A system dynamics-based scenario analysis of residential solid waste management in Kisumu, Kenya

The problem of solid waste management presents an issue of increasing importance in many low-income settings, including the progressively urbanised context of Kenya. Kisumu County is one such setting with an estimated 500 t of waste generated per day and with less than half of it regularly collected. The open burning and natural decay of solid waste is an important source of greenhouse gas (GHG) emissions and atmospheric pollutants with adverse health consequences. In this paper, we use system dynamics modelling to investigate the expected impact on GHG and PM2.5 emissions of (i) a waste-to-biogas initiative and (ii) a regulatory ban on the open burning of waste in landfill. We use life tables to estimate the impact on mortality of the reduction in PM2.5 exposure. Our results indicate that combining these two interventions can generate over 1.1 million tonnes of cumulative savings in GHG emissions by 2035, of which the largest contribution (42%) results from the biogas produced replacing unclean fuels in household cooking. Combining the two interventions is expected to reduce PM2.5 emissions from the waste and residential sectors by over 30% compared to our baseline scenario by 2035, resulting in at least around 1150 cumulative life years saved over 2021–2035. The contribution and novelty of this study lies in the quantification of a potential waste-to-biogas scenario and its environmental and health impact in Kisumu for the first time.

Human Development Index (HDI), below the national average at 0.56 (County Government of Kisumu, 2019). The 2019 population census indicates that the county has a population of about 1,156,000 people (KNBS, 2019). Population has been growing at a rapid rate of about 2.3% per year and is expected to continue to grow at over 2% per year until 2030(United Nations, 2019. Rapid urbanisation and changing consumption patterns, together with poor environmental management, have turned MSW into an alarming crisis for Kenya, manifest in the commonly overflowing dumpsites in the cities which are cause for environmental and health hazards (Awuor et al., 2019). As with many urban areas in the Global South, Kisumu is struggling with an overflowing dumpsite as well as consequent environmental and health risks associated with improper disposal of MSW (Sibanda, Obange and Awuor, 2017). Kisumu County generates about 500 tonnes of solid waste per day 1 (Oyake-Ombis, 2017) out of which, based on estimates we obtained from local actors in the system, only about 40% is collected for disposal at the city's open landfill (see Appendix A, unsightly garbage heaps scattered around the city, see Figure 2 right photo) or in drainage systems (resulting in frequent flooding of neighbourhoods with waste and sewage water) (Gutberlet et al., 2017;Sibanda, Obange and Awuor, 2017). The County has developed and revised an Integrated Solid Waste Management Plan (KISWAMP) (County Government of Kisumu, 2017), but this has so far failed to result in a transformation of the state of MSWM in Kisumu (Awuor et al., 2019). In line with Kenya's strategic target of reducing GHG emissions by 30% by 2030, as pledged at COP-21 in Paris 2015 with a strong focus on increasing access to renewable energy (Dalla Longa and van der Zwaan, 2017), Kisumu County's KISWAMP (County Government of Kisumu, 2017) discusses the potential in waste-to-energy (WtE) technologies. Currently, a wide range of such technologies exist. These are broadly categorised as thermal (e.g., incineration, pyrolysis, gasification) and biological (e.g., aerobic composting or anaerobic digestion/biogasification) (Moya et al., 2017). We assert that incineration, which is the most widely used method (Fernández-In Kenya, in the city of Naivasha, 76 km from Nairobi, a 2.4 MW commercial biogas plant, with a cost of $6.5 million and an annual treatment capacity of 50,000 tonnes of organic waste, inaugurated in 2017 and is reportedly the largest grid-connected biogas power plant in Africa, meeting the power needs of 6,000 rural homes (Roopnarain and Adeleke, 2017;Kemausuor, Adaramola and Morken, 2018). In this paper, however, rather than proposing to use biogas from waste to generate electricity, we explore the option of making the biogas directly accessible to households for use in cooking.
Currently, close to 80% of households in Kisumu use traditional biomass fuels (mainly wood and charcoal) for cooking (KNBS, 2019, p. 336). Indoor air pollution caused by traditional cooking is today's most important environmental health risk and second-largest risk factor in all categories in Eastern SSA (Lim et al., 2012). Women and children are disproportionately at risk of health issues caused by indoor air pollutants. Furthermore, the use of wood and charcoal for cooking is a major driver of deforestation and GHG emissions (Carvalho et al., 2019). Evidence shows that using alternative cook stoves significantly reduces indoor air pollution, and numerous studies demonstrate the link between reductions in household air pollution and improved respiratory health (Anderman et al., 2015). Tumwesige et al. (2017) monitored real-time PM2.5 and CO concentrations in 35 households in Cameroon and Uganda where biogas and firewood (or charcoal) were used and found that fully switching to biogas for cooking reduces both CO and PM2.5 concentrations to below WHO recommended limits. Although no direct evidence on the health benefits of households switching to biogas is available, comparable studies of households switching to LPG suggest that such a shift could bring respiratory and cardiovascular health benefits of the order of 20-25% In summary, it appears that anaerobic digestion of biowaste to produce biogas for use in household cooking holds great potential in reducing waste to landfill and associated externalities (e.g., pollutant and GHG emissions, groundwater contamination), while simultaneously improving indoor air quality and related health outcomes. Within this context, the purpose of this study is therefore to explore the idea of a transition towards anaerobic digestion of Kisumu's organic fraction of MSW and the use of the produced biogas in household cooking on the levels of waste accumulating in landfill or waste scattered elsewhere, on waste related GHG emissions, on air pollutant concentrations, and on related health impacts. The novelty and importance of this paper lies in the quantification of a potential waste-to-biogas scenario and its environmental and health impact in Kisumu for the first time.
Existing studies on the impacts of WtE technologies in other contexts-e.g., Ayodele, Ogunjuyigbe and Alao (2017) in Nigeria, Chaya and Gheewala (2007) in Thailand, Evangelisti et al. (2014) in the UK, and Rigamonti, Grosso and Giugliano (2010) and Cremiato et al. (2018) in Italy-tend to take a static Life Cycle Assessment (LCA) approach. Considering that the waste system involves distinctly dynamic processes, such as the accumulation, depletion and degradation of stocks of waste, static methods do not appear up to the task of informing policymaking in this area, where investments are often large-scale with long timeframes in mind. Thus, for various reasons, the primary method used in this study is system dynamics (SD). Firstly, a key advantage of SD over common spreadsheet waste management models such as LCA is the dynamic nature of SD models, versus the static optimization in spreadsheet-based methods (Adamides et al., 2009;Inghels and Dullaert, 2011).
The rest of the paper is structured as follows. In the next section, the methodology used in this study is described. Subsequently, in Section 3, the results from our scenario analyses are visualised, compared and contrasted. The paper concludes in Section 4 with a brief discussion of the results, implementation challenges and study limitations. This manuscript is accompanied by three Appendices including a full documentation of the model, list of model parameters, and detailed specification of the scenarios. The paper is accompanied by an online supplement containing a folder with the model and all scenario runs.

