to assess the effect of future climate on emissions. 122

5

3) Link a global network of experimental sites to provide suitable data for testing, tuning and 123

validation of models and their derivatives across different crops, management strategies, soil 124

types, and climates. 125

4) Develop protocols for model development, application, calibration and evaluation with the 126

aim of providing an unprecedented level of detail in describing models and simulations. 127

5) Allow network members to exchange information, experience and data and provide a 128

forum for model development for future needs. 129

Users: four types of users are identified, viz; 1) researchers working on model development, 130

2) researchers using models for various outputs, 3) students who want to be trained in 131

ecosystem modelling, 4) researchers interested in policy making, based on modelling 132

outcomes. 133

Content and database management system: GRAMP will allow users to link databases for 134

use by the modelling community. The GRAMP platform contains a list of management 135

system and a database system which are searchable by region, crop etc. GRAMP will host a 136

set of links to global databases like NitroEurope(C1 and C3 database), CarboEurope, 137

GRACEnet and REAP databases (Del Grosso et al., 2013) etc., with associated metadata. It 138

also contains a web-GIS linked mapping system with a reference library, a database system 139

and training materials (case studies, demos, videos). 140

141 Functionality and outputs: The web platform will host the existing ecosystem models with a 142

version control system. This will allow users and model developers to create version specific 143

documentation. All the models entering the platform need to develop a model tree with 144

documentation (Figure 1). GRAMP describes the performance of different model versions, 145

6

which allows users to identify changes, and the implications of those changes on output 146

variables. 147

2.2 GRAMP platform design: 148

149

The website was built upon Python's Django open source web framework. Django is a free 150

and open source web application framework, written in Python programming language. Use 151

of Django eases the creation of complex, database-driven websites like GRAMP. The 152

website has a custom-built user authentication system which implements the Django 153

Guardian project for multiple tiers of permissions depending on a user's GRAMP affiliation 154

and offers several tools to implement management of those permissions. After initial access, 155

the users are allowed to enter any of four categories: 156

1. Data records system: The database system has been classified into four categories: (i) 157

project resources (ii) web resources (iii) model version records and (iv) application records. 158

The project resources will include links to a global wide database, and metadata associated 159

with each experimental dataset. Users can also add new database links to other databases by 160

following the standard protocol provided on the website. Field databases identified by the 161

collaborators will be collected from various sources, harmonized (where possible) and placed 162

in the database system. Project resources store the records, for example of measured 163

emissions of GHG from different ecosystems, which would be suitable for the further 164

development, calibration or evaluation of the models. It will also store the records of a 165

centralized database that is harmonized with clear and full attribution of the sources of the 166

data, authorship, measurement methods, referencing, etc. Web resources provide links to data 167

without harmonization. There are several good experimental databases in existence (e.g. 168

Croplands research database : Liebig et al., 2013; Australian N2O 169

Network: http://www.n2o.net.au etc.,), so direct links will be provided in this category. 170 7

Model version records keep the summary of model versions in a specific format that are used 171

in the modelling portal. Application records are the bibliographic references which are 172

classified according to a set criterion and linked to almost every other entity in the database; 173

the corresponding information will be made accessible to all users. 174

2. Model repository: A repository where models can be stored and accessed with a detailed 175

description of the most relevant processes, authors, version history etc. The repository uses 176

version-control tools. This will also provide version-specific documentation, which is easily 177

accessible, complete, standardized, mutually comparable and transferable to different 178

applications. The database is accessed via a web interface which allows modellers to search 179

and download different versions of the models in the form of ready-to-compile software. 180

Modellers can also add their own models to the existing repository. This also provides best-181

practice guidelines, on-line tutorials and links to modelling and data provider research 182

groups, and their associated publications. 183

3. Model application: Model performance with different model versions is documented in 184

this category. Different statistical performance indicators are used to compare the 185

performance of different versions of model. Model performance is also assessed by 186

considering biological meaning (processes), in addition to statistical significance. Model 187

versions that constantly fail to predict known patterns, or those that generate implausible 188

estimates will be viewed as untenable for given applications. 189

4. Research & education: This category provides the training manuals, videos, tutorials for 190

new users, and provides FAQs. Users are allowed to interact in the forums and raise 191

questions and get help from worldwide colleagues to solve questions. Tools are provided for 192

blogging, which allow experienced users, developers and other researchers to communicate 193

