WPS6188 Policy Research Working Paper 6188 Climate Change, Agriculture and Food Security in Tanzania Channing Arndt William Farmer Kenneth Strzepek James Thurlow The World Bank Development Research Group Agriculture and Rural Development Team September 2012 Policy Research Working Paper 6188 Abstract The consequences of climate change for agriculture economy-wide model of Tanzania. The authors find that, and food security in developing countries are of serious relative to a no-climate-change baseline and considering concern. Due to their reliance on rain-fed agriculture, domestic agricultural production as the principal channel both as a source of income and consumption, many of impact, food security in Tanzania appears likely to low-income countries are considered to be the most deteriorate as a consequence of climate change. The vulnerable to climate change. This paper estimates analysis points to a high degree of diversity of outcomes the impact of climate change on food security in (including some favorable outcomes) across climate Tanzania. Representative climate projections are used scenarios, sectors, and regions. Noteworthy differences in in calibrated crop models to predict crop yield changes impacts across households are also present both by region for 110 districts in the country. The results are in turn and by income category. imposed on a highly-disaggregated, recursive dynamic This paper is a product of the Agriculture and Rural Development Team, Development Research Group. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://econ.worldbank.org. The authors may be contacted at channing.arndt@econ.ku.dk , whfarmer@gmail.com, strzepek@mit.edu, and thurlow@wider.unu.edu. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Climate Change, Agriculture and Food Security in Tanzania Channing Arndt, William Farmer, Kenneth Strzepek and James Thurlow1 Keywords: climate change; agriculture; food security; crop model; CGE model; Tanzania. JEL Codes: O440, Q180, Q540 Acknowledgements: Thanks to Will Martin for leadership in the overall effort and valuable comments. Seminar participants in Dar Es Salaam and the World Bank also provided useful comments. All errors and omissions are the responsibility of the authors. The research on which this study is based was supported by the Trust Fund for Environmentally and Socially Sustainable Development (TFESSD). 1 Arndt: University of Copenhagen (channing.arndt@econ.ku.dk); Farmer: Tufts University (whfarmer@gmail.com); Strzepek: Massachusetts Institute of Technology (strzepek@mit.edu); Thurlow: United Nations University - World Institute for Development Economics Research (thurlow@wider.unu.edu). 1. Introduction The consequences of climate change for agriculture and food security are of serious concern, not least because food supplies are already inadequate and poverty is severe in many low-income countries, particularly in Africa. Moreover, developing countries are generally considered to be most vulnerable to climate change, mainly due to their reliance on rain-fed agriculture. Previous studies linking climate change to food security have typically used agricultural crop models (see, for example, Parry et al., 2004). Their predictions range from precipitous declines in yields for major African food crops (Schlenker and Lobell, 2010) to more modest reductions (Lobell et al., 2008; Nelson et al., 2010), and even to improvements (Butt et al., 2005). Previous studies have, however, suffered from at least one of four limitations. First, they often provide global or regional assessments. Yet climate change is expected to vary widely within continents and even countries, and so adaptation policies require higher-resolution information, possibly even at sub-national levels (Lobell et al., 2008). Secondly, despite considerable uncertainty surrounding future climate change, some studies rely on only a few climate projections (see, for example, Butt et al., 2005). Thirdly, many calibrated agronomic crop models exclude “autonomous adaptation� that may offset at least some climate change damages. Finally, previous studies typically measure direct or partial equilibrium production changes, but may exclude indirect and general equilibrium effects, including price and household income changes and inter-sectoral linkages. Since food security depends on both food availability and accessibility, it is inadequate to measure production changes without considering, for example, the impacts of climate change on households’ incomes (Parry et al., 2004; Ahmed et al., 2009). In this paper we estimate the impact of climate change on agricultural production in Tanzania using detailed sub-national crop models. Four projections are drawn from available general circulation models (GCM) to reflect a range of possible temperature and precipitation changes by mid-century. These climate projections are then used in calibrated crop models to predict crop yield changes, which are in turn imposed on a highly-disaggregated, recursive dynamic economy-wide model of Tanzania. This model captures indirect effects and permits (some) autonomous adaptation. The economic model allows us to evaluate the availability (production) and accessibility (income) dimensions of food security. In the next section, we describe the selected climate scenarios and the crop modeling framework used to translate climate conditions into crop yields. We then describe the economy-wide model and present the results from our simulated baseline and climate change scenarios. We conclude by summarizing our results and identifying areas for further research. 2 2. Climate Change and Agricultural Crop Yields Selecting climate change scenarios General Circulation Models (GCM) produce a wide range of future climate change scenarios, especially when examined at the country-level (see Solomon et al. 2007). Apart from differences in the science of modeling global climate systems, there is also uncertainty about other key variables such as how the global economy will evolve in coming decades. To account for this, GCMs typically employ different “emission scenarios� based on assumptions about future populations, technological advances, and global agreements to reduce carbon emissions. To capture a range of possible climate change realizations, we select four projections with different temperature and precipitation outcomes averaged over all land areas of the country. The scenarios employed are presented in Table 1. In addition to temperature and precipitation deviations, the table presents the Climate Moisture Index (CMI) (Willmott and Feddema, 1992), which is an indicator of a region’s aridity, at the national level. The CMI depends on average annual precipitation (P) and potential