New land use change scenarios for Brazil: refining global SSPs with a regional spatially-explicit allocation model

The future of land use and cover change in Brazil, in particular due to deforestation and forest restoration processes, is critical for the future of global climate and biodiversity, given the richness of its five biomes. These changes in Brazil depend on the interlink between global factors, due to its role as one of the main exporters of commodities in the world, and the national to local institutional, socioeconomic and biophysical contexts. Aiming to develop scenarios that consider the balance between global and local factors, a new set of land use change scenarios for Brazil were developed, aligned with the global structure Shared Socio-Economic Pathways (SSPs) and Representative Concentration Pathway (RCPs) developed by the global change research community. The narratives of the new scenarios align with SSP1/RCP 1.9, SSP2/RCP 4.5, and SSP3/RCP 7.0. The scenarios were developed combining the LuccME spatially explicit land change allocation modeling framework and the INLAND surface model to incorporate the climatic variables in water deficit. Based on detailed biophysical, socio-economic and institutional factors for each biome in Brazil, we have created spatially-explicit scenarios until 2050, considering the following classes: forest vegetation, grassland vegetation, planted pasture, agriculture, mosaic of small land uses, and forestry. The results aim at regionally detailing global models and could be used both regionally to support decision-making, but also to enrich global analysis.

1 Introduction 1 Land available for agricultural expansion is an increasingly scarce resource in several 2 global regions [1,2]. The agricultural frontier's expansion, currently concentrated in the 3 tropics [3,4], affects the regulation of the hydrological and climatic regime, local 4 socioeconomic relations and generates a great biodiversity loss. This context could be 5 aggravated if we consider population increase and demand for food 2050 projections 6 (25% and 40%, respectively) [5]. 7 In this context, global models and scenarios, particularly the ones quantified with 8 Integrated Assessment Models (IAMs), that represent complex interactions and 9 July 24, 2021 1/14 feedback on a long term scale between the socioeconomic system (including climate 10 policies) and the natural system [6], playing a key role in helping us to understand the 11 impacts and consequences of agricultural expansion in different regions. In Brazil, for 12 example, this process over the last few decades has contributed to the country 13 consolidating worldwide as one of the main commodity-exporting countries, whether 14 agricultural or mineral. One of the key impacts of this process is the loss of natural 15 vegetation in the Amazon and Cerrado biomes. On the other hand, other areas in 16 Brazil, such as the Atlantic Forests, are undergoing a forest transition process [7]. The 17 integration and understanding of the factors that influence land use and land cover 18 changes (LUCC) in Brazil in different regions are important in order to define indicators 19 for guiding public policies to establish sustainable development strategies. 20 Global models and scenarios may fail to capture the regional land change dynamics 21 as they do not always include local factors, regional narratives, the national political and 22 institutional structure, and the dynamics and magnitude of intraregional drivers, which 23 determine the demand for land. In addition, the information used in most global models 24 is aggregated for comparability between large regions, such as continents, etc. [8][9][10]. In 25 this sense, Dala-Nora [8] points out that a balance between global and local factors is 26 necessary since the integration between these complex factors that operate on a global 27 and regional scale through extensive flow networks can change the structure and 28 consistency of land use change scenarios. In addition,VanVuuren [11] points out that 29 studies that examine phenomena of a more precise scale should consider more detailed 30 information (eg, geographic characteristics, land use patterns, or the location of cities). 31 In this paper, we present a new set of land change scenarios for Brazil, aligned with 32 the global Shared Socio-economic Pathways (SSPs) and Representative Concentration  [16][17][18][19][20]) and adopting them as a reference allows us to better link our scenarios to 37 the global context. The regionalized scenarios developed here aim to represent the 38 diversity of processes linked to land use change in the Brazilian territory. The modeling 39 approach considers Brazilian biomes' interregional socio-ecological differences, including 40 a more detailed analysis scale, without losing their relationship with global relations. 41 We model changes in natural vegetation, large and small-scale agricultural lands, and 42 planted forests. These land change processes are directly related to regional and local 43 factors, and the global context still plays a significant role in these processes. The regional scenarios were quantified by the LuccME modeling framework [21].