8

with the audience. GRAMP also has the capabilities to organize Webinars, which allow 194

scientists across the world to attend web-based seminars. 195

196

Figure 1. A schematic representation of the GRAMP network 197

198

2.3 Data record system under GRAMP 199

200

2.3.1 Project resources 201

We developed a simple template for researchers to document research projects that have 202

measured emissions of GHGs from agricultural land, which could be suitable for the 203

development, calibration or evaluation of models. The template is a Microsoft Word 204

document that uses named fields for automatic extraction of the data. This will enable 205

automatic generation of web-site pages from the records. The template will be available for 206

9

download from the web-site, to allow researchers to submit formatted records of their 207

projects for inclusion in the GRAMP database. 208

The template collates project information on (i) project location and duration (ii) contact 209

details for the coordinator and organisation (iii) description of work done and method used 210

(iv) published papers and reports (v) site measurements available for input to the ecosystem 211

models such as site climate, soil properties, land use and grazing practices, fertiliser and 212

manure inputs (vi) if site measurements are available, the type of site measurement 213

parameters, (vii) expert opinion on best use of the dataset. To demonstrate use of the 214

template, we have completed examples for 6 national and 2 European scale projects which 215

are available on GRAMP (section 3). 216

2.3.2 Web resource records 217

A set of searchable ‘card’ records are created to summarize existing web resources relevant to 218

measurement and modelling of GHG emissions that would be of interest to users of the 219

different models. Each record is formatted according to a template, and can be stored in a 220

relational database for easy search. Each web resource record provides a description of the 221

purpose of the web site and the types of information available, along with contact information 222

and any restrictions on data access. A total of 50 web resource records have been prepared to 223

date, based on the standard template format. In the future, further records may be added by 224

the user community using this template. 225

2.3.3 Model version records 226

GRAMP allows a set of searchable ‘card’ records to be created, summarising versions of the 227

model that can be used in the modelling portal. Each model record will be a formatted record, 228

stored in a relational database, and as such, each record follows a standard template format. 229

10

Each model record includes a description of model version, an explanation where possible of 230

its link to the original model form; details of any modifications and version numbers; and a 231

general description of any validation and specific data requirements. The biopic provides 232

pointers to the home-page where the model executables and manuals can be downloaded, if 233

available, and also provides citations of key papers describing each model version. As an 234

example we produced eighteen model records for versions of the DNDC model by combining 235

literature searches, web searches and DNDC community expertise. 236

2.3.4 Application records 237

This section contains a database of papers published in peer-reviewed journals that describe 238

the development or application of the model. Each paper was classified according to a set 239

criterion to enable the database to be searched for previous applications of the model to areas 240

of interest defined by land use and region, and types of study outcome, such as a regional 241

emissions inventory or an improved process description. For each publication, we have 242

produced a study record. Each study record contains 12 classes (Figure 2). The Web Portal 243

will display the list of papers, and the links to the source journals, as the paper abstracts are 244

generally copyrighted and cannot be displayed. We have classified all of these papers into 245

eight categories (Figure 2). The classification will allow users of the web portal to rapidly 246

identify papers that are relevant to their needs. The classification system anticipates other 247

