The Impact of Climate Variability and Change on Economic Growth and Poverty in Zambia

Transcription

1 The Impat of Climate Variability and Change on Eonomi Growth and Poverty in Zambia James Thurlow, Tingju Zhu and Xinshen Diao 1 International Food Poliy Researh Institute Deember We are grateful to Len Abrams, Rimma Dankova and Marus Wishart for their tehnial advie throughout the projet. A number of people in Zambia provided information, for whih we are grateful. These inlude Angel Daka, Klaus Droppelmann, Paavo Eliste, John Fynn, Alex Mwanakasale, Peter Sheppard, Henry Sihembe, George Sikuleka, Timothy Stephens, and Mike Weber. We also thank Zhe Guo for providing GIS assistane and Vida Alpuerto for other researh assistant to the projet. The study is funded by the World Bank. 1

2 The Impat of Climate Variability and Change on Eonomi Growth and Poverty in Zambia 1. Introdution Zambia is a low-inome ountry with a history of errati eonomi growth. Some of the uneven performane of the eonomy has been driven by unsustainable poliies, adverse global onditions, and pronouned shoks from maroeonomi reforms (Resnik and Thurlow, forthoming). While the ountry has performed well sine the late 1990s, with positive eonomi growth and poverty redution (see Figure 1), growth in agriulture remains volatile despite improvements in the poliy environment. This an, at least in part, be attributed to high rainfall variability, or more generally limate variability, in the ountry. Indeed some of the more substantial delines in eonomi growth over the last three deades have ourred during major drought years. Climate variability is espeially important for the agriultural setor, whih is heavily dependent on rainfall due to the ountry s limited irrigation apaity. Climate variability may also undermine attempts to redue poverty, sine most of Zambia s poor population live in rural areas and are heavily dependent on agriultural inomes. Overoming limate variability therefore poses a signifiant hallenge to maintaining agriultural growth, signifiantly reduing poverty, and ahieving the Millennium Development Goals. It also heightens onern over potentially negative impats from limate hange. Together limate variability and limate hange plaes onsiderable pressure on Zambia s government to improve inentives for farmers and for the private setor to invest in infrastruture and improve produtivity. Within this ontext, a number of key poliy relevant questions emerge: What is the eonomi ost of limate variability for both agriultural and national prodution? How does limate variability affet household welfare and poverty at the national level? Whih regions in the ountry are most vulnerable to limate variability? Will limate hange exaerbate or dampen variability and what are its long-term impliations for eonomi growth and poverty redution in Zambia? This paper addresses these questions through an integrated framework linking together various hydrologial, rop simulation and eonomi models that draw on Zambia s historial data. The next setion reviews temporal, inter-temporal and spatial rainfall patterns in Zambia in the 33 years between 1975 and Setion 3 estimates the impat of Zambia s historial limati patterns on rop yields using hydrologial and rop yield simulation models. Setion 4 ombines the results from the hydro-rop models and analyzes the impat of limate variability on eonomi growth and poverty redution over the next 10 years using an eonomywide model. Setion 5 assesses how limate variability and its 2 In the following analysis for rop growing season the data period is referred to as beause rop growing season spans two alendar years. 2

3 Annual growth rate (%) eonomi impats may be exaerbated or dampened by limate hange over the next years. The final setion summarizes our findings and suggests poliy responses. Figure 1.1: Annual total and agriultural GDP growth rates, Total GDP Agriultural GDP Flood Drought Notes: World Bank (2008) for , and national aounts for (CSO, 2008). 2. Climati harateristis of Zambia The high plateau on whih Zambia is loated ensures that the ountry has a moderate limate, with summer temperature rarely exeeding 35 C. However, rainfall is unevenly distributed throughout the year, with the majority onentrated in the six months from November to April. This leaves the remaining months almost dry. Aordingly, Zambia has three seasons: 1) a rainy season in summer from November to April; 2) a ool dry winter season from May to August; and 3) a hot dry season in September and Otober. Hene, for most rain-fed rops the growing season is the rainy season. Moreover, muh of the ountry s soioeonomi life is dominated by the onset and end of the rainy season, and the amount of rain it brings. Agro-limati harateristis in any ountry or sub-national region are primarily determined by intrayear distributions and inter-year variations in rainfall and temperature. The rains of Zambia are brought by the Intertropial Convergene Zone whih is loated north of the ountry in the dry season. It moves southwards in the seond half of the year, and northwards in the first half of the year. Given Zambia s altitude, its temperatures are lower than oastal regions at similar latitudes. Analysis of the auses of Zambia s limate and agriulturally-important weather events are beyond the sope of this study. For this study, monthly weather observations were made available by the Zambia Meteorologial Department, and data for at 30 weather stations were used (see the red markers in Figure 3

