CROSS Harmonization & HPC modelization of FOREST Datasets


Pilot  CAMBRIC  is  aimed  to  show  a  map  of  quality  and  quantity  of  wood  in  current  state  and  the evolution among next decades, for the Spanish territory. Two main tasks are required, calculate stocks and simulate its evolution. Herein below we will describe steps accomplished to yield a simulator able to perform this task as well as handle and transform required databases.

As a preliminary requirement, a simulator capable of running on HPC Calendula is required. There is already a web-based simulator, SIMANFOR, capable of running various types of models, based on .NET and C # technology. For this, it was decided to completely reprogram SIMANFOR to meet the Cross-Forest Project.

SIMANFOR  is the core  of  simulation and is  able  to  be  deployed  in  three  different  environments,  i) desktop  PC,  ii)  cloud  with  the  ability  of  run  in  different  nodes,  and  iii)  HPC,  similar  to  cloud  but compatible  with  MPI  and  OpenMP  communication.  Technology is typed Python 3.6  within  Dask  (Dask Development Team 2016) to support MPI programming (The Open MPI Project ©2004-2020 and Dask Development Team 2016).

SIMANFOR simulator is able to handle different input formats like CSV, EXCEL or JSON, what is defined in simulation input file.  Output  formats  can  be  the  same  type  as  input,  and  are  compatible,  so  an output can be used as input in a different simulation.

Simulator output can be tuned to represent the amount of wood available for each industry and year within a specific area.  These outputs stored in the project  database in  LOD  format,  are easily browsable by the sector stakeholders. A very important issue concerning our simulations is the definition of the stand evolution and the different interventions are being performed. Each scenario is stored as JSON file, and stored in the project database. Growth and yield models codification are constitutive key of the simulations, which allow simulator perform changes that occur in each lapse of time. These changes are in size, diameter and height, or in health status.  The  model  should  be  able  to  predict  mortality  or  ingrowth  of  trees.  A suitable  model  includes  the  following  equations:  Stand  Quality  Curves,  Diameter  and  height growth, Mortality, Ingrowth and Taper equations. Ibero models have been developed for Pinus pinaster and for P. sylvestris and programmed for previous version of SIMANFOR, and now transcribed to new SIMANFOR in Python.

For  evaluation  purposes  a  non-real  test  is  set  up,  based  on  a  specific  real-based  randomly generated  set  of  one  hundred  plots  with  one  hundred  trees  per  plot.  This kind of data provides a balanced source of data, without null data or any other kind of data that can disturb results. The  scenario  for  wood  production  is  used  to  evaluate  this  plots,  in  which  trees evolve from 20 to 150 years with 6 thinnings at 20, 30, 40, 50, 65 and 80 years age. Performance measured in computing time compares different versions of SIMANFOR:  current online version, linux-PC, windows-PC and HPC (without parallelization). New SIMANFOR is able to provide more tailored outputs, but for comparing purposes only total commercial volume is being analysed.