evapotranspiration (PET). A climate is classified as semi-arid (semi-humid) and then arid (humid) as PET increases (decreases) relative to precipitation. The CMI is defined as: A CMI of -1 is very arid whereas a CMI of +1 is very humid. [Insert Table 1] The scenarios are labeled WET, DRY, COOL, and HOT. The COOL scenario is relative to the other scenarios, and projects a mean average temperature increase of 1.1 degrees Celsius by 2041-50 compared to 1.9 degrees Celsius in the HOT scenario. Three of the four scenarios project an increase in precipitation reflecting an analysis of climate futures for Tanzania conducted by the Tyndall Center (2010). While precipitation rises in three scenarios, the CMI remains fairly constant in two of the three because rising temperatures increase PET. It is of some interest to note that the WET scenario for Tanzania is the same scenario identified by the World Bank as the driest scenario globally out of all 56 possible scenarios considered for analysis (World Bank, 2010). This is a reminder that local or national conditions can differ drastically from broader global averages. Overall, the GCMs suggest that Tanzania’s climate will become warmer, although the extent of warming varies by GCM. In addition, the change in average precipitation may be positive or negative. In these senses, climate in Tanzania becomes more uncertain as a result of climate change. 3 To develop a baseline “no climate change� scenario, we use historical daily climate data for 1997-2006 (i.e., mean, minimum and maximum surface temperatures and precipitation) retrieved from the NASA POWER database (Stackhouse, 2010).2 A random 50-year baseline climate sequence was generated using bootstrapping techniques based on the historical annual data. In other words, one year of climate is drawn from the historical data fifty times. This technique preserves intra-annual correlations (and higher moments) but not inter-annual correlations. For the purposes of the crop modeling, the distinct wet and dry seasons in Tanzania limit the impact of inter-annual variation in soil moisture at the start of the growing seasons rendering this approach an acceptable compromise in the face of existing data limitations. Our baseline scenario therefore assumes that future weather patterns will retain the characteristics of recent historical climate variability. The principal purpose of the baseline scenario is to provide a counterfactual for the climate change scenarios. For the purposes of projecting future climate in a given region (Tanzania), it is useful to note that the GCMs are calibrated to reasonably reproduce historical climate on a global basis and predict future climate under alternative levels of global greenhouse gases. The historical predictions of a given GCM for temperature and precipitation for a specific region of the globe, such as Tanzania, will not perfectly match the historical data. In other words, output from a given GCM for historical periods may, for example, consistently under-predict temperature and over-predict precipitation. This is also true with respect to higher order moments and time series properties of the climate series. In sum, while the GCMs provide information with respect to long run trends in average temperatures and potential trends in average precipitation, it is less clear that the outputs of the GCMs provide useful information with respect to daily variation in climate (Schlosser, 2011). As a result, it is inappropriate to directly take the raw output from GCMs and use these in, for example, crop models for Tanzania estimated on the basis of historical data. Some manipulation of the GCM output is required. Here, we combine the long run temperature and precipitation changes predicted by each GCM with historical data on daily climate variation. Specifically, we calculate a ten-year moving average of the percentage monthly changes in temperature and precipitation between the selected climate change scenario as predicted by the GCM and a baseline no climate change scenario from the same GCM. So, the percentage change in temperature for January 2030 is the average percentage change in temperature recorded by the GCM for the periods January 2026 to January 2035. We then overlay these changes on the baseline historical scenario. For example, when the moving average from a GCM predicts a five percent increase in January temperature and a three percent decrease in precipitation for a given year, then the January baseline temperatures and precipitations (drawn from the historical record) are augmented by those percentages. 2 A longer historical data series would be desirable. 4 This procedure produces the four “synthetic� climate projections mentioned above (i.e., WET, DRY, COOL, and HOT). This method retains many aspects of the time series properties of historical climate variability (i.e., the historical baseline, which remains constant across all scenarios) and overlays future climate changes. It is important to point out that this procedure has implications for variability. For months with predicted increases in temperature and/or precipitation, peaks will be accentuated. Predicted temperature almost never declines; however, some months experience drying trends in nearly all GCMs and one GCM predicts an overall drying trend (see Table 1). When the GCM predicts drying, historical dry periods are deepened and tend to be extended. Based on these climate scenarios, we next assess the implications of climate change for crop production. Crop and water balance models We use a generic crop model called CLICROP to simulate the impact of the baseline and climate change scenarios on rain-fed and irrigated crop yields and on irrigation water demand (Strzepek et al., forthcoming). CLICROP was specifically designed to capture climate change impacts since it models water stress from both insufficient and excess water supply (measured daily). Yield reduction due to retarded root growth resulting from excess water is known as “water- logging�. Water-logging reduces yields via oxygen loss and root growth hindrance (see Sieben, 1964). The inclusion of water-logging and crop-specific parameters is an extension over simpler models, such as the FAO’s CROPWAT. Moreover, CLICROP’s daily time scale allows it to capture the shorter but higher intensity rainfall expected in Eastern Africa (Solomon et al., 2007). The effects of the atmosphere (i.e., temperature and precipitation) are modeled indirectly in CLICROP via evapotranspiration (Allen et al., 1998) and infiltration to the soil layers (based on soil properties). Soil composition is considered at each site and is used to calculate soil moisture in each soil layer, including the moisture allowed to percolate into deep soil