47
LuccME is a spatially explicit dynamic modeling structure for LUCC developed at the 48 Institute for Space Research -INPE. This approach makes it possible to delineate the 49 spatial patterns of land use and land cover classes based on the components of (a) 50 Demand, that is the amount/intensity of changes in each use that is intended to be 51 allocated over time [22,23]; (b) Potential, which corresponds to the adequacy that a 52 given cell in the cellular space has to change with each step of time, complete, in this 53 case, using the Spatial Lag regression model [23][24][25] and (c) Allocation that spatially 54 and interactively distributes the LUCC according to the previous components (demand 55 and potential), based on competition between classes of land use in each cell. The land 56 use and cover data used in the LuccMEBR comes from IBGE [26]. We chose this 57 database due to its national scope, periodicity (2000, 2010, 2012, and 2014), and classes. 58 Also, IBGE data is consistent with other regional mappings as TerraClass The area is occupied by grassland vegetation subject to grazing and other low-intensity anthropic interference. To be continued.  2.2 Scenario assumptions: from global to regional 78 The SSPs are based on five different development paths for societal trends ( We use the LuccME "PreComputedValues" component, in which we externally calculate 105 demand and report to the model the expected area for each land use class annually, in 106 the period 2000 to 2050 (Tabela 4) . As described above, we use the amount of change 107 projected by IMAGE and adjusted to the IBGE land use and land cover classes for the 108 SSP1/RCP1.9, SSP2/RCP4.5, and SSP3/RCP7.0 combinations, to generate the annual 109 demand for each land use class in each scenario, between 2015 to 2050.

110
Equation 1 presents the calculation of the annual change C ca for each class of land 111 use and land cover in area unit: where C ca corresponds to the annual change, in area, of the land use class L c where D cat k corresponds to the annual demand, in area, of a given land use class L c 118 in a given year t k , calculated from the sum of the class area in the previous year t k−1 119 and the annual change C ca .  Forest Code is not respected, and deforestation control measures are discontinued. Protected areas are not fully implemented and protected.

Roads network
No major Federal or State roads were built after 2020.
Same as Scenario C, but accompanied with measures to avoid uncontrolled occupation.

Rural settlements
Existing settlements are maintained, and nonconventional ones (sustainable) are well protected.
Existing settlements are maintained, but the nonconventional ones (sustainable) are less protected than Scenario A.
Rural settlements are canceled, and their areas become private property.

Protected areas -PAs
Maintenance of the 2016 protected areas network. Fully protected.
Same as Scenario A in terms of area, but less protected in more densely occupied areas. The LuccME component is used to determine the potential occurrence of a given land 126 use cover class, as well as the PotentialCSpatialLagRegression (Equation 3), which 127

Decrease in the extension and level of protection of the UCs from 2030. Maintenance only of Regularized and Approved Indigenous Lands
July 24, 2021 7/14 is based on and adapted from the spatial regression model (Spatial Lag) [23][24][25] Pot cxy t = %RegL cxy t − %L cxy t−1 : {P ot c xy t ∈ − 1 ≤ P ot cxy t ≤ 1}, where Pot cxyt corresponds to the potential for the occurrence of a given land use  (Fig. 2). 144 For that, the FillCell plugin was used [29]. The use of cellular space made it possible 145 to homogenize the factors described above, regardless of their origin format (vector data, 146 matrix data, etc.), aggregating them in the same space-time basis, through operators according to the regions of the Brazilian territory were also considered. The allocation 167 process for each type of land use or land cover can be described using equation 4.

168
L cxy t = L cxy t−1 + P ot cxy t * IT F c , where the amount of area allocated from a given class of land use L c at a given xy 169 location in the cell plane at time t is determined in an iterative process of the sum of 170 L cxy at time t-1 and the potential Pot cxyt , multiplied by an adjustment factor 171 proportional to the difference between the allocated area, the reported demand, and the 172 direction of the change ITF c .
where NS corresponds to the level of similarity between the real and simulated 184 maps at a given resolution i ; j is the window considered; n establishes the number of 185 windows to be considered; textitc is the number of cells in a resolution k (i*i); and 186 dif real = % real tf -% real ti and dif sim = % sim tfinal -% real initial being ti and tf the 187 initial and real years, respectively, considered in the validation.  [32], the result obtained from this process has greater consistency between the 210 different spatial and temporal scales of interest. In addition, the development and study 211 of regional scenarios help policymakers and the scientific community to develop robust 212 strategies in the face of uncertain futures, evaluating and improving the feasibility, 213 flexibility, and concreteness of their actions [33][34][35][36][37]. Analyzing the dynamics of LUCC (Table 5), according to the scenarios considered, 222 has been observed, Agriculture, the Mosaic of occupation, as well as Grassland 223 vegetation, will continue in the same direction, regardless of the scenario considered. In 224 relation to the other classes, it can be seen that the Sustainable development scenario is 225 distinguished from the others, as well as in the spatial pattern observed in Fig 4. 226 values observed in the other scenarios. As seen in Fig 4 and  Mosaic of occupation, regardless of the scenario considered, the increase will occur with 246 greater intensity in the Middle of the road and Strong inequality scenarios, whereas in 247 Agriculture the increase will be 259,892 and 303,781 km 2 , respectively, and in the 248 Mosaic of occupation this increase will be 1,366,687 and 1,450,867 km 2 , respectively.