4 2.1). 3 These 30 meteorologial stations are loated in five distinguished agroeologial zones that later form the spatial unit for our analysis. Figure 2.1: Zambia s agroeologial zones, meteorologial stations and Thiessen polygons Notes: Thiessen Polygons were reated to define the influening domain of eah of the thirty meteorologial stations. Red triangles mark the loations of meteorologial stations. 2.1 Spatial distributions of annual rainfall and evaporation Figure 2.2 shows average annual rainfall at the 30 weather stations in Zambia during interpolated to 1km pixels. There is a downward gradient of annual rainfall from the north to the south of the ountry, with the highest rainfall in the northwest and northeast (generally above 1200 mm) and the lowest in the southwest (generally below 800 mm). Referene evapotranspiration (ET O ) is the evapotranspiration potential or atmospheri water demand of a referene grass (Allen et al., 1998). This measure provides the basis for estimating rop water requirements, whih is essential information if we are to estimate how hanges in rainfall or water availability affet rop yields. We alulate monthly ET O using weather data from the 30 stations. Mean annual ET O values were then interpolated to 1km pixels, as shown in Figure 2.3. The results indiate that, in ontrast to the north-to-south delining trend of rainfall, ET O inreases from the north to south, espeially in the southwest part of the ountry. This suggests rainfall is lowest in areas where rops water requirements are highest, thus exposing rain-fed agriulture in the south to onsiderable risks of yield losses or even rop failure during droughts. 3 There is a tradeoff between station overage and the lengths of observations of seleted stations. The hoie of 30 stations data for the period was thus a balane between ross-setional and time-series overage. 4

5 Figure 2.2: Average annual rainfall, Soure: Authors alulations using historial rainfall data from the Zambia Meteorologial Department. Notes: Generated from 1km pixel averaged annual rainfall interpolated from 30 weather stations. Figure 2.3: Average annual referene evapotranspiration, Soure: Authors alulations using historial rainfall data from the Zambia Meteorologial Department. Notes: Generated from 1Km pixel averaged annual ETo interpolated using ETo alulated for thirty weather stations. The main objetive of this study is to analyze the impat of limate variability on Zambia s eonomy of whih agriulture is an important omponent. Thus, it is neessary to aggregate the ountry into five agroeologial zones, eah of whih has different limate patterns. Zone I overs most of Southern provine and parts of Lusaka and Eastern provines. Zone IIa is subdivided into two smaller zones beause of differenes in rainfall patterns (see disussion below). Zone IIa1 overs the apital ity Lusaka and the eastern parts of Central provine, while Zone IIa2 inludes the western parts of Central provine and most of Eastern provine. Zone IIb omprises almost entirely Western provine. Finally, Zone III, 5

6 whih is the largest one in terms of geographi size, inludes the Copperbelt, North Western, Luapula and Northern provines. In this study, rainfall and other meteorologial data were aggregated to these five zones, taking into onsideration the influening domain of eah weather station (Figure 2.1). Zonal level average annual rainfall is presented in Figure 2.4, whih shows a similar pattern as in Figure 2.2, with the highest rainfall in the northern Zone III, and the lowest rainfall in the southern Zone I. Zone IIa1 has a mean annual rainfall of 818 mm, whih is only slightly higher than Zone I. By ontrast Zones IIa2 and IIb have higher annual rainfall at 941mm and 930mm respetively. These differenes highlight the importane of separating the eastern and western parts of Zone IIa, espeially sine most of Zambia s eonomi ativity takes plae near Lusaka. 4 Figure 2.4: Average annual preipitation by agroeologial zone, (mm) Soure: Authors alulations using historial rainfall data from the Zambia Meteorologial Department. 2.2 Intra-annual distribution and inter-annual variations of rainfall and evaporation The distribution of rainfall and temperatures during a year determines the growing season of annual rops and influenes their yields, espeially those ultivated under rain-fed onditions. Figure 2.5 shows monthly rainfall distributions in Zambia s five agroeologial zones. It shows that most of the rainfall in eah of the five zones is onentrated between November and April (the rainy season), with virtually no rainfall from May to September (the dry winter season). Although the 20-40mm rainfall in Otober marks the end of the dry season, the depletion of soil moisture during the dry season may prevent immediate rop planting in this month. 4 Appendix C desribes the eonomi struture of the five agroeologial zones. 6

7 Preipitation (mm) Figure 2.5: Mean monthly preipitation in agroeologial zones, Jan Feb Mar Apr May Jun Jul Aug Sep Ot Nov De I IIa1 IIa2 IIb III Soure: Authors alulations using historial rainfall data from the Zambia Meteorologial Department. Compared to the intra-year distribution of rainfall, referene evapotranspiration (ET O ) shown in Figure 2.6 is low during rainy months and high during dry months. For all zones, the lowest ET O is in Deember to February (around 100 mm) when wind speed is lowest and rainfall and relative humidity are highest for the year. The highest ET O values are in September, ranging from about 160 mm in Zone III to 200 mm in Zone IIb, with the other zones in between. During this month temperature and wind speed are highest and relative humidity is lowest for the year. Interestingly, a seond peak of ET O is seen in May, whih is also onfirmed by the observed open-water pan evaporation data for most of the 30 weather stations. For Zone III, relative humidity during May is lower and wind speed is higher than during the two neighboring months, whih may explain the small peak of ET O in May. This is not seen in other zones. 5 Year-to-year variations in rainfall are generally high in zones with low rainfall and low in wetter zones towards the north of the ountry. Figure 2.7 gives the normalized standard deviation (i.e., oeffiient of variation) of annual rainfall for the period The dry Zones I and IIa1 have the highest interannual rainfall variability, with oeffiient values of and respetively. Assuming that the annual rainfall amount follows a normal distribution, this implies that in Zone IIa1, for example, there is about a 30 perent probability that rainfall in any given year is 20 perent (i.e., 170 mm) higher or lower than the mean rainfall level of 820 mm shown in Figure 2.4. This indiates potential drought or flood events depending on the distribution of the rainfall defiit or the intensity of the rainfall surplus in a partiular rainy season. A higher oeffiient of variation for Zone IIa1 inidates this zone to have higher inter-annual rainfall variation than Zone I, despite the latter having lower average annual rainfall. 5 A thorough explanation of the bimodal intra-year distribution of ET O needs further analysis. 7