Methods
The aims and scope of this study were determined based on a series of eight focus group discussions in Kisumu during July 2019 with representatives from Kisumu County's Department of Environment, the local industry, non-government groups, community-based organisations, academia and resident associations. These discussions, which were audio-recorded and later transcribed, provided context information of the current waste management situation and diverse stakeholder perspectives about it (Salvia et al., 2021). Our scenario definitions were also informed by these discussions.
Multiple methods are combined for the purpose of this study. First, the central method applied is SD (Sterman, 2000), which is introduced in the following sub-section 2.1. In sub-section 2.2, a description of the SD model follows. As seen in Appendix B, where all parameter assumptions used in the SD model and their sources are listed, the primary source for parametrising the model has been existing academic papers, national and international databases and industry publications. Data for certain parameters specific to the state of SWM in Kisumu, such as the city's current waste J o u r n a l P r e -p r o o f collection capacity or estimates of the current stock of waste in the city's landfill, were obtained in correspondence with the Kachok dumpsite manager and Kisumu county officials.
Second, emission factors used to calculate GHG emissions were obtained according to the IPCC guidelines (IPCC, 2006), as described in Section 2.3. Third, the method for estimating ambient and household PM2.5 concentrations is described in Section 2.4. Fourth, these estimates are fed into a life table health impact assessment model (as described in Section 2.5). This Methods section concludes with a description of our scenarios.
In their review of the main existing approaches to GHG accounting in waste management, including national accounting, corporate level accounting, life cycle assessment, and carbon trading methodologies, Gentil, Christensen and Aoustin (2009) emphasise the importance of transparency in GHG accounting concerning aspects such as waste type and composition, time period considered, GHGs included, choice of system boundaries, etc. Following this guideline, full transparency is followed in describing the method and the model in the following sub-sections, and in more detail in the Appendices. This being an initial, high-level, aggregate model, it has several limitations, as discussed later in Section 4.3.

System Dynamics and Its Past Applications to SWM
System dynamics is a method based on computer simulation where a model of the cause-and-effect relationships of a real-world complex system is built, parametrised and validated using real-world information. The sources of such information can be varied and can include not only those available in numerical datasets and scientific literature, but also those gleaned from the mental models of experts (Forrester, 1987).

J o u r n a l P r e -p r o o f
Thanks to its strengths in bringing together knowledge from a variety of fields in an integrated framework and in tackling dynamically complex problems, SD has been widely applied to the problem of MSWM in the past. In terms of quality, papers applying SD to SWM are very mixed. The history of such applications goes back around three decades, starting with Mashayekhi (1993) who uses an SD model capturing major interactions between different socioeconomic and environmental factors to study the problem of solid waste disposal in New York. Later, and within the context of a lower-income country, Sudhir, Srinivasan, & Muraleedharan (1997) propose an SD model for the study of the potential consequences of various structural and policy alternatives for a sustainable urban SWM system for a typical metropolitan city in India, and conclude by recommending the allocation of waste management funds in proportion to the requirements of collection, disposal and processing, as opposed to prioritising short-term interests such as only collection of waste. Still within the context of India, Talyan, Dahiya, Anand, & Sreekrishnan (2007) use an SD approach to quantify CH 4 emissions from MSW disposal under various scenarios in Delhi. Their model shows that an improved waste management system, involving the introduction of composting, biogasification, and refuse-derived fuel, would significantly reduce CH 4 emissions over time despite an increase in waste generation. Sufian & Bala (2007) build an SD model for SWM in the city of Dhaka, Bangladesh, the results of which show that in order to improve environmental outcomes, it is not sufficient to increase budget for waste collection capacity, but this needs to be accompanied by increasing the budget for treatment, mirroring the finding of Sudhir et al. (1997). This mindset informs the current study as well.
Within the context of Kisumu, Gutberlet et al. (2017) apply a combination of action net theory and systems thinking to build a map of the waste management system in Kisumu with all its actors, actions, processes and interconnections. Their main conclusion is that -new waste initiatives should build on existing waste management practices already being performed within informal settlements J o u r n a l P r e -p r o o f by waste scavengers, waste pickers, waste entrepreneurs, and community-based organizations (Gutberlet et al., 2017, p. 106).‖