layers. Water balances and the upward flow of soil water are then measured. As rain falls on the given soil, certain amounts are allowed to run off, infiltrate and percolate through deep layers (in addition to the demands of evapotranspiration). Crops are then allowed to draw what water that they can from the soil layers. Crop yields are estimated using the approach proposed by Allen et al. (1998) with the additional possibility that yields are reduced when excess water results in submersion. For this examination, the effects of CO2 fertilization are not considered in our analysis. As a result, we may overestimate yield losses caused by climate change. Recent free-air carbon enrichment (FACE) studies contradict the results of earlier closed-laboratory experiments that suggested the presence of strong positive productivity effects for major crops due to higher CO 2 concentration levels (Long et al, 2006). At the same time, however, the validity of FACE results has been questioned (Tubiello et al, 2007). The debate remains unresolved at present. 5 CLICROP was run at a 1°×1° resolution (i.e., 111 square kilometer grids in a country measuring 945,000 square kilometers). Separate models were developed for the 9 major crops of Tanzania (i.e., cassava, groundnuts, maize, millet, potatoes, sorghum, soybeans, sweet potatoes and wheat). Predicted yields for each sub-national region are calculated as the sum of overlaid gridded results weighted by geographic area. CLICROP was calibrated to information on soil parameters from the FAO Soils Database (e.g., field capacity, wilting point and saturated hydraulic conductivity) (FAO-UNESCO, 2005). Regional information on crop growing seasons and planting dates were provided by Sacks et al. (forthcoming). Finally, crop locations were based on You et al. (2006) for the year 2000, and crop parameters were drawn from Allen et al. (1998) and Doorenbos and Kassam (1979). Overall, the process is detailed with more than 13 million potential simulation sets.3 Crop modeling results Although CLICROP analysis was conducted for nine crops, we focus on the results for maize for purposes of exposition. Maize is the principal food crop in Tanzania, representing 35 and 45 percent of calories consumed by poor urban and rural households, respectively (Pauw and Thurlow, 2010). Tanzanian farmers allocate about one-third of their crop land to growing maize, mainly without the use of irrigation. Deviations in dry-land maize yields therefore provide a first-cut indicator of food availability in Tanzania (Thornton et al. 2009). In addition, the implications of climate outcomes for maize are likely to be similar for sorghum and millet, making the coverage of caloric sources and land use, particularly for more vulnerable populations, even broader. Table 2 summarizes the deviations in mean maize yields from a “no climate change� baseline scenario for the 10-year period 2041-2050. The administrative regions are grouped into similar agro-climatic zones based on Fan et al. (2010). There is a high degree of variation both across the four climate scenarios considered and across sub-national regions. For example, maize yields in the Northern Zone are projected to increase substantially in the WET scenario, but decrease by similar amounts in the HOT and DRY scenarios. Varied impacts also occur across regions but within the same scenario. For example, average maize yields in the WET scenario are projected to increase by 15 percent in Manyara in the Northern Zone, but decline by 12 percent in Tabora in the Central Zone. Nevertheless, a few regularities emerge. Maize yields are generally more favorable under the COOL and WET scenarios than under the HOT and DRY scenarios. In addition, yield declines are much more prevalent across regions and scenarios than are yield increases. In particular, under the HOT and DRY scenarios, yields rise in only a few regions and these increases are in all instances small. Finally, the coastal islands remain virtually unaffected in all climate scenarios. 3 The product of approximately 8500 grids times nine crops times 43 crop years times four climate scenarios. 6 [Insert Table 2] Figures 1 and 2 provide insights into the geographic correlation of results. They show maize yields for each scenario for 110 districts in Tanzania. The darker lines show the boundaries of the larger administrative regions presented in Table 2. Once again, there is substantial regional variation in our results. For example, the COOL scenario shows significant yield increases in the Northern Zone (i.e., in the districts surrounding Mount Kilimanjaro), while yields decline slightly in the Southern Coast and Southern Highlands. By contrast, yields are damaged though-out the country in the HOT scenario, with particularly strong negative impacts in the Northern and Lake Victoria regions. The WET and DRY scenarios contrast similarly. The WET scenario shows mean yields increasing around Kilimanjaro and its southern slope while they decrease dramatically in the western regions around Lake Tanganyika. [Insert Figures 1 and 2] In summary, there are strongly heterogeneous impacts across the four climate scenarios. However, as expected, there is some regional correlation in results. Climate outcomes favorable (unfavorable) to maize farmers in a particular region are also likely to favor (harm) maize farmers in neighboring regions. At the same time, geographical impacts can vary dramatically across scenarios, with some scenarios producing favorable outcomes while others resulting in pronounced negative impacts. Obviously, from a national food availability perspective, the impacts on yields in the major producing regions are more important. Maize is not equally important in all parts of the country and for all household groups (e.g., poor/non-poor and farm/non-farm). Since our objective is to evaluate the economic implications of climate change for agriculture as a whole and for broadly-defined food security, we employ, in the next section, an economy-wide model of Tanzania in order to evaluate these impacts. 