Paper No. 3
Presentation Time: 9:30 AM

SPATIAL DOWNSCALING IN CLIMATE MODELS: AN APPLICATION FOR GENERATION OF ALCOHOL PRODUCTION SCENARIOS IN BRAZIL


LINS, Daniela Barbosa da Silva, Center of Meteorological and Climatological Researches in Agriculture- CEPAGRI, University of Campinas- UNICAMP/Brazil, Inácia de Souza Veloso, 71 São Luiz, Itu/São Paulo, 13304277, Brazil, ZULLO JR, Jurandir, Center of Meteorological and Climatological Researches in Agriculture- CEPAGRI, University of Campinas- UNICAMP/Brazil, Cepagri - Cidade Universitária Zeferino Vaz Campinas, Campinas, 13083-970, Brazil and FRIEDEL, Michael J., Center for Computational and Mathematical Biology, University of Colorado, Campus Box 170, PO Box 173364, Denver, CO 80217-3364, danielablins@gmail.com

The Brazilian agribusiness has an oscillatory behavior as a function of market demand resulting from the expansion/contraction of cultivation areas. The current demand for sugarcane biofuels is a main driving force of changes in the Brazilian landscape. Some models incorporate climate as a factor to understand the dynamics of sugarcane transformation. The integration of climate data faces challenges in associated with downscaling from the coarse scale of global circulation models to the fine scales of applied studies. In this study, we relate data from different spatial scales for the prediction of spatial climate dynamics. The downscaling approach is is developed and validated using Eta climate model outputs from 2011 to 2090 with 40 km (24.85 miles) resolution and observed daily mean temperature and precipitation from meteorological stations (www.agritempo.com.br) in the 1991-2010 period. A type of unsupervised artificial neural network, Self Organizing Map, and cross-validation approach will be used to compare empirical downscaling techniques for climate data. Observed progresses will contribute to bring machine-learning techniques as a valid tool when dealing with multivariate data in the downscale context.
Handouts
  • GSA_final.pdf (1.1 MB)