Model Description
The full model documentation is provided in Appendix A -Full Model Documentation. In this section, a high-level schematic overview of the model is shown in Figure 3. The model consists of four inter-connected sectors: (1) Waste Collection, (2) Biogas, (3) Landfill, and (4) Scattered Waste.
Variables calculated in one sector are often used as inputs in another sector. In the first sector, which captures waste collection, indicators such as total waste generated, total waste collection capacity, proportion of waste collected and greenhouse gas emissions due to waste transport are calculated. In particular, total food waste collection capacity becomes a key input to the Biogas Sector, as a constraint on biogas production capacity along with the cumulative capacity of the biogas facilities, together determining total biogas generated. Subsequently, the savings in GHG emissions resulting from a switch to clean biogas for cooking are calculated and accumulate in the stock of cumulative savings in GHG emissions due to products of anaerobic digestion.
A by-product of the biogas plants is digestate, which can be used as fertiliser, either directly or upon further processing into compost. This organic fertiliser reduces the need for inorganic fertiliser use in the region, potentially countering another source of GHG emissions. However, there is substantial uncertainty around the extent of such savings (Møller, Boldrin and Christensen, 2009). Cecchi et al. (2011) estimate these savings in the range of 30-40 kg-CO 2 t -1 while cautioning that fugitive CH 4 and N 2 O emissions when digestate is applied on land, ranging from 0 to 50 and from 30 to 60 kg-CO 2 t -1 respectively, can cancel out any savings (Cecchi et al., 2011). The aggregate result will depend on the exact operating conditions and is likely to be small (Møller, Boldrin and Christensen, 2009). Therefore, any digestate-related GHG saving or load is disregarded in this model. Similarly, J o u r n a l P r e -p r o o f assuming that any fugitive CH 4 emissions from the biogas plant are flared, such emissions are not accounted for.

Figure 3 -Overview of model sectors and interlinkages
Next, the waste that remains and that is not used for biogas production is transported to landfill, as captured in the Landfill Sector, given our mixed waste collection constraints (coming from the Waste Collection Sector). The accumulation of food and non-food waste in landfill, together with any reductions in the waste mass via open burning, natural decomposition and informal waste-picking are captured in the Landfill Sector. Furthermore, emissions of different types of GHGs as a result of burning and decomposition, including carbon dioxide (CO 2 ), methane (CH 4 ) and black carbon (BC), are also calculated, along with the annual and cumulative savings in GHG emissions (both from landfill waste and from scattered waste, as imported from Scattered Waste Sector). Various emission factors for food and non-food waste required for these calculations are derived based on best available evidence, as described in Section 2.3. A key feature of the model is that the food and nonfood contents of the waste that remains after biogas production and is disposed of are dynamically J o u r n a l P r e -p r o o f calculated. This leads to outcomes which are not immediately evident without using simulation, as we will see in the results (Section 3).
Similarly, the Scattered Waste Sector captures the accumulation, depletion and emission processes for food and non-food waste which is not collected due to the constraints of our waste collection fleet capacity and is structured in the same way as the Landfill Sector. Besides GHG emissions, particulate matter (PM 2.5 ) emissions from both landfill and scattered waste are also calculated in this sector, which are then used for estimating the potential effects of our scenarios on population health, according to the method described in Section 2.5.
With regards to the boundaries of the model, based on Gentil et al.'s (2009) proposed upstreamoperating-downstream framework for GHG accounting in waste management, in the ‗indirect upstream' category, in the model we have accounted for emissions from waste transport; in the ‗direct operating' category, we have accounted for landfill and scattered waste emissions (CH 4 from decomposition and CO 2 and BC from burning), and in the ‗indirect downstream' category, we have accounted for savings resulting from the biogas substituting biomass in household cooking. These boundaries for the model can be considered in compliance with Møller, Boldrin and Christensen's (2009, p. 823) conclusion that -irrespective of the employed technology, as long as the produced biogas is utilized for energy substitution, the indirect downstream emissions are the most important factor. Direct emissions at the AD facility and indirect upstream emissions play less important roles.‖

Development of emission factors
We use emission factors from the GAINS model (Amann et al., 2011(Amann et al., , 2020  Emission factors for black carbon (BC) and PM 2.5 are adopted from Akagi et al. (2011) andChristian et al. (2010) and are in line with the emission factors used by Klimont et al. (2017) and Wiedinmyer, Yokelson, & Gullett (2014). The emission factors are 8.74 tonne/kt waste burnt for PM 2.5 and 0.65 tonne/kt waste burnt for BC. These emission factors are for mixed waste and are not representative of Kisumu's particular waste composition. J o u r n a l P r e -p r o o f  Table 2 shows the background information needed to carry out the estimation of the emission factors.

Estimation of ambient and household PM 2.5 concentrations
The PM 2.5 annual emissions obtained based on the above emission factor are converted into ambient PM 2.5 concentrations using a simplified version of the atmospheric calculations in the GAINS model (Amann et al., 2020) which themselves rely on a linearized representation of full atmospheric chemistry transport model simulations. GAINS contains atmospheric transfer coefficients from all source pollutants for PM 2.5 in Kenya to a 0.1° receptor grid. As detailed in Appendix A (Section iv), we developed an integrated atmospheric transport coefficient from near-ground emissions of PM 2.5 J o u r n a l P r e -p r o o f in Kisumu to ambient PM 2.5 concentrations in Kisumu, which is then applied to the respective emissions from residential combustion and MSW burning to estimate their impacts.
For household PM 2.5 concentrations, we used an approximation method with a high level of uncertainty, described in detail in Appendix A (Section ii), which is based on empirical measurements reported in Muindi et al. (2016, p. 7 Table 3) on mean levels of indoor PM 2.5 concentrations in households using different cooking fuel types.

