Interdisciplinary Project

(1)

AI Driven Solutions for Effects of Slash and

Burn on Madagascar: An Interdisciplinary

Approach

Course | Interdisciplinary Project Assignment | Final Report Expert Supervisor | A. Uilhoorn Date | 18/12/2020

Word Count | 7987

Jaimy Langenakker Political Sciences 11771356 Joost de Fretes Artificial Intelligence 11979771 Lucas Jonk Earth Sciences 11641444

(2)

Abstract

This research examines the biophysical as well as the socio-economic and political harmful effects of slash and burn in Madagascar and has subsequently tried to provide an Artificial Intelligence (AI) driven technique that would reduce these effects. The complexity of this problem is rather extensive. Therefore, this paper employs four different scientific disciplines and combines them to overcome these complexities. This paper discusses several fundamental theories, which have been used to determine the relevant indicators for analyzing different AI driven techniques. Subsequently, all indicators are scaled in order to effectively evaluate proposed AI techniques. The analysis is

performed by using a white box problem analysis. The results show that technique 1 ‘field irrigation using soil sensors’ scores relatively low’. Technique 2 ‘robotic support for weeding’ also scores relatively low. Finally, technique 3 ‘informational support by image processing’ scores relatively high and has the best overall score and would in this specific case be the best possible AI driven technique in order to reduce the harmful effects of slash and burn in Madagascar.

(3)

1. Introduction

Madagascar is facing many challenges including soil degradation, deforestation and

underdevelopment. These challenges are all aggravated by slash and burn, which is a traditional farming method in Madagascar that has been practiced for centuries (Jarosz, 1993). Formerly, this method was appropriate for agricultural cultivation, due to the abundance of land, the absence of economic capital and minimal technology (Jarosz, 1993). Today however, the rate in which the method is practiced is alarming due to extensive population growth (Randrianarison et al., 2016). Thereby, slash and burn has become unsustainable.

Slash and burn can be considered unsustainable for numerous reasons. In the process, trees and other vegetation are slashed down and then burned off. This creates a fertile layer of soil suitable for farming. However, this fertility only lasts for a short period of time since intensive agriculture causes the soil fertility to decrease rapidly (Buol, 2003). This soil degradation forces farmers to move on to the next patch of forest (Vågen et al., 2006). Hereby, slash and burn causes not only soil degradation but also is the primary cause for deforestation (Scales, 2014).

Deforestation itself causes problems including habitat loss and GHG emissions. Madagascar only has 9.9 percent of original primary vegetation left (Myers et al., 2000). Deforestation leads directly to habitat loss (Clark, 2012). Since Madagascar is a biodiversity hotspot and home to many endemic species (Goodman & Benstead, 2005), habitat loss could result in extinction (Brooks et al., 2002). Moreover, habitat loss causes a decrease in ecosystem services (Dobson et al., 2006), which are important to human livelihood. Ecosystem services in tropical forests entail provisioning services such as providing food and timber, regulating services such as managing soil erosion and cultural services such as attracting tourism (Dave et al., 2017). Deforestation thus forms a threat to Malagasy flora and fauna and social well-being through loss of ecosystem services.

Another important effect of deforestation are the greenhouse gasses (GHGs) emitted. When vegetation is cut and burned off, different GHGs are released into the atmosphere. For instance, Carbon stored in trees is released into the atmosphere as CO2 when these trees are cut and burned off (Kotto-Same et al., 1997). This makes slash and burn in Madagascar responsible for large CO2 emissions in the atmosphere (Vieilledent et al., 2018). These GHGs contribute to the greenhouse effect which warms the Earth’s surface and thus enhances climate change (Rosa & Dietz, 2012). Climate change itself again

intensifies biodiversity loss (Malcolm et al., 2006). Since the rates of slash and burn and deforestation are high, slash and burn is a major driver to global climate change.

While the harmful effects of slash and burn are evident, the method continues to be widely used in Madagascar. This can be partially explained by the effects of the legacies of colonialism. The colonization of Madagascar by France in 1896 led to regional fragmentation, uneven economic development and environmental pressures in Madagascar (Jarosz, 1993). The French government decided to restrict traditional agricultural practices, including slash and burn, in order to save its forests (Raik, 2007). However, the Malagasy interpreted the ban on traditional practices as a strategic act of the colonizers. Hereby, slash and burn became a symbol of resistance from farmers against the state (Jarosz, 1993). Madagascar is currently experiencing low levels of development. Economic performance is relatively low and governance is ineffective (World Bank, n.d.). The continuation of slash and burn as an unsustainable practice only poses more challenges for achieving higher levels of sustainable development.

(5)

The aforementioned problems caused by slash and burn and the origin of the method are complex and require an interdisciplinary approach. This paper will conduct a literature review of theories and concepts from four different disciplines, namely: political sciences; AI; earth sciences; and physics. With this knowledge, a set of requirements defining sustainable agriculture will be presented. These in turn will be used to grade possible AI driven techniques. AI has established itself as a field of research that is able to help and support many industries. Its implementation into agricult ure is a relatively young field but nevertheless shows a lot of promising results (Jha et al., 2019). Therefore, the research question of this paper is ‘How can AI driven technological advancements improve the local stress factors caused by slash and burn practices in Madagascar?’. Our objective is to find and grade technological advancements provided by AI that can aid in increasing the sustainability of the agricultural industry in Madagascar.

(6)

2. Theoretical Framework

The following four main concepts and theories, which are: slash and burn; postcolonial development; sustainable agriculture; agriculture in AI, are important to obtain a better understanding of our

problem. Those concepts ultimately help us to find the answer to our research question.

2.1 Slash and Burn Theory

Burning the ground to prepare it for agriculture is known as the ‘slash and burn’ technique, it is a cheap and common way to add nutrients to the soils (Otto & Anderson, 1982). If practiced moderately and if the land is given enough time to recover, slash and burn is not a burden on the environment (Otto & Anderson, 1982).

Slash and burn is a technique that can be practiced to gain land from forests or other wooded areas. It is a popular technique all around the world since very little and basic equipment is needed (Randrianarison et al, 2016). Because of this, it is a very cheap way of cultivating land into farmland and is therefore popular in poor rural areas like Madagascar. Slash and burn is mostly common in the tropical areas, and most of the nutrients needed for plant growth like P and N, are stored in the vegetation and not in soils. This is because the soils are very toxic because of low pH values, and the nutrients do not fare well in low pH environments (Fagaria & Baligar, 2008). Therefore, the soil is not very fertile, by burning the vegetation the stored nutrients are released into the soil that is now suitable for agricultural practice (Gade, 1996).

The added nutrients by burning vegetation allow for only a few years of cultivation, after which the nutrients are depleted and crop yields fall (Buol, 2003). Farmers abandon the site and move on to the next patch of land, leaving the abandoned site to recover in vegetation and restore nutrients (Buol, 2003). This is an essential part of sustained slash and burn practices, whereby excessive use of patches of land results in permanently damaging the soil and causing it to be exhausted of any nutrients (Otto & Andersen, 1982). However, with Madagascar's population growth, the pressure on sufficient food supplies has increased to the point that farmers cannot afford to let the land fully recover and thus exhaust the soil so that it is not suitable for future agriculture (Randrianarison et al., 2016; CIA World Factbook, 2020).