GHG models, and other types of models. All the papers that belong to one model version are 248

linked to the model tree (Figure 2). 249

11

250

Figure 2: Description of the database structure describing the linkage between publications, 251

their classification and the model to which they refer within GRAMP. 252

Here we present a bibliography associated with DNDC model as an example. Papers were 253

identified by searching for the term ‘*DNDC*’ in the ‘Web of Knowledge’ and ‘Scopus’ 254

search engines. A total of 248 papers were identified. All these papers are categorized 255

according to the classification system presented above. The papers collectively provide 256

trends in DNDC model development and application. As shown in Figure 3a, the majority of 257

research papers published have used the original DNDC model version. DNDC was initially 258

developed in the USA, it has been used and tested extensively in Asia (Figure 3b), followed 259

by Europe and North America. DNDC has been applied in many land uses, but the majority 260

of applications have been in croplands, followed by agricultural grasslands and paddy fields 261

(Figure 3c). DNDC has primarily been used for GHG quantification and soil C and N 262

dynamics, as shown in Figure 3d. Sixty eight percent of literature focused on quantification 263

of environment fluxes under present-day land management practices, such as fertiliser inputs, 264

livestock grazing regime and crop rotations – at field, farm or landscape scale. Only 15% 265

studies focused on quantification of the impact of changing climatic rainfall and temperatures 266

on different ecosystems (Table 1). 267

268

269

12

270

271

272

273

274

275

276

277

278

13

279

280

281

Figure 3. Percentage of publications that used (A) different version of DNDC (B) different 282 regions (C) different land use and for (D) different research purposes. 283

284

285

286

14

No. Name Description % of Papers

1 Development, integration and testing

Detailed description and testing of new algorithms for improved process representation. 25.0

2

Measurement and verification

Comparison of model outputs with measured fluxes at plot and field scale for verification and calibration of the model parameters. 57.0

3 Inter comparison

Comparison of the abilities of different models or model versions to reproduce measured fluxes 16.0

4 Sensitivity and uncertainty

Analysis of the sensitivity of model outputs to varying the scale and range of input data and internal model parameters. 27.0

5

Scenario evaluation

Application of the model to calculate the impact of, for example, a change in land management or climate change on simulated fluxes. 34.0

287

Table 1. Percentage of papers which cover different aspects of model use, development and 288 testing. 289

290

291

292

293

294

295

296

297

15

2.4 Model tree and repository 298

Ecosystem model construction is an iterative process in which the modeller often develops a 299

number of models or variants due to changes in the underlying assumptions made about the 300

system. The number of assumptions and simplification of the system, increases or decreases 301

depending on the contemporary understanding of the system and objective of the model. As a 302

result, a number of model representations will emerge, with one of these ultimately being 303

used for the desired purpose. During this refinement of models, the changes that are made to 304

the model normally diverge from the original design or process of the model. There is a need 305

for continued documentation which explains how each model version differs, and why each 306

was created. Ultimately, the modelling community is interested to know how the existing 307

model was changed to justify the creation of a new model, or model version. To improve 308

current modelling practice, GRAMP describes a framework for developing a “Model Tree”, 309

in which other tools such as a model repository work together for greater productivity and 310

transparency. 311

A Model Tree is a hierarchical collection of models which provide many different 312

representations of the same system. These are collated in a manner which focuses on the 313

similarities and differences between each model in the collection. The specific differences 314

between individual models are recorded as model members. The use of Model Tree and 315

model families makes it possible to store a large number of models of the same system, 316

improving understanding of the system and allowing reuse of concepts or ideas. Each version 317

in the Model Tree is associated with the model repository. The aim of the GRAMP model 318

repository is to provide access to an up-to-date collection of ecosystem models or model 319

versions. This model repository ensures that the model is curated, which is important to 320

ensure that the model is able to accurately reproduce the published results. This tool brings 321

together a rich set of features for the analysis, management and usage of large sets of process 322

16

models. The repository holds models along with conceptual metadata, rather than as 323

mathematical equations or programming language code. The conceptual representations of 324

models in metadata enhance the use and improve the understanding of models by various 325

stakeholders. 326

327

2.5 Model performance 328

329

Linking detailed model description with model performance might help in improving process 330

understanding and detecting the origin of some model errors. Most of the time model 331

calibration is carried out by trial and error or by using optimization techniques. Both of these 332

methods are designed to search the parameter space for combination of parameters which 333

provides the best fit. There is sufficient information provided in the literature on general 334

aspects of model structure but little is presented about the values of model parameters. 335

