Hungry cities: how local food self-sufficiency relates to climate change, diets, and urbanisation

Urban areas are defined by connected built-up urban areas derived from land-cover information. For this purpose, orthodromic spatial clustering (Kriewald et al 2016) is applied, a refined approach based on the city-clustering algorithm (CCA) approach from Rozenfeld et al (2008). The CCA approach involves a clustering parameter l determining up to which distance urban cells (built-up area) are connected with neighboring cells, i.e. urban cells within that distance are assigned to the same city cluster. After the identification of urban clusters (UC), they are joined with point-based population information. By using ESA CCI Land Cover data (v1.6.1) (ESA Land Cover CCI project team 2016) and GRUMP settlement points (CIESIN/IFPRI/CIAT 2011, United Nations 2014) the method presents an automated, fast, and systematic way to identify UC globally. The area ${A}_{i}^{\mathrm{UC}}$ of an identified UC i is subsequently surrounded by a so-called PU region, whose area ${A}_{i}^{\mathrm{PU}}$ is defined as

$$\begin{eqnarray}&&{A}_{i}^{\mathrm{PU}}=\displaystyle \frac{{\rho }_{i}}{{\rho }_{\mathrm{ref}}}\cdot {A}_{i}^{\mathrm{UC}},\end{eqnarray} \tag{ 1 }$$

with ρ_i as population density of UC i. The population density ρ_i is divided by a fixed reference population ρ_ref = 1000 residents km⁻². The latter scales the PU land demand per resident to 0.1 ha. This amount is motivated by the FAO (1993) which assumes 0.07 ha of arable land per capita as an absolute minimum requirement needed for individual food self-sufficiency (Myers 1999). For the estimation of the actual PU agricultural potential, we kept the shares of current and future land use pattern constant in order to avoid an additional transformation of natural areas into farmland. Nevertheless, the average size of PU areas will increase from 225 km² in 2010 to 378 km² in 2050. The averaged width of the PU area will therefore increase from 5.6 km in 2010 to 8.0 km in 2050.

Considering only cities with >100 000 residents leads to an identification of 4121 city clusters in 164 countries, which were included in the analysis. These city clusters cover 2.536bn urban residents or 71% of the total urban population in 2010. For details refer to supplementary information, sections 2 and 3 (available online at stacks.iop.org/ERL/14/094007/mmedia).

To quantify the potential yield in any PU_i the output of the GAEZ model is employed (IIASA/FAO 2012). The GAEZ model combines geo-referenced global climate, soil and terrain data with matching procedures to identify crop-specific limitations. With the help of additional simple crop models, it provides the maximum potential crop yields for different agricultural production systems defined by irrigation type. For a more detailed explanation of the GAEZ model (see Fischer et al 2012).

For our analysis, the actually arable land for each PU is identified as the first step. For eight FAO food categories (and the items therein, e.g. for cereals: wheat, rice, maize, etc), namely animal products (considering also embodied calories based on Pradhan et al 2013), cereals, fruits, oil crops, pulses, roots and tubers, sugar crops, and vegetables, the current and the potential maximum yield is calculated by a linear programming optimization method for the years 2010 and 2050. Based on these results the amount of the potential calories produced per m² has been calculated. As nourishment styles locally are quite different, national diets were also considered.

For the future development of cities, national urban growth scenarios are considered for each city in the respective countries (United Nations 2014). While the city size is extended (urban sprawl effect) assuming constant density as derived from the penultimate section, the PU area is calculated as above, while keeping the land-use pattern constant as for the year 2010. The future calorie demand per capita is estimated on the basis of Hic et al (2016), while the diets itself are considered as constant. For the climate impact on potential yields for any representative concentration pathway (RCP) scenario the ensemble mean of the GAEZ model output for five climate models is used.