2.2 Postcolonial Development

The continuation of slash and burn in combination with Madagascar’s population growth implies serious consequences for the future development of Madagascar. While many African countries strive for development (Ukaga & Afoaku, 2005), Madagascar has failed to achieve more development and it is likely to fail again in the future if it is to follow its current trajectory. Development is a broad concept with different definitions which can be approached from a variety of perspectives (Grieco et al., 2019). According to Grieco et al. (2019), economic, human and sustainable development are the most well-known approaches of development. The latter one, sustainable development has become more significant recently, with the introduction of the Sustainable Development Goals (SDGs) by the United Nations (UN) in 2015. The SDGs consist of 17 goals which are displayed in figure 1.

According to Robert et al. (2005), sustainable development can be described as an approach of development which aims at progressing both human and economic needs, while staying within planetary boundaries. This paper will use sustainable development as its approach of development because of the connection with sustainable agriculture, which will be further discussed in section 2.3.

(7)

Postcolonial theory states that the former relationships between the colonizers and colonized are hierarchical in nature, and still has effects, such as unequal socio-economic development between them (Jackson & Sørensen, 2016). This can be explained by these postcolonial legacies. For instance, Acemoglu et al. (2001), argue that former colonies with a high disease burden currently experience lower levels of development. Their analysis suggests that the mortality risk by diseases in colonies determined if colonizers would settle. If the rate of mortality was high, few would settle. This resulted in extractive and predatory institutions, instead of institutions that are more supportive of

development, resulting in lower levels of development. Another example is the type of economic activity that colonizers practiced. According to Sokoloff & Engerman (2000), the type of economic activity is an important factor for the development of effective institutions and the distribution of wealth. The economic activities are determined by factor endowments, such as climate and soils, but also the size and distribution of the population (Sokoloff & Engerman, 2000). Furthermore, Nunn & Wantchekon (2011), articulate that slavery has led to poor economic performance present-day. Slavery was accompanied by raiding and kidnapping, which resulted in either ethnic fragmentation and mistrust or a decay of legal institutions, both producing poor economic performance (Nunn & Wantchekon, 2011). In conclusion, events that occurred during the colonial period, still affect today's development in colonized countries. Madagascar is also subject to the legacies of postcolonialism. It experienced significant population growth (31 percent increase from 2009-2019) and reached a population of 27 million citizens in 2019 (FAO, 2020). However, while approximately 70 percent of the population is operating within the agricultural sector (Harvey et al., 2014), about 41.7 percent was still undernourished (FAO, 2020). Moreover, about 80 percent of the population lives beneath the poverty line of 1.25 US dollar per day (Harvey et al., 2014). The second goal of the SDGs, ‘Zero Hunger’ is thus specifically relevant for this case as well as in general because it is inherently linked to the economy, society and the environment and is of importance for achieving the SDG agenda as a whole (Gil et al., 2019). According to the United Nations (n.d.), sustainable food production is crucial to realize the aims of this goal.

Figure 1.

Overview of the 17 Sustainable Development Goals.

(8)

2.3 Sustainable Agriculture

As mentioned in the previous section (2.2) this paper focuses on sustainable development. Moreover, sustainable agriculture is a key aspect in realizing the second goal of the SDGs. According to The Food and Agriculture Organization (FAO), sustainable agriculture can be defined by use of five key principles balancing social, economic and environmental dimensions of sustainability (FAO, 2014). These principles are:

1) improving efficiency in the use of resources

2) conserving, protecting and enhancing natural ecosystems 3) protecting and improving rural livelihoods and social well-being 4) enhancing the resilience of people, communities and ecosystems 5) promoting good governance of both natural and human systems

However, this paper will not include principle five: ‘promoting good governance of both natural and human systems’, as this is not within the scope of this proposal. Good governance refers to a country’s political and legal systems whereby transparency and consistency is of importance. Hereby, a country takes action against corruption and ensures the protection of property (Grieco et al., 2019). While good governance is relevant for development and technological innovation, this is considered as a limitation in our model, because it is not likely that an AI driven solution will be able to promote good

governance.

The four principles relate to the subchapters in the following ways: 1) Improving efficiency in the use of resources

Soils

Tropical soils are toxic and infertile because of a low pH. (<5) (Fageria & Baligar, 2008). Increasing the pH. would simultaneously increase the fertility of the soil, increasing pH., over longer periods of time would increase the durability of a soil (Fageria & Baligar, 2008; Burrel et al., 2016). Therefore, farmers can stay on a patch of land longer and do not need to move to a new patch of forest. This is a more sustainable way of agriculture than intensive slash and burn.

2) Conserving, protecting and enhancing natural ecosystems Deforestation and ecosystem services

Slash and burn in Madagascar is the key driver for deforestation and it is estimated that already 90 percent of the Malagasy primary forests have been cut (Myers et al., 2000). Deforestation leads directly to habitat loss (Clark, 2012) which causes a decrease in ecosystem services (Dobson et al., 2006), which are important to human livelihood. Ecosystem services in tropical forests entail provisioning services such as providing food and timber, regulating services such as managing soil erosion and cultural services such as attracting tourism (Dave et al., 2017). Because Madagascar is home to many endemic species, habitat loss could result in extinction (Brooks et al., 2002). Moreover, tropical deforestation is the main source of GHG emissions (Corbera et al., 2016), which contribute to the greenhouse effect and therefore warm the Earth’s surface (Rosa & Dietz, 2012). This global warming again enhances biodiversity loss (Malcolm et al., 2006). Deforestation and GHG emissions are therefore inseparable and must both be understood in relation to another.

(9)

Greenhouse Gasses

The most important gasses emitted when vegetation is slashed and burned are carbon dioxide (CO2), methane (CH4), nitrous oxide (N2O), carbon monoxide (CO) and nitrogen oxides (NOx) (Silva et al., 2011). They highlight that other research (Fearnside, 2000) states that GHGs are also emitted when secondary vegetation is again slashed and burned. If after the initial slashing and burning vegetation is allowed to regrow, CO2 that was released by the initial burning will be reincorporated into the new, secondary vegetation. In theory, this method then would not contribute to variations of CO2 in the atmosphere. However, Fearnside (2000) states this would only be true if these two amounts of GHGs were in equilibrium, which is rarely the case. Therefore, the burning of secondary vegetation will indeed contribute to GHGs emitted by slash and burn.

In order to properly understand GHG emission units and make a comparison, units of different GHGs are mostly measured in a CO2 equivalent. This makes it possible to state that the most important anthropogenic GHGs in the atmosphere are CO2 and CH4 (Boucher et al., 2009). To get the CO2 equivalent, the emissions of each gas are multiplied by the Global Warming Potential (GWP). The GWP is the absorbed heat by a GHG, as a multiple of the heat absorbed if it were CO2. The GWP takes into consideration that some GHGs absorb more heat and therefore have a greater impact on global warming. For the five gasses listed before, the lifetime and GWP for 20 and 100 years are given in Table 1. As previously discussed, CO2, CH4 and N2O are GHGs themselves, CO and NOx are not, but have effects on GHGs. For N2O this effect is quite straightforward, which makes it possible to estimate its lifetime. NOx itself has an extremely short lifetime, however, due to its effects on other GHGs it is still an important contributor.