Without this information it is difficult to assess whether the lack of fit is due to the 336

inadequacy of model structure or due to poor parameter choice. This information also helps 337

in improving scientific interpretation and transparency in model analysis. 338

3 Pilot study of GRAMP using the DNDC model 339

340

We present here a case study with the DNDC (DeNitrification-DeComposition) model to 341

demonstrate the functionality and utility of the major features of the GRAMP tree and model 342

repository. Prototyping with the DNDC model presented in this paper demonstrates its 343

feasibility, as well as an outlook to the further developments of GRAMP. We piloted this 344

study with the DNDC model due to its wide-spread use throughout the world. To develop a 345

DNDC Model Tree under GRAMP, we reviewed DNDC model versions and documented the 346

17

important chronological changes made to the model. We reviewed papers published in peer-347

reviewed journals that describe the development or application of the DNDC model. Each 348

paper was classified according to criteria to enable the database to be searched for previous 349

applications of the DNDC model, to areas of interest defined by land use and region, and 350

types of study outcome, such as a regional emissions inventory or an improved process 351

description. A total of 248 papers were identified for this study. The aim was to build a 352

Model Tree to identify the major processes in each version of the model. The ability within 353

GRAMP to create an easily exchangeable model tree knowledgebase is relevant in this 354

respect. 355

3.1 DNDC model families 356

Several standalone versions evolved from DNDC, sharing most of the sub-models of the 357

original DNDC. Many standalone versions of DNDC were regionalized by incorporating 358

regional-specific management or parameterization of the model (Figure 4). There were 359

several versions of DNDC developed during the last few decades. Many of these 360

modifications have been incorporated into the latest standalone versions of DNDC (Giltrap et 361

al., 2010.). There are several standalone versions of DNDC, the most stable of which have 362

been reviewed and tracked through GRAMP. Constructing models in this manner enables the 363

modeller to retain various representations of DNDC in one location. This simple change in 364

model typology dramatically improves the model repository by eliminating most of the 365

repetition in modelling. 366

18

367

368

369

370

371

372

373

374

375

376

377

378

379

380

381

382

383

384

385

386

387

388

389

390

391

Figure 4: Schematic diagram of the DNDC extended family. By detailed literature review, we identified the following standalone versions of DNDC: 1) PnET-N-DNDC 2) Crop DNDC 3) Wetland DNDC 4) Rice DNDC 5) Forest DNDC 6) Landscape-DNDC 7) Forest DNDC-Tropica 8) Manure-DNDC 9) Mobile-DNDC 10) NZ-DNDC 11) DNDC - EUROPE 12) EFEM-DNDC 13) NEST-DNDC 14) BE-DNDC 15) DNDC-CSW 16) UK-DNDC.

19

3.2 Example of DNDC model performance 392

393

In an attempt to evaluate the current state of the DNDC crop model as an example we present 394

a meta-analysis of 363 modelling studies published in the peer-reviewed literature between 395

1990 and 2013. GRAMP has the user interface to display the model with associated 396

simulation results. The model performance tab shows the systematic goodness-of-fit 397

assessment of the original models, i.e., plots in which simulated values were visually 398

compared with observed data. The model performance window will have the capacity to 399

show graphs comparing modelled and observed values in various formats. Under GRAMP a 400

diagram has been devised that can provide a concise statistical summary of how well daily or 401

annual field observations match the model simulations in terms of their correlation, their root-402

mean-square difference, and the ratio of their variances. Representing the results in this form 403

is especially useful in evaluating complex biogeochemical models. It will also be capable of 404

showing the location of these field sites on world maps. This process helps in identifying the 405

parts of the model that needs to be improved. This is an important tool to evaluate the current 406

state of ecosystem models and rigorously assesses what the model can or cannot predict. This 407

