Die Doktorarbeit untersucht, wie sich Mensch-Umwelt-Interaktionen am Beispiel von Abholzung und Landnutzungsänderungen im Amazonas analysieren und modellieren lassen. Die Abholzung tropischer Wälder bedroht die Stabilität artenreicher Ökosysteme, lokaler Wettergeschehen und des globalen Klimas. Drei Hauptteile erforschen das Thema mit Konzepten der theoretischen Physik und Netzwerktheorie. Der erste Teil gibt einen kritischen Überblick über Modellansätze, die Entscheidungen und menschliches Verhalten beschreiben. Agentenbasierte Netzwerkmodelle ergeben sich als vielversprechender Ansatz um sozial-ökologische Systeme zu modellieren. Der zweite Teil identifiziert Muster in satellitengestützten Landbedeckungsdaten im brasilianischen Amazonas. Basierend auf der Theorie der Markov-Ketten werden Übergangsraten zwischen verschiedenen Typen von Landbedeckung berechnet und Übergangsmatrizen für Teilgebiete mit Clusteralgorithmen verglichen. Angrenzende Teilgebiete weisen ähnliche Übergänge auf. Die identifizierten Cluster decken sich mit Erkenntnissen aus Feldstudien. Auf Grundlage der geschätzten Übergangsrate ergeben sich Projektionen für die Entwicklung der Landbedeckungsanteile. Der dritte Teil entwickelt ein agentenbasiertes Modell um zu untersuchen, unter welchen Bedingungen die Intensivierung der Viehhaltung im Amazonas die Abholzung reduzieren kann. Das Modell kombiniert ökologische, ökonomische und soziale Prozesse und modelliert Landnutzungsstrategien mit Heuristiken. Die Modellanalyse zeigt, dass eine Intensivierung die Abholzung nur dann verringert, wenn der lokale Viehmarkt saturiert. Unter anderen ökonomischen Bedingungen kann Intensivierung die Abholzung erhöhen. Die Arbeit demonstriert, dass eine Kombination von Methoden aus der Theorie komplexer Systeme mit sozialwissenschaftlichen Theorien zu einem besseren Verständnis der emergenten Dynamik sozial-ökologischer Systeme führen kann – eine Grundvoraussetzung, um solche Systeme nachhaltig zu bewirtschaften. / This thesis investigates how to model and analyze human-nature interactions using the example of deforestation and land-use change in the Brazilian Amazon. Deforestation of tropical forests threatens the stability of species-rich ecosystems, local weather patterns, and global climate. The three main parts of the thesis study different aspects of this topic using concepts from theoretical physics and network theory. The first part reviews modeling approaches to human decision making and behavior. From the review, networked agent-based models emerge as promising tools to capture the dynamics of social-ecological systems such as the land system. The second part of the thesis combines Markov-chain and cluster analyses to detect patterns in satellite-derived land-cover maps of the Brazilian Amazon. I compute transition rates between different land-cover types and apply clustering algorithms to find spatial patterns. The analysis shows that neighboring subregions undergo similar transitions and identifies clusters corresponding to findings from field surveys. Markov-chain models, parameterized with the transition rates, are used to compute land-cover projections. In the third part, I develop an agent-based model to investigate under which conditions the intensification of cattle ranching can reduce deforestation in the Amazon. The model captures stylized environmental, economic, as well as social processes, and uses heuristic decision theory to represent different land management strategies. A detailed analysis reveals that fast intensification can only lower deforestation rates if local cattle markets saturate. Under other economic conditions intensification may increase deforestation. The contributions of this thesis demonstrate that combining modeling tools from complexity science with social-science theories allow better understanding the emergent dynamics of social-ecological systems, which is a prerequisite for their sustainable management.

The increasing demand for food and fiber, the need for climate change mitigation and adaptation as well as for environmental protection impose severe challenges on land systems worldwide. Solutions to support the transformation towards a sustainable development of land systems are needed. One response to the multiple challenges is the introduction of policy options aimed at steering land use activities towards a bundle of societal goals. However, it is difficult to empirically foresee the effectiveness and unintended consequences of policy options prior to their deployment. A second response is environmental education because human consumption behavior, among other factors, strongly influences natural ecosystems. However, it is a non-trivial task to develop effective communication strategies for complex topics such as sustainable land management. In both cases, modeling can help to overcome the different obstacles along the way. In this thesis, dynamic process-based social-ecological models at the individual scale are developed and analyzed to study effectiveness and unintended side effects of policy options, which promote agricultural management strategies and were intentionally designed to cope with multiple societal challenges. Two case studies of political intervention are investigated: the promotion of perennial woody crops in European agricultural landscapes for a sustainable bioeconomy and governmental supplementary feeding programs to cope with climate risks in pastoral systems in drylands. These two case studies are complemented by the development of a serious online game on sustainable land management in general that bridges the gap between land use modeling and environmental education. Simulation results of this thesis provide insights into (i) the performance of the politically promoted agricultural management strategies in meeting various intended goals such as poverty alleviation or the maintenance of biodiversity and ecosystem services, (ii) the emergence of unintended (environmental and social) side effects such as land use conflicts, land degradation or cost explosion and (iii) the mitigation of such side effects by appropriately adjusting the design of the policy options. These insights are enabled by representing temporal as well as spatial variability in the developed models. Furthermore, different mechanistic approaches of transferability analyses based on stylized landscapes are developed and applied. They enable to check whether and in what respect policy impacts actually differ substantially between regional contexts, to identify what regional factors steer the impact and to derive indicators for grouping regions of similar policy impacts. Finally, based on a conducted survey-based evaluation and experiences from various applications, the value of the developed serious game for environmental education is revealed and discussed.Altogether, this thesis contributes to model-based decision support for steering transformation towards the sustainable development of land systems in an appropriate way. This is done by developing appropriate social-ecological modeling approaches, by performing specific policy impact analyses in two transformative agricultural systems using these models and by providing a model-based communication tool for environmental education.