3. Economy-wide Impacts and Food Security Economy-wide model The crop modeling results discussed in the previous section are passed down to a recursive dynamic computable general equilibrium (DCGE) model of mainland Tanzania, which estimates the economic impact of the baseline and climate change scenarios, including indirect or economy-wide linkages between the agricultural and nonagricultural sectors. Our model belongs to the structural-neoclassical class of CGE models (see Dervis et al., 1982). DCGE models are well-suited to analyzing climate change. First, they simulate the functioning of a market economy, including markets for labor, capital and commodities, and therefore can evaluate how changing economic conditions are mediated via prices and markets. Secondly, DCGE models ensure that all economy-wide constraints are respected, which is crucial for long-run climate change projections. Finally, CGE models contain detailed sector breakdowns and provide a 7 “simulation laboratory� for quantitatively examining how the individual impact channels of climate change influence the performance and structure of the whole economy (see Lofgren et al., 2004 for a detailed exposition of the base modeling framework adapted for this analysis). Economic decision-making in the DCGE model is the outcome of decentralized optimization by producers and consumers within a coherent economy-wide framework. A variety of substitution mechanisms occur in response to variations in relative prices, including substitution between factors, between imports and domestic goods, and between exports and domestic sales. Production and trade function elasticities were drawn from Dimaranan (2006). The Tanzania model contains 28 activities or sectors, including 12 agricultural subsectors (see Pauw and Thurlow, 2010). Six factors of production are identified: three types of labor (unskilled, semi-skilled and skilled), agricultural land, livestock, and capital. Agricultural activities and land are distributed across the 20 administrative regions of mainland Tanzania. This sectoral and regional detail captures Tanzania’s economic structure and influences model results. Table A2 in the appendix outlines the disaggregation and regional characteristics of the model while Figure A1 provides a map of agro-climatic zones. Climate change affects agricultural production, economic growth and household incomes in the DCGE model via predicted annual yield deviations for rain-fed crops estimated by CLICROP. The DCGE then determines how much resources should be devoted to each crop given their profitability relative to other activities. This reallocation of resources permits some autonomous adaptation by farmers and nonagricultural producers. For example, representative farmers in each region within the DCGE model allocate their land and capital between crops based on long-run rates of technical change and climate change. However, farmers are unable to anticipate weather conditions for a particular season, and so once planted, land cannot be reallocated even if weather patterns are not as expected. The representative producer in our model therefore corresponds to a “typical farmer� (see Füssel and Klein, 2006), who does not assume that historical weather patterns will persist indefinitely, but also does not have perfect foresight of future climate change. Rather they adapt their behavior based on the gradual realization of climate change. The long timeframe over which climate change will unfold implies that dynamic processes are important (Arndt et al., forthcoming). The recursive dynamic specification of our CGE model allows it to capture annual changes in the rate of physical and human capital accumulation and technical change. So, for example, if climate change reduces agricultural production in a given year, it also reduces income and hence savings. This reduction in savings displaces investment and lowers production potential and economic growth. Given our long-run focus, our macroeconomic “closure� assumes that changes in aggregate absorption are proportionally distributed across nominal private and public consumption and investment via distribution neutral changes in savings rates. Government savings are flexible, tax rates are fixed, and the real exchange rate adjusts to maintain an exogenously determined current account 8 balance. In summary, our DCGE model is well suited to capture path dependent effects within a consistent macroeconomic framework. Baseline scenario In order to estimate the economic impact of climate change for Tanzania, we first specify a baseline scenario that reflects development trends, policies and priorities in the absence of climate change. The baseline provides a reasonable trajectory for growth and structural change of the economy from 2007 to 2050 that can be used as a basis for comparison. Economic growth in the DCGE model is determined by rates of factor accumulation and technical change. For population and labor supply, we assume that Tanzania’s population will continue to grow but at a decelerating rate (i.e., 2.0 percent today falling to 0.3 percent by 2050). We assume that the expansion of cultivated crop land will slow such that growth in agricultural production becomes increasingly dependent on the adoption of improved technologies rather than land expansion. As described earlier, the crop models use historical climate data to define year-on-year yield fluctuations in the baseline for each crop and region. Exogenous long-term agricultural productivity growth is set at 0.8 percent per year in agriculture and 1.2 percent in non-agriculture. Improvements in the education levels of Tanzania’s workforce are assumed to continue, with supply and productivity rising faster for skilled and semi-skilled workers than for unskilled workers (i.e., at 2.0 and 1.5 percent per year, respectively, compared to 0.5 percent). Under the above assumptions, Tanzania’s economy gradually develops, with agriculture’s contribution to gross domestic product (GDP) falling from 27.8 to 14.1 percent during 2007- 2050. Overall, per capita GDP grows at an average 2.2 percent per year in the baseline, leading to significant improvements in average household welfare. Economy-wide modeling results The DCGE model uses the crop yield results from Section 2 to estimate the economy-wide impacts of climate change. We first discuss the macroeconomic results from the model, which are summarized in Table 3. We focus on changes in “absorption�, which is the broadest measure of national welfare. Absorption tracks an economy’s use of goods for household consumption (C), investment (I), and government expenditure (G). Absorption is closely related to GDP growth. Formally, absorption (A) is defined as A=C+I+G. Recalling that GDP=C+I+G+X-M, where X is exports and M is imports, we can write that A=GDP+M-X. In other words, absorption is the volume of goods produced by the economy plus the goods that foreigners supply to the economy (imports) less the goods sent out to foreigners (exports). One advantage of measuring outcomes based on aggregate absorption is that it is less sensitive to our choice of “closure rules�, which influence the relative sizes of absorption’s C, I and G components. [Insert Table 3] 9 Table 3 reports average annual growth rates of real per capita absorption over the entire 2007-2050 simulation period. Changes in annual growth rates are small, with discernable