Scholarly article on topic 'Prioritizing climate-smart livestock technologies in rural Tanzania: A minimum data approach'

Prioritizing climate-smart livestock technologies in rural Tanzania: A minimum data approach Academic research paper on "Agriculture, forestry, and fisheries"

CC BY-NC-ND
0
0
Share paper
Academic journal
Agricultural Systems
OECD Field of science
Keywords
{"Trade-off analysis" / "Crop-livestock systems" / Tanzania / "Ruminant model" / "Climate-smart agriculture" / "Food security"}

Abstract of research paper on Agriculture, forestry, and fisheries, author of scientific article — Kelvin M. Shikuku, Roberto O. Valdivia, Birthe K. Paul, Caroline Mwongera, Leigh Winowiecki, et al.

Abstract Crop-livestock production systems play an important role in the livelihoods of many rural communities in sub-Saharan Africa (SSA) but are vulnerable to the adverse impacts of climate change. Understanding which farming options will give the highest return on investment in light of climate change is critical information for decision-making. While there is continued investment in testing adaptation options using on-farm experiments, simulation models remain important tools for ‘ex-ante’ assessments of the impacts of proposed climate-smart agricultural technologies (CSA). This study used the Ruminant model and the Trade-offs Analysis model for Multi-Dimensional Impact Assessment (TOA-MD) to assess how improved livestock management options affect the three pillars of CSA: increased productivity, improved food security, and reduced greenhouse gas (GHG) emissions. Our sample was stratified into: 1) households with local cow breeds (n =28); 2) households with improved dairy cow breeds (n =70); and 3) households without dairy cows (n =66). Results showed that the predicted adoption rates for improved livestock feeding among households with improved dairy cows (stratum 2) were likely to be higher compared to households with only local cows (stratum 1). Both households with local cows and those with improved cows had increased income and food security. However, overall poverty reduction was only modest for households with local cows. Expected methane emissions intensity declined with adoption of improved livestock feeding strategies both in stratum 1 and stratum 2, and greater impacts were observed when households in stratum 2 received an additional improved cow breed. Providing a cow to households that were not keeping cows showed substantial economic gains. Additional research is, however, needed to understand why those farms currently do not have cows, which may determine if the predicted adoption rates are feasible.

Academic research paper on topic "Prioritizing climate-smart livestock technologies in rural Tanzania: A minimum data approach"

ARTICLE IN PRESS

AGSY-02140; No of Pages 13

Agricultural Systems xxx (2016) xxx-xxx

Prioritizing climate-smart livestock technologies in rural Tanzania: A minimum data approach

Kelvin M. Shikuku a,d'*, Roberto O. Valdiviac, Birthe K. Paula,d, Caroline Mwongera a, Leigh Winowieckie, Peter Läderach b, Mario Herrerof, Silvia Silvestrig

a International Center for Tropical Agriculture (CIAT), Kenya b International Center for Tropical Agriculture (CIAT), Colombia c Oregon State University, United States d Wageningen University, The Netherlands e World Agroforestry Research Center (ICRAF), Kenya

f Commonwealth Scientific and Industrial Research Organization (CSIRO), Australia g Center for Agriculture and Biosciences International (CABI), Kenya

ARTICLE INFO ABSTRACT

Crop-livestock production systems play an important role in the livelihoods of many rural communities in sub-Saharan Africa (SSA) but are vulnerable to the adverse impacts of climate change. Understanding which farming options will give the highest return on investment in light of climate change is critical information for decisionmaking. While there is continued investment in testing adaptation options using on-farm experiments, simulation models remain important tools for 'ex-ante' assessments of the impacts of proposed climate-smart agricultural technologies (CSA). This study used the Ruminant model and the Trade-offs Analysis model for MultiDimensional Impact Assessment (TOA-MD) to assess how improved livestock management options affect the three pillars of CSA: increased productivity, improved food security, and reduced greenhouse gas (GHG) emissions. Our sample was stratified into: 1) households with local cow breeds (n = 28); 2) households with improved dairy cow breeds (n = 70); and 3) households without dairy cows (n = 66). Results showed that the predicted adoption rates for improved livestock feeding among households with improved dairy cows (stratum 2) were likely to be higher compared to households with only local cows (stratum 1). Both households with local cows and those with improved cows had increased income and food security. However, overall poverty reduction was only modest for households with local cows. Expected methane emissions intensity declined with adoption of improved livestock feeding strategies both in stratum 1 and stratum 2, and greater impacts were observed when households in stratum 2 received an additional improved cow breed. Providing a cow to households that were not keeping cows showed substantial economic gains. Additional research is, however, needed to understand why those farms currently do not have cows, which may determine if the predicted adoption rates are feasible.

© 2016 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license

(http://creativecommons.org/licenses/by-nc-nd/4.0/).

Article history: Received 15 August 2015 Received in revised form 31 May 2016 Accepted 7 June 2016 Available online xxxx

Keywords: Trade-off analysis Crop-livestock systems Tanzania Ruminant model Climate-smart agriculture Food security

1. Introduction

In sub-Saharan Africa (SSA), climate change is projected to have a negative impact on smallholder livestock production systems, which play an important role in the livelihoods of rural communities (Tubiello et al., 2007; Thornton et al., 2009). The risks posed by climate change are more severe for populations most dependent on crop and livestock production for overall household food security (Battisti and Naylor, 2009). Further concerns have been raised about the negative environmental impacts of increased greenhouse gas (GHG) emissions, for

* Corresponding author at: International Center for Tropical Agriculture (CIAT), Kenya. E-mail address: k.m.shikuku@cgiar.org (KM. Shikuku).

example on exacerbating the subtropical drought occurrences (IPCC, 2007; Easterling et al., 2007). Addressing climate change has, therefore, become tremendously urgent both from an adaptation as well as a mitigation perspective (Vermeulen et al., 2013; Paris Agreement (http://unfccc.int/paris_agreement/items/9485.php), IPCC).