Table 1.

Gas emissions from burning biomass with their corresponding lifetime and GWP for 20 and 100 years.

Formula nameName Lifetime1 _{GWP for 20 years}1 _{GWP for 100 years}1

CO2 Carbon dioxide 30-95 years1 1

CH4 Methane 12 years 84 28

CO Carbon Monoxide 2-3 months6 2

NOx Nitrogen oxides <1 day2 30-333 7-103

N2O Nitrous oxide 121 years 264

265

(10)

3) Protecting and improving rural livelihoods and social well-being & 4) Enhancing the resilience of people, communities and ecosystems Food Security

Principle three and four are combined here, since the protection and improvement of rural livelihoods and social well-being as well as the enhancement of the resilience of people, communities and ecosystems, can be assigned to food security. Food security can be defined as a “situation that exists when all people, at all times, have physical, social and economic access to sufficient, safe, and nutritious food that meets their dietary needs and food preferences for an active and healthy life” (Schmidhuber & Tubiello, 2007, p.1). Food security is connected with the development of countries since it is essential for achieving the second SDG: Zero Hunger. (Gil et al., 2019). Food security can be achieved through four dimensions which are availability, stability, access and utilization

(Schmidhuber & Tubiello, 2007). Firstly, the availability of food is related to the capacity of the agricultural system to supply enough food to meet demand (Schmidhuber & Tubiello, 2007). Secondly, the stability refers to the resilience of individuals related to external shocks, such as a decrease in income or climate variability (Schmidhuber & Tubiello, 2007). Thirdly, access concerns the access of individuals to sufficient entitlements to obtain appropriate food and nutrition

(Schmidhuber & Tubiello, 2007). These entitlements do not have to be solely monetary, they can also for instance be traditional rights (Schmidhuber & Tubiello, 2007). Finally, utilization can be described as the last step in the process, namely food safety and quality, which needs to be sufficient in order to be healthy enough to consume (Schmidhuber & Tubiello, 2007). Obtaining food security is a

challenge on itself, but it is contested even more by the effects of climate change (Harvey et al., 2014).

2.4 AI in Agriculture

When discussing Artificial Intelligence (AI) in agriculture, a selection of techniques must be made since it is a broad field of research. For the scope of this research two types of AI techniques will be taken into account: practical support and informational support. The former is perhaps a more

futuristic approach where self-driving equipment or robots aid the farmers in their work. Informational support, focusses on programs that help process data be it from images or observation and could thus teach and advise the farmer on what can be done to improve situations. The second implementation does not actively take action but has the potential to help the farmers make more informative and perhaps more sustainable decisions.

When looking at practical support or hands-on implementation of AI techniques Talaviya et al. (2020) present multiple uses of robotics in agriculture. They discuss the application in two fields: irrigation and weeding. Since technologies in these aspects can decrease th e excess use of water, pesticides, herbicides while maintaining the fertility of the soil while improving both quality and productivity (Talaviya et al., 2020). For irrigation algorithms are used that take into account the water content in the field, meteorological data and other factors specified by the farmer to come up with the most efficient irrigation strategy (Zhen et al., 2010). Controlling invasive weeds requires actual robots to move across the fields and uses computer vision techniques to evaluate whether or not a plant is a weed or a crop (Talaviya et al., 2020). Once identified the three most common ways for the robot to take action are: spraying pesticides/herbicides, using electrical discharge or physically removing the weeds.

For informational support the goal is to provide the farmers with better data of their fields allowing them to make more informed decisions. In addition, it could also help educate the people working in the industry. Anami et al. (2020) present an algorithm which detects stresses to the crop in paddy

(11)

fields (rice fields) based on pictures of the crops. The algorithm successfully identified different stress factors with an accuracy of 92.89 percent. Identifying different stress factors to the crop will help in taking the proper actions in order to combat them and thus reduce the amounts of crops lost. The same type of research has been done in identifying different diseases in the crops, where Ferentinos (2018) used a deep learning model with a 99.53 percent success rate in identifying the corresponding plant and disease.

Another powerful aspect in which AI could help is raising awareness. Satellite image processing allowed for more detailed mapping of the deforestation rates and could show the farmers the negative effects of certain actions on a large scale (Agarwal et al., 2005).

(12)

3. Problem Definition

As portrayed in the introduction our main problem is the negative impact of slash and burn on the local environment of Madagascar. Therefore, the research question of this paper is ‘How can AI driven technological advancements improve the local stress factors caused by slash and burn practices in Madagascar?’.

3.1 Why is our problem complex or wicked?

According to Boulton & Allen (2007), complexity can be characterized as something that is

8 | Robustness & Resilience 9 | Instability

10 | Adaptiveness

When examining our own research problem, we find that most of these properties are important. 1 | Connectivity

Our own research problem is interconnected through the atmosphere, biosphere, hydrosphere and anthroposphere. For instance, emissions of GHGs in Madagascar could have effects on other states through the atmosphere. The same logic applies to the biosphere. When ecosystems are affected in one area, other areas could be affected as well. Deforestation in Madagascar that results in the

disappearance of species in its ecosystem could have profound impacts on other living species of flora and fauna, as well as an impact on humans. This logic is relevant for the hydrosphere as well, for instance with the pollution of fluvial components such as rivers, oceans and lakes, that cross borders. Finally, Madagascar is part of a globalized world economy, which acts as a system. They are

interconnected through the structures of supply and demand within the global economy, which could affect local prices, income and developmental outcomes. Moreover, the structure of the agricultural system on Madagascar does not imply easy fixes, since interference on one level does not necessarily mean that this could also apply to other levels of the system.

4 | Emergence

Emergence, interesting, non-obvious consequences of low-level properties, better understood in their own right (Rammelt, 2020). This is visible in our own research because the issue at hand here is better understood in its own right and not definable by one discipline. Emergence is an interdisciplinary approach to the complex issue at hand in our research. Our four disciplines are a way to evalu ate, describe and interpret the issue in a wider framework to gain a better understanding.

(13)

5 | Observer Dependence

Within our research problem we can examine observer dependence. For instance, when we look at the main issue of our research, slash and burn. When studying this phenomena from a single perspective, the explanations and effects of it would be insufficiently explored. However, when combining our different views, we can develop a complete definition of slash and burn, its explanations and its effects.

6 | Path Dependence

This entails that history matters. For our research problem it does as well. Especially the colonial history that remains important in the decisions today. For instance, the legacies of colonialism have resulted in poor economic and political performance on Madagascar today and also the continuation of slash and burn as a popular agricultural method.

8 | Robustness & Resilience

The Malagasy ecosystem has been exposed for a long time to unsustainable agricultural methods, soil erosion and global warming. This caused the ecosystem to change which makes it resilient. However, there are some consequences that the ecosystem may not cope with itself, such as habitat loss. This may result in losing many plant and animal species before collapsing.