reductions for only the HOT and DRY scenarios. However, even small reductions in growth rates accumulate over time. For example, by the end of the 2040s, national absorption is 0.77 and 1.7 percent below the baseline in the HOT and DRY scenarios, respectively. This is consistent with the larger and more widespread reductions in crop yields experienced under these two scenarios (see Table 2). We measure the total economic damages caused by climate change via the agricultural sector as the cumulative loss or deviation in national absorption from the baseline using a five percent annual discount rate. The largest damages occur in the DRY scenario, where the total discounted loss throughout the 2007-2050 period amounts to US$13 billion (measured in 2007 prices). This amount is two-thirds of Tanzania’s GDP in 2007. By contrast, total discounted absorption rises in the WET scenario by US$3 billion suggesting possible gains from climate change for Tanzania.4 Table 3 also decomposes economic damages across time periods. Despite the escalating biophysical effects of climate change on agricultural yields towards the end of our simulation period, applying a five percent discount rate means that a significant share of the economic costs or benefits of climate change will accrue over the next two decades. This is because crop yield reductions are often temporary during a bad year and can immediately rebound in the subsequent year if the season’s climate improves. In other words, crop yield losses in a given year are usually temporary, unlike damages to assets which may have lasting effects, such as roads damaged by flooding. This partly explains why agricultural damages are more evenly distributed across time periods. However, if we did not discount effects further into the future, then the costs or benefits of climate change would be larger and more heavily weighted towards the middle of the century. Moreover, it should be noted that most GCMs predict a pronounced aggravation of climate change impacts during the second half of the century. Were the time horizon of our analysis extended beyond 2050, then later periods would begin to exhibit progressively stronger impacts. One of the advantages of CGE models is their ability to decompose national impacts to the sector and regional levels. Table 4 reports deviations in real GDP from the baseline in 2046- 2050 for different sectors. Since agriculture is our only impact channel through which climate effects economic growth, it is not surprising that this sector exhibits the largest changes in our four scenarios. However, agriculture provides important inputs into downstream sectors, such as agro-processing. For example, agricultural GDP is 11.5 percent below the baseline in the DRY scenario by the end of the 2040s. This reduces the supply of raw inputs (e.g., grain) to 4 Other impacts, such as increased frequency and intensity of flooding events, were not modeled for the case of Tanzania. The impacts of these events are potentially large (Arndt et al., forthcoming) and could easily overwhelm the benefits of enhanced moisture to crops. 10 downstream agro-processing sectors (e.g. milling), causing their GDP to contract by 7.8 percent. However, not all sectors are adversely affected, even in the DRY scenario. For example, food imports increase in order to offset declining domestic production in the DRY scenario (see Table 3). Accordingly, some traders in the service sector benefit from higher demand for their services. Despite the expansion of services, forestry and fishery, which are only indirectly affected by climate change in our analysis, the net effect of climate change is a significant reduction in national GDP in the HOT and DRY scenarios and a slight decrease/increase in the COOL and WET scenarios, respectively. [Insert Table 4] Table 5 presents deviations in real agricultural GDP from the baseline for the different regions in the model. It should be noted that the DCGE model is for mainland Tanzania only, and so does not reflect changes on the coastal islands (i.e., Pemba and Zanzibar). Almost all regions are adversely affected in the HOT and DRY scenarios, with particularly large reductions in agricultural GDP in the Northern and Central Zones and around Lake Victoria. These regions represent a large share of Tanzania’s agricultural sector, and so a drop in their production has national-level implications. Similarly, while overall agricultural production rises in the WET scenario, it hides significant regional variation. While production increases in the Northern Zone and Northern Coast, it falls in most other regions, including around Lake Victoria. There are also differences in regional outcomes even within agro-climatic zones, such as within Southern Coast. Such pronounced regional variation underscores the need for sub-national assessments, especially for designing policy responses to climate change. [Insert Table 5] Households in the DCGE model are affected by climate change via changes in both agricultural incomes and consumer prices. Households can adapt to these changes by reallocating their resources (e.g., land, labor and capital) towards less-affected sectors or occupations (e.g., non-farm activities). However, if agricultural production falls as a result of climate change then consumer prices for agricultural products will likely increase. Producers may then allocate more of their resources towards climate change affected sectors in order to take advantage of higher prices. This will certainly be the case for farmers in regions that are less adversely affected by climate change. Households’ adaptation decisions therefore involve production and demand considerations, both of which are captured in a general equilibrium model. Table 6 reports deviations in households’ real food consumption expenditure from baseline by 2046-2050. Changes in food consumption in each scenario are less pronounced than changes in agricultural GDP. Two factors drive this result. First, Tanzania is able to import food to replace falling domestic supplies. For example, the 11.5 percent decline in national agricultural production in the DRY scenario (see Table 4) is partially offset by a 37.1 percent increase in net food imports (see Table 3), leaving national food consumption to fall by 8.0 11 percent (see Table 6). Secondly, the model assumes that transport systems are sufficiently developed in Tanzania by 2050 that food is effectively traded in national markets. This means that falling production and excess demand in certain regions can be supplied by producers in other regions. In this way, market forces will distribute changes in national food consumption across regions and household groups. [Insert Table 6] The impact of