Human population in SSA is expected to more than double to 2.4 billion by 2050 (UN, 2015). Furthermore, demand for animal products is expected to increase given not only the growth in population, but also higher incomes, increased urbanization, and change in dietary preferences (Delgado et al., 1999; Thornton et al., 2007). The rising demand for livestock products will require an associated increase in farming systems productivity (Staal et al., 2001). Current livestock systems in SSA have low overall performance, which results in low herd productivity

http://dx.doi.org/10.1016/j.agsy.2016.06.004

0308-521X/© 2016 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

ARTICLE IN PRESS

K.M. Shikuku etal. / Agricultural Systems xxx (2016) xxx-xxx

and high GHG emission intensity (Herrero et al., 2015). The reasons for this are mainly associated with poor feed quality, which in turn determines low digestibility, and poor animal health (Herrero et al., 2015).

Scarcity of resources, impacts of climate change, and the increase in demand for livestock products have made some traditional coping mechanisms less effective (Sidahmed, 2008). Increased resource-use efficiency is, therefore, a necessary component for environmental sustain-ability of the livestock sector (FAO, 2013). There is an urgent need to develop and implement 'climate-smart' livestock management options that can achieve the triple win scenario of increasing productivity, adapting and building resilience to climate change, and mitigating climate change through reduction of GHG emissions (FAO, 2013; Steenwerth et al., 2014; Lipper et al., 2014). Improved livestock breeds, together with improved forage quality and diet supplementation are among the livestock management strategies that have high potential to address poor livestock system performance and reduce emissions intensity (FAO, 2013). With well-functioning markets and improved livestock management practices, incentives and opportunities can be created for farmers and other stakeholders to invest in adequate livestock intensification. Adoption of improved livestock management practices might, however, present trade-offs, if for example, income increases and poverty declines but net GHG emissions increase as households shift from no-livestock systems to livestock keeping. Considering the complexity of livestock systems in developing countries, Thornton et al. (2009) suggested that improved management strategies should be based on a combination of factors, including feed and nutrition, genetics and breeding, health and environment, with different combinations for different systems.

The objectives of this study were: (i) to analyze the synergies and trade-offs of adopting improved feeding practices and livestock breeds in Lushoto district of Tanzania; and (ii) to assess how alternative livestock management options can increase productivity, improve food security, and reduce methane emissions intensity. The study used household surveys, stakeholder consultations, and secondary data as

inputs into livestock and economic models to assess trade-offs of economic and environmental outcomes of improved livestock feeding strategies and improved breeds.

2. Methodology

2.1. Study area

The study was conducted in Lushoto district of Tanzania, which lies within the western Usambara Mountains in northeastern Tanzania (Fig. 1). This region has a unique and diverse history as it is one of the 25 global biodiversity hotspots due to the high number of endemic species (Critical Ecosystem Partnership Fund, 2005) while also having a rich agricultural history of cultivation on the steep mountainous slopes (Jambiya, 1998). The major economic activity in the study area is agriculture, and major crops include: maize, beans, potatoes, cassava, vegetables (such as tomatoes, cabbages, peppers), coffee and temperate fruits (such as avocados and peaches) (Silvestri et al., 2015; Lyamchai et al., 2011). Farming systems across Lushoto district are diverse, and include mixed crop-livestock systems, intensive horticultural systems, extensive cereal systems, perennial cropping systems such as coffee, among others (Kristjanson et al., 2012, Lyamchai et al., 2011). However, off-farm employment continues to be an important income-generating activity (Jambiya, 1998; Lyamchai et al., 2011). Within Lushoto district, keeping livestock is a common household practice, with about 84% of households producing small livestock (goats and sheep) and 43% producing large livestock (cattle), and 45% producing fodder (Lyamchai et al., 2011). For decades, Lushoto district has received national and international attention to curb the high soil erosion rates and conserve indigenous forests (Watson, 1972; Tenge et al., 2005). However, low productivity, small farm sizes, high soil erosion rates, and migration out of farming continue to challenge farming systems in Lushoto (Tenge et al., 2005; Okoba et al., 2007; Foerch et al., 2011; Wickama et al., 2014; Winowiecki et al., 2015).

Fig. 1. Study site: Lushoto district, Tanzania.

ARTICLE IN PRESS

K.M. Shikuku et al. / Agricultural Systems xxx (2016) xxx-xxx

2.2. Household survey data collection

This study used household survey data collected in 2012 under the CGIAR research program on Climate Change, Agriculture, and Food Security (CCAFS) (Rufino et al., 2012). Data were collected using a stratified sampling strategy, described in detail by Rufino et al. (2012). Three main agricultural production systems were used to stratify the sampling design within Lushoto district: i) farm households growing staple crops only; ii) farm households growing staple crops plus horticultural crops such as tomatoes and cabbages; and iii) farm households growing staple crops plus cash crops such as coffee and tea. Within each of the identified production systems, seven representative villages were randomly selected. A total of200 households were sampled from the selected villages. Household interviews were conducted to capture data on: household composition; crop and livestock production activities; household assets; agricultural inputs and labour use for cropping and livestock activities; utilization of agricultural products including sales and consumption; and off-farm employment and other sources of livelihoods such as remittances and subsidies. The final sample size used for analysis was 164 farm households for which data was complete to estimate the simulation models.

Given the evidence that livestock ownership is correlated with income levels (Claessens et al., 2012; Homann-Kee Tui et al., 2016), we stratified our sample of farm households into three strata according to the type of livestock owned: i) stratum 1 comprised of households with local cow breeds only; ii) stratum 2 comprised of households with improved cow breeds only; and iii) stratum 3 comprised of households without cows. The number of farm households that had both local and improved breeds in our sample was very small hence this category was not included in the analysis. Out of the 164 farm households, 17% owned local cows only, 43% owned improved breeds only, and 40% did not own cows.