(14)

4. Interdisciplinary Integration

4.1 Justification

According to Menken & Keestra (2016), complexity can be seen as a propulsive effort behind interdisciplinarity. This complexity requires interconnectivity in order to effectively react to change (Boulton & Allen, 2007). Because of the complex character of the research we wish t o conduct, an interdisciplinary research is a perfect fit. As mentioned, the complex nature of this problem is down to the five properties as defined in the problem definition section above. However, these five properties are viewed by four different disciplines and thus represent much more than just the initial five properties. Because of the interdisciplinarity of the paper and the different views that the disciplines provide we are able to gain more depth, insight and understanding about the issue at han d.

4.2 Integration Techniques

This proposal mainly uses extension as defined by Menken & Keestra (2016), to seek overlap between disciplines and to bridge the gaps between the disciplines. This way, adding to each other and

reinforcing statements and ideas from different perspectives in order to provide a well-rounded and interdisciplinary research paper.

Table 2.

Cross table of disciplines and their overlaps.

Artificial Intelligence

Earth Sciences Physics Political Sciences

Artificial Intelligence

x 1 | ‘Both

technologies rely on sensors (AI) placed in the field which measure water content and other chemicals of the soil (ES)’.

Data collection (AI) on GHG emissions (PH) 1 | ‘The use of computers (AI) itself in farming is not so common especially not in underdeveloped countries (POL)’. Earth Sciences 1 | ‘Mapping and the gathering of data within Earth sciences are influenced by the technological advancements in AI.’. x 1 | ‘Physics are a key component in understanding and explaining soil processes’. 1 | ‘Soil characteristics are of influence on how a country develops and evolves over time’.

(15)

Physics Data collection (AI) on GHGs x 1 | Ecosystem services (PHY) in tropical forests entail provisioning services such as food and timber (POL)

Political Sciences

1 | Low development (POL) and general costs of AI solutions as a challenge (AI). 1 | Economic activities (POL) are determined by factor endowments, such as soils (ES). 2 | Population growth (POL) on Madagascar has led to more land degradation (ES) 3 | ‘Tavy’ ban (POL) backfired, leading to more slash and burn practices (ES) 1 | Sustainable development aims at progressing economic and human needs (POL) while staying within planetary boundaries (PHY) 3 | Economic activities (POL) are determined by factor endowments, such as climate (PHY).

4 | Population growth (POL) has resulted in more deforestation (PHY). 5 | Unequal economic development and spatial fragmentation (POL) resulted in difficulties for the environment (PHY)

6 | Low EPI (PHY) scores can be explained by lack of (good)

(16)

economic prosperity and economic

(17)

5. Selected Methods and Data

5.1 Data

This research is a literature review. Most of the data necessary to answer the research question is already available in individual scientific papers. The structure of this research will be a timeline showing the different stages of slash and burn and the key concepts that cause concern. The literature used to construct the framework will be the basis of our analysis and help us identify different concepts and theories related to slash and burn. Each discipline will present indicators that have an impact on the future of slash and burn in Madagascar. These indicators are soil toxicity, GHG

emissions and food security. The indicators in turn consist of measurable variables with which they are graded. For soil toxicity this is the pH value of the soil, GHG emissions in potential decrease in GHG. Food security consists of availability, access, stability and utilization. Moreover, the specifics of each indicator and variable will be presented in the deconstruction per variable. For each AI driven solution, these variables will be considered to determine which solution is best.

Besides these variables, we take into account World Governance Indicators (WGI), digital

infrastructure, education and general costs, not as a variable, but rather as a context in which potential solutions should fit. WGI includes voice and accountability, political stability and absence of violence, government effectiveness, regulatory quality, rule of law and control of corruption (World Bank, n.d.). Madagascar scores far below average: especially the scores on the rule of law and government

effectiveness are relatively low and have even decreased over the past years (World Bank, n.d.). The absence of good governance in the country could be an obstacle in successfully implementing an AI technique. Moreover, Madagascar's digital infrastructure, such as access to the internet is not great, not a lot of inhabitants are tertiary or secondary educated and not a lot of money is available for

implementing these types of techniques. These indicators are relevant; however, we do not expect our solutions can change them. We therefore identify them as Malagasy specific limitations.

Table 3.

Overview of the different Malagasy specific limitations.

Malagasy Specific limitations World Governance Indicators:

• Voice and accountability • Political stability • Absence of violence • Government effectiveness • Regulatory quality • Rule of law • Control of corruption Digital infrastructure Education General costs

(18)

The goal of this paper is to produce AI driven technological advancements that can help improve the local stress factors in Madagascar. These variables form the sustainability benchmarks that allows us to evaluate the found solutions. This research will aim at finding a solution that takes all of the variables and desired values into account.

5.2 White Box Problem

The analysis can therefore be described as a white box problem shown in Figure 2. A white box problem analysis looks at how different inputs behave inside of the ‘box’ in order to find the best combination of inputs to achieve your desired output or goal. The different inputs for this box are slash and burn, different AI aided technologies and our Malagasy Specific Limitations context since this is constant and is not affected by possible AI solutions. The situation where no AI aided technologies are used, so only slash and burn as input is present, will be our base situation which we hope to improve by adding the technological input. Inside the box these inputs are then graded with our identified variables for example: a proposed solution decreases the amount of GHGs produced. This solution will then score better on that variable and have a more positive effect on the output. The output is a goal we want to achieve in this case it is sustainable agriculture as defined earlier on. This type of analysis allows us to see which type of technique will eventually get closest to our goal of sustainable agriculture in Madagascar.

Figure 2.

Visualization of the white box problem.

(19)

5.3 Indicators

5.3.1 Soil toxicity

Soil toxicity is derived from the variable pH., and is linked to the fertility of the soil. Fertile soils contain an abundance of nutrients like P, N and Ca2+_{and are low in nutrients like Al}3+ _{(Fagaria &} Baligar, 2008). Toxic soils are not able to contain these nutrients and therefore these nutrients are washed away, and so are not available for plant growth (Fagaria & Baligar, 2008). The pH. indicates the toxicity of a soil and thus also the amount of available nutrients as mentioned above.

A higher pH. is necessary to reduce the toxicity in the soil and increase the available nutrients in the soil. Therefore, a higher pH. means more plant growth. This is preferable for farming since this would result in higher yields (Raboin et al., 2016). This is proven by the effects of applying liming and biochar. Liming is the addition of calcium and magnesium rich material to raise the pH. value of the soil. Biochar is the addition of pyrolyzed charcoal in order to raise the pH. value of the soil. As seen in figure 3, biochar and liming show similar trends regarding the increase of pH., available P and Ca2+_, and the decrease in Al3+.

Simultaneously, figure 3 shows bean yield doubled and maize yield increased almost 50 percent, meaning the soil did in fact become more fertile and thus less toxic as a result of either biochar or liming. Rice fares better in a lower pH. environment but the decrease in rice is overshadowed by the increase in beans and maize that result in an overall increase in crop yield. Thus, in order to improve soil toxicity, the pH. value has to be increased.

Figure 3.