tool can show statistically significant trends of the model performance. 408

Despite the heterogeneity of the modelling studies examined with respect to model 409

complexity, type of ecosystem modelled, spatial and temporal scales, and model development 410

objectives, this study revealed statistically significant trends of the DNDC model 411

performance. Here we present the predictions of N2O emissions by the DNDC crop model as 412

expressed by the coefficient of determination (r2). As shown in Figure 5 & 6, predictions of 413

cumulative annual N2O emissions improved over several versions. Our analysis is limited by 414

the number of samples and heterogeneity in these modelling studies. 415

20

416

417

418

419

420

421

422

Figure 5 : Measured and modelled total or annual N2O sorted by model version, extracted 423

data from publications. 424

425

426

427

0

2

4

6

8

10

12

14

16

18

20

0 5 10 15 20

mod

elle

d N

2O [k

g N

2O-N

ha-1

yr-1

] [k

g N

2O-N

ha-1

seas

on-1

]

measured N2O [kg N2O-N ha-1 yr-1] [kg N2O-N ha-1 season-1]

all data (r2=0.84, n=192)

DNDC (Li et al. 1992)(r2=0.89, n=10)DNDC (Li et al. 2002)(r2=0.85, n=19)DNDC 7.1 mod (r2=0.73,n=6)DNDC 7.2 (r2=0.24, n=18)

DNDC 7.2 (org+mod)(r2=0.32, n=23)DNDC 8.3P (r2=0.44, n=25)

DNDC 8.9 (r2=0.67, n=23)

DNDC 9.1 (r2=0.81, n=4)

21

428

429

430

431

432

433

434

435

436

437

438

439

440

441

442

443

444

Figure 6: Measured and modelled N2O emissions (A) Annual and (B) daily total N2O sorted 445

by model land use, extracted data from publications. 446

0

5

10

15

20

0 5 10 15 20

mod

elle

d N

2O [k

g N

2O-N

ha-1

yr-1

] [k

g N

2O-N

ha-1

seas

on-1

]

measured N2O [kg N2O-N ha-1 yr-1] [kg N2O-N ha-1 season-1]

all data(r2=0.84,rmse=6.12,n=192)crop (r2=0.91,rmse=6.2,n=93)

rice(r2=0.96,rmse=1.7, n=29)

grassland(r2=0.7,rmse=9.2, n=46)

A

0

5

10

15

20

25

30

35

0 10 20 30

mod

elle

d N

2O [g

N2O

-N h

a-1 d

-1]

measured N2O [g N2O-N ha-1 d-1]

all data(r2=0.62,rmse=5,n=173)grassland(r2=0.14,rmse=0.7,n=21)grassland + pasture(r2=0.91,rmse=4.6,n=15)rice(r2=0.71,rmse=20.4,n=3)forest(r2=0.45,rmse=4.6,n=139)B

22

Registered users of GRAMP can upload the simulated results to an existing database. It is 447

anticipated that database will grow over a period of time and give a snapshot of model 448

performance. In this analysis, daily N2O emissions were poorly modelled (r2), indicating that 449

the performance of DNDC model declines as we move from annual to daily time step (Figure 450

6A & 6B). This model performance tool can be used to summarize the relative merits of a 451

collection of different models or to track changes in performance of a model as it is modified. 452

4 Discussion 453

The modeller’s task is to identify or develop an appropriate model or methodology for a 454

given modelling objective (Wagener et al., 2003). Experience shows that identifying or 455

developing a best methodology is difficult due to several different conceptualizations of 456

ecosystems, which may yield equally good results. This ambiguity has serious implications 457

for models and limits the applicability of ecosystem models for the simulation of land use or 458

climate-change scenarios, or for regionalization studies (Moore and Clarke, 1981). There is a 459

rapidly growing literature on ecosystem models predicting soil C (Liu et al., 2009; Smith et 460

al., 2010), N dynamics (Bell et al., 2012; Giltrap et al., 2010; Thorburn et al., 2010) , GHG 461

emissions (Hutchings et al., 2007; Smith W. N et al., 2008), ecosystem services (Schröter et 462

al., 2005) and climate change mitigation (Del Prado et al., 2013), from different ecosystems 463