2.3. Ruminant model and model inputs

The Ruminant model (Herrero et al., 2002) was used to simulate milk production and methane (CH4) emissions from enteric fermentation. A dynamic component of the model estimates feed intake and supply of nutrients to the animal from knowledge of the fermentation kinetics and passage of feed constituents (carbohydrate and protein) through the gastrointestinal tract. A static component of the model determines the animal's response to nutrients in terms of growth and milk production. Validations have been carried out for more than 80 tropical and temperate diets and the results suggest that the model has the required accuracy not only as a research tool but also for providing decision support at the farm level (Herrero, 1997). Among other uses, the model has previously been applied to estimate CH4 emission factors of tropical livestock (Herrero et al., 2013), and in various modeling studies across SSA (Bryan et al., 2013; Waithaka et al., 2006; Zingore et al., 2009).

Feeding data from the household survey could not be used to construct individual diets per household due to the lack of high quality information on quantity and types of feeds throughout the year. Therefore, two uniform baseline diets ('dry' and 'wet' seasons) were constructed for local and crossbred dairy cattle for the entire study population. The diets were based on previous participatory, community level feeding system assessments (Morris et al., 2014; Mangesho et al., 2013).

The main livestock system in the area was intensive cut and carry (zero-grazing), where livestock is kept in pens year round. All collected and purchased feed is provided in situ to the cows (Morris et al., 2014). The different feed baskets for the dry and wet seasons were recorded. There are two rainy seasons lasting for a total of eight months, while the dry seasons stretch over four months of the year. Naturally occurring green forages are collected from roadsides and constitute the primary component throughout the year. Maize bran was the only

purchased feed in the area and was only used in small amounts (500 g day-1) by about 30% of farm households throughout the year, although cotton seed cake and sunflower cake were also locally available. Another source of feed was cultivated fodder, mainly Napier grass (Pennisetum purpureum), although the acreage was low due to land shortage and lack of knowledge on cultivated fodder. Seasonality of feed availability is high — especially cultivated fodder and natural forages, and maize residues which is most abundant only after harvest. Overall, cows were underfed due to land shortage for growing and collecting fodder, and total fresh weight of daily feed was estimated to lie between 40 and 60 kg. Farmers did not report different feeding for local or crossbred cows (Mangesho et al., 2013).

Table 1 summarizes the diets used in the Ruminant model. In the dry season, the diet comprised approximately 20% Napier grass (P. purpureum), 35% collected natural grasses, 40% maize residues, and 5% maize bran concentrate. The total amount of feed corresponded to 12.9 kg dry weight (DW) per cow per day. For the wet season, the diet comprised 40% Napier grass, 40% collected grass, 15% maize residues, and 5% maize bran concentrate, totaling 14.4 kg DW per cow per day. For the Ruminant model, local cow body weight was set to 250 kg, while a crossbreed was assumed to be 350 kg.

Farmers of village innovation platforms in Lushoto identified growing improved forages and supplementing with concentrate feeds as two of the most promising interventions to improve their livestock feeding, in addition to adequate feed rations and feed conservation (Lukuyu et al., 2015). Based on these findings, we developed two livestock feeding options to be simulated using the Ruminant model (Table 1). The design of the livestock feeding options was based on the kinds of data available. We linked qualitative and quantitative information to generate a minimum dataset that allowed us to use the TOA-MD model. The first feeding option (hereafter option 1) represented an improvement in feed quality, while keeping the feed quantity almost constant. We assumed that the contribution of Napier grass to the diet increased from 21.7% to 31% in the dry season and to 54% in the wet season. In the dry season, the increase in Napier grass was expected to reduce the contribution of low quality natural grasses from 40% to 28% and that of maize residues from 38.8% to 27%. In the wet season, the increase in Napier was expected to reduce the contribution of low quality natural grasses from 40% to 27% and the one of maize residue from 14% to 7%. We further increased the contribution of concentrate feeding from 6% to 14% in the dry season and from 5.5% to 12% in the wet season, adding higher quality sunflower seedcake to the mix.

The second feeding option (hereafter option 2) assumed an improvement of the feed quality in the diets similar to the first feeding scenario, and an increase in the quantity from 12.9 kg to 17.0 kg dry matter (DM) in the dry season and 14.4 kg to 21.8 kg DM in the wet season per cow and day. Under this scenario, we further increased the contribution of protein concentrates from 6.2% to 24% in the dry season and from 5.5% to 19% in the wet season. Maize bran was combined with cotton seed instead of sunflower seed in this scenario.

In order to deal with uncertainty and lack of heterogeneity in the feeding data, we conducted sensitivity tests where we assumed an increase and decrease in the total quantity of feed in the three diets (baseline and the two feeding options). We kept the percentage contributions of the different feed constituents constant (Tables 7, 8, 9, and 10). We assumed that any additional inputs to livestock production (including Napier grass) would be purchased and not produced on-farm due to small sizes of land and farm orientation for self-consumption of crops. Therefore, it was assumed that there would be no changes to land allocation.

2.4. Trade-off Analysis for Multi-Dimensional Impact Assessment (TOA-MD) model

The Trade-off Analysis model for Multi-Dimensional Impact Assessment (TOA-MD) is a parsimonious model that is used to simulate

ARTICLE IN PRESS

4 K.M. Shikuku etal / Agricultural Systems xxx (2016) xxx-xxx

Table 1

Livestock diet data used in the Ruminant model.