Representative relationship between the yield of the maize, bean and upland rice crops and soil properties (pH, soil available P, soil exchangeable Al and soil exchangeable Ca).

(20)

Figure 4.

Illustrates the relation between pH. value and the crop yield in kg ha-1 regarding rice, maize and beans.

Source: Raboin et al. (2016). Table 4.

The table shows the Ph values related to the grading scale that is used.

-- - + ++

pH. 4.4 pH. 4.6 pH. 4.8 pH. 5.0

The indicator soil toxicity is measured based on the variable: pH. value. As can be seen in figure (4) an increase in pH. value results in higher crop yields regarding maize and beans and an increase in total crop yield. The data collected shows that a pH. value around 5 is preferable.

(21)

5.3.2 Food Security in Madagascar

As mentioned before, food security consists of four dimensions, including availability, stability, access and utilization. These dimensions can be seen as the main indicators for the variable food security. In the following tables, an overview is given of the relevant indicators for measuring food security and the current status in Madagascar in comparison with the global average. In addition, information from a survey and from a report of the World Bank (n.d.), will be taken into consideration for determining the variable and scale for measuring the dimensions. Not all information from these sources is relevant for this research. For the same reason, the fourth dimension of food security, utilization, is left out in table 5, because of irrelevant indicators for this research, that were provided by the FAO (2020). Relevant information from other sources for measuring food utilization is presented below in the next section. A further elaboration on missing data will be provided in section 7.2.

Table 5.

Three of the four food security dimensions and their indicators.

Indicator (unit) Value in

Madagascar Reference value (world) Year Source 1 | Availability

Average dietary energy supply adequacy (%) 88 119 2017-2019 FAO (2020) Average value of food production

(I$ per caput)

137 313 2014-2016 FAO (2020)

Value in Madagascar Reference value (world) Year Source 2 | Access

GDP per capita - in purchasing power equivalent (I$)

1646.2 16950.8 2019 FAO (2020)

Value in Madagascar Reference value (world) Year Source 3 | Stability

Cereal import dependency ratio (%) 17.7 -1.8 2015-2017 FAO (2020)

(22)

Per capita food production variability (%)

5.3 1.6 2015 FAO

(2020)

Per capita food supply variability (%)

31 4 2017 FAO

(2020)

Scale for Food Security

Availability

According to Harvey et al. (2014), food availability in Madagascar is compromised by lack of sufficient land. About 68 percent of farmers under slash and burn for rice production only possesses up to 1 ha of land. It is stated that 75 percent of all households could not suffice in their household food demand. The obvious solution then is providing more land for farmers. However, Madagascar has a rapidly growing population and land is scarce, and as mentioned before, the country struggles with serious environmental issues, such as deforestation. According to the FAO (2020), food availability can be measured by the average dietary energy supply (DES) adequacy and the average value of food production in international dollars per capita. A m ore sustainable option to increase food availability would be to increase yield on the existing patches of land. When the yield increases, more food can be produced which means there is more food available. This can be translated into direct availability, whereby the produced food is consumed directly by the farmers themselves, or indirect availability, whereby products are sold in return for economic resources which allow households to purchase other food. This will likely result in higher levels of dietary energy supply. Moreover, increased yield will result in higher values for produced food. Thus, leading to improved food availability.

Table 6.

Scale for food availability measured by yield (kg ha-1) in percentages.

-- - +

(23)

Access

According to the FAO (2020), access to food can be measured by the use Purchasing Power Parity (PPP) GDP per capita. This means that Madagascar’s GDP per capita is converted to international dollars by use of PPP (FAO, 2020). Currently, about 82 percent of the rural population in Madagascar lives beneath the international poverty line of 1.25 US Dollar per day (Harvey et al., 2014). By

increasing the PPP GDP per capita, individuals simply have more monetary resources to secure proper food and nutrition. The scale will work with availability of monetary resources, because less expenses also results in a wider availability of monetary resources.

Table 7.

Scale for food access measured availability of monetary resources.

-- - +

Decrease No change Increase

Stability

According to Harvey et al. (2014) there are six agricultural risks in Madagascar: Significant disease outbreak; severe pest damage; loss of crops during storage; cyclones; severe flooding; and severe drought. In figure x a more detailed overview is given of the percentage of affected farmers by agricultural risks, the frequency of those risks and the losses in crop yields and income. These risks cohere with the indicators of the FAO for food stability. By reducing agricultural risks, the cereal import dependency ratio; the per capita food production variability; and the per capita food supply variability, are expected to decline as well. Reducing the agricultural risks will lead to lower food and income losses and lead to less food imports, which is beneficial for food stability. Moreover, if these agricultural risks are lowered, food production and supply will be less variable. Agricultural risks can be reduced by providing information. For instance, providing more information on the shelf life of crops, could possibly reduce the number of crops that are lost during storage.

Figure 5.

Summary of the risks to rice production experienced by smallholder farmers and the impacts of these risks on rice yields and household income.

(24)

Table 8.

Scale for food stability measured by ‘ticking boxes’ of providing information.

--

-

+

++

No extra information Providing information on 1 agricultural risk Providing Information on 2 agricultural risks Providing information on 2+ agricultural risks Utilization

According to Schmidhuber & Tubiello (2007), food utilization concerns the quality and safety of food. By improving food quality, more food could potentially be sold. Improving food quality does thereby not only increase utilization but also have a positive effect on the availability and access. By

improving food safety, thus producing healthier products, less people will become sick by the food they eat, or food could be more nutritious, demanding for less products and increasing food availability.

Table 9.

Scale for food utilization measured by ‘ticking boxes’ of improving food quality or food safety.

--

-

+

++

Reduced food quality or reduced food safety

No change Improved food quality or improved food safety

Improved food quality and improved food safety

5.3.3 Deforestation and GHGs

As stated in the theoretical framework, the most important GHGs emitted from slash and burn are CO2, CH4, N2O, CO and NOx (Silva et al., 2011). The resulting emission factors are given in Table 10. It is notable that CO2, CH4 and N2O are GHGs themselves, whereas CO and NO2 are not. CO is however responsible for enhancing concentrations of CH4 and tropospheric Ozone (O3), which is a

GHG itself. NOx are multiple nitrogen oxides, including nitric oxide (NO) and nitrogen dioxide (NO2).

NOx interacts with trace gasses and has an effect on GHGs, however, it is most important due to playing a catalytic role in the production of tropospheric O3 (Lammel & Grassl, 1995). Moreover, NOx affects stratospheric H2O and even CO2 through all sorts of different processes (IPCC, 2013), which are interesting but require further reading which is not possible due to time limitations of this research. The data in Table 10 is more significant when compared to deforestation rates in Madagascar and given in CO2 equivalents.

Table 10.

Emission factors (mean and standard deviation) for biomass burning in tropical forests in g kg-1 (Silva et al.,

2011).