Health impact assessment
We estimated the effect of changes in exposure to ambient and household PM 2.5 on mortality in Kisumu under each scenario using life tables based on the IOMLIFET model (Miller and Hurley, 2003) programmed in R (version 3.5.1, R Foundation for Statistical Computing, MA, USA). The effects of changes in PM 2.5 were modelled by applying to the life tables the Global Burden of Disease (GBD) Integrated Exposure-Response functions relating long-term PM 2.5 exposure to mortality risk from five causes -ischaemic heart disease (IHD), chronic obstructive pulmonary disease, stroke, lung cancer and lower respiratory infections (LRI) (Apte et al., 2015). The functions for IHD and stroke varied by age.
The life tables were set up using age-and gender-specific population and cause-specific mortality data for Kenya from the GBD's GHDx tool for the closest available year of data to the study period (2017). The national-level population data was downscaled to represent the population of Kisumu.
Single-year-of-age mortality rates were calculated from 5-year rates via one-way spline interpolation using the MS Excel add-in, SRS splines (version 2.5, SRS1 Software LLC, MA, USA).

J o u r n a l P r e -p r o o f
We combined ambient and household PM 2.5 as a time-weighted average, assuming that men and women in Kisumu spend 50% and 80% of their time indoors at home, respectively. To account for delays in changes in mortality risk following air pollution exposure reductions, we incorporated cessation lags for each outcome. These were exponential functions parameterised using evidence from studies of smoking cessation (Lin et al., 2008) and assumptions about disease progression over time. For IHD and lung cancer, we assumed the full effect would be reached after 15-20 years, with shorter lags for COPD, stroke and LRI.
The outputs from the life tables are life years lived by the population over the study period. Solid waste may give rise to other forms of adverse health impact but in the analysis presented in this paper, we concentrate only on those arising from contamination of the outdoor air by fine particles (PM 2.5 ) arising from burning of solid waste.

Description of scenarios
In this study, we simulate four different scenarios as summarised in Table 3. The scenarios were developed in close connection to planned developments of Kisumu City regarding waste management strategies (County Government of Kisumu, 2017) and designed to account for local structural factors as well as international guidelines.
In our (1) Baseline (business-as-usual) scenario, we assume only a gradual increase in the mixed waste collection transport fleet, in line with recent trends. Waste volume at the dumpsite is mainly managed through open burning (as the existing mechanical compactor is insufficient and usually non-operational due to inadequate maintenance). At the same time, since most of the waste is composed of moist organic matter, combustion occurs only on the surface and does not significantly reduce waste volume (Awuor et al., 2019). This open burning is a major contributor to emissions of J o u r n a l P r e -p r o o f

Journal Pre-proof
GHGs and atmospheric pollutants (Forbid et al., 2011). In scenario (1b) Ban on Burning, we assume the enforcement of a gradual regulatory ban on the open burning of waste in landfill. 2 The ban on open burning is one of the Global Waste Management Goals set out by the United Nation's Environment Programme (UNEP) in the Global Waste Management Outlook (Wilson et al., 2015).
This being a major change in SWM practices in Kisumu, in consultation with local county officials we assume that the Ban takes place over an extended period of eight years, bringing the fraction of waste annually burnt in landfill from the current 23 percent per year (Onyango and Kibwage, 2008) gradually down to zero. This Ban is assumed to be enforced only in dumpsite at this stage.
In the (2)  In terms of substrate provision, these plants would need to be supplied with source-separated organic fraction of MSW. The decentralised approach has the advantage of minimising the distance travelled for transporting the waste to treatment facilities (Gebreegziabher et al., 2014). We assume that a separate collection system for food waste is gradually built up to match the plants' expanding waste treatment capacity. The collection and transportation of the food waste shall be done by specialpurpose handcarts, capable of accessing narrow alleyways in the informal settlements and operated by waste collectors formally employed by the City-perhaps recruited from among current informal actors in the sector, in line with Gutberlet et al.'s (2017) context-specific recommendation of building improved SWM practices on existing ones.
As outlined and justified earlier in the Introduction, we assume that the produced biogas will then be bottled and distributed to households at filling stations for use in cooking instead of currently prevalent biomass and kerosene (KNBS, 2019, p. 336). A distributed set of facilities makes the filling stations more easily accessible for households while providing jobs to the local community.
Based on the assumption of a 3,000 tonne per year treatment capacity, a yield of 100 m 3 per tonne of food waste (Veeken (2005) cited in Müller (2007, p. 26, Table 3)), and an average household need of 262.5 m 3 biogas per year for cooking (see Appendix B for sources and calculation), each facility is J o u r n a l P r e -p r o o f expected to provide cooking fuel for around 1,150 households. A recent working paper by Twinomunuji et al. (2020) suggests that, in the SSA region, biogas-based cooking fuels would compete favourably in price with other commercial fuels, including LPG. While highlighting the promise in such initiatives, they furthermore identify several barriers towards widespread interest in bottled biogas in Africa, which will be discussed later in Section 4.2.
Finally, in scenario (2b)   J o u r n a l P r e -p r o o f

Results
In this section, we use simulation to gain insight into the likely future developments in the dynamics of waste accumulation, associated GHG emissions, PM 2.5 concentration and consequent health outcomes under the described sets of scenario assumptions. We will start by comparing projected trends in waste accumulation under the Baseline and Biogas scenarios in the first sub-section and continue by comparing GHG emissions under the two scenarios in the following sub-section. Next, we will look at results from the Ban on Burning scenario and the Combined scenario. The last two sub-sections deal with projections related to changes in PM 2.5 and the resulting health impacts. Biogas scenario, landfill food waste is projected to reach less than 60% of its Baseline value by 2035.