Dry season (4 months) Wet season (8 months)

Baseline Improved quality Improved quality of diet plus Baseline Improved quality Improved quality of diet plus of diet increased quantity of feed of diet increased quantity of feed

Base scenario

Local grass (kg DM/day) 4.50 3.60 4.50 5.80 4.00 5.80

Napier grass (kg DM/day) 2.60 4.00 5.00 5.80 8.10 11.00

Maize residue (kg DM/day) 5.00 3.50 3.50 2.00 1.00 1.00

Maize bran (kg DM/day) 0.70 1.50 3.00 0.70 1.50 3.00

Sunflower cake (kg DM/day) 0.10 0.30 0.00 0.10 0.30 0.00

Cotton cake (kg DM/day) 0.00 0.00 1.00 0.00 0.00 1.00

Sensitivity analysis with +10% variation

Local grass (kg DM/day) 4.95 3.96 4.95 6.38 4.40 6.38

Napier grass (kg DM/day) 2.86 4.40 5.50 6.38 8.90 12.10

Maize residue (kg DM/day) 5.50 3.85 3.85 2.20 1.10 1.10

Maize bran (kg DM/day) 0.77 3.30 3.30 0.77 1.65 3.30

Sunflower cake (kg DM/day) 0.11 0.00 0.00 0.11 0.33 0.00

Cotton cake (kg DM/day) 0.00 1.10 1.10 0.00 0.00 1.10

Sensitivity analysis with —10% variation

Local grass (kg DM/day) 4.05 3.24 4.05 5.22 3.60 5.22

Napier grass (kg DM/day) 2.34 3.60 4.50 5.22 7.30 9.90

Maize residue (kg DM/day) 4.50 3.15 3.15 1.80 0.90 0.90

Maize bran (kg DM/day) 0.63 1.35 2.70 0.63 1.35 2.70

Sunflower cake (kg DM/day) 0.09 0.27 0.00 0.09 0.27 0.00

Cotton cake (kg DM/day) 0.00 0.00 0.90 0.00 0.00 0.90

Note: Base runs of baseline and scenario diets were repeated with a +10% and -10% variation to test sensitivity of the approach.

potential technology adoption rates, ex-ante impact assessment, and ecosystem services analysis, across heterogeneous farm populations and for different types of households (Antle and Valdivia, 2011; Antle et al., 2014). The TOA-MD model has key features that make it appropriate for assessment of technologies for climate smart agriculture (CSA). The model represents the whole farm production system which can be composed of a crop sub-system containing multiple crops, a livestock subsystem with multiple livestock species, an aquaculture sub-system with multiple species, and the farm household characteristics (e.g. household size and off-farm income). Furthermore, the TOA-MD is a model of a population of farms, not a model of an individual or "representative" farm. Accordingly, the fundamental parameters of the model are population statistics — means, variances and correlations of the economic variables in the models and the associated outcome variables of interest. With suitable biophysical and economic data, these statistical parameters can be estimated for an observable production system. Another unique feature of the TOA-MD model is its parsimonious, generic structure, which means that it can be used to simulate any farm system. One virtue of this model design is that, unlike many large, complex simulation models, it is easy to address the inherent uncertainty in impact assessments by using a set of minimum data and sensitivity analysis to explore how results change with the relatively small number of model parameters (Antle et al., 2010).

The TOA-MD model simulates the proportion of farms that utilizes a baseline system (e.g. system 1) and the proportion of farms that would adopt an alternative system (e.g. system 2) within defined strata of the population. The model then predicts an adoption rate for each stratum of the population, using the assumption that farmers are economically rational and adopt practices that are expected to provide the highest economic return. Accordingly, this predicted adoption rate should be interpreted as the proportion of farms for which the new system's practices are economically feasible, after correcting for the opportunity costs associated with the technology (Antle and Valdivia, 2011). Positive opportunity costs of an improved technology would discourage adoption. If there are institutional or behavioral factors that constrain adoption -such as limited access to financial resources, and risk aversion - then this predicted adoption rate is likely to be an upper bound on the actual adoption rate that is observed. Based on the predicted rate of adoption,

the TOA-MD model also simulates economic, environmental and social impact indicators for the sub-population of adopting farms, the sub-population of non-adopters, and the entire population. The TOA-MD can also simulate supply curves for ecosystem services associated with agricultural systems and assess impacts of environmental change, such as climate change, with or without adaptation. Further details on the impact assessment aspects of the model are provided in Antle (2011) and Antle et al. (2014).

We applied the TOA-MD following the approach described by the Agricultural Model Inter-comparison and Improvement Project (Rosenzweig et al., 2015; AGM1P, www.agmip.org) to integrate livestock simulation models with economic models for impact assessment (Antle et al., 2015a). Returns to crop production for the baseline system were calculated as a product of amount produced and the farm-gate price of the crop product, and summed over all crops that the household produced. Prices for crop and livestock products were derived from the median of estimated village prices by farmers and verified by comparing with observed prices in the village. Variable cost of crop production included farmers' estimates of cash expenses for production during the observed year, including land preparation and farm inputs (herbicides, pesticides, seeds, and fertilizer). Returns to livestock production were separated into two: one for milk production and another one for other livestock products (manure and eggs). Revenue from milk was calculated from the average milk yield per cow, the number of cows, a lactation period of 150 days for local cows and 270 days for improved breeds, and the village price for milk. The value of manure was calculated from the amount of manure produced (kg/per farm), adjusted by a utilization factor of 0.7 to account for the proportion of manure used for fertilizing the fields (Homann-Kee Tui et al., 2015), and village price for manure. The value of the number of animals sold, given away and consumed was calculated based on village prices. Variable cost of livestock production comprised the cost of feeding and pests control (both internal and external pests). Our analysis assumed a feeding season length of eight months in the wet season and four months in the dry season. The economic indicators used in this paper are: farm income (USD/year) and the income-based poverty rate, defined as the proportion of the population living under 1.25 USD/day/person. The environmental indicator is the methane emission intensity (CH4/1 of milk per year) and was obtained from the Ruminant model output.

