For CONTROL-E40 (Figure 7(c)) the negative bias extends along the Andes into the things Peruvian Amazonia, northern Bolivia, and parts of Brazilian Amazonia and central Brazil. Positive biases are few and small for the latter run.During the equinox seasons (MAM, Figures 6(b), 7(b) and SON, 6(d), 7(d), resp.) important regional differences are also present. In particular during autumn, for CONTROL-EC4, an extended cool bias extends from the Andes and Altiplano, the northern Gran Chaco and the Selva Paranaense, with minor warm biases near the Equator and over the Argentine Humid Pampas. For CONTROL-E40 the negative bias practically spans all the Brazilian territory, and only limited warming appears over the Humid Pampas.

During spring (Figures 6(d) and 7(d)) the cool bias is mostly confined to the Andes and to the Altiplano in both runs, though for CONTROL-E40 the bias does extend into the low lands of Amazonia and Chaco alongside the mountain range. For both runs sizeable warm biases appear in the Humid Pampas, as well as over northeastern Brazil.RCMs in general, partly due to the inherent problems in the source AGCMs, have similar problems [23�C25]. In this case a particular feature is the strong positive bias in summer, mostly over the Pampas region. AM10 and [23, 25] using PRECIS with HadAM3P did not obtain as large a bias there. Hence, this feature could in principle be due to problems in the ECHAM4 Baseline driver, due to land surface process representation in the AGCM. Yet, because it is also present in ERA-40-driven runs, this suggests that this feature could be at least partially due to the domain considered.

In winter AM10 shows a warm bias in the northeast corner of our domain, albeit a weaker one, which however is not present in CONTROL-E40, suggesting that this feature could be related to the AGCMs driving PRECIS. However, comparison with AM10 biases over the domain shows that their run has a cold bias, mainly in summer, over most of Brazil.3.3. Model Response to Land-Use ChangesBoth ERA-40- and ECHAM4-driven PRECIS can thus reasonably represent during the 1960�C2000 period, in agreement with the previous RCM literature, the main climate features within the chosen domain. Therefore, using both of these drivers, PRECIS is now applied to study the model’s response to deforestation/land-use change on the domain as given in mean seasonal precipitation and temperature simulations.

The runs span the same period 1960�C2000, with the original land-use scenario included in the RCM (CONTROL-E40 and CONTROL-EC4 runs), the 2002 state of land-use [16] labeled MAP1 runs (MAP1-E40 and MAP1-EC4, resp.), and a hypothetical vegetation/land/use map proposed by Nepstad et al. [7] simulating AV-951 the state of Amazonia in 2030, as shown in Figure 1, labeled MAP2 runs (MAP2-E40 and MAP2-EC4, resp.).3.3.1.