(De Gryze et al., 2010). As these models develop, the challenges of information accessibility, 464

data comparability and unification of existing methods become more prevalent. New research 465

approaches must be developed to support decision-making for the management of ecosystems 466

and natural resources (Parker et al., 2002; Spielman et al., 2009; Walker, 2002). 467

GRAMP is an open-source platform, where scientists can collaborate freely and share 468

data. GRAMP allows the creative and productive powers of numerous individuals and 469

research groups to be harnessed with the common goal of quantifying GHG emissions and 470

23

simulation of soil C & N dynamics across broad geographic regions and multiple spatial 471

scales. It is an integrated, web-accessible knowledge base that allows temporally and spatially 472

explicit data to be linked to dynamic simulation models. Anyone can participate by 473

registering on the site as model users or as developers. It provides various services, such as: 474

version control, code sharing, modelling tools sharing and support organizing online training 475

sessions, tutorials and webinars. It allows greater interactions among different scientific 476

communities across the world who are interested in the study of soil C and N dynamics and 477

climate change. 478

In addition, the GRAMP meta-database resource provides information for researchers on the 479

existence and availability of data applicable to a wide range of agricultural and environmental 480

questions. The metadata base has proved useful for many applications and is freely available 481

for many more via the GRAMP web portal. Working on a common platform using 482

standardized models should enable the harmonisation of many existing methodologies. 483

5 Conclusions and future outlook 484

485

The aim of GRAMP is to develop a web resource that will serve as a central hub for 486

information on agriculture GHG emission modelling. GRAMP is anticipated to increase the 487

modelling research capacity and to accelerate improved reliance on models to predict GHG 488

emissions and test mitigation practices. GRAMP will bring greater transparency in model 489

development and application, which will help in the advancement of ecosystem modelling. 490

GRAMP will collect and document a comprehensive and standardized set of metadata for 491

ecosystem model simulations. Using this web-platform, the modelling community, along 492

with end users, can build well-documented models and harmonise existing methodologies. 493

The metadata archive and model repository will provide a much more comprehensive and up-494

24

to-date description of ecosystem models than is typically available in journal articles or 495

reports. The open-source community managed GRAMP as a metadata repository is 496

anticipated to spur the development of cutting edge modelling techniques. GRAMP will 497

advance the fundamental understanding of C-N interactions at different scales, and improve 498

the interaction between modellers, experimentalists and users, to synthesize solutions in the 499

problem areas of model application and validation. GRAMP will act as a global 500

communication tool between research teams and model users, specifically interested in the 501

measurement and modelling of GHG mitigation. 502

Acknowledgements: 503

The authors are grateful to the UK Government Department of Environment Farming and 504

Rural Affairs (DEFRA), working within the framework of the Global Research Alliance on 505

Agricultural Greenhouse gases for supporting this project. PS is a Royal Society-Wolfson 506

Research Merit Award Holder 507

References: 508

Bell, M.J., Jones, E., Smith, J., Smith, P., Yeluripati, J., Augustin, J., Juszczak, R., Olejnik, 509 J., Sommer, M., 2012. Simulation of soil nitrogen, nitrous oxide emissions and 510 mitigation scenarios at 3 European cropland sites using the ECOSSE model. Nutr. 511 Cycl. Agroecosystems 1–21. 512

De Gryze, S., Wolf, A., Kaffka, S.R., Mitchell, J., Rolston, D.E., Temple, S.R., Lee, J., Six, 513 J., 2010. Simulating greenhouse gas budgets of four California cropping systems 514 under conventional and alternative management. Ecol. Appl. 20, 1805–1819. 515