ARTICLE IN PRESS

K.M. Shikuku et al. / Agricultural Systems xxx (2016) xxx-xxx

2.5. Income-based food security indicator

Changes in technology that would improve crop and livestock productivity and farm incomes have a direct effect on food security. Several studies have used a range of food security indicators based on calories, dietary diversity, or wealth (e.g. asset ownership) but focused mostly to access and availability. Antle et al. (2015b) proposed a new method for measuring food security that can be used for technology impact assessment and can be constructed with data usually available. The Income-Based Food Security Indicator (IBFS) is similar to the Foster-Greer-Thorbecke (FGT) poverty indicators, but instead of comparing income to a poverty line, the IBFS estimates the income required to purchase a food basket that meets nutritional requirements and then compares this food security income requirement to the household's per capita income (Antle et al., 2015b). In this study we use the IBFS indicator to assess the impact of improved livestock feeding scenarios on food security. Following Antle et al. (2015b), we estimated the IBFS following three steps. First, the share of income devoted to food purchases was obtained from existing literature and specified as a parameter, Sf. Second, the cost (Cf) of a nutritionally adequate food basket was estimated using data on per capita nutritional requirements (macronutri-ents and micronutrients), typical food consumption patterns, and the nutritional content of foods. Third, the IBFS threshold was defined as Tf = Cf/Sf, indicating the amount of income needed per person for the purchase of a nutritionally adequate diet. If a household's per capita income is less than Tf, then the household cannot afford a nutritionally adequate food basket. We then used the TOA-MD to estimate the potential adoption of the proposed technologies and the changes in per-capita and farm income. The model then uses this information and the IBFS threshold to estimate the percentage of food insecure households.

2.6. Characterizing the farming systems in Lushoto

In the Tanzanian farming systems considered in this study, the current or base system (from now called System 1) consisted of households that were not feeding livestock on improved diets. System 1 was parameterized using the distribution of farm characteristics observed in a subset of the population including farm size, household size, off-farm income, and net returns from crop and livestock activities. Farmers in our sample produced several staple crops (maize, beans, cassava, yams, potatoes); horticultural crops (cabbages, tomatoes, peppers, cucumbers, onions), and cash crops (coffee and tea). In the base system, current livestock feeding (for farms that owned cattle) included local grass, Napier grass, and maize residue. Herd composition and herd size remained unchanged in the base system. The average milk yield was about 2 l/day for farms with local cows and 5 l/day for farms with improved breeds.

The other important parameters in the TOA-MD model were the correlations between the economic returns to the activities within each system and the correlation between the returns of the base system and the alternative system (Antle, 2011). We parameterized System 2 (i.e. the alternative or proposed technology) using simulated milk yields from the Ruminant model. First, we calculated relative yields - a ratio of simulated yield with improved feeding or improved feeding with an improved breed to simulated yield with current feeding. The relative yields were then multiplied by the actual milk output for System 1 to obtain milk output for System 2 (Antle et al., 2015a,b). Multiplying the milk output by the average market price of milk, we obtained returns from milk for System 2. In order to calculate the cost of milk production for System 2, we first took the ratio of cost of milk production in System 1 to returns from milk production for System 1 and multiplied this adjustment factor with the returns to milk production from System 2. We acknowledge that using average price and livestock production cost data reduces heterogeneity in returns to milk production. There is, however, heterogeneity in herd sizes as well as in other livestock production activities. Our approach therefore follows the minimum data approach

as described by Antle and Valdivia (2006) using the TOA-MD to simulate economic and environmental impacts. Methane emissions for TOA-MD simulations were calculated using Ruminant model's input data on emissions, the lactation period, and the total quantity of milk produced.

2.7. Scenario design

Using the livestock feeding options described in Section 2.3, we developed a set of scenarios that combined the livestock feeding strategies with increasing herd size and different assumptions regarding the cost of acquiring additional cows. Construction of these scenarios was informed by existing evidence to indicate that with market access and improved infrastructure, improved breeds produce higher milk yields resulting in positive economic gains for farmers and improvement in food security (Henderson et al., 2016). Design of the scenarios was also consistent with efforts by development projects such as the East Africa Dairy Development project (EADD) that promote adoption of improved breeds and improved livestock feeding accompanied by better market access and infrastructure. Economic modeling, therefore, began by assessing economic and environmental impacts of the two options considered in the Ruminant model, that is, improved quality of livestock diets only and improved quality of diets plus increased feed amounts. Because of the two options considered in Ruminant model, estimations of economic and environmental impacts were conducted separately for 1) improved feeding quality only and 2) improved feeding quality with increased livestock feed quantity. The analysis was then extended to include five additional scenarios. In scenario 2, farm households in all the three strata received an improved cow breed and the cost was subsidized at 100%, that is, farm households did not pay the purchasing price for the cow. Scenario 3 involved provision of an improved cow breed to farm households and requiring them to pay 25% of the purchase price. Similar to scenario 3, scenarios 4 and 5 provided farm households with an improved cow breed for which they paid one-half and three-quarters of the purchase price, respectively. The last scenario (scenario 6) provided an improved cow breed to farm households in all the three strata for which they paid full price but received a loan at 18% interest rate with a repayment period of three years. Table 2 presents the different scenario considered in the economic analysis.

3. Results: impact assessment of adoption of improved livestock systems

Table 3 displays summary statistics by strata for farms in our sample. These statistics are the base for the TOA-MD model parameters. The average household size was relatively lower in stratum 3 (4.26) compared with stratum 1 (5.00) and stratum 2 (4.96). On average, stratum 3 had the highest proportion (68%) of the population that had earned income from non-agricultural sources. Farm households in stratum 1 earned, on average, 47.5% more income from non-agricultural sources than those in stratum 1 and 7.2% higher than those in stratum 2. Net crop returns was, however, about 46% lower in stratum 3 compared to stratum 1 and 14.5% lower compared to stratum 2. Stratum two had 90% higher net returns from milk compared with strata one, which is expected due to the higher milk productivity from improved breeds. This stratum also had the highest net returns from other livestock products. These statistics suggest that farmers in stratum 1 focused more on crop production, farmers in stratum 2 were more involved in livestock production, while farmers in stratum 3 relied more on non-agricultural activities for a source of livelihood.