Stocks of Waste: Baseline and Biogas Scenarios
This is not surprising because as more and more of the food waste (57% by 2035) is used for biogas production, there is less food waste being transported to landfill, to the point that the flow of food waste into the stock comes close to the aggregate outflows due to decomposition and burning, keeping landfill food waste relatively stable. Conversely, there is a relatively higher accumulation of non-food waste in landfill, as the waste that is left after biogas production to be transported to landfill becomes more non-organic in nature, with the non-food content ratio (not shown here) going from around 37% initially to 58% by the end of the simulation period in the Biogas scenario, while it stays roughly constant in the Baseline simulation.  Projected GHG emissions resulting from scenarios 1 and 1b are shown in Figure 6. The behaviour of total CO 2 eq methane emissions due to waste decomposition (panel A) can be understood by referring to the two graphs on the left hand-side of Figure 5. With waste being transported increasingly to Similarly, black carbon emissions due to waste burning rise at a decreasing rate in the Baseline scenario, while they stay fairly stable under the Biogas scenarios, cut by about 33% by 2035 as compared to Baseline. Since the BC emission factor assumed for all three types of waste is the same, the change in emissions in our scenarios cannot be the result of a redistribution of waste among the various stocks (food/non-food landfill/scattered waste) but is rather the result of a reduction in the sum total amount of the waste that is disposed of due to the recycling of a part of the total waste for biogas production.
On the bottom left (panel C), we can see that total direct CO 2 emissions due to waste burning do not change in the Biogas scenario compared to Baseline, with the two curves fully overlapping. This is because, as mentioned in Section 2.3, these emissions are a product of non-food waste only, and total non-food waste does not change under the Biogas scenario, rising slowly with population as it does in Baseline.
J o u r n a l P r e -p r o o f As a result of this reduction in emissions throughout the 15 years of the simulation as shown in the above figures, as well as many households being able to switch from fossil fuels to renewable biogas for cooking and the resulting digestate from the biogas production process replacing an equivalent amount of inorganic fertiliser, we expect to see a substantial cumulative saving in GHG emissions in the Biogas scenario, as shown in Figure 7. Simulation suggests that by 2035, each year around 9 million m 3 of biogas can be generated in this way, providing cooking fuel for 8-9% of total households in Kisumu county. Total cumulative savings in emissions reach 700,000 tonnes of CO 2 eq

D) Total CO 2 eq emissions from waste
Baseline Biogas J o u r n a l P r e -p r o o f by 2035. Two thirds of these savings come from households switching to biogas, with one third resulting from the reduction of waste in landfill and scattered waste.

Ban on Burning Scenario
Based on what we saw in Figure 6, it becomes clear that potentially significant improvements in total emissions are undermined by the lack of any improvements in direct CO 2 emissions from burning.
Therefore, if we are to make more substantial and sustainable improvements in GHG emissions, we it would lead to waste piling up more rapidly. In total, by 2035, we expect total landfill waste to be 2.3 times higher than the Baseline scenario. Mentally simulating the aggregate outcome of this intervention for total emissions is not straightforward because on the one hand landfill waste is growing faster but on the other hand emissions due to burning are reduced to zero in landfill.
Simulation can help here by providing a projection for future emissions, as shown in Figure 9.

Combined Scenario
Having seen the significant potential of this intervention for reducing emissions, we will now investigate the expected outcome of combining this with our Biogas scenarios, identified as Scenario 2b in Furthermore, it would be of interest to investigate the share of each individual intervention in the resulting cumulative savings in GHG emissions. This is visualised in Figure 10 below. As can be J o u r n a l P r e -p r o o f seen, the largest contribution (42% of total in 2035) is derived as a result of the biogas produced replacing unclean fuels in the community's kitchens. On top of that there are significant savings (30% of total in 2035) thanks to the gradual enforcement of a ban on the open burning of waste, pointing to the crucial importance of enforcing such measure for reducing emissions. Next, we expect substantial savings (20% of total in 2035) in emissions associated with recycling part of the organic waste, diverting it away from landfill and into biogas production. Also interesting is the nonnegligible portion of the savings (8% of total in 2035) that cannot be contributed to any individual intervention alone and is rather the synergistic outcome of simultaneous implementation of all interventions (the portion shown in black in Figure 10). As we saw earlier (Figure 9), the ban on burning policy alone significantly reduces emissions due to burning but at the same increases emissions due to waste decomposition, due to the higher levels of accumulated waste. Therefore, combining this intervention with the Biogas scenario which helps decrease the accumulation of food waste gives results that are superior to simply superimposing improvements from each separate intervention. Therefore, a ban on open burning together with the biogas production intervention helps maximise potential benefits.

J o u r n a l P r e -p r o o f
Journal Pre-proof

Share of individual interventions in emission savings
Household switching to biogas Ban on burning Recycling organic waste Synergy Baseline scenario, as a result of growth in population, this total is expected to rise by over 50% to 12.6 µg/m 3 . The Biogas scenario stands at a total of 10.8 µg/m 3 by 2035, 14% lower than Baseline, with improvements coming from both sources (a transition to biogas for cooking as well as less waste being burnt). The Ban on Burning scenario brings a slightly more substantial reduction of 18% compared to Baseline, with all of this reduction naturally deriving from less waste burning (which only takes place in places other than landfill in this case). As expected, the highest reduction results from combining the two interventions, which brings total PM 2.5 concentration from the two sources down to 8.7 µg/m 3 , over 30% lower than Baseline, and only 5% higher than the present level, despite the nearly 40% projected rise in population over the period. Given the time lags between changes in exposure and health outcomes, the estimated improvements are expected to grow substantially larger over time and would be greater than presented here if we were to extend the follow up period.

Discussion and Conclusions
In this section, we will start with a summary of our findings and continue with a discussion of some of the implementation challenges of our proposed policies and conclude by briefly enumerating some of the limitations of this study and suggesting avenues for further research. With regards to air pollutant emissions and concentrations, combining the two interventions is expected to bring total PM 2.5 emissions from the residential and waste sectors down by over 30% compared to Baseline by 2035; a level only 6% higher than present, despite the nearly 40% projected rise in population over the period. Furthermore, the model estimates a potential improvement of around 10% in indoor air PM 2.5 concentrations by 2035 as a result of a fraction of households (8.2%) being able to switch to biogas for cooking, as well as improved ambient air quality. This mirrors the qualitative but empirical findings of Clemens et al. (2018), who report that 45%-91% of users in the Africa Biogas Partnership Program reported reduced eye problems and respiratory symptoms. Our health impact assessment suggests that these combined improvements in exposure can be expected to result in nearly 1,150 cumulative life years saved by 2035, with an additional ~220 years or more added to those savings every year by that point.

Implementation Challenges
In our modelling and analysis, we did not consider potential difficulties in the implementation of the interventions considered. Kemausuor, Adaramola and Morken (2018) present a comprehensive review of barriers towards the uptake of biogas technology in Africa and maintain that, given the large initial investment costs, financing is at the heart of the barriers to extended uptake of biogas.
Therefore, this study is part of a larger multi-partner effort to obtain funding for the described wasteto-biogas initiative from an international green climate fund. Other barriers identified by Twinomunuji et al. (2020) based on their case studies in Uganda and Ghana include varying enforcement of regulations, uncertainties around user experience with biogas including cooking preferences, and lack of in-country expertise. Furthermore, there are safety issues around operation J o u r n a l P r e -p r o o f of biogas installations having to do with the toxicity and the combustibility of biogas which can cause fires and explosions, although the associated risks are lower than chemical plants (Trávníček and Kotek, 2015).
In addition, transitioning towards our particular preferred scenario (Scenario 2b. Biogas + Ban on Burning) would require planning for and investing in the filling stations needed to make the product available to households, which poses an important technical and organisational challenge. It would also call for significant behavioural changes by households and other actors involved in the system.
Firstly, households would need to sort their organic waste for collection. This has been identified as an ongoing challenge in Kisumu over several decades (Henry, Yongsheng and Jun, 2006;M. Aurah, 2013;Sibanda, Obange and Awuor, 2017;Awuor et al., 2019), although some household waste is sorted for composting and informal waste picking (Sibanda, Obange and Awuor, 2017). Field studies suggest there is an interrelated set of barriers to efficient waste sorting at scale. One is that households and public spaces in the city lack segregated bins (Sibanda, Obange and Awuor, 2017;Awuor et al., 2019). Where they are available, waste types are still often mixed either at the point of disposal, or when the bins are emptied and waste transported to the dumpsite (Sibanda, Obange and Awuor, 2017;Awuor et al., 2019). Knowing this may undermine households' motivation to segregate waste. This might be further compounded by disagreement among stakeholders about who is responsible for the city's solid waste management, and a perceived mismatch between the government's expectations of the public and the public's willingness to participate in waste management (Schlueter, 2017).
Secondly, our combined scenario would require households to switch to and sustain the use of biogas as a cooking fuel. Despite the health, climate and economic advantages of switching from traditional to cleaner cooking fuels, studies in Kenya and other low-and middle-income settings indicate that J o u r n a l P r e -p r o o f such considerations do not necessarily drive sustained adoption (Jonušauskait, 2010;Rupf et al., 2015;Puzzolo et al., 2016;Chalise et al., 2018;Hamid and Blanchard, 2018;Thompson et al., 2018). Barriers identified among rural Kenyan communities to the sustained adoption of biogas included a lack of information and understanding about its use, benefits and cost-efficiency compared to traditional fuels (Ndereba, 2013;Hamid and Blanchard, 2018).
For both sorting waste and switching fuels, tools for designing and implementing behaviour change interventions may help achieve these transitions. Systems methods can also be used to understand the wider network of actions needed to support these changes (Gutberlet et al., 2017).

Limitations
In building the model used in this study we have made a number of simplifying assumptions. For example, we have assumed that waste generation per household will stay constant over our simulation period of 15 years. However, Olang et al. (2018) have demonstrated that the amount of waste generated per household for Kisumu is dependent on factors such as household size and income. The model can be improved by incorporating these drivers based on any existing future projections for income and household size and by allowing waste generated per household to vary based on these.
Another key limitation of the model has to do with its choice of boundaries concerning the GHG accounting aspect, which includes only those components believed to be the most significant. The upstream-operating-downstream framework suggested by Gentil, Christensen and Aoustin (2009) includes several other components that, albeit less important in scale, represent useful potential additions to our model. These include leaked N 2 O and CH 4 emissions from the biogas plant and digestate-related considerations (including fugitive and transport emissions and mineral fertiliser substitution savings).
Certain limitations are imposed on this study by the generally poor availability of data in the context of Kisumu. For instance, our estimation of PM 2.5 emissions and particularly ambient concentrations resulting from them are subject to considerable uncertainty. While the GAINS model has been validated against ambient PM 2.5 observations globally (Amann et al., 2020), we are not able to J o u r n a l P r e -p r o o f provide ground truthing of estimated PM 2.5 concentrations in Kisumu due to the lack of ambient PM 2.5 monitoring data there.
In addition, as explained in Appendix A (Section ii), the parameters we have used to estimate the average household PM 2.5 concentration due to cooking are necessarily simplifications. Such estimates are obtained using a simplified method outlined in Appendix A (Section iv) and our focus is solely on the potential for biogas in reducing pollutant concentrations. The methodology for evaluating changes in indoor air PM 2.5 concentration can be improved if empirical data on household air pollution for the context of Kisumu becomes available.
Moreover, with regards to capturing the health impacts of our scenarios, we have limited our analysis to the effects of particulate matter, while the risks associated with for instance contamination of Lake Victoria or flooding as a result of drainage systems being blocked by waste or the risks of vectorborne disease from breeding in water deposits in the waste are not considered, and therefore our reported health impact results are likely to be underestimates.
Lastly, concerning our Biogas scenario, while we have assumed the provision of substrate only from household food waste, a potentially promising alternative could involve an industrial symbiosis scenario where MSW is co-digested with waste from breweries operating in Kisumu. Under such scenario, the resulting biogas could be used not only for the required heat in the brewing process but also to produce electricity for the grid. There is an abundance of studies exploring the potential in codigestion of brewery waste, although most studies appear to be in experimental and pilot stages (Tewelde et al., 2012;Murunga et al., 2016;Gunes et al., 2019).

