Modelling Land-Climate Interactions in Amazônia under Uncertainty
Moore, Michigan State University, email@example.com
Walker, Michigan State University, firstname.lastname@example.org
Arima, Hobart and William Smith Colleges, email@example.com
Silva, UFPA, firstname.lastname@example.org
Pfaff, Duke University, email@example.com
Robalino, Columbia University, firstname.lastname@example.org
Land-climate interactions in the Amazon basin are of mounting concern, given ongoing processes of agricultural conversion in the region, and possible impacts on climate, which could impose serious stressors on the basin’s remaining forest. To investigate these possible impacts, we linked a spatial econometric model (SEM) to a regional climate model (RCM), and explicitly incorporated variability into the modelling framework in order to make climate projections that take uncertainty into account. In essence, we treated the basin as a large collection of Bernoulli experiments, and used a Monte-Carlo approach in generating probability distributions of climate variables, as opposed to point estimates, thereby enabling assessments of statistical significance in observed changes. With the SEM, we projected a range of likely land covers for the Amazon, based on two development scenarios - “restraints-on-development” and “business-as-usual” - as well as total forest removal, and generated a set of possible land covers associated with each scenario, using the probabilistic features of the SEM. In this way we were able to explicitly represent variability in future land covers. We then conducted Monte Carlo-type simulations with the RCM (RAMS 4.4) driven by these landscapes for five different years, employed to introduce atmospheric variability. Absent restraints on development, we find that certain areas can expect annual rainfall declines of 3-5% that persist in spite of introduced uncertainty. These declines are strongly tied to key landscape features. Results indicate that land cover and land use change associated with major roads leads to locally reduced rainfall, in a more persistent way than ENSO events or other annual atmospheric features. For the case of total deforestation we found an average annual decline in rainfall of 10-20% across the entire basin.