Land-use and land-cover changes are driven by a multitude of interdependent socio-economic, environmental and natural processes. Analysis of these complex nonlinear processes requires new approaches to integrating and rescaling of models, data, and decision-making procedures between various scales. For example, socio-economic data and scenarios often exist at the national level, which does not permit the analysis and modeling of possibly alarming heterogeneities among the regions, in particular, income disparities between population groups. Scenarios of global and national agricultural production do not reflect conditions, uncertainties and potential outputs of individual farmers.

The main issue addressed within the project that develops aggregation (upscaling) and disaggregation (downscaling) techniques is to narrow the spatio-temporal mismatch between scenarios, models outputs, available data, decisions and the scales required for the policy analysis and implementation. The mismatch of scales creates a major source of uncertainties, which calls for the identification of proper indicators, new measures of uncertainties and risks, and goodness criteria for aggregation and disaggregation results. In particular, estimation of global processes consistently with local data and, conversely, local implications emerging from global tendencies challenge the traditional statistical estimation methods and criteria. The traditional methods are based on the ability to obtain observations from unknown and true probability distributions. For the downscaling and upscaling applications, we often have only very restricted samples of real observations. The real sampling model may not be known completely or may be incorrectly specified.

Additional observations may be difficult or costly to obtain. In this project, sequential downscaling and upscaling procedures are developed to solve the problem of data scarcity and incompleteness and to achieve the required spatio-temporal heterogeneities. To represent information in locations, the procedures rely on an appropriate optimization principle, e.g., cross-entropy maximization, and combine the available samples of real observations in the locations with other “prior” hard and soft data (expert opinion, scenarios), pseudo-sampling models, evidences on the related variables that exist in the form of interdependent observable, partially observable or indirectly observable and non-observable variables on all scales. Of particular interest is distribution-free non-parametric estimation of spatio-temporal interdependencies among the variables.

In practical applications, the choice of prior distribution is in the core of the downscaling methodology, ultimately determining the success of these procedures. A key issue is treatment of uncertainties in priors and the parameters of available constraints. In fact, the main idea is not the precise estimation of a prior but rather the design of pseudo-sampling models based on proper characterization of the real but unobservable values enabling to generate such a prior distribution which will ensure the robustness, in a sense, of the downscaling solution.

The developed sequential rescaling procedures (Fisher et al., 2004, Fisher et al., 2006) can be used in a variety of practical situations. Extensive testing of this procedure for downscaling of agricultural production, consistent with national statistics and compatible with various geographical and technical ancillary sources of information, has demonstrated that the iterative downscaling procedures are converging fast, allow for great geographical detail and are very flexible in model specification and detail.
Downscaling methods are essential elements in various projects of the LUC Program and are seen as a key ingredient of spatial modeling. Several tasks have been identified for developing, extending and applying this important and novel methodology:

• Research on specification of ‘prior’ distribution. This will be carried out for countries where detailed sub-national statistics are available, such as for China, Brazil, Europe, or United States, to test the results of different hypotheses on forming a ‘prior’.
• Sustainable livestock production allocation consistently with increasing demands and available economic-demographic, environmental, health, and resource constraints and risks at locations.
• Regional and global feed balances.
• Development of distribution-free non-parametric estimation procedures for data harmonization and formal treatment of imprecision and data gaps in geographical layers of GIS, remote sensing data, and statistical input data.

Responsible for this page: Elisabeth Kawczynski
Last updated: 16 Oct 2009