8 Referene Evapotranspiration (mm) Figure 2.6: Mean monthly referene evapotranspiration in agroeologial zones, Jan Feb Mar Apr May Jun Jul Aug Sep Ot Nov De I IIa1 IIa2 IIb III Soure: Authors alulations using historial rainfall data from the Zambia Meteorologial Department. Figure 2.7: Annual preipitation oeffiient of variation by agroeologial zone, Soure: Authors alulations using historial rainfall data from the Zambia Meteorologial Department. Notes: Standard deviation normalized by mean, indiating inter-annual variations of rainfall in an agro-eologial zone. Figure 2.8 shows inter-year rainfall variations in eah of the five zones, while Figure 2.9 shows the perentage deviations of annual rainfall from their means. Annual dry and wet flutuations appear to share the same rhythm in major drought or wet years in Zones I, IIa1, IIa2 and IIb, exept for Zone III. This indiates that drought events an be nationwide in some years, making the ountry less apable of mitigating its onsequenes through its own efforts. The wet Zone III shows only moderate inter-year rainfall variation, with most years lying within 10 perent of the mean and rarely falling below 1100 mm. By ontrast, rainfall deviations in Zones I, IIa1 and IIb frequently exeed 20 perent of the mean, and even approah 30 perent defiits in major drought years. 8

9 Annual Preipitation (mm) Annual Preipitation (mm) Figure 2.8: Annual preipitation of agroeologial zones, Zones I, IIa1 and IIb I IIa1 IIb 1700 Zones IIa2 and III IIa2 Soure: Authors alulations using historial rainfall data from the Zambia Meteorologial Department. III 9

10 Deviation from Mean (%) Deviation from Mean (%) Figure 2.9: Annual deviation in preipitation from long-term mean for agroeologial zones 50 Zones I, IIa1 and IIb I IIa1 IIb 50 Zones IIa2 and III IIa2 III Soure: Authors alulations using historial rainfall data from the Zambia Meteorologial Department. 10

11 Potential evapotranspiration of maize in its growing season are shown in Figures Maize is hosen for the disussion beause it is Zambia s most important annual rop. Clearly the amplitudes of interyear potential evapotranspiration variations are muh smaller than rainfall. Nevertheless, like rainfall, the variations of potential evapotranspiration are large in Zones I, IIa1 and IIb but small in Zones IIa2 and III. Even for zones with large variations, the differene in maize evapotranspiration between any two years is well within 100 mm. For Zone III, the largest differene in two years is only around 30 mm. Potential evapotranspiration of maize during the growing season in eah of the zones is found to be inversely orrelated with zonal rainfall. These opposite deviations of rainfall and potential evapotranspiration in the same year are partiularly apparent in major drought years (1986/87, 1991/92, 1994/5 and 2004/05) and major wet years (1977/78, 1988/89, 1992/93 and 2006/07). 7 This implies that larger-than-usual amounts of water are needed by rops during years when rainfall is low. On the other hand, rop water requirements are usually below normal in years when rainfall is high, with flooding and water-logging impliations. 2.3 Distribution of droughts and wet events based on drought index analysis Droughts are omplex phenomena and an index is often used to measure their severity. For agriulture, drought indies usually take into aount the amount of soil water available to the rop rather than fousing only on rainfall defiits. For this study we use a Palmer Z Index (Palmer 1965; Allen 1984) as a drought severity metri, sine it is a better indiator of monthly soil moisture onditions than other indies. The Palmer Z index is based on the supply-demand onept of soil water balane, taking into aount more than just rainfall defiits at speifi loations. It also provides a standardized measure of moisture onditions, thus allowing for omparisons aross loations and time periods (Palmer 1965). The annual drought indies disussed below are in fat a wet season (November-Marh) average of the monthly Palmer Z-index. This is beause the wet season overlaps with the growing period of most rainfed rops, inluding maize. A negative value for the Palmer-Z-index indiates dry onditions, while a positive value indiates wet onditions. Unfortunately, the monthly meteorologial data used in this study do not provide suffiient information to identify floods, whih usually our over muh shorter periods of time suh as a few days of heavy rains (World Bank, 2008). However, the wet onditions indiated by a high positive value of the Palmer Z index might suggest possibilities of flooding. Monthly Palmer Z indies were alulated for for the five agroeologial zones, and were then averaged over the wet season to reate annual drought indies for eah zone for the harvest years. As shown in Table 2.1, threshold values for the index were hosen in order to ategorize the growing seasons into severe drought years (-1.5), moderate drought years (-0.5), normal years, moderately wet years (0.5) and very wet years (1.5). 6 See Appendix A. 7 The harvest year is usually assoiated with drought/floods, despite the growing season spanning two alendar years. For example, the drought that took plae during the season orresponds to 1992 for other data soures (e.g., national aounts). We follow this onvention when the season is not identified expliitly in the text. 11

12 Figure 2.10: Growing period potential evapotranspiration of maize in agro-eologial zones, Zones I, IIa1 and IIb ET I IIa1 IIb Zones IIa2 and III ET Soure: Results from the hydro-rop model. IIa2 III 12

13 Table 2.1: Number of years of simultaneous limati event ourrenes aross agroeologial zones, Number of zones simultaneously affeted Severe drought (Z a -1.5) Moderate drought (-1.5 < Z -0.5) Normal (-0.5 < Z 0.5) Moderately wet (0.5 < Z 1.5) Very wet (Z > 1.5 ) (1994) (1978: I, IIa1, IIa2, IIb) 3 1 (1992: I, IIa1, IIb) (1995: I, IIa1; 2005: I, IIa1) 1 1 (1987: IIa1) (1981: I, IIb) (1979: III; 1989: IIa1; 1997: IIa1; 2004: 2b) Note: Averaged monthly Palmer Z Index in maize growing period from November to Marh. The monthly Palmer Z index shows how monthly moisture onditions depart from normal, refleting short-term drought and wetness (Palmer 1965; Alley 1984). 13

14 Apart from a drought s severity measure, its spatial extent an prove as important for agriulture. Simultaneous drought onditions over large areas an greatly redue a ountry s ability to mitigate negative outomes and provide food assistane to drought-striken regions. Table 2.1 gives the frequenies of drought/wet events taking plae simultaneously aross agroeologial zones. It shows that the worst drought event during happened in the 1991/92 growing season, when three out of the five zones experiened a severe drought (i.e, Zones I, IIa1, IIb). Both Zones I and IIa1 had severe droughts during the 1994/95 and 2004/05 seasons, while only Zone IIa1 had a severe drought in 1986/87. These four seasons represent the major drought years in the historial rainfall data. Moderate droughts our more often than severe droughts and usually affet larger areas. Table 2.1 shows that all five agroeologial zones had moderate droughts during the 1993/94 season. This is the only ountrywide moderate drought observed in the 32-year period between 1976 and There were four other rainy seasons during when four of the five zones were simultaneously affeted by moderate droughts; four seasons when only three of the five zones were simultaneously affeted; and four seasons when only two of the five zones were simultaneously affeted. However, the situation where only one agroeologial zone is under moderate drought onditions happens more frequently. There were seven seasons with moderate droughts for one of the five zones, indiating a reourrene interval of eah 4.5 years. Normal weather (i.e., a Palmer Z index between -0.5 and 0.5) never happens simultaneously in more than three of the five agroeologial zones during the 32-year study period, indiating that Zambia is prone to extreme weather events. There were six seasons when normal weather simultaneously ourred in three of the five zones; eleven seasons when normal weather simultaneously happened in two zones; and ten seasons when normal weather happened in only one zone. Moderately wet and very wet events appear to be less frequent than droughts, and they never simultaneously happened in all five zones. The indies show that there were four seasons when moderately wet onditions simultaneously ourred in four of the five zones; five seasons when moderate drought simultaneously happened in three zones, and two seasons when moderate drought simultaneously happened in two zones. The hanes that only one zone is moderately wet is onsiderably higher (i.e., nine seasons out of a total of 32 years). Finally, as with severe droughts, very wet onditions rarely our simultaneously in Zambia. Only in the 1977/78 season did four zones simultaneously experiene very wet onditions (i.e., Zones I, IIa1, IIa2 and IIb). Moreover, only in the 1980/81 season were two zones simultaneously affeted (i.e., Zones I and IIb). Four seasons were found to have only one agroeologial zone under very wet onditions during the study period. Some onlusions an be drawn from the index statistis in Table 2.1. First, Zambia is prone to droughts and floods, with a high hane that at least one agroeologial zone experienes abnormal weather events in any given year. Seondly, the entral and southern regions of the ountry (i.e., Zones I, IIa1 and IIb) are partiularly prone to both droughts and floods. By ontrast Zone III has fairly stable weather onditions, with no severe droughts having taken plae over the last three deades and only one season 14

15 identified as very wet (i.e., 1978/79). Thirdly, it seems drought events are ourring more frequently, with almost all of the ountry s severe droughts taking plae during the seond half of the study period. 3. Modeling the biophysial impat of limate variability using a hydro-rop model Two types of models were used to analyze eonomi impats of variability and limate hange. An integrated hydro-rop (HC) model was first used to predit soil water balanes and rop yield responses in Zambia s five agro-eologial zones. This information was then applied into a dynami eonomywide omputable general equilibrium (DCGE) model, whih estimated impats on prodution at the subsetor, zonal and national level and on household inomes and poverty. This setion desribes the hydro-rop model and the estimation of rop yield losses resulting from limate variability. Quantitative information on rop yield responses to water defiieny is ruial for evaluating the eonomi impats of limate variability and hange on agriultural prodution. Climate variability s impat on agriulture is espeially important in Zambia sine this is the primary impat hannel through whih the broader eonomy is affeted (World Bank, 2008). There are different approahes to estimating rop yields under varying weather onditions. One approah is to use proess-oriented rop growth models to simulate rop growth and yields, taking into aount detailed biophysial proesses from the root extration of water and nutrients to plant photosynthesis and yield formation. However, these kinds of models require extensive soil and rop information, and are typially onstruted separately for different rops. Given the lak of data in Zambia, we adopt a semi-empirial approah and develop a hydro-rop model that inludes two stages: the first stage is to simulate atual evapotranspiration using a soil water balane model for rop root zone similar to that reommended in Allen et al. (1998); and the seond stage is to estimate rop yield responses to water defiits using an empirial rop water prodution model originally proposed by Jensen (1968). Together these two stages form an integrated hydro-rop model that is used in this study. The output of the hydro-rop model beomes an input into the eonomywide model in order to assess the eonomi impat of limate variability. For this study, a dynami omputable general equilibrium (DCGE) model has been developed and this model, together with the hydro-rop model, is applied to analyze eonomi impats of limate variability and hange. The DCGE model ontains information on different setors, fators and households in eah of the five agro-eologial zones. For example, the model distinguishes between 34 different setors, half of whih are rop and livestok subsetors and the other half are industrial and servie subsetors. Prodution in eah agriultural subsetor was disaggregated aross zones using distrit-level data provided by the Ministry of Agriulture, Food and Fisheries. We also separate small/large-sale and rural/urban agriultural prodution. So the model aptures differenes in ropping and livestok patterns driven by agro-eologial onditions and by farm land endowments. This disaggregation of prodution allows the DCGE model to apture how limate onditions vary aross zones, and how limate affets rops differently aording to their agronomi harateristis. The eonomywide model s harateristis and simulations are desribed in detail in Setion 4. 15

16 3.1 The hydro-rop (HC) model The soil water balane module of hydro-rop model regards rop root zone as a buket with water flowing in through rainfall, and irrigation if appliable, and water leaking away as ET, runoff and deep perolation. The first step in the soil water balane is to estimate rop water requirements, whih are normally expressed by the rate of potential evapotranspiration (ET). The United Nation s Food and Agriulture Organization (FAO) has developed a pratial proedure to estimating rop water requirements (Allen et al., 1998) whih has beome a widely aepted standard. Owing to the diffiulty of obtaining readily available and aurate field measurements, the estimation of rop water requirements are derived from estimating rop evapotranspiration aording to referene rop and limati onditions. The water requirements of a given rop are derived through a rop oeffiient that integrates the ombined effets of rop transpiration and soil evaporation into a single rop oeffiient. The potential ET for a partiular rop is estimated using potential ET of a referene rop, usually alfalfa, and a alibrated rop oeffiient whih onverts the referene ET to potential ET of the rop, under given limati onditions. With rainfall and potential ET,the soil water balane model takes into aount the water flowing into the rop s root zone from preipitation (assuming no irrigation) and the water leaving the rop s root zone due to evapotranspiration, surfae runoff and deep perolation. It then balanes these flows using the rop s root zone available water apaity (AWC) as storage. If the soil water ontent is above a threshold level of AWC then atual evapotranspiration takes plae at the potential rate. But if it is not, then atual evapotranspiration is stressed by soil moisture. Surfae runoff and deep perolation our if the estimated end-of-period soil water ontent exeeds AWC.Details of the soil water balane model and the rop water prodution model are given in Appendies A and B. There is a rih literature on rop yield responses to water defiits (e.g. Jensen, 1968; Doorenbos and Kassam, 1979). However, the hoie of whih rop water prodution models to use for eonomi evaluations is usually onstrained by the omplex nature of these models and the availability of model parameters. Thus, simpler estimation models, suh as the FAO linear yield response model (Doorenbos and Kassam, 1979) and Jensen s model (Jensen, 1968), are usually preferred to more omplex models. The FAO model is a pratial approah to measuring yield responses to water supply. It is relatively simple; requires ommonly available limati, water, soil and rop data; and is applied widely with aeptable levels of auray. 8 However, omparing observed yields and FAO model simulations for Zambia suggests that rop losses in the ountry are espeially sensitive to seasonal water defiits. Therefore, it is not appropriate to use the FAO model, whih uses relative evapotranspiration for the entire growing season, and thus annot apture monthly limate variations and rop yield sensitivities in Zambia. Instead we used the Jensen rop water prodution model 9. Crop water sensitivity index values were estimated using OLS regression and based on FAO yield response fators for eah growing stage (FAO, 1979). These were then mapped to eah month in rops growing periods using the umulative 8 Equation 2.8 in Appendix A. 9 Equation 2.9 in Appendix A. 16