In order to estimate CO₂ emissions from urban food transport a two-step approach was applied. First, the PU area PU_i around any city cluster UC_i is calculated. By doing so urban fringes are successively increased by 10 km steps and subsequently the potential agricultural production is calculated based on the actual land use (see supplementary information). This is repeated until the urban hinterland can meet the food demand of the city (minimum daily calories requirements times urban residents and year). Second, the food distance is calculated as the product of the great-circle distance and the food supply from each grid cell in the buffer to the corresponding UC_i. Finally, the CO₂ emissions for such an optimized transport are estimated by applying emission factors for inland (120 g CO₂/tkm) and maritime (13.5 g CO₂/tkm) food transport (Cristea et al 2013) and compared with numbers from the literature referring to the status quo. Air transport is not included in this study, as transport data are not reliable, nevertheless, the emission factor for air transport is the highest (approx. 680 g CO₂/tkm). By using the great circle distance (shortest possible distance) between a production site and an urban cell direct routes over land and via sea transport were considered. For details refer to SI, section 5.

To quantitatively estimate local food production the non-overlapping potential of PU agriculture for urban centers is calculated. For this purpose, the shares of actual land use pattern have been used and kept constant for the future. The CCA approach (see section 2.1 for details) makes it possible to use remote sensing and land use data in order to investigate PU agriculture and its consequences. As the CCA defines cities from a functional point of view, closely neighbored cities or suburban regions are considered as one. Examples are twin cities like Cairo/Giza or Tokyo/Yokohama, which are considered as one UC. By globally applying this approach 4121 urban entities, each with more than 100 000 residents have been identified covering approximately 71% of urban residents in 2010. Analyzing the PU areas in detail, they are composed of 63% farmland, 31% natural areas and 6% of non-arable areas. For the purpose of the paper, the area for PU agriculture is estimated on the minimum of arable land needed per capita to ensure food self-sufficiency (see section 2.1). This implies that under optimal conditions (soil quality, climate, diets) cities could nourish their inhabitants. Considering the actual land use, we estimated that PU areas globally cover an area of 2.15 million km² comprising approximately 8.5% of the global arable land available in 2010. As shown by table 1 with this amount of land 30.6% of all urban residents can be nourished considering current food production, diets, and land use in 2010 (figure 2, upper panel). Applying more advanced agricultural practices to close yield gaps would improve the situation further, as in this case 35.3% of the global urban population could be nourished self-sustaining with the help of PU agriculture (table 1, figure 2, middle panel). However, on the global scale the situation for single cities is quite diverse showing that PU agriculture is not a 'silver bullet' solution for all cities. However, the difference between figure 2 upper panel and lower panel indicates that an optimized and local production may solve nourishment problems for many cities. As cities in Eastern, Western Africa and in Central India are currently not self-sufficient in terms of local food production a strong improvement could be achieved by closing yield gaps. The reason why North-American cities, for example, are characterized by a very low potential (as an average only 20% of residents can be sustained) is rooted in the fact that cities are mostly dominated by natural vegetation in PU areas (see, figure 5, SI). Although PU agriculture could be extensified around North-American cities, it is in general not desirable to convert all natural areas into farmlands, as this would mobilize even more greenhouse gas emissions from land conversion and the subsequent agricultural production. Moreover, such a strategy would reduce biodiversity as well. The opposite situation can be observed for Indian cities for which PU regions provide up to 90% of the actual food demand, while cities on the Arabian peninsula and on the Western coast of South America have a limited potential due to its desert environment. The remaining challenging questions, namely how the impact of climate change on yields, diet change and urban growth would change the potential for PU agriculture is displayed as a composite effect in figure 2, lower panel. In this case, nearly all cities will face a decreasing nourishment potential, with a few exceptions in higher latitudes or mountain regions. The most prominent changes are visible for the Indo-Gangetic plain, along the river Nile, South-Eastern China, and for the whole African continent. In other words, if PU agriculture is suggested as a solution for the future, one has to take into account urban growth, climate change and changing lifestyles (e.g. diet styles, food waste) as otherwise, the dynamics could foil positive effects. In order to discuss the most prominent factors for 19 world regions, we would like to refer to figure 3(b). For 16 out of 19 regions urban growth will have the largest impact on future PU agriculture. The strongest impact can be observed in Western, Middle, and Eastern Africa. Only in Southern Europe and Northern Africa is climate change the most constraining factor for the PU food production potential, while dietary pattern change and the subsequent demand change is most important in Eastern Europe. For 9 regions diet changes have the second largest impact on relative urban food supply, namely in Australia, Eastern, Southern, South-Eastern Asia, Northern Europe, Northern, Central and South America. Also, in relative terms, climate change is not the dominant factor for PU agriculture by 2050. This holds for the RCP_2.6 and RCP_8.5 scenarios. However, an exception is Southern Europe and Northern and Western Africa, where the optimal situation will be shrinking considerably.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