climate change on incomes and food security therefore depends on three household characteristics. First, climate change has region-specific implications, with some regions benefitting from improved conditions while others are adversely affected. Secondly, climate change will affect crops differently, and so changes in households’ agricultural incomes will depend on their cropping patterns and their ability to reallocate farm resources between farm activities. Finally, agriculture generates only part of households’ incomes and food comprises only part of their consumption basket. Climate change will therefore affect households differently based on their income and consumption patterns. In sum, despite endogenous market-based adaptation, there are still significant differences in outcomes across household groups and regions. For example, lower-income households experience larger declines in per capita food consumption than higher-income households in the COOL and DRY scenarios. This is because poorer households are typically more reliant on agriculture for their livelihoods, and because they spend a larger share of their incomes on food. Likewise, while all regions around Lake Victoria experience similar reductions in agricultural GDP (see Table 5), household food consumption declines by 10.5 percent in Kagera and by only 6.7 percent in Shinyanga. This is because households in Shinyanga region are less dependent on agricultural incomes, and are more heavily engaged in non-farm activities than are households in Kagera. Finally, food consumption amongst non-farm households also declines in the DRY scenario, due to rising food prices and falling real incomes (i.e., due to falling demand for nonagricultural products). 4. Conclusions Relative to a no climate change baseline and considering domestic agricultural production as the principal channel of impact, food security in Tanzania appears likely to deteriorate as a consequence of climate change. This relative decline comes about through reductions in agricultural production, principally food production, due to increases in temperature and changes in rainfall patterns. In the DRY scenario, average agricultural production levels are more than 10 percent below the levels of a hypothetical no climate change scenario by mid-century. This reduced productive capacity also limits growth in exports and growth in household incomes 12 hence reducing the overall capacity of the economy to obtain and distribute food from international markets. It is important to point out that the results do not point to an absolute decline in the levels of food security indicators, such as total agricultural production and household purchasing power. Rather, the rate of improvement in these indicators in three out of four scenarios is reduced. In addition, in one scenario, projected changes in climate are favorable to agricultural production and food security. Overall, the analysis points to a high degree of diversity of outcomes across climate scenarios, sectors, and regions. The methodology applied is well-suited to considering the implications of climate change for growth, development, and ultimately food security. However, while the economic modeling framework is comprehensive, the treatment of climate change within the modeling framework is not. Climate change could impact food security through numerous additional channels beyond reductions in potential yields. These additional channels merit particular attention in future research. We will briefly discuss three. First, the increase in intensity of rainfall due to climate change has the potential to increase the frequency and intensity of flooding events (Arndt et al, 2011). As recent events in Pakistan illustrate, flooding can be highly destructive. Not only does flooding cause a spike in food insecurity in the short run, it frequently wipes out economic infrastructure, such as transport networks, with potentially long term implications for production and growth (Chinowsky, 2011). Second, the results presented reiterate the importance of long run accumulation (Arndt et al., forthcoming). If rates of growth decline even slightly over long periods of time, this decline eventually leads to significant economic impacts. In this context, it is important to recall that the (assumed) underlying rate of agricultural productivity growth is the same across all climate scenarios. It is certainly conceivable that climate change could reduce the expected rate of underlying agricultural productivity growth for any given level of effort devoted to new technology generation and adoption. Finally, all world prices are assumed to be constant. If the climate outcomes in the DRY scenario also resulted in reduced (increased) production globally, then the impacts of the reduction in agricultural production would be magnified (mitigated) by increased (decreased) prices for food commodities on world markets. Overall, while significant progress has been registered, we as a field remain at nascent stages in our understanding of the implications of climate change – across the multiplicity of possible dimensions, and especially for food security in vulnerable low-income countries like Tanzania. 13 References Ahmed, S.A., Diffenbaugh, N.S. and Hertel, T.W. “Climate Volatility Deepens Poverty Vulnerability in Developing Countries,� Environmental Research Letters 4(3) (2009): 8pp. Allen, R.G., Pereira, L.S., Raes, D. and Smith, M. “Crop evapotranspiration: Guidelines for computing crop water requirements,� Irrigation and Drainage Paper 56, Food and Agriculture Organization of the United Nations, Rome (1998). Arndt, C., S. Robinson and D. Willenbockel. “Climate Change and Development Strategy: A Case Study of Ethiopia,� Global Environmental Change (forthcoming). Arndt, C., K. Strzepek, F. Tarp, J. Thurlow, C. Fant, L. Wright. “Adapting to Climate Change: An Integrated Biophysical and Economic Assessment for Mozambique,� Sustainability Science 6(1) (2011): 7-20. Butt, T.A., McCarl, B.A., Angerer, J., Dyke, P.T. and Stuth, J.W. “The Economic and Food Security Implications of Climate Change in Mali,� Climatic Change 68 (2005): 355-378. Chinowsky, P. S., C. Hayles, A. Schweikert, N. Strzepek. “Climate Change as Organizational Challenge: Comparative Impact on Developing and Developed Countries,� Engineering Project Organization Journal 1(1) (forthcoming). Dervis, K, de Melo, J. and Robinson, S. General Equilibrium Models for Development Policy, Cambridge University Press, London (1982). Dimaranan, B.V. (ed.) Global Trade, Assistance, and Production: The GTAP 6 Data Base, Center for Global Trade Analysis, Purdue University, Indiana (2006). Doorenbos, J. and Kassam, A.H. “Yield response to water,� Irrigation and Drainage Paper 33, Food and Agriculture Organization of the United Nations, Rome (1979). FAO-UNESCO. 