17 sensitivity index method (Tsakiris, 1982; Kipkorir, 2002). Figure 3.1 shows the value of the monthly rop water sensitivity index for maize, sorghum and root rops. Figure 3.1: Crop water sensitivity index in Jensen rop water prodution funtion for maize, sorghum and root rops Maize Sorghum Root 0.0 Nov De Jan Feb Mar Apr 3.2 Impat of limate variability on rop yields HC model results Figure 3.2 shows the simulated atual evapotranspiration for the maize growing season in the five agroeologial zones. Modulated by rop water requirements and soil moisture ontents, atual evapotranspiration in the three drier zones (i.e., Zones I, IIa1 and IIb) show rising and downward patterns similar to the rainfall patterns in Figure 2.8. This is beause the limiting fator in these zones is the available soil water ontent. However, atual evapotranspiration in the wetter zones (i.e., Zones IIa2 and III) follow the potential evapotranspiration patterns in Figure 2.10, beause soil water ontent does not stress evapotranspiration in these zones (i.e., it is ontrolled by atmospheri water demand). Simulated relative maize yields for are shown in Figures 3.3 for the five agroeologial zones. Relative yields are defined as the ratio of the simulated atual yield to the maximum yield ahievable without any water stress. Maize is again hosen as an illustrative example beause it is Zambia s major rop. However, relative yields for other rops follow similar patterns to maize. For Zones I and IIa1, the worst maize yield redution ourred in the 1991/92 season when estimated relative yield losses were 77 perent and 65 perent, respetively, of normal yields. While Zone IIb had a 40 perent yield loss in 1991/92, its worst season was in 1994/95 when there was a 54 perent yield loss for maize. Considerable yield losses were also found in other seasons for the drier Zones I, IIa1 and IIb. These inlude 1994/95 for Zones I, IIa1 and IIb; 1976/77 for Zones IIa1 and IIb; and 1983/84, 1986/87 and 2001/02 for Zone IIa1. The relative yield redutions during these seasons range from 30 to 50 perent. Figure 3.3 also shows that there were no major yield losses aused by droughts in Zones IIa2 and III. For Zone III, in partiular, its highest relative yield redution was 14 perent in 1991/92. For Zone IIa2, the highest relative yield loss was 20 perent, ourring in 1988/89. In general, Zone IIa2 is muh less drought-prone than Zones I, IIa1 and IIb, and drought-indued rop yield losses in zone III are rare. 17

18 Figure 3.2: Growing period atual evapotranspiration of maize in agroeologial zones, Zones I, IIa1 and IIb ETa I IIa1 IIb 420 Zones IIa2 and III ETa Soure: Results from the hydro-rop model. IIa2 III 18

19 Table 3.1 provides information to help in understanding the nature and onsequenes of droughts/wet events. The table shows the ranges of growing season rainfall; relative yield losses; and the frequeny of weather events. It also shows the rop water requirement satisfatory index (WRSI), whih is the ratio of atual to potential evapotranspiration during the growing season (Verdin et al., 2005). Eah of these indiators is separated aross agroeologial zones and event ategories. For example, the range of growing season rainfall in Zone I for all severe drought years is mm. The table indiates that for the three drier zones (i.e., Zones I, IIa1 and 11b) there were only 7 or 8 years out of 32 years (i.e., ) that are identified as normal years in whih the amount of rainfall in growing seasons is in the normal range. This indiates that there is a 75 to 80 perent hane in any given year that there is a drought or too muh rain in these three zones. Moreover, over the 32 years, three severe droughts ourred in Zone I, four in Zone IIa1, and one in Zone IIb, and there was no severe drought in the other less dry zones (i.e., Zones IIa2 and IIb). During severe drought years, Zones I and IIa1 experiened the largest redutions in relative yields (i.e., and perent respetively). The average relative water defiit (i.e., WRSI) is usually not as high as relative yield losses during severe droughts. Rather it was abnormally low rainfall during ritial growing stages that aused substantial yield losses. Ultimately, Table 3.1 illustrates again that Zones I, IIa1 and IIb are drought-prone, while major drought damage is rare in Zones IIa2 and III. Despite being the wettest zone in the ountry, very wet weather event is also rare in Zone III, with only one ourrene during the study period. It should be noted that the models used in this study are for drought impat assessments, and therefore rop yield losses resulting from floods or water-logging are not assessed. This is primarily beause floods are usually highly loalized and with short durations. Hene, any damage assessment requires highresolution data that is not readily available in Zambia. However, drought damage is more important than flood damage for Zambian agriulture (World Bank, 2008). Moreover, the analysis of wet events in Tables 2.1 and 3.1 provides some measurement of possible flood events, and shows that rop yield losses an also our in wet years due to the uneven distribution of seasonal rainfall. 19

20 Figure 3.3: Maize relative yield in agroeologial zones, Zones I, IIa1 and IIb Relative Yield I IIa1 IIb 1.0 Zones IIa2 and III Relative Yield Soure: Results from the hydro-rop model. IIa2 III 20

21 Table 3.1: Palmer Z drought index-based weather lassifiation and the ranges of derived limati and agronomi statistis, Severe drought year (Z a -1.5) Moderate drought year (-1.5 < Z -0.5) Normal year (-0.5 < Z 0.5) Moderately wet year (0.5 < Z 1.5) Very wet year (Z > 1.5 ) Zone I Growing period rainfall (mm) Maize WRSI b (%) Maize yield loss (%) Frequeny d Zone IIa1 Growing period rainfall (mm) Maize WRSI (%) Maize yield loss (%) Frequeny Zone IIa2 Growing period rainfall (mm) e Maize WRSI (%) Maize yield loss (%) Frequeny Zone IIb Growing period rainfall (mm) Maize WRSI (%) Maize yield loss (%) Frequeny Zone III Growing period rainfall (mm) Maize WRSI (%) Maize yield loss (%) Frequeny Notes: (a) Average monthly Palmer Z Index in maize growing period (November-Marh). Palmer Z index shows how monthly moisture departs from normal onditions and reflets short-term drought or wetness (Palmer 1965; Alley 1984); (b) WRSI is water requirement satisfatory index and is the ratio of total atual to total potential rop evapotranspiration (see Setion 2.3); () Perentage maize yield loss estimated from hydrologial model; (d) Number of years of ourrenes during Using the monthly averaged Palmer Z index and the -1.5 threshold did not identify year 1992 as a severe drought year in Zone I where estimated maize yield redution in this study (about 77%) was the severest among all five agro-eologial zones, although the index suessfully identified the other severe drought years, 1995 and 2005 for zone I. The reason is that there was only 22 mm of rainfall fallen in February, 1992, whih is about 13% of average rainfall in this month. The extreme rainfall shortage ourring in ritial maize growing stage resulted in dramati yield loss in As a fix, we moved 1992 from the moderate drought ategory to severe drought for zone I. (e) Only one very wet event was found in zone IIa2 during the period so no need is reported. The same is true for severe drought in Zone IIb and very wet event in Zone III. 21

22 4. Climate variability and eonomi growth ombining the HC and DCGE models The previous setions examined historial trends in limate variability aross Zambia s five major agroeologial zones (Setion 2) and estimated their annual impat on rop yields over the last three deades using a hydro-rop model (Setion 3). Drawing on the hydro-rop model results and limate variability data, this setion assesses the potential impat of future limate variability on eonomi growth and household inomes and poverty using the dynami CGE model developed for this study. 4.1 The dynami omputable general equilibrium (DCGE) model Climate variability not only affets rop yields, harvested areas, and agriultural prodution, but also has impliations for the entire eonomy. Moreover, spatial variation in rainfall patterns means that suh effets vary aross sub-national regions. We therefore developed an eonomywide model with five agroeologial zones. 10 The model ontains detailed information on prodution, onsumption and trade. There are totally 34 different prodution setors, half of whih are agriultural rops and livestok. These setors are defined for the five agroeologial zones as five representative rural farm groups, together with large-sale and urban farm groups. The tehnologies of these representative farm groups (together with the nonagriultural prodution funtions) are alibrated using distrit level prodution data from the Crop Foreasting Surveys (CFS) and the 2004 Living Conditions Monitoring Survey (LCMS4). Eah farm group an hange ropping and livestok patterns and engage more or less intensively in nonfarm ativities. Laborers an also migrate to urban enters and nonagriultural jobs (both are modeled expliitly). This model setup therefore aptures some autonomous adaptation to limate hange. We do, however, limit the extent to whih representative produers adjust ropping patterns in response to short-term limate variability. We assume that land alloations are determined at the start of the season, and, one planted, farmers annot realloate land to different rops during the growing season (i.e., land alloations in short-run are exogenous). This assumption is appropriate sine farmers typially annot predit and respond to limate variation one land in planted. They an, however, realloate mobile resoures, suh as labor, and influene the level of prodution. While substitution between fators (labor, land and apital) depends on relative osts, the model distinguishes between self-employed agriultural workers; unskilled workers (working in both agriulture and non-agriulture) and skilled workers (in non-agriulture only). Information on employment and wages is from LCMS4, while labor supplies expand over time exogenously aording to demographi projetions. Capital is immobile aross setors, and, after aounting for annual depreiation, is supplemented by past investments alloated aording setors relative profitability. This is the reursive dynami feature of the DCGE model. Total fator inputs are then ombined with intermediate inputs (e.g., fertilizer and fuels) to produe a total level of output. These setor-speifi prodution tehnologies are taken from a 2006 Zambian soial aounting matrix (SAM). Produers in eah setor/region deide how muh output to supply to national domesti and foreign markets based 10 The eonomywide model used is a reursive dynami omputable general equilibrium (DCGE) model developed by Thurlow (2005). Appendix C presents the equations and parameters of the DCGE model. 22

23 on relative pries. Sine Zambia is a small ountry, we assume that world pries are fixed, but are influened by hanges in the real exhange rate, whih adjusts to maintain the urrent aount balane. In order to apture how different households are affeted by limate variability and hange, the model inludes 15 representative household groups by zones, rural and urban, and by the size of farms (small rural, large-sale rural and urban farmer households). Inome and expenditure patterns vary aross the 15 household groups. This is important for distributional hange sine inomes generated during farm/nonfarm prodution arue to different households depending on their loation and fator endowments. Households in the model reeive fators inomes and then pay taxes, save and make transfers to other households. Remaining inomes are used to onsume goods and servies, whih an either be purhased loally or imported depending on relative pries. Taxes are olleted by the government, whih also onsumes goods and servies, while pooled savings are used to finane investment spending. Total demand interats with prodution and trade (i.e., supply) to determine pries in the model. This full speifiation of supply/demand for ommodities and fators using prodution and utility funtions is the general equilibrium feature of the DCGE model. Finally, in order to retain as muh information as possible on households inome and expenditure patterns, the DCGE model is linked to a mirosimulation module based on LCMS4. Changes in onsumption for eah representative household in the DCGE model are used to adjust the level of ommodity expenditure of orresponding households in LCMS4. Consumption levels are then realulated in the survey, and standard poverty measures are re-estimated. The model distinguishes domesti markets from trade with rest of world for the most important trading ommodities. However, international pries an affet domesti supply and demand through imperfet substitution with domesti supply as well as through the real exhange effet. The linkages among different agriultural subsetors and between agriultural and nonagriultural setors are fully aptured through inome generation and household expenditures and through use of intermediate inputs. Beause suh model setup, the eonomywide impats of limate variability in Zambia an be aptured. Detailed desription of data applied to represent the Zambian eonomy in 2006, the maroeonomi and setoral struture of the prodution, is taken from national aounts and rop prodution data, while information on household inomes and expenditures is drawn from 2004 Living Conditions Monitoring Survey. Note that the agriultural season used for the alibrated base year of the model is 2005/06. Appendix B provides a detailed desription of 4.2 Combining the HC and DCGE models and designing senarios of alternative rainfall sequenes The DCGE model is used to simulation of 10 years period of limate variability between 2006 and 2016 and to assess the eonomi impat of suh variability in this period. It is impossible to aurately predit Zambia s future annual rainfall pattern. It is, however, possible to identify a range of possible patterns using historial data. Many methods have been developed to simulate the future potential rainfall patterns. In our study we use an 'index sequential method (Prairie et al, 2006). Given 32 years of historial rainfall data for the period , we an draw 32 different ten-year rainfall sequenes 23

24 (i.e., 32 different starting years for eah ten-year onseutive sequene). This method preserves the original inter-annual orrelations inherited in the observed limate data. We inlude all 30 of the possible sequenes in our analysis and thus apture the full distribution of past limate variability. This approah has the advantage of greatly reduing the number of possible rainfall senarios, while also apturing any interdependenies between rainfall patterns aross onseutive years. Thus, the starting point (or the base-year) of the DCGE model is 2006, whih is the most reent year for whih all of the neessary data is available and all model s parameters thus are alibrated to this year.we first apply the DCGE model to a senario with normal rainfall for eah of the ten years between 2006 and 2016 (i.e., during the simulated 10 years there are no adverse effets from limate variability on the eonomy). Beause of this insulation from limate variability, rop yields are assumed to be stable. Moreover, the level of yields and the alloation of land are assumed to grow steadily aording to estimated yield potentials drawn from field trials and historial trends in land expansion (see Thurlow et al., 2008). We all this senario normal rainfall without limate variability (or normal rainfall in short). We then develop 30 senarios eah of whih has a period of 10 years. These senarios are developed to simulate the eonomi impat of different rainfall patterns disussed above (i.e., using the index sequential method to reprodue various historial rainfall patterns for the next 10 years). The link between various rainfall patterns and the eonomywide model is through the imposition of rop yield shoks in the model. These yield shoks are onsistent with the hydro-rop model results disussed in the previous setion. This is done by rops for eah of the five agroeologial zones and is repeated for all 30 ten-year rainfall sequenes. As rainfall patterns are unpreditable for farmers and droughts often our after planting, farmers usually are unable to hange their rop land alloation to avoid yield loss in response to drought. Aordingly, in the model we assume that land alloation within eah year is fixed by rop. Given that more than 80 perent of rain-fed rop areas are for maize, and droughtresistant rops, suh as assava and sorghum, have signifiant spatial patterns, this assumption seems to be more reasonable than a flexible land alloation assumption that is ommonly used in many other CGE models. While farmers are unable to realloate land into other rops, they still an deide to realloate other inputs, inluding labor, apital and intermediate inputs in response to limate variability. For example, labor and apital may swith into other agriultural and nonagriultural ativities, suh as partiipating in some off-farm ativities as workers, in response to hanges in agriultural pries and nonagriultural employment opportunities resulting from the drought. Apart from imposing the yield shoks drawn from the rop model, we also aount for other transmission hannels through whih extreme rainfall variation affets the agriultural setor and the rest of the eonomy. These additional non-yield impats only our when there is a severe drought or flood event. 11 The various impat hannels are summarized in Table Major drought events are defined for a partiular agro-limati zone as being years in whih the water response index is two standard deviations or more below the mean for the period of 30 years ( ). For Zone 2a1 this ourred in the following seasons: 1983/84, 1986/87, 1991/92, 1994/95 and 2001/02. Due to a lak of data to support the modeling of floods, only the 2007/08 season is identified as a major flood event. 24

25 Table 4.1: Climate variability and severe drought/flood event impat hannels assumed in the eonomywide model Impat hannel Affeted setors Desription of impat All 10 years in eah of the 30 sequenes Crop yields Rainfed rops Level of yield redued based on the rops WRSI Years with severe drought events that reflet rainfall patterns in 1983/84, 1986/87, 1991/92, 1994/95 and 2001/02 Crop area expansion Rainfed rops Crop land expansion that would take plae in a normal year is eliminated in the drought year and remains at zero in the immediate post-drought year Livestok stoks Livestok setors Livestok stoks deline in the drought year and the growth in these stoks gradually returns to normal year rates with diminishing lagged effets for two subsequent years Physial apital aumulation All setors Capital depreiation rates are inreased in the drought year and gradually return to normal year levels with diminishing lagged effets for two subsequent years Major flood year (2006/07) Agriultural land expansion Crop setors Land area under ultivation delines in the flood year and only returns to preflood levels in the immediate post-flood year Note: WRSI is water response satisfatory index (see Setion 2.3). Details of eah impat hannel are provided in Appendix A. 25