So far only a limited amount of urban hinterland was used to estimate the PU food production potential. In order to estimate the effect of food transport on fossil fuel use, one has to figure out how much land is needed to nourish all citizens of an urban agglomeration. For this purpose, the so-called urban foodshed concept has been defined. Again the land use share is kept constant, but the urban hinterland is stepwise increased until it can produce sufficient food for all urban residents. Figure 4, left panel, shows urban foodsheds and the associated transport distance of food into the identified UCs. A major result is that most foodsheds are larger than 5000 km² (figure 4, top-right panel) which relates to a circle with a diameter of approximately 80 km. Consequently, at least inter-local food transport and trade is unavoidable and in many cases, even global food transport trade would be needed to nourish cities (e.g. D'Odorico et al 2014). Nevertheless, global food transport is an issue which needs particular concern. It has been shown that global food transport is increasing considerably in recent decades (Smith et al 2005) and therefore it is still worthwhile to estimate sector effects as climate change mitigation options have to start in sectors.

Zoom In Zoom Out Reset image size

Weber and Matthews (2008) estimated that 11% of total agricultural emissions could be related to food transport, while others argue that emissions from food transport are neglectable, because of food production via low-intensity farming methods and shipping it around the globe is always better than producing food locally using high-intensity agriculture (Avetisyan et al 2014, Dalin and Rodriguez-Iturbe 2016). Indeed emissions factors are much lower for maritime shipping (see SI and Fitzgerald et al 2011). Although global accounting of emission factors in food transport is somewhat uncertain, the foodshed approach allows us to estimate the mitigation potential in the food transport sector assuming that local food production could be a sustainability strategy for cities (figure 4, left bottom panel). Moreover, in this study we applied the lowest values (see SI) in order not to overestimate effects and excluded air transport, although the latter is increasing over-proportional due to the transport of more luxury food.

In this paper, we also assume an optimized transport scenario which applies the shortest distances from food production sites into the urban foodsheds. On the basis of the applied approach, our findings estimate a global emission of around 0.15 Gt CO₂/yr for a distance optimized food transport scenario. Taking into account that the total emissions from agriculture (including production, storage, packaging) amounts up to 19%–29% of total emissions and that additional 5%–10% accounts for transport and other postprocessing activities on food (e.g. Vermeulen et al 2012), approximately 14–18 Gt CO_2eq. emissions are assumed to be released from the agricultural sector in 2007. Considering the Weber and Matthews (2008) estimate that food transport is actually responsible for approx. 1.5–2.0 Gt of CO_2eq. emissions per year our optimized estimation shows the emissions saving potential. Namely, that on a global scale an optimized food transport can reduce transport emissions by approximately a factor of 10. Although an approximate saving of roughly 1 Gt CO_2eq. seems to be small one can argue that climate change, population growth and dietary pattern may further increase emissions in the future, as e.g. food demand is increasing and urban centers are unlikely to optimize PU food production. Therefore, if future food transport will rely on fossil fuel driven engines, our approach highlights that agricultural production in the near vicinity of consumers could help to reduce emissions considerably.