2005. Soil Map of the World, United States Department of Agriculture, Washington D.C. Fan, S., Nyange, D. and Rao, N. “Public Investment and Poverty Reduction in Tanzania: Evidence from Household Survey Data�, in Morgues, T. and Benin, S. (eds.) Public Expenditures for Agricultural and Rural Development in Africa, Routledge, London (2012). Füssel, H-M. and Klein, R.J.T. “Climate Change Vulnerability Assessments: An Evolution of Conceptual Thinking�, Climatic Change 75 (2006): 301-329. 14 Lobell, D.B., Burke, M.B., Tebaldi, C., Mastrandrea, M.D., Faclon, W.P. and Naylor, R.L. 2008. “Prioritizing Climate Change Adaptation Needs for Food Security in 2030�, Science 319: 607-610. Lofgren, H., Robinson, S. and El-Said, M. “A Standard Computable General Equilibrium (CGE) Model in GAMS�, International Food Policy Research Institute, Washington, DC (2002). Long, S.P., E. A. Ainsworth, A. D. B. Leakey, J. Nösberger, D. R. Ort. “Food for Thought: Lower-Than-Expected Crop Yield Stimulation with Rising CO2 Concentration,� Science 312(5782) 2006: 1918-21. Nelson, G.C., Rosegrant, M.W., Richard Robertson, J.K., Sulser, T., Zhu, T., Ringler, C., Msangi, S., Palazzo, A., Batka, M., Magalhaes, M., Valmonte-Santos, R., Ewing, M. and Lee, D. Climate Change: Impact on Agriculture and Costs of Adaptation, Food Policy Report, International Food Policy research Institute, Washington DC (2010). Parry, M.L., Rosenzweig, C., Iglesias, A., Livermore, M. and Fischer, G. “Effects of climate change on global food production under SRES emissions and socio-economic scenarios,� Global Environmental Change 14 (2004): 53-67. Pauw, K. and Thurlow, J. “Agricultural Growth, Poverty and Nutrition in Tanzania,� Discussion Paper 947, International Food Policy Research Institute, Washington DC (2010). Sacks, W.J., Deryng, D., Foley, J.A. and Ramankutty, N. “Crop planting dates: An analysis of global patterns,� Global Ecology and Biogeography (forthcoming). Schlenker W. and Lobell D.B. “Robust negative impacts of climate change on African agriculture,� Environmental Research Letters 5 (2010): 8pp. Schlosser, A. Personal communication. Joint Program for Global Change, Massachusetts Institute for Technology (February 2011). Sieben, W.H. “Relation of drainage conditions and crop yields on young light clay soils in the yssellake polders,� Van Zee tot Land 40 (1964). Solomon, S., Qin, D., Manning, M., Chen, Z., Marquis, M., Averyt, K.B., Tignor, M. and Miller, H.L. (eds.). Contribution of Working Group I to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change, Cambridge University Press, Cambridge, United Kingdom and New York, USA (2007). Stackhouse, P.W. NASA Climatology Resource for Agroclimatology Daily Averaged Data (Evaluation Version), National Aeronautics and Space Administration, USA (2010). . 15 Strzepek, K., C. Fant, W. Farmer, A. Gueneau, and A. Schlosser. “CliCrop: Modeling the effects of climate change on agriculture,� Working Paper, United Nations University’s World Institute for Development Economics Research, Helsinki Finland (forthcoming). Thornton, P.K., Jones, P.G., Alagarswamy, G. and Andresen, J. “Spatial variation of crop yield response to climate change in East Africa�, Global Environmental Change 19(1) (2009): 54-65. Tubiello, F., J.S. Amthor, K.J. Boote, M. Donatelli, W. Easterling, G. Fischer, R.M. Gifford, M. Howden, J. Reilly, and C. Rosenzweig. “Crop Response to Elevated CO2 and World Food Supply: A Comment,� European Journal of Agronomy 26(3) (2007): 215-23. Tyndall Center. Data and images downloaded from www.tyndall.ac.uk (2010). Willmott, C. J. and J. J. Feddema. “A More Rational Climatic Moisture Index ,� The Professional Geographer 44(1) (1992): 84-88. World Bank. Economics of Adaptation to Climate Change: Synthesis Report. Washington DC (2010). You, L., S. Wood, U. Wood-Sichra. “Generating global crop maps: from census to grid,� Selected paper, IAAE (International Association of Agricultural Economists) Triennial Conference, Gold Coast, Australia (2006). Appendix [Insert Table A1] [Insert Figure A1] [Insert Table A2] 16 Table 1: Projected national climate changes in Tanzania by 2041-2050. Scenario General Emissions Average change from baseline, 2040-2050 ID circulation model scenario Temperature Precipitation CMI (GCM) (SRES) (°C) (%) (absolute) HOT ncar_ccsm3_0 a1b 1.87 5.67 +0.0565 COOL ncar_pcm1 a1b 1.13 5.37 +0.0157 WET csiro_mk3_0 a2 1.43 13.3 +0.0243 DRY ukmo_hadgem1 a1b 1.49 -11.14 -0.0853 Source: Own calculations using GCM results obtained from the Fourth Assessment Report online archive. 17 Table 2: Changes in mean annual dry-land maize yields, 2041-2050. Change from baseline (%) HOT COOL WET DRY Northern zone Arusha -15.05 4.06 12.66 -23.1 Kilimanjaro -13.61 1.87 11.81 -15.55 Manyara -13.46 3.18 15.28 -16.64 Tanga -11.29 1.31 8.12 -6.84 Southern highlands Iringa -3.2 -2.72 1.51 -5.51 Mbeya 0.25 -3.02 -3.12 -4.58 Ruvuma -2.23 3.74 5.51 -5.17 Northern coast Dar Es Salaam -1.37 -0.42 0.97 -0.03 Morogoro -4.49 -3.11 4.15 -5.49 Pwani -6.25 -1.83 3.64 -5.09 Southern coast Lindi -3.03 -2.97 -2.45 -4.29 Mtwara 0.01 -4.65 -7.47 0.19 Lake Victoria Kagera -4.64 -3.35 -9.35 -16.48 Mara -6.94 -1.19 -0.33 -16.91 Mwanza -6.18 1.11 -2.62 -18.27 Shinyanga -8.14 3.52 -2.76 -19.78 Western zone Kigoma 2.03 -7.56 -7.73 -14.27 Rukwa 0.69 -4.57 -8.39 -8.63 Central zone Dodoma -9.45 1.15 13.46 -13.25 Singida -6.79 0.74 -0.89 -9.8 Tabora -4.64 -0.91 -12.38 -14.74 Coastal islands Kaskazini Pemba -0.9 -0.52 2.58 -2.39 Kaskazini Unguja -1.37 -0.42 0.97 -0.03 Kusini Unguja -1.37 -0.42 0.97 -0.03 Mjini Magharibi -1.37 -0.42 0.97 -0.03 Source: Results from CLICROP models for Tanzania. 18 Figure 1: Mean annual dry-land maize yield changes for HOT and COOL scenarios, 2041-2050. HOT scenario COOL scenario Source: Results from CLICROP models for Tanzania. 19 Figure 2: Mean annual dry-land maize yield changes for WET and DRY scenarios, 2041-2050. WET scenario DRY scenario Source: Results from CLICROP models for Tanzania. 20 Table 3: Macroeconomic results. Baseline HOT COOL WET DRY Average annual real per capita absorption growth rate, 2007-50 (%) 2.74 2.72 2.73 2.74 2.70 Deviation from baseline - -0.02 0.00 0.00 -0.04 Average annual undiscounted value of absorption, 2046-50 (US$ billions, 2007 prices) 95.42 94.69 95.27 95.51 93.82 Deviation from baseline - -0.73 -0.16 0.08 -1.60 Deviation as a share of baseline (%) - -0.77 -0.16 0.09 -1.71 Accumulated discounted deviation in absorption from baseline, 2007-50 (US$ billions, 2007 prices) - -4.21 -0.91 3.03 -12.70 Accrued during 2010s - -1.10 -0.45 0.53 -3.38 Accrued during 2020s - -0.94 0.11 0.81 -3.61 Accrued during 2030s - -0.74 -0.18 0.73 -2.67 Accrued during 2040s - -1.00 -0.23 0.26 -2.46 Deviation in average annual net food imports from baseline, 2046-50 (%) 21.34 3.65 -6.49 37.13 Source: Results from the DCGE model for Tanzania. 