J o u r n a l P r e -p r o o f
Notwithstanding the above limitations, we maintain that, with respect to orders of magnitude and the relative performance of scenarios, our results are still valid and can be useful as a basis for policy planning over the medium term in the area of solid waste management in Kisumu. Findings can also provide informative background for policy planning in similar contexts.
In summary, the analysis presented in this paper demonstrates that a move towards recycling food

J o u r n a l P r e -p r o o f Appendix A -Full Model Documentation
In this appendix, the formulation and parametrisation of the SD model is explained in detail. The model and the simulation runs are available as online supplementary material to this paper. The model is built in Vensim, a widely used SD simulation software package. The whole 120-plusvariable model is presented sector by sector, with visual snapshots to aid understanding. The following table elaborates the colour-coding and other information needed to interpret the diagrams. Grey arrows Initial condition setting.
As outlined earlier in the body of the paper, the model consists of four inter-connected sectors: (1) Waste Collection, (2) Biogas, (3) Landfill, and (4) Scattered Waste. In this section, the four sectors will be described in detail.

J o u r n a l P r e -p r o o f i) Waste Collection Sector
This sector, as depicted in Figure 13, carries out simple accounting operations and involves no dynamic complexity (such as feedback loops, delays or accumulations). It is, however, useful in capturing the waste collection process within the case study in a visual and aggregate way.
Specifically, two different types of waste collection capacity are modelled: Firstly, the currently existing mixed waste collection trucks which transport the waste to landfill without any segregation of waste; and secondly, potential specialised waste collection capacity for separated food waste, whereby a number of waste handcarts would collect food waste from households (including in narrow alleyways of informal settlements where poor access prohibits the use of trucks) and take it to decentralised biogas production facilities, as modelled in the next sector. Total mixed/food waste collection capacity in Figure 13 is the number of trucks/handcarts multiplied by the average capacity of the vehicles. The only currently existing capacity for waste collection consists of five mixed-waste trucks (with an average capacity of about 45 tonnes per day 3 ). Total waste collection capacity, which is the sum of all existing capacity plus any future added capacity, determines the proportion of waste collected, as a key indicator, on the right-hand side of the diagram. Currently, this capacity stands at about 225 tonnes per day, which is about 43% of the total approximately 522 tonnes per day of waste generated. Any remaining waste that is not collected is assumed to be inappropriately disposed of in open pits or scattered on roadsides and elsewhere, as modelled later in the Scattered Waste Sector.
Total GHG emissions due to waste transport, which is accounted for in total GHG emissions from waste as seen later, is also calculated in this sector based on a constant level of GHG emissions per