516 Del Grosso, S.J., White, J.W., Wilson, G., Vandenberg, B., Karlen, D.L., Follett, R.F., 517

Johnson, J.M.F., Franzluebbers, A.J., Archer, D.W., Gollany, H.T., Liebig, M.A., 518 Ascough, J., Reyes-Fox, M., Pellack, L., Starr, J., Barbour, N., Polumsky, R.W., 519 Gutwein, M., James, D., 2013. Introducing the GRACEnet/REAP Data Contribution, 520 Discovery, and Retrieval System. Journal of Environment Quality 42, 1274. 521 doi:10.2134/jeq2013.03.0097 522

523

25

Del Grosso, S.J., Ojima, D.S., Parton, W.J., Stehfest, E., Heistemann, M., DeAngelo, B., 524 Rose, S., 2009. Global scale DAYCENT model analysis of greenhouse gas emissions 525 and mitigation strategies for cropped soils. Glob. Planet. Change 67, 44–50. 526

Del Prado, A., Crosson, P., Olesen, J.E., Rotz, C.A., 2013. Whole-farm models to quantify 527 greenhouse gas emissions and their potential use for linking climate change mitigation 528 and adaptation in temperate grassland ruminant-based farming systems. animal 7, 529 373–385. 530

Foresight 2009 Global Food and Farming Futures. See 531 http://www.foresight.gov.uk/OurWork/ActiveProjects/FoodandFarmingFutures/Fooda532 ndfarmingProjectHome.asp. 533

Giltrap, D.L., Li, C., Saggar, S., 2010. DNDC: A process-based model of greenhouse gas 534 fluxes from agricultural soils. Agric. Ecosyst. Environ. 136, 292–300. 535

Hutchings, N.J., Olesen, J.E., Petersen, B.M., Berntsen, J., 2007. Modelling spatial 536 heterogeneity in grazed grassland and its effects on nitrogen cycling and greenhouse 537 gas emissions. Agric. Ecosyst. Environ. 121, 153–163. 538

Jørgensen, S.E., 1999. State-of-the-art of ecological modelling with emphasis on 539 development of structural dynamic models. Ecol. Model. 120, 75–96. 540

Li, C., 2007. Quantifying greenhouse gas emissions from soils: Scientific basis and modeling 541 approach. Soil Sci. Plant Nutr. 53, 344–352. 542

Liebig, M.A., Franzluebbers, A.J., Richard, G., Cellier, P., Rice, C., Shafer, S.R. 2013. 543 Quantifying greenhouse gas mitigation potential of cropland management practices: A 544 review of the GRA croplands research group greenhouse gas network. Meeting 545 Abstract. P. 112. In Abstracts of technical papers, 2013 joint meeting of the Canadian 546 Soil Science Society, Manitoba Soil Science Society, and Canadian Society of 547 Agricultural and Forest Meteorology, Winnipeg, MB. 22-25 July 2013. Canadian Soil 548 Science Society, Pinawa, MB 549

Liu, D.L., Chan, K.Y., Conyers, M.K., 2009. Simulation of soil organic carbon under 550 different tillage and stubble management practices using the Rothamsted carbon 551 model. Soil Tillage Res. 104, 65–73. 552

Moore, R.J. and Clarke, R.T., 1981.A distribution function approach to rainfall-runoff 553 modelling Water Resour. Res. 17(5), 1376-1382. 554

Parker, P., Letcher, R., Jakeman, A., Beck, M.B., Harris, G., Argent, R.M., Hare, M., Pahl-555 Wostl, C., Voinov, A., Janssen, M., Sullivan, P., Scoccimarro, M., Friend, A., 556 Sonnenshein, M., Barker, D., Matejicek, L., Odulaja, D., Deadman, P., Lim, K., 557 Larocque, G., Tarikhi, P., Fletcher, C., Put, A., Maxwell, T., Charles, A., Breeze, H., 558 Nakatani, N., Mudgal, S., Naito, W., Osidele, O., Eriksson, I., Kautsky, U., Kautsky, 559 E., Naeslund, B., Kumblad, L., Park, R., Maltagliati, S., Girardin, P., Rizzoli, A., 560 Mauriello, D., Hoch, R., Pelletier, D., Reilly, J., Olafsdottir, R., Bin, S., 2002. 561 Progress in integrated assessment and modelling. Environ. Model. Softw. 17, 209–562 217. 563

