Robust Rescaling Methods for Integrated Water, Food, Energy Security Management under Uncertainties and Risks

This project will develop new robust, non-Bayesian probabilistic cross-entropy based rescaling techniques to improve the integration of models and data used in the study of food, water, and energy security under inherent uncertainties and risks.

Improving our understanding of the complex processes involved in managing food, water, and energy security requires new research methods to integrating and rescaling of models, data, and decision-making procedures between various scales.

For example, scientists analyzing water security issues may use a hydrological model that requires inputs which are much finer than the resolution of, say, the economic or climatic models generating these inputs. With respect to food security, aggregate national or regional projections derived with global economic land use planning models give no clue regarding potentially critical local production-demand imbalances.

Many practical studies analyzing regional developments use traditional cross-entropy maximization as an underlying downscaling principle. Cross-entropy theory provides a powerful tool for spatially-detailed data estimation when prior information in locations may not be observed directly. However, the cross-entropy downscaling relies on a single prior distribution. In reality, prior distributions depend on various “environmental” parameters which may be uncertain. Therefore, instead of a uniquely defined prior there is a feasible set of these distributions.

The novelty of this project is that the estimation of local changes consistently with available aggregate data is formulated as probabilistic inverse problem in the form of, in general, stochastic non-convex cross-entropy minimization model requiring the development of proper stochastic optimization (STO) procedures. The STO enables new downscaling procedures, i.e., non-Bayesian probabilistic cross-entropy based disaggregation techniques, to derive local estimates that are, in a sense, robust with respect to all priors from the feasible set. Important feature of the procedures is that they treat two main cases of priors: compound priors and non-Bayesian priors.

The new procedures will be applied to downscale aggregate regional model projections of land use changes in Ukraine, China, and Brazil from IIASA’s GLOBIOM (Global Biosphere Management) model. The methodology will also be used to harmonize alternative land cover maps and to create hybrid maps for the GEO-wiki project.

From aggregate regional

to grid cell level

Collaborators

This work brings together experts from Ecosystems Services and Management Program, Advanced Systems Analysis Program (Systemic Risks and Robust Solutions Project), Institute of Economics and Forecasting and Institute of Cybernetics at the Ukrainian National Academy of Sciences.


Print this page

Last edited: 28 January 2015

CONTACT DETAILS

Tatiana Ermolieva

Research Scholar

Ecosystems Services and Management

T +43(0) 2236 807 581

Yurii Yermoliev

Institute Scholar

Advanced Systems Analysis

T +43(0) 2236 807 208

International Institute for Applied Systems Analysis (IIASA)
Schlossplatz 1, A-2361 Laxenburg, Austria
Phone: (+43 2236) 807 0 Fax:(+43 2236) 71 313