The impact of migration on tropical deforestation: An agent-based modeling approach for Guraferda, Southwest Ethiopia

Tropical deforestation is an on-going and urgent global issue, for which proximate and underlying causes are still the subject of academic debate. Rural-rural migration is a known underlying cause, but very little research is dedicated to the exact human-environment interactions that are involved in this deforestation process. Agent-based modeling is a relatively recent spatial modeling technique that can be used to formally bring together qualitative and quantitative information and perform simulations under various assumptions and conditions. The main novelty of agent-based modeling is that the behavior and decision rules of groups of actors are modelled directly, making it a bottom-up modeling approach. This thesis hypothesizes that agent-based modeling is a valid and useful tool to organize the knowledge about the various relations and interactions that link rural-rural migration with tropical deforestation, and to assess the relative importance of driving forces.
To falsify this hypothesis, a field study to Guraferda (Southwest-Ethiopia), a rural area that has experienced a large inflow of state-sponsored and spontaneous migrants, has been conducted. 100 interviews with local farmers and 9 interviews with other persons of interest have been conducted to gather information on the differential livelihood of migrants and natives and to find out how migration has caused deforestation directly and indirectly. An agent-based model, called AMFI (Agent-based Migration Forest Interactions) is developed, in which the quantitative and qualitative knowledge gathered in the fieldwork is presented in a formal, mathematical way. The model uses a forest map, a digital elevation model and demographic data as input, and uses this to simulate the quantity and allocation of deforestation in the study area. Three agent types are defined: native, migrant and investor agents. Migrant and native agents are given a ruleset to decide where to settle and occupy land. This ruleset is based on preferences of yield, elevation, forest cover and population density of a pixel. Lastly, 11 forest maps dating between 1973 and 2013 are created by classifying Landsat imagery, using an unsupervised isocluster classification method and rigorous post-processing. The resulting maps are used to complement the qualitative descriptions of the human-environment interactions, and to assess the quality of the output of the AMFI-model.
Results show that forest cover in the villages of Guraferda has decreased from 53% in 1973 to 36% in 2013. The forest cover reached a minimum around 2003, and slightly increased again in later years, as a consequence of the expansion of coffee forests and forest protection and reforestation programs by a centralized government. Guraferda experienced, and is still experiencing in-migration in 5 different modes: (1) state-sponsored resettlement by the socialist Derg regime, (2) spontaneous migration, (3) intra-regional sponsored resettlement, (4) intra-zonal sponsored resettlement and (5) intra-woreda sponsored resettlement. This has led to direct deforestation, as migrants occupy patches of land, and indirect deforestation. The latter denotes the various processes through which migration changes social, cultural and economic reality, such as technology spillovers towards the native population, or the opening up of the region to investors.
The AMFI model, loaded and parameterized with data gathered in the field and assumptions derived from existing literature, is able to simulate the quantity of deforestation between 1973 and 2012 in 13 villages in Guraferda. The model is very sensitive to the quality of the demographic input data, which causes it to overestimate deforestation during the Derg resettlement, where this data is of a lesser quality. The allocation of deforestation is not simulated accurately at the pixel level, but aggregated at the village level simulations are in close agreement with observations. After 2003, when reforestation is observed, AMFI fails to produce correct results, because the logic behind these reforestations is not implemented in the AMFI framework. AMFI shows that it is possible to organize complex qualitative data and use it in a spatially explicit way to simulate land cover. Furthermore, it reveals knowledge gaps. Further improvements of AMFI should focus on better yield estimations at the pixel level. Also parameterization is shown to become difficult with increasing model complexity. In light of these issues, AMFI is most useful as an exploratory tool. AMFI not only brings together what is already known about the migration-deforestation conundrum, but also serves as a placeholder in which better information and new insights can be loaded and tested. MSc thesis in the program of Geography (KU Leuven and VU Brussels, Belgium)