Rivington, M., Koo, J., 2010. Report on the meta-analysis of crop modelling for climate 564 change and food security survey. 565

Rotmans, J., 2009. Three decades of integrated assessment: the way forward. introductory 566 keynote paper, in: AgSAP Conference. 567

Russell, M.H., Layton, R.J., 1992. Models and Modeling in a Regulatory Setting: 568 Considerations, Applications, and Problems. Weed Technol. 6, 673–676. 569

Schröter, D., Cramer, W., Leemans, R., Prentice, I.C., Araújo, M.B., Arnell, N.W., Bondeau, 570 A., Bugmann, H., Carter, T.R., Gracia, C.A., 2005. Ecosystem service supply and 571 vulnerability to global change in Europe. Science 310, 1333–1337. 572

26

Shepherd, A., Wu, L., Chadwick, D., Bol, R., 2011. 1 A Review of Quantitative Tools for 573 Assessing the Diffuse Pollution Response to Farmer Adaptations and Mitigation 574 Methods Under Climate Change. Adv. Agron. 112, 1. 575

Smith, J., Gottschalk, P., Bellarby, J., Chapman, S., Lilly, A., Towers, W., Bell, J., Coleman, 576 K., Nayak, D., Richards, M., others, 2010. Estimating changes in Scottish soil carbon 577 stocks using ECOSSE. I. Model description and uncertainties. Clim. Res. 45, 179–578 192. 579

Smith, P., 2013. Delivering food security without increasing pressure on land. Glob. Food 580 Secur. 2, 18–23. 581

Smith, P., Albanito, F., Bell, M., Bellarby, J., Blagodatskiy, S., Datta, A., Dondini, M., 582 Fitton, N., Flynn, H., Hastings, A., others, 2012. Systems approaches in global change 583 and biogeochemistry research. Philos. Trans. R. Soc. B Biol. Sci. 367, 311–321. 584

Smith, P., Gregory, P.J., 2013. Climate change and sustainable food production. Proc. Nutr. 585 Soc. 72, 21–28. 586

Smith, P., Martino, D., Cai, Z., Gwary, D., Janzen, H., Kumar, P., McCarl, B., Ogle, S., 587 O’Mara, F., Rice, C., Scholes, B., Sirotenko, O., Howden, M., McAllister, T., Pan, G., 588 Romanenkov, V., Schneider, U., Towprayoon, S., Wattenbach, M., Smith, J., 2008. 589 Greenhouse gas mitigation in agriculture. Philos. Trans. R. Soc. B Biol. Sci. 363, 590 789–813. 591

Smith, W.N., Grant, B.B., Desjardins, R.L., Rochette, P., Drury, C.F., Li, C., 2008. 592 Evaluation of two process-based models to estimate soil N2O emissions in Eastern 593 Canada. Can. J. Soil Sci. 88, 251–260. 594

Spielman, D.J., Ekboir, J., Davis, K., 2009. The art and science of innovation systems 595 inquiry: Applications to Sub-Saharan African agriculture. Technol. Soc. 31, 399–405. 596

Thorburn, P.J., Biggs, J.S., Collins, K., Probert, M.E., 2010. Using the APSIM model to 597 estimate nitrous oxide emissions from diverse Australian sugarcane production 598 systems. Agric. Ecosyst. Environ. 136, 343–350. 599

Wagener, T., McIntyre, N., Lees, M.J., Wheater, H.S., Gupta, H.V., 2003. Towards reduced 600 uncertainty in conceptual rainfall-runoff modelling: dynamic identifiability analysis. 601 Hydrol. Process. 17, 455–476. 602

Walker, D.H., 2002. Decision support, learning and rural resource management. Agric. Syst. 603 73, 113–127. 604

605

27