CO2 CO CH4, NOx N2O

1626 ± 39 101 ± 16 6.6 ± 1.8 2.26 ± 1.26 0.2 ± 0.1

(25)

In 1950 Madagascar had already lost more than half of its original cover (Harper et al., 2007) and the forest cover was estimated at 160,000 km2_{. In 2000, an additional 56 percent (from 1950) of tropical} forest was lost and resulted in a forest cover of 99 000 km2_{. This gives an average deforestation rate of} 1 percent per year between 1950 and 2000. For tropical rainforest the biomass for 1 ha forest equals 509,000 kg ha-1_{(Yamakura et al., 1986) and thus 50,900,000 kg km}-2_{. This results in an average} deforestation rate in biomass of 814,400,000 t year-1. This rate multiplied by the emission factors in Table 10 gives the annual GHG emissions of slash and burn in Madagascar shown in Table 11. Multiplied by the GWP from the theoretical framework results in the CO2 equivalent emissions which are also given in Table 11.

Table 11.

Annual GHG emissions for slash and burn in Madagascar.

CO2 CO CH4, NOx N2O Total emissions in t 1,324,200,000 82,500,000 5,370,000 1,840,000 162,000 CO2 eq emissions for 20 years in t 1,324,200,000 495,000,000 451,080,000 55,200,000 - 60,720,000 42,768,000 CO2 eq emissions for 100 years in t 1,324,200,000 165,000,000 150,360,000 12,880,000 - 18,400,000 42,930,000

Under the Paris Agreement, Madagascar aims to reduce 30 Mt CO2 emissions of GHGs by 2030, equating 14 percent of total national emissions (UNFCCC, 2016). They state this reduction can be achieved by actions including reforestation, enhanced forest and grassland monitoring, climate-smart rice farming techniques, increased hydropower and solar energy, sustainable cookstoves and energy efficiency.

This paper also aims to find an AI driven solution that reduces the GHG emissions, preferable a decrease of 14 percent. To compare different AI driven solutions regarding their GHG emissions, precise emissions for each solution are desirable yet difficult to obtain. However, an estimation of decrease or increase of GHGs can be made by means of the aforementioned actions. The scale for GHG emissions is shown in Table 12.

Table 12.

Scale for GHG emissions.

-- - + ++ GHG emissions increase No change GHG emission reduction directly or indirectly GHG emission reduction ≥14%

(26)

6. Results

6.1 What Techniques?

Not every technique AI can provide is usable within the framework of Madagascar. As with most cases specific requirements need specific solutions in order to be useful and reach its full potential. For this reason, three techniques have been selected based on the Malagasy specific limitations. Since in order to be applicable in Madagascar the technique has to fit within this framework. The selected techniques are:

• Field irrigation using soil sensors. • Robotic support for weeding.

• Informational support by image processing.

Each of the techniques have their individual advantages and drawbacks but moreover this in the following chapter. As aforementioned these techniques will be graded based on the variables soil toxicity, food security and reduction in GHG.

6.2 Grading AI Driven Techniques

6.2.1 Technique 1: Field irrigation using soil sensors

As defined in the theoretical framework AI could offer solutions either from a practical or informational support. The first solution proposes practical support for irrigation. The main crop grown in Madagascar is rice, close to 4 million tons was produced in 2015 (FAO, 2020). For the AI to make the biggest impact would be to help this industry. Implementation of subsurface AI controlled drip irrigation within these fields could help reduce the total amount of water needed greatly (Qualls et al., 2001). Even though the test was being executed back in the late 90’s Qualls et al. (2001) proof of concept showed an overall decrease of $7,628 ($12,375 in 2020) spent on water in the period from April to September in Boulder, Colorado. Stating that the operating of the equipment was really easy and that new users only needed 15 minutes to understand the full system. A downside was that the system needed 3 years of calibration in order to operate efficiently. The sensors proved to be reliable since during both the calibration and testing period only two sensors broke and the total repair costs were less than $270 ($438 in 2020) and only taking 15 minutes per sensor to replace (Qualls et al., 2001).

However, using Qualls et al. (2001) as inspiration but adapted to how Madagascar irrigates its rice fields with a more traditional technique of different canals through the fields (Potten, 1983). So, a system designed to work with these conditions would be preferable. Arvind et al. (2017) proposes a system that uses the same sensors but instead operates different mechanical valves allowing flow of water between fields. In addition, this system uses data from existing databases to determine whether to irrigate or not. This eliminates the long calibration period of 3 years and makes the system effective the moment it is turned on (Arvind et al., 2017). In addition, the system can take into account weather activity and other parameters set up by the user. However as stated by Arvind et al. (2017) the system only takes action when the operator or farmer agrees to it. The central computer will advise or alert, via notifications to a phone for example, the farmer on what action to take and only executes when the farmer allows the system to. This same principle should be applicable for Madagascar.

(27)

Table 13.

Scores for Technique 1.

Indicators Score Justification

Soil toxicity -- Soil toxicity is not altered towards a higher pH value.

Food Security Availability Access Stability Utilization -

- There is no change in yield.

+ Increased monetary resources by lowering costs.

- There is one potential agricultural risk averted. There is no direct

information on drought, however, by use of smart water systems like these, water is more efficiently used, resulting in better coping mechanisms for farmers.

- There is no change in food quality or food safety.

GHG emissions

+ GHG emissions are not directly decreased but agricultural improvement may take away the need to slash and burn other land and therefore emits less GHGs.

(28)

Technique 2: Robotic support for weeding

Secondly, the practical support for weeding. As aforementioned these techniques do need a robot of sorts to go over the field and use computer vision in order to identify the weeds from the crops. Again, with a focus on rice paddies different possibilities exist. One that could work in Madagascar is a robot known as “Robo-Ducky”, this robot swims over the field and simulates actual ducks (Nakamura et al., 2016). The movement of ducks has proven to prevent invasive seeds of weeds to sprout since the ducks disrupt the water. These “ducks” can be deployed when a camera sees an increase in seeds polluting the field (Nakamura et al., 2016). Maruyama et al. (2014) laid down the basis for this concept by doing field tests. This was done on a field of 20 by 20 meters and showed that for effective weeding only 2-3 robots where needed running 60-90 minutes a day for two months in the growing season. After the two months the rice plants will have become too big for the robot to swim through and the robots could damage the crops (Maruyama et al., 2014).

The improvement of Nakamura et al. (2017) is that they allow their robots to be guided by geolocation software such as presented in the book of Thomasson et al. (2019). Thomasson et al. (2019) shows in great detail how big farming equipment can be completely autonomous moving across many acres of fields. It can therefore be deduced that moving multiple “Robo-Duckies” around should not be a challenge. The software behind autonomous geolocation however is rather costly and needs a lot of data in order to work properly. This includes ground mapping and satellite imagery both are cost and time intensive procedures (Thomasson et al., 2019).

Table 14.

Soil toxicity -- Soil toxicity is not altered towards a higher pH. value. Food Security Availability Access Stability Utilization -

+ Increase of yield by less weeds.

+ Increase of yields goes together with increased monetary resources. -- No extra information provided on agricultural risks.

- No change in food quality or safety. GHG

emissions

+ Again, there is no direct link to GHGs, however, agricultural improvements may cause GHG emissions to decrease.