ii) Biogas Sector
The second sector of the model includes another set of accounting equations for keeping track of the portion of waste that isor rather could be, in the futurerecycled into biogas and fertiliser. The sector is presented in two diagrams: Figure 14 shows the structure where savings in GHG emissions are calculated.
Starting with Figure 14, the cumulative number of biogas facilities is a ‗policy variable', which means that it is a user-determined external input to the model that is based on our scenario assumptions (as described in ). The resulting total biogas capacity, based on an average waste processing capacity per biogas facility, together with the concurrent restriction of total food waste collection capacity (imported from the previous sector), gives total food waste treated for biogas production.
This determines total biogas generated based on a constant food waste to biogas yield factor, assumed equal to 100 m 3 of biogas per tonne of food waste. This is based on the figure provided by Veeken (2005) cited in (Müller, 2007, p. 26 Carvalho et al., 2019 (p. 173). The result is then multiplied by PM 2.5 emissions factor by cooking fuel type (also from Carvalho et al. (2019, p. 172) to give the total PM 2.5 emissions due to cooking. This total is added to emissions related to waste burning in the Scattered Waste Sector to obtain ambient PM 2.5 concentration from cooking and waste burning.
The same piece of structure also gives a rough estimate for the average household PM 2.5 concentration due to cooking. This estimate is based on empirical measurements reported in Muindi et al. (2016, p. 7    All remaining waste which is collected but not treated for biogas production is sent to the landfill as much as the current waste collection capacity allows. The part which is not collected due to a lack of capacity is assumed to be disposed of inappropriately and scattered or dumped anywhere other than the main landfill, as modelled in the next sector. A fraction of the waste dumped in landfill or elsewhere is burnt to reduce volume (Klimont et al., 2017, p. 8700). Waste burning emits CO 2 into the atmosphere, while the decomposition of organic waste emits CH 4 , both of which are greenhouse gases. The dumping of the waste in landfill and the resulting GHG emissions are captured via the structure introduced in this sector. Figure 15 shows the structure of this sector, which is a key part of the model as it captures the dynamics of the accumulation of waste as well as the potentially changing composition of the waste in landfill. Landfill waste is disaggregated into the two stocks of landfill food and nonfood waste, as the two types have different profiles in terms of GHG emission potentials. The two stocks are similarly configured in terms of inflows and outflows.
The stocks are initialised according to estimates of total amount of waste currently existing in the city's main landfill, which is estimated by dumpsite management at around 140,000 tonnes. This amount is split between the stocks of food and non-food landfill waste initially based on Aguko et al. (2018, p. 6), who estimate empirically that 51.8% of Kachok's waste content is organic.
Total mixed waste to landfill is the lesser value between total mixed waste collection capacity and the total waste that remains after recycling some of it for biogas (i.e. total waste generated minus total food waste treated for biogas production). These variables are J o u r n a l P r e -p r o o f imported from the Waste Collection Sector. A useful indicator calculated here is the percentage of food waste recycled for biogas production, i.e. total food waste treated for biogas production divided by food content of waste generated.
The split of the total mixed waste to landfill going into each of the two food/non-food stocks depends on a dynamic food content ratio of waste to be disposed. This ratio varies depending on how much of the food waste generated by households is used for biogas production. In order to calculate the food content ratio of waste to be disposed we first take out total food waste being treated for biogas production from food content of waste generated (which is 63% of total waste generated in the Kisumu context), obtaining the food waste left after biogas production. Dividing this by the sum of the same plus non-food content of waste generated (37% of total waste generated) gives the dynamic food content ratio of waste to be disposed. As we will see later in our scenarios, this ratio will naturally go down as we start to recycle a part of the food waste into biogas. Multiplying this ratio by total mixed waste to landfill gives total food waste to landfill, with the rest flowing into the landfill non-food waste stock.
As for the outflows, each stock has an outflow of waste burning. The fraction of waste burnt every year is assumed equal to 23% for all three stocks based on Onyango & Kibwage (2008) cited in (Gutberlet et al., 2017, p. 113 year, for slow-degrading waste such as paper, which constitutes one third of non-food waste in Kisumu. Therefore, non-food waste decomposition fraction is assumed equal to one third of 5% or 1.67% per year. In the case of landfill non-food waste, an additional outflow of waste is captured which represents the landfill non-food waste informally recycled by scavengers who contribute towards recycling plastics, bottles, cans and metallic objects (Awuor et al., 2019). Based on our consultation with Kisumu county's waste officials, there are currently around 80 such informal workers, each collecting on average about 25 kg of waste every day. In the future, the number of informal waste-pickers is assumed to grow according to the projected population growth rate. Since the composition of the waste will change under our scenarios, we assume that the capacity of each waste-picker is a function of the non-food content ratio of the waste in landfill. We assume that this capacity stays at its current value of an estimated 25 kg/day per person under the current non-food content ratio. However, if the non-food content ratio goes down to zero or up to 100%, in conjunction with that, it is assumed that the waste-pickers' average capacity goes respectively down to zero or up to twice the current capacity (linearly, in both cases).
Next, we are going to calculate waste-related GHG emissions for landfill waste based on the outflows of waste burning and decomposition. This is done by multiplying the amount of food/non-food waste that is burnt or decomposed each year by the respective emission factors for CO 2 , CH 4 , and BC. The structure for making these calculations is shown in Figure 15.  The other important contributor to global warming which results from the incomplete combustion of waste is black carbon (BC). BC is a carbonaceous aerosol (Klimont et al., 2017) with a global warming potential 5 (GWP) of 460 based on a 100-year time horizon (IPCC, 2007). The emission factor used here to estimate BC emissions is 0.65 kg BC per tonne of waste burnt (Akagi et al., 2011, p. 4047  The final sector captures the accumulation of waste anywhere other than in landfill, such as on the roadside or in open pits, along with the resulting GHG and PM 2.5 emissions. This sector ( Figure 16) is constructed very similarly to the previous one, in the sense that scattered waste is conceptually divided between the two stocks of scattered food and non-food waste, with similarly configured inflows and outflows.
What remains of total waste generated after subtracting total food waste treated for biogas production (Biogas Sector) and total mixed waste to landfill (Landfill Sector) constitutes total waste inappropriately disposed of, which finds its way into one of the two stocks, depending on the dynamic food content ratio of waste to be disposed as calculated in the Landfill Sector.
The stock of scattered waste is initialised in relation to the stocks of landfill waste. Based on an initial total waste generated of around 522 tonnes per day in 2021 and an initial total mixed waste collection capacity of around 225 tonnes per day, it is estimated that initially around 225/522=43% of the waste is being collected and the remaining 57% is inappropriately disposed of in places other than the landfill. Therefore, it is considered a fair assumption that the ratio of initial scattered waste to initial landfill waste should also be close to 57/43. Given the rough estimate of 140,000 tonnes for initial landfill waste, we reach an estimate of 185,000 tonnes for initial scattered waste. As before, each stock has two outflows of decomposition and burning with the same fractions previously used for landfill waste.
Resulting GHG emissions are calculated in the same way and with emissions factors equal to those of the respective types of landfill waste, except for CH 4 emissions which are calculated using different emission factors. This is because in the case of scattered waste there is J o u r n a l P r e -p r o o f assumed to be a lower level of compacting and therefore weaker anaerobic conditions resulting in lower CH 4 emission factors for scattered waste as compared to landfill waste.
CH 4 emission factors for scattered food and non-food waste are assumed to be 10.13 and 8.85 kg CH 4 per tonne of waste decomposed respectively (see Section 2.3). Total scattered waste emissions is calculated by summing up direct CO 2 emissions due to waste burning, CO 2 equivalent black carbon emissions due to scattered waste burning, and CO 2 equivalent CH 4 emissions due to scattered waste decomposition.
Additionally, in this sector total PM 2.5 emissions due to waste burning is calculated by summing up the burning rates of different types of waste (food/non-food; landfill/scattered) and multiplying by waste burning PM 2.5 emission factor. This factor is set to 8.74 kg PM 2.5 per tonne of waste burnt based on (Klimont et al., 2017, p. 8700) for all different types of waste as an approximation. Total PM 2.5 emissions due to cooking is imported from the Biogas Sector.
Each and then take a population-weighted average across all grid cells in the city to derive an integrated coefficient from Kisumu to itself. Thereby, we take two simplifying assumptions: 1) that only primary PM emissions play a role and local secondary particle formation can be neglected, and 2) that the contribution from such sources in other Kenyan cities to ambient PM in Kisumu is low. Assumption 1 may lead to a small underestimation of the coefficient, while assumption 2 may lead to a small overestimation. Given the large uncertainties in emissions, these simplifications seem justified. J o u r n a l P r e -p r o o f Food waste to biogas yield factor 100 m 3 per tonne Veeken (2005) cited in (Müller, 2007, p. 26,