Application note High-performance computing for climate change impact studies with the Pasture Simulation model Jean-André Vital a , Michael Gaurut a , Romain Lardy a , Nicolas Viovy b , Jean-François Soussana c , Gianni Bellocchi a,⇑ , Raphaël Martin a a Grassland Ecosystem Research Unit, French National Institute for Agricultural Research (INRA), 5 Chemin de Beaulieu, 63039 Clermont-Ferrand, France b Laboratoire des Sciences du Climat et de l’Environnement (LSCE), CE L’Orme des Merisiers, 91191 Gif-sur-Yvette, France c Scientific Direction in Environment, French National Institute for Agricultural Research (INRA), 147 rue de l’Université, 75338 Paris, France article info Article history: Received 17 December 2012 Received in revised form 30 July 2013 Accepted 5 August 2013 Keywords: Climate change High-performance computing Job launcher Pasture Simulation model (PaSim) Pixel-based simulation abstract High-performance computing technology permits to efficiently achieve high-performance throughputs for intensive CPU load applications. We describe the development of an integrated tool for climate change impact studies on grassland ecosystems running with pixel-wise data. The pixel-based Pasture Simula- tion model (PaSim) is suited to work with a NetCDF format of input and output files. It includes the par- allel job launcher, which dispatches individual jobs to execute simulations. In a case study covering metropolitan France, we demonstrate how this approach is configured and used to evaluate the impact of climate change on grassland productivity. Over 10,000 pixels of 8  8 km resolution, we report 25 h to complete the simulation on a cluster machine (TITANE) with 200 processors, which is a speedup of 200. Ó 2013 Elsevier B.V. All rights reserved. 1. Introduction The need to store, process and distribute large volumes of data makes the use of high-performance computing an effective solu- tion to support the application of numerical modelling for regional and large-scale impact studies (e.g. Sloot et al., 2005). As in the case of climate change impact studies on natural and managed ecosys- tems, development of the global and regional climatic scenarios (resolved on spatially-gridded outputs) makes the number of sim- ulations prohibitive (Lin et al., 2004). This is even truer when high- resolution studies (fine scale of cells) are run over large territories (e.g. from regions to continents). High-performance computing stands at the core of computa- tional sciences (Fiore and Aloisio, 2011). Based on vector or parallel processors, it essentially consists of supercomputer hardware capable of providing an order or more of magnitude higher com- puting power than that offered by desktop workstations (after Tur- ton and Openshaw, 1998). A high-performance enabling approach could be hampered by the software and hardware complexity. The build-up of high performance-aware software solutions is thus a challenge for the analysis currently demanded by agricultural and environmental sciences. For instance, testing sets of adapta- tion (and mitigation) measures to reduce adverse effects of climate change requires exploring a high dimensional management sce- nario space (Smit and Skinner, 2002), and is a typical problem for which high-performance computing solutions are necessary. In the agro-environmental domain, scale changes and model linking methods are evolving (e.g. Ewert et al., 2011) as the com- plexity in modelling shifts towards applications at progressively larger scales. However, just a few of the great amount of biophys- ical models developed are engineered for distributed computing to meet the needs of high-resolution regional or continental simula- tions. A high-computing approach was documented by Nichols et al. (2011) with EPIC (Environmental Policy Integrated Climate) model for simulation of grain yields by crop at high-resolution spa- tial scales in central Wisconsin (USA). Zhao et al. (2012) developed high-performance computing capabilities to model high-resolution wheat production and soil carbon/nitrogen dynamics in Australia with APSIM (Agricultural Production Systems SIMulator) model. We addressed this matter with the Pasture Simulation model (PaSim, INRA-UREP, 2012, APP ID: IDDN.FR.001.220024.000.R.P. 2012.000.10000), originally developed by Riedo et al. (1998) to sim- ulate grassland systems. The development of PaSim is documented with respect to distributed simulations of climate change impacts. Grasslands will likely need supplemental studies about climate change impacts for their prominent role as carbon sinks and in providing food/feed services (e.g. Graux et al., 2012). Within this perspective, our primary purpose was to produce a new version of PaSim by mainly adapting and linking it to high-performance 0168-1699/$ - see front matter Ó 2013 Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.compag.2013.08.004 ⇑ Corresponding author. Tel.: +33 4 73624866; fax: +33 4 73624457. E-mail address: gianni.bellocchi@clermont.inra.fr (G. Bellocchi). Computers and Electronics in Agriculture 98 (2013) 131–135 Contents lists available at ScienceDirect Computers and Electronics in Agriculture journal homepage: www.elsevier.com/locate/compag