(29)

Technique 3: Informational support by image processing

Thirdly, the informative support by analysing data. The goal of these techniques is to help educate and inform the farmers on what they can do in order to increase yield, sustainability and longevity. Anami et al. (2020) showed a great algorithm that is able to identify different stress actors on paddy crops such as rice by analysing pictures of the field and crops. They achieved this by using a deep-learning algorithm on a dataset containing 30.000 pictures of 5 types of paddy crops. This resulted in the identification of 12 stress categories including healthy plants (Anami et al., 2020).

Figure 6.

Shows a classification tree containing 12 stress categories for the farmer to act upon after uploading images of their field.

Source: (Anami et al., 2020).

After testing, Anami et al. (2020), saw an average accuracy of identifying the correct stress factor with the picture of 92.89 percent. Having access to such a program could greatly benefit the farmers by showing what they have to do in order to be more successful. They can see when actions such as irrigation, liming or spraying pesticides are needed. This will allow farmers to greatly extend the longevity of their fields since they can easily see what additives are needed in order to improve the soil quality for a given crop. It is therefore important not to forget that this type of implementation is there as a support for the farmer and unlike the previously stated techniques which actively take action in improving the water efficiency or yield for example. Sudhakara et al. (2017) conducted research on increasing productivity of rice paddies by implementing the System of Rice Intensification (SRI). SRI “is an agro-ecological methodology for increasing the productivity of irrigated rice by changing the management of plants, soil, water and nutrients” (SRI, 2020). Using SRI saw improvements in both yield and irrigation management (Sudhakara et al., 2017). Such a method relies heavily on data that could be provided by the program Anami et al. (2020) present.

(30)

Table 15.

Soil toxicity ++ Possible increase in pH. through liming when identified as a stress factor. Food Security Availability Access Stability Utilization +

+ Increase in rice yields.

+ Increase in rice yields goes together with increased monetary resources. ++ Provides information on two + agricultural risks. Namely on pests and

droughts. The water stress of ‘submerged’ is not the same as the risk of flooding and is therefore not taken into account.

+ Increase in quality.

GHG emissions

+ Educating farmers and longevity will reduce the need to

6.3 Summary of Results

Table 16.

Overview scores per technique.

Indicators Scores: Technique 1 Scores: Technique 2 Scores: Technique 3

Soil toxicity -- -- ++

Food security - - +

GHG emissions + + +

Table 16 presents the individual scores of the different techniques placed next to each other. It clearly shows that technique 3 has reached the overall best score whereas both technique 1 and 2 have failed to reach desirable grades for soil toxicity and food security.

(31)

7. Conclusion, Discussion and Recommendations

7.1 Conclusion

This research tried to provide an answer to the main question which is stated as: ‘How can AI driven technological advancements reduce the local stress factors caused by slash and burn practises in Madagascar?’. This research analysed three different AI techniques by use of a white box problem analysis. Within the white box three different variables were used to test the three different AI techniques. The variables, soil toxicity, food security and GHG emissions were all derived from scientific literature and subsequently, a scale was created in order to test these techniques. Besides these variables, Malagasy Specific Limitations were provided as limiting conditions. Technique 1: ‘Field irrigation using soil sensors’, had mixed scores. It does not alter soil toxicity towards a higher pH value. Moreover, while this technique increases monetary resources by lowering costs, and thus food access, it does not improve food availability, stability or utilization. However, this technique is likely to lower GHG emissions. The second technique ‘Robotic support for weeding’ also had mixed scores. This technique does also not alter soil toxicity towards a higher pH value. And while this technique leads to more food availability and more food access, it does not lead to improved food stability and utilization. However, this technique also may cause GHG emissions to decrease. The analysis suggests that technique 3: Informational support by image processing, would be most

preferable in the case of Madagascar. This technique would allow for more optimal soils by increasing the pH by liming. Moreover, it would lead to increased food security by increasing yields and thereby monetary resources, as well as providing information on agricultural risks and lowering their impact, and increasing the quality of the products. Furthermore, this technique will lead to lower GHG emissions. All together, the effects of this technique would reduce the local stress factors the most, caused by slash and burn practices in Madagascar.

7.2 Discussion & Recommendations

Regarding soil toxicity, food security and GHG emissions, the informative support by analysing data techniques would propose a solution. However, there were many limitations in our research including lack of available data and prior research on AI techniques. For determining the indicators for soil toxicity, we used crop yields of rice, maize and beans. Other research showed an increase in pH. resulted in increased yields for maize and beans, while rice yields decrease. Rice is the crop that produces the highest yields per ha, and is an essential part of the Malagasy diet. However, the total crop yield of rice, maize and beans increase, suggesting that the loss in rice yield is overshadowed by large increases in maize and bean yield. More research on various rice strains related to soil pH. values might provide valuable information on how to improve rice yields in higher pH. soils.

Another limitation is missing data or data not within the scope of our research. Data that was important for creating a scale for food security was not available in the datasets that were used, or were irrelevant for this research. For instance, the FAO includes the variable ‘people usin g at least basic drinking water services’ for indicating food utilization. This is not something that would be relevant for our research, however it is for their definition of food security. Data limitation was also a problem for determining GHG emissions. There has not yet been research on emissions for the specific AI techniques discussed in this paper. This made it hard to measure the effects of the

techniques of GHGs. For further research of our proposed technique, research on emissions would be interesting. We would recommend conducting additional research to fill the existing data limitations and gaps. Moreover, field research is necessary to determine the feasibility of the proposed technique

(32)

8. References

Acemoglu, D., Johnson, S., & Robinson, J. A. (2001). The colonial origins of comparative development: An empirical investigation. American economic review, 91(5), 1369-1401.

Agarwal, D. K., Silander Jr, J. A., Gelfand, A. E., Dewar, R. E., & Mickelson Jr, J. G. (2005). Tropical deforestation in Madagascar: analysis using hierarchical, spatially explicit, Bayesian regression models. Ecological modelling, 105-131.

Anami, B. S., Malvade, N. N., & Palaiah, S. (2020). Deep learning approach for recognition and classification of yield affecting paddy crop stresses using field images. Artificial Intelligence in

Agriculture., 12-20.

Arvind, G., Athira, V. G., Haripriya, H., Rani, R. A., & Aravind, S. (2017). Automated irrigation with advanced seed germination and pest control. 2017 IEEE Technological Innovations in ICT for

Agriculture and Rural Development (TIAR), 64-67.

Boucher, O., Friedlingstein, P., Collins, B., & Shine, K. P. (2009). The indirect global warming potential and global temperature change potential due to methane oxidation . Environmental Research

Letters, 4(4), 044007.

Boulton, J., & Allen, P. (2007). The complexity perspective. In V. Ambrosini, M. Jenkins, & N. Mowbray (Eds.), Advanced strategic management: A multi-perspective approach (2 ed.). Palgrave Macmillan

Brooks, T. M., Mittermeier, R. A., Mittermeier, C. G., Da Fonseca, G. A., Rylands, A. B., Konstant, W. R., Flick, P., Pilgrim, J., Oldfield, S., Magin, G. & Hilton‐Taylor, C. (2002). Habitat loss and extinction in the hotspots of biodiversity. Conservation biology, 16(4), 909-923.