21 Table 4: Sectoral results. Initial Deviation in average annual real GDP from GDP share baseline, 2046-50 (%) (%) HOT COOL WET DRY Total GDP 100.00 -0.89 -0.19 0.17 -1.93 Agriculture 27.82 -5.42 -1.19 1.10 -11.51 Cereals 8.31 -6.05 -0.59 1.25 -10.48 Maize 4.42 -5.91 -0.45 1.23 -10.89 Root crops 3.27 -2.44 -1.54 -1.40 -10.13 Pulses and oilseeds 2.71 -2.80 -0.99 0.03 -8.75 Horticulture 5.19 -5.44 -3.22 3.45 -13.92 Export crops 2.79 -6.53 -1.03 1.79 -11.37 Livestock 5.55 -6.96 -0.06 0.04 -13.18 Forestry and Fisheries 4.02 0.20 0.03 0.03 0.50 Mining 3.94 0.00 0.00 0.00 0.01 Manufacturing 8.84 -2.05 -0.38 0.26 -4.42 Food processing 4.58 -3.57 -0.62 0.32 -7.84 Construction and energy 10.33 -0.02 0.00 0.02 -0.01 Services 45.05 0.13 0.03 -0.03 0.23 Source: Results from the DCGE model for Tanzania. 22 Table 5: Regional results. Initial Deviation in average annual real agricultural GDP GDP share from baseline, 2046-50 (%) (%) HOT COOL WET DRY National (all regions) 100.00 -5.42 -1.19 1.10 -11.51 Northern zone 21.85 -9.76 -1.89 8.51 -15.51 Arusha and Manyara 10.12 -13.18 -0.69 7.07 -18.89 Kilimanjaro 5.95 -7.88 -4.16 9.49 -15.23 Tanga 5.77 -5.58 -1.66 10.06 -9.75 Southern highlands 15.72 -2.14 -1.88 -0.84 -6.33 Iringa 3.36 -3.35 -2.04 1.49 -6.45 Mbeya 7.53 -0.44 -2.64 -3.08 -6.46 Ruvuma 4.83 -4.15 -0.46 1.27 -6.02 Northern coast 13.48 -3.59 -1.32 5.66 -9.23 Dar Es Salaam 0.62 2.20 1.89 5.79 -5.71 Morogoro 8.85 -4.11 -1.85 4.98 -9.37 Pwani 4.00 -3.34 -0.62 7.18 -9.47 Southern coast 3.95 -1.88 0.65 0.10 -6.49 Lindi 1.94 -4.56 -0.11 3.41 -12.48 Mtwara 2.00 1.38 1.57 -3.92 0.77 Lake Victoria 30.21 -6.24 -0.12 -3.53 -13.61 Kagera 7.60 -4.67 -0.58 -7.90 -13.78 Mara 4.95 -5.18 -0.94 -1.16 -12.14 Mwanza 8.99 -6.84 -0.65 -3.14 -13.28 Shinyanga 8.66 -7.66 1.29 -1.41 -14.66 Western zone 6.96 -0.57 -4.04 -5.32 -9.47 Kigoma 4.64 -0.36 -4.18 -4.92 -10.50 Rukwa 2.32 -1.00 -3.78 -6.08 -7.48 Central zone 7.83 -5.39 -0.11 0.05 -10.19 Dodoma 3.07 -6.43 0.80 8.20 -10.05 Singida 2.09 -4.95 0.25 -0.91 -7.36 Tabora 2.68 -4.54 -1.46 -8.66 -12.60 Source: Results from the DCGE model for Tanzania. 23 Table 6: Household food consumption results. Initial food Deviation in average annual real per capita food consumption consumption from baseline, 2046-50 (%) (US$ p.c.) HOT COOL WET DRY National (all households) 355 -3.57 -0.78 0.33 -7.95 Farm 303 -3.59 -0.83 0.35 -8.04 Non-farm 582 -3.53 -0.65 0.30 -7.75 Quintile 1 114 -3.72 -1.06 0.39 -8.54 Quintile 2 193 -3.58 -0.92 0.35 -8.11 Quintile 3 272 -3.62 -0.83 0.36 -8.15 Quintile 4 390 -3.55 -0.81 0.35 -7.96 Quintile 5 805 -3.54 -0.67 0.31 -7.75 Northern zone 425 -3.69 -0.90 0.17 -8.42 Arusha and Manyara 430 -3.86 -0.91 0.23 -8.53 Kilimanjaro 477 -4.04 -0.98 0.40 -9.16 Tanga 382 -3.17 -0.83 -0.12 -7.64 Southern highlands 268 -3.64 -0.65 0.33 -7.79 Iringa 260 -3.81 -0.63 0.23 -8.14 Mbeya 214 -3.62 -0.64 0.70 -7.49 Ruvuma 377 -3.53 -0.68 0.01 -7.82 Northern coast 459 -3.18 -0.80 0.28 -7.14 Dar Es Salaam 789 -3.10 -0.89 0.36 -7.33 Morogoro 324 -3.45 -0.77 0.35 -7.42 Pwani 649 -2.91 -0.80 0.17 -6.74 Southern coast 412 -3.53 -0.91 0.34 -8.18 Lindi 452 -3.34 -0.94 0.40 -7.93 Mtwara 379 -3.71 -0.88 0.28 -8.43 Lake Victoria 236 -3.60 -0.96 0.54 -8.17 Kagera 226 -4.36 -1.62 1.14 -10.51 Mara 399 -3.53 -0.78 -0.09 -8.40 Mwanza 218 -3.32 -0.64 0.41 -7.18 Shinyanga 200 -3.21 -0.76 0.55 -6.72 Western zone 285 -3.98 -0.64 0.25 -8.64 Kigoma 277 -3.79 -0.76 0.19 -8.51 Rukwa 294 -4.19 -0.51 0.31 -8.79 Central zone 228 -3.59 -0.78 0.45 -7.86 Dodoma 204 -3.16 -0.94 0.08 -7.37 Singida 267 -3.78 -0.75 0.62 -8.08 Tabora 228 -3.85 -0.65 0.69 -8.18 Source: Results from the DCGE model for Tanzania. 24 Table A1: Disaggregation of the Tanzania DCGE model. Agricultural sectors Maize; Sorghum; Millet; Rice, wheat & barley; Cassava; Root crops; Pulses; Oilseeds; Horticulture; Export crops; Livestock; Other agriculture Nonagricultural sectors Mining; Meat processing; Maize milling; Rice milling; Other milling; Other food processing; Export crop processing; Chemicals; Machinery; Other manufacturing; Electricity; Water distribution; Construction; Trade and transport; Other private services; Public services Sub-national regions (for Arusha; Coast; Dodoma; Dar es Salaam; Iringa; Kagera; Kigoma; Kilimanjaro; agricultural sectors only) Lindi; Mara; Mbeya; Morogoro; Mtwara; Mwanza; Manyara; Rukwa; Ruvuma; Shinyanga; Singida; Tabora; Tanga Factors Primary school and uneducated labor; Secondary school educated labor; Tertiary educated labor; Agricultural capital; Mining capital; Nonagricultural capital; Agricultural crop land (by region); Livestock capital (by region) Households Farm households (by region and national per capita consumption expenditure quintile); Rural nonfarm households (by national per capita consumption expenditure quintile); Urban nonfarm households (by national per capita consumption expenditure quintile) 25 Figure A1: Agro-climatic regions of Tanzania 26 Table A2: Land and population distribution across regions and farm households. All Nonfarm households Farm households house- All Urban Rural All North South North South Lake Western Central holds Zone high- coast coast Victoria zone zone lands Population (1000) 31,683 5,890 3,590 2,301 25,793 2,517 3,963 3,657 1,690 7,889 1,972 4,105 Number of households 6,393 1,360 878 482 5,033 480 938 698 431 1,273 355 859 Household size 5.0 4.3 4.1 4.8 5.1 5.2 4.2 5.2 3.9 6.2 5.6 4.8 Per capita exp. (US$) 329 558 687 356 277 339 288 372 355 214 324 209 Poverty rate (%) 40.0 24.9 15.8 39.2 43.5 40.9 33.5 38.4 37.6 49.0 54.1 45.8 Poor population (1000) 12,679 1,468 567 901 11,211 1,030 1,328 1,405 635 3,868 1,066 1,879 Share of poor (%) 100.0 11.6 4.5 7.1 88.4 8.1 10.5 11.1 5.0 30.5 8.4 14.8 Harvest area (1,000 ha) - - - - 8,209 1,004 951 1,919 605 2,597 411 722 Average farm land (ha) - - - - 1.63 2.09 1.01 2.75 1.40 2.04 1.16 0.84 Maize - - - - 0.53 0.74 0.41 1.04 0.27 0.52 0.44 0.34 Sorghum and millet - - - - 0.18 0.11 0.04 0.15 0.00 0.31 0.05 0.34 Other cereals - - - - 0.12 0.17 0.05 0.43 0.00 0.15 0.02 0.00 Roots - - - - 0.24 0.04 0.15 0.22 0.82 0.38 0.10 0.01 Pulses and oilseeds - - - - 0.29 0.47 0.14 0.57 0.16 0.31 0.43 0.14 Horticulture - - - - 0.13 0.35 0.12 0.22 0.03 0.14 0.07 0.00 Export crops - - - - 0.13 0.21 0.11 0.12 0.13 0.24 0.03 0.01 Source: Authors’ calculations using data from Pauw and Thurlow (2010). Notes: Population data is from HBS 2000–2001 (NBS 2002). Per capita expenditure is based on consumption spending from the 2007 social accounting matrix. The poverty line identifies the bottom two per capita expenditure quintiles as poor. 27