Spatially Specific Land Cover Econometrics & Integration With Climate Prediction: Scenarios of Future Landscapes and Land-climate Interactions
Eustáquio J Reis IPEA/DIMAC (SA-PI)
The proposed project will advance our ability to anticipate the sustainability impacts of development in the Amazon basin in two ways, building on past and ongoing LBA efforts. First, it will extend basin-scale empirical modeling of deforestation to the pixel-level to test for spatial elements of the driving processes, and to provide an econometric model for predicting future Amazonian land cover in considerable spatial detail. And second, it will integrate land cover impacts – generated by the pixel-level econometric model – with a regional climate model in order to predict climate impacts associated with a set of development scenarios for the Amazon Basin. The pixel-level land cover model will probabilistically predict Amazonian landscapes as a function of the spatial distribution and times paths of observed economically-relevant drivers of deforestation. Using disaggregate pixel data facilitates the construction of a model that can accommodate situations in which unobserved drivers are spatially correlated and also in which their variation occurs at aggregate spatial scales beyond the individual pixel. It also permits tests, and corrections, for spatial contagion of deforestation. Finally, a pixel-level analysis generates many observations, which enhances statistical power and allows the construction of spatially detailed land cover projections. The regional climate model, i.e., the Regional Atmospheric Modeling System (RAMS), requires this level of detail in projected land cover as an input. The proposed project will predict climate for scenarios developed in its predecessor (LC-24: A Basin Scale Econometric Model for Projecting Future Amazonian Landscapes) as follows. First, the econometric model of land cover will produce a set of deforestation probabilities, associated with the individual development scenarios, covering the entire basin. Each pixel, with scenario-specific probability, will be treated as a Bernoulli trial, and probability functions of GIS software will produce hundreds of realizations of the full basin landscape. These landscapes will serve as input for RAMS, which will be executed for as many times (per scenario) to produce climate outcomes (e.g. rainfall, max temperature) for an entire year. These will then be used to produce probability density functions (i.e., frequency histograms) of key variables (e.g., total yearly rainfall), with associated estimates of distributional parameters such as mean value (μ) and variance (σ2). This explicitly addresses uncertainty by developing measures of central tendency (μ) and dispersion (σ2) for the estimated climate impacts of the development scenarios.
Robert T. Walker Michigan State University (US-PI)