Buol, S. W. (2003). Formation of soils in North Carolina. Papers Commemorating a Century of Soil Science. Soil Science Society of North Carolina, Raleigh, 42-43.

Burrell, L. D., Zehetner, F., Rampazzo, N., Wimmer, B., & Soja, G. (2016). Long-term effects of biochar on soil physical properties. Geoderma, 282, 96-102.

Central Intelligence Agency. (2020, 10 9). The World Factbook, Africa: Madagascar. Retrieved from The World Factbook: https://www.cia.gov/library/publications/the- world-factbook/geos/ma.html Clark, M. "Deforestation in Madagascar: Consequences of population growth and unsustainable agricultural processes." Global Majority E-Journal 3.1 (2012): 61-71.

Corbera, E., Estrada, M., & Brown, K. (2010). Reducing greenhouse gas emissions from deforestation and forest degradation in developing countries: revisiting the assumptions. Climatic change, 100(3-4), 355-388.

Dave, R., Tompkins, E. L., & Schreckenberg, K. (2017). Forest ecosystem services derived by smallholder farmers in northwestern Madagascar: Storm hazard mitigation and participation in forest management. Forest Policy and Economics, 84, 72-82.

(33)

Dobson, A., Lodge, D., Alder, J., Cumming, G. S., Keymer, J., McGlade, J., Mooney, H., Rusak, J.A., Sala, O,. Wolters, V., Wall, D., Winfree, R., Xenopoulos, M. (2006). Habitat loss, trophic collapse, and the decline of ecosystem services. Ecology, 87(8), 1915-1924.

Fageria, N. K., & Baligar, V. C. (2008). Ameliorating soil acidity of tropical Oxisols by liming for sustainable crop production. Advances in agronomy, 99, 345-399.

FAO. (2020, 17 July). FAO - Food Security Indicators [Dataset]. FAO.

http://www.fao.org/fileadmin/templates/ess/foodsecurity/Food_Security_Indicators_17Jul2020.xlsx

Fearnside, P. M. (2000). Global warming and tropical land-use change: greenhouse gas emissions from biomass burning, decomposition and soils in forest conversion, shifting cultivation and secondary vegetation. Climatic change, 46(1-2), 115-158.

Ferentinos, K. P. (2018). Deep learning models for plant disease detection and diagnosis. Computers and Electronics in Agriculture, 311-318.

Food and Agriculture Organization (2014). Building a common vision for sustainable food and agriculture: principles and approaches. Retrieved from http://www.fao.org/3/a-i3940e.pdf

Gade, D. W. (1996). Deforestation and its effects in highland Madagascar. Mountain Research and Development, 101-116.

Gil, J. D. B., Reidsma, P., Giller, K., Todman, L., Whitmore, A., & van Ittersum, M. (2019). Sustainable development goal 2: Improved targets and indicators for agriculture and food security. Ambio, 48(7), 685-698.

Goodman, S.M. & Benstead, J.P. (2005) Updated estimates of biotic diversity and endemism for Madagascar. Oryx 39.1 73-77.

Grieco, J., Ikenberry, J. G., & Mastanduno, M. (2019). Introduction to International Relations (2nd edition). London: Macmillan Education UK.

Harper, G. J., Steininger, M. K., Tucker, C. J., Juhn, D., & Hawkins, F. (2007). Fifty years of deforestation and forest fragmentation in Madagascar. Environmental conservation, 325-333. Harvey, C. A., Rakotobe, Z. L., Rao, N. S., Dave, R., Razafimahatratra, H., Rabarijohn, R. H., ... & MacKinnon, J. L. (2014). Extreme vulnerability of smallholder farmers to agricultural risks and climate change in Madagascar. Philosophical Transactions of the Royal Society B: Biological Sciences, 369(1639), 20130089.

Jackson, R. H., & Sørensen, G. (2016). Introduction to International Relations (6th edition). Oxford University Press.

Jarosz, L. (1993). Defining and explaining tropical deforestation: shifting cultivation and population growth in colonial Madagascar (1896–1940). Economic Geography, 69(4), 366-379.

(34)

Jha, K., Doshi, A., Patel, P., & Shah, M. (2019). A comprehensive review on automation in agriculture using artificial intelligence. Artificial Intelligence in Agriculture, 2, 1-12.

Kotto-Same, J., Woomer, P. L., Appolinaire, M., & Louis, Z. (1997). Carbon dynamics in slash -and-burn agriculture and land use alternatives of the humid forest zone in Cameroon. Agriculture,

Ecosystems & Environment, 65(3), 245-256.

Lammel, G. & Grassl, H. (1995). Greenhouse effect of NO X. Environmental Science and Pollution

Research, 2(1), 40-45.

Malcolm, J. R., Liu, C., Neilson, R. P., Hansen, L., & Hannah, L. E. E. (2006). Global warming and extinctions of endemic species from biodiversity hotspots. Conservation biology, 20(2), 538-548. Maruyama, A., & Naruse, K. (2014). Feasibility study of weeding robots in rice fields inspired by natural ducks. In Proceedings of the 8th international conference on bioinspired information and

communications technologies (pp. 378-381).

Menken, S. & Keestra, M. (2016). An introduction to interdisciplinary research. Amsterdam University Press.

Myers, N., Mittermeier, R. A., Mittermeier, C. G., Da Fonseca, G. A., & Kent, J. (2000). Biodiversity hotspots for conservation priorities. Nature, 403(6772), 853-858.

Myhre, G., D. Shindell, F.-M. Bréon, W. Collins, J. Fuglestvedt, J. Huang, D. Koch, J.-F. Lamarque, D. Lee, B. Mendoza, T. Nakajima, A. Robock, G. Stephens, T. Takemura and H., Zhang. (2013)

Anthropogenic and Natural Radiative Forcing. Climate Change 2013: The Physical Science Basis. Contribution of Working Group I to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change. https://www.ipcc.ch/

Nakamura, K., Kimura, M., Anazawa, T., Takahashi, T., & Naruse, K. (2016). Investigation of weeding ability and plant damage for rice field weeding robots. 2016 IEEE/SICE International

Symposium on System Integration (SII), 899-905.

Nunn, N., & Wantchekon, L. (2011). The slave trade and the origins of mistrust in Africa. American Economic Review, 101(7), 3221-52.

Otto, J. S., & Anderson, N. E. (1982). Slash-and-burn cultivation in the Highlands South: A problem in comparative agricultural history. Comparative Studies in society and History, 24(1), 131-147.

Potten, D. (1983). Irrigation in Madagascar: a review.

Raik, D. (2007). Forest management in Madagascar: An historical overview. Madagascar Conservation & Development, 2(1).

Raboin, L. M., Razafimahafaly, A. H. D., Rabenjarisoa, M. B., Rabary, B., Dusserre, J., & Becquer, T. (2016). Improving the fertility of tropical acid soils: liming versus biochar application? A long-term comparison in the highlands of Madagascar. Field Crops Research, 199, 99-108.

Rammelt, C. (2020). Weblecture 5: Complexity [Powerpoint]. Retrieved from https://canvas.uva.nl/courses/pages/weblectures