As shown in Table 3 farm households with no livestock had lower farm income compared to those with cattle. Furthermore, farms households with improved cattle had higher incomes from livestock production compared to those with local breeds, probably because of higher milk yields. As mentioned earlier, we assumed that additional inputs

ARTICLE IN PRESS

6 K.M. Shikuku etaL / Agricultural Systems xxx (2016) xxx-xxx

Table2

Description of scenario simulated in TOA-MD model.

Baseline feeding Improved quality of diet Improved quality of diet plus

increased quantity of feed

TOA-MD scenario Scenario description Stratum With local With improved With local With improved With local With improved

cows cows cows cows cows cows

Improved feeding for farms with local cows Stratum 1 Yes No Yes No Yes No

Scenario 1 Improved feeding for farms with improved cows Stratum 2 No Yes No Yes No Yes

No change Stratum 3 No No No No No No

Improved feeding for farms with local cows + 1 IB cow at zero cost Stratum 1 Yes No Yes No Yes Add 1IB

Scenario 2 Improved feeding for farms with improved cows + 1 IB cow at 0% cost Stratum 2 No Yes No Yes No Yes, add 1IB

Add 1 IB cow with improved feeding. Stratum 3 No No No No No Add 1 IB cow

PC of cow = 0%

Improved feeding for farms with local cows + 1 IB cow at 25% cost Stratum 1 Yes No Yes No Yes Add 1IB

Scenario 3 Improved feeding for farms with local cows + 1 IB cow at 25% cost Stratum 2 No Yes No Yes No Yes, add 1IB

Add 1 IB cow with improved feeding. Stratum 3 No No No No No Add 1 IB cow

PC of cow = 25%

Improved feeding for farms with local cows + one improved 1 IB cow at 50% PC Stratum 1 Yes No Yes No Yes Add 1IB

Scenario 4 Improved feeding for farms with local cows +1 IB cow at 50% PC Stratum 2 No Yes No Yes No Yes, add 1IB

Add 1 IB cow with improved feeding. Stratum 3 No No No No No Add 1 IB cow

PC of cow = 50%

Improved feeding for farms with local cows + 1 IB cow at 75% PC Stratum 1 Yes No Yes No Yes Add 1IB

Scenario 5 Improved feeding for farms with local Stratum 2 No Yes No Yes No Yes, add 1IB

cows + 1 IB cow at 75% PC

Add 1 IB cow with improved feeding. Stratum 3 No No No No No Add 1 IB cow

PC of cow = 75%

Improved feeding for farms with local

cows + 1 IB cow at full PC but HH receives credit at 18% interest and with a repayment Stratum 1 Yes No Yes No Yes Add 1IB

period of 3 years

Improved feeding for farms with local Stratum 2 No Yes No Yes No Yes, add 1IB

Scenario 6 cows + 1 IB cow at full PC but HH receives credit at 18% interest and with a repayment period of 3 years

Add 1 IB cow with improved feeding. PC of Stratum 3 No No No No No Add 1 IB cow

cow = 100% but HH receives credit at 18%

interest and with a repayment period of 3 years.

Notes: IB means improved breed and PC means purchase cost.

to livestock production (including Napier grass) would be purchased and not produced on-farm. Hence, we assume no changes in cropland allocation. Several studies have shown that if smallholder farmers face higher levels of uncertainty, they will allocate less land to a new technology (Feder, 1980; Smale et al., 1995; Pannell, 2008). The area allocated to the technology is, therefore, expected to increase if absolute risk aversion is decreasing (Feder 1980; Smale et al., 1995; Pannell 2008).

Changes in land allocation for dominant crop enterprises can only be expected after some period of time when farmers become confident on benefits of the new technology (Feder 1980; Smale et al., 1995; Pannell 2008). Although we did not consider any possible feedback between livestock production and crops (e.g. increased use of manure on crops that may affect crop yields), we acknowledge that crop-livestock interactions play an important role in farming systems. However, to

Table 3

Summary statistics for system one, by strata.

Stratum 1 (N = 28)

Stratum 2 (N = 70)

Stratum 3 (N = 66)

Proportion in the population 0.17

Household size (total number of people in a household) 5.00

Farm size (ha) 1.02

Annual non-agricultural income (USD) 127.33

Herd size (number of cows) 1.50

Annual net returns from crops (USD) 580.51

Annual net returns from milk production (USD) 48.51

Annual net returns from other livestock products (USD) 60.76

- 0.43 - 0.40 -

1.56 4.96 1.79 4.26 1.90

0.81 0.89 0.37 0.91 0.71

197.41 224.89 375.80 242.33 498.60

0.64 1.70 1.10 0 -

787.16 364.78 472.07 312.05 424.51

20.64 484.43 313.61 0 -

79.29 94.00 85.54 12.63 44.71

Notes:

1) Stratum 1 comprises farm households with local cows only.

2) Stratum 2 comprises farm households with improved dairy cows only.

3) Stratum 3 comprises farm households without cattle.

4) 1 TSh was equivalent to 0.0006 USD at the time of data collection.

ARTICLE IN PRESS

K.M. Shikuku et al. / Agricultural Systems xxx (2016) xxx-xxx

keep our analysis focused on the impacts of improved livestock feeding options our approach allowed us to maintain same crop yields and returns to crops production both under system one and system two.

3.1. Results of the ruminant model

Table 4 presents results of the Ruminant model. As shown, expected milk yield was highest for farm households with improved cows that were fed improved diets. Milk yield increased by 42% when farm households fed their improved breeds on improved diets compared to baseline diets. Feeding local cows on improved diets also increased milk yield but the yield was about 29% lower compared to that obtained with improved breeds. Together, these results suggest that improved feeding is critical for increased productivity and can contribute to improved food and nutrition security. The results further confirm that payoffs are higher when improved diets are coupled with improved breeds due to increased efficiency in utilization of the feed. Table 4 also shows that methane emission intensities were higher for farm households with local cows compared to those with improved breeds of dairy cows under baseline diets. Improved feeding strategies lowered methane emission intensity of both types of breeds.

3.2. Results of TOA-MD modelling

32.1. Predicted adoption rates

Fig. 2 presents the predicted adoption rates at the population level (aggregated over the three strata) under the improved livestock diets option (i.e. livestock feeding option 1) while Fig. 3 presents predicted adoption rates under the improved diets plus quantity of feed option (i.e. livestock feeding option 2). In the first case, the average projected adoption rates by the TOA-MD model ranged from about 59 to 84% while in the second case the average projected adoption rates by the TOA-MD model ranged from about 64 to 84%. These results suggest higher adoption rates when both quality and quantity of feed improves, indicating that net returns are higher compared to when only quality is improved.

Table 5 summarizes the adoption rates predicted by the TOA-MD model for each stratum of the population when the quality of the livestock diet was improved holding quantity constant. As shown, predicted adoption rate for farm households with local cows only (i.e. stratum 1) ranged between 42% and 65%. The rate was higher for farm households with improved cows only (i.e. stratum 2) and ranged from 67% when diets were improved holding herd sizes constant to 87% when diets were improved and farm households received an improved breed of cow at zero purchase cost. These results suggest that, on average, net returns are higher when farm households in stratum 1 and stratum 2 improve the quality of livestock diets than if they rely on current feeding. The higher adoption rates for farm households in stratum 2 compared to those in stratum 1 indicate that net returns are higher for the former. As shown in Table 5, adoption rate for farm households in stratum 3 when diets are improved (i.e. scenario 1) is not reported. This is because we had no ability to compare stratum 3 with a baseline. It is also for this reason that adoption rates were higher than those in stratum 1 and stratum 2 when farm households in stratum 3 received an improved breed cow. Table 6 presents the predicted adoption rates by stratum when both the quality and quantity of the diets improved. As

shown, the predicted adoption rates were higher in this scenario compared to varying quality of diets alone for farm households in stratum 2 but not for those in stratum 1.

Providing an improved breed of dairy cow at zero purchase cost increased the predicted adoption rate of improved feeding by 13% for farms with local cows only (i.e. stratum 1) and 30% for farms with improved cattle only (i.e. stratum 2) when compared with introduction of improved feeding only. Projected adoption rates, however, declined when farmers were required to pay part of the purchase price of the improved cow. These lower adoption rates were noteworthy for farm households in stratum 1. On the other hand, farm households that already own improved breeds (i.e. those in stratum 2) might be able to afford an upfront payment hence the higher predicted adoption rates. Conditions improved when farm households had access to credit implying that even farmers in stratum 1 might afford to cover the upfront cost of the improved cow breed. Even though the adoption rates were expected, given the design of the scenarios, it is important to highlight the importance of policies or programs that would allow farmers to adopt the proposed technologies, which as we show below, may have important consequences for food security and the environment. Given the predicted adoption rates, the projected outcomes of income, food security, poverty rate, and methane emissions for each stratum were calculated.

3.2.2. Simulated impacts on income, food security, poverty, and methane emissions

Results of TOA-MD in Table 5 showed that at the predicted adoption rates for improved quality of feed, expected income would increase for farm households in stratum 1 and stratum 2. As shown, TOA-MD projected an increase in annual income from USD 728 to USD 968 representing a 33% increase for farm households in stratum 1. Expected economic payoffs were higher for households in stratum 2 and were projected to increase by about 48% from USD 1116 to USD 1432 with adoption of improved diets only. Results further showed an increase in income when farm households were provided with an improved cow breed at zero purchase price. Although positive, expected gain in income was lower when households were required to pay a fraction of the purchase price and was lowest when households paid 75% of the purchase price. However, access to credit increased income when households paid full cost of purchase for the improved cow breed. Income level for household that accessed credit was about equal to that obtained when the purchase price was subsidized by 50%. These findings suggest that a policy aiming to provide access to credit may, in part, eliminate some of the constraints that small farmers have to invest on their farms. Consistent with studies that aim to understand indicators that farmers use to prioritize climate smart agricultural practices (Mwongera et al., 2015; Shikuku et al., 2015), this finding reinforces the idea that cost plays an important role in the decision making process of farm households. Projections of TOA-MD showed that, the increase in income was highest for farm households in stratum 3. As already mentioned, this stratum did not have a baseline outcome, which partly explains the high economic gains. Although it is difficult to make direct comparisons between stratum 3 and the others, the findings obtained are insightful as they suggest considerable economic gains for farm households in stratum 3 if they were to adopt an improved cow breed.

Daily milk production and emission (methane from enteric fermentation) across seasons for local and improved breeds with baseline and improved livestock feeding.

Livestock type

Local cow breed Improved cow breed

Base Improved quality Improved quality plus quantity Base Improved quality Improved quality plus quantity

Milk production (l/cow/day) 2.83 5.10 6.53 3.80 7.00 9.37

Methane emission (l/l of milk/day) 125.60 145.40 148.80 162.40 190.80 200.10

ARTICLE IN PRESS

K.M. Shikuku et al. / Agricultural Systems xxx (2016) xxx-xxx

-5000000

Fig. 2. Predicted population adoption rates by scenario (Ruminant model scenario — improved quality of livestock diets only). Notes:

In terms of poverty rates, defined as the percentage of the population living below a poverty line of USD 1.25/day, improvement in the quality of livestock diets reduced poverty both in stratum 1 (i.e. households that owned local cows only) and stratum 2 (i.e. households that owned improved breeds) but was only modest in the former. While poverty reduction remained modest when households in stratum 1 received an improved cow breed at zero purchase cost, households in stratum 2 observed a greater decline (48%) in poverty when they received an improved cow breed. A similar pattern to that observed for income was seen in poverty when households were required to pay for the improved cow breed. Although the gains in terms of decline in poverty rates were high even when households in stratum 2 paid for the improved cow breed, these gains reduced and were lowest when households paid 75% of the purchase cost. Subsidizing the cost of the improved cow breed at 50% yielded gains in poverty reduction

equal to those achieved when farmers paid full price but accessed credit. Although the predicted adoption rate for farm households in stratum 3 was higher compared to stratum 2, the expected decline in poverty rate was lower in stratum 3 than in stratum 2 when an improved cow breed was provided. This might be due to the distribution of income in stratum 3 that the TOA-MD could not capture.

Although households in stratum 1 gained less in terms of expected reduction in poverty, there were substantial gains in food security when they adopted improved quality of livestock feeds. Expected improvement in food security was higher for farm households in stratum 1 compared to those in stratum 2. Increased milk productivity due to better feeding and the addition of an improved breed might lead to an increase in farm income, which farms with local cows can use to satisfy their food requirements. Reducing food insecurity between 20% and 37%

-6000000

Fig. 3. Predicted population adoption rates by scenario (Improved quality of livestock diets plus increased amount of feed). Notes:

ARTICLE IN PRESS

K.M. Shikuku et al. / Agricultural Systems xxx (2016) xxx-xxx

Table 5

Predicted adoption rates and simulated Impacts on net returns, food insecurity, poverty rates, and methane emissions intensity, by scenario1 1

Please refer to Table 2 for the definition of the scenarios. and stratum: Ruminant model scenario = Improved quality of diet only.

Indicators Stratum (type of farm) Base system Improved systems Scenario 1 Scenario 2 Scenario 3 Scenario 4 Scenario 5 Scenario

Predicted adoption rate (%) 1 — Only local cattle n/a 58.00

65.52 57.93 50.03 42.12 51.30

2 — Only improved cattle n/a 67.00

87.03 84.39 81.41 78.10 81.91

3 — No cattle n/a n/a

99.93 99.75 99.19 97.69 99.32

1 — Only local cattle 728.00 33.10 41.20 33.15 26.11 20.09 27.17

2 — Only improved cattle 1116.00 47.70 86.30 79.05 71.99 65.20 73.10

Average farm income (USD/year) 3 — No cattle 324.77 n/a 218.00 190.67 163.08 135.76 167.49

Food insecurity (%) 1 — Only local cattle 21.66 —30.42

— 36.15 — 30.47 — 25.07 — 20.13 — 25.90

2 — Only improved cattle 13.90 — 13.86

— 30.09 — 26.79 — 23.64 — 20.63 — 24.07

3 — No cattle 23.87 n/a

— 74.28 — 69.96 — 64.68 — 58.27 — 65.60

Poverty rate (%) 1 — Only local cattle 92.89 — 6.40

— 8.00 — 6.36 — 4.76 — 3.50 — 4.98

2 — Only improved cattle 80.13 — 14.10

— 47.59 — 44.41 — 41.16 — 37.91 — 41.69

3 — No cattle 92.95 n/a

— 37.20 — 31.72 — 26.19 — 20.79 — 27.07

Methane emission intensity (l CH4/l of milk/year) 1 — Only local cattle 27.77 — 23.19

— 13.36 — 12.21 — 10.91 — 9.51 — 11.13

2 — Only improved cattle 24.70 — 24.00

— 52.96 — 50.94 — 48.70 — 46.28 — 49.11

1. n/a means not applicable.

2. For the scenarios (1,2,3,4,5,6), average income, poverty rate, food insecurity, and average GHG emission intensity are expressed as percentage changes compared to the baseline.

3. Negative figures show a percentage decrease in an indicator compared to the base scenario. 4.1 TSh was equivalent to 0.0006 USD at the time of data collection.

with these combined strategies is promising for farm households in stratum 1.

As shown in Tables 5 and 6, improved feeding and improved breeds could potentially provide mitigation co-benefits. Improved feeding strategies decreased methane emission intensity in both local and improved breeds. Consistent with the results of the Ruminant model, results ofTOA-MD showed no difference in emissions reduction when farm households improved quality of diets only vis-à-vis when then they improved quality and increased quantity fed. The magnitude of reduction in emission intensity was also similar between stratum 1 and stratum 2 when improved feeding was adopted. Provision of an improved cow breed, however, on average yielded greater gains in emission intensity reduction for farm households in stratum 2 than those in stratum 1. The reason is that these results are aggregating the emissions from both the local breed cow that farmers already own and the new improved breed they would acquire under any of the scenarios.

Although results showed considerable economic gains for farm households in stratum 3 in terms of income, poverty, and food security, introducing an improved cow breed in this stratum implies shifting non-cattle producers to dairy production. Therefore, this results in an overall net increase in GHG emissions meaning that for stratum 3, adoption might create trade-offs and not a triple win.

Tables 7,8,9, and 10 report results of a sensitivity analysis, using the adoption rate, poverty rate, threshold food security indicator, and methane emission intensity. As noted above, quantity of feed was increased and decreased by 10%. As shown the adoption rate, poverty, methane emissions, and food security indicators did not deviate more than 11% when the feed quantity was varied individually from — 10% to 10%.

Fig. 4, shows that the population adoption rates did not deviate more than 11% when feed quality and quantity was varied from —10% to 10%.

Although results predict that adoption of improved livestock feeding and improved breeds might improve food security and reduce poverty with net reductions in methane emission intensity, barriers to adoption might hinder uptake. As shown, for example, economic and environmental gains were lower when farm households in all the strata paid for the purchase price of the improved cow breed. Providing access to credit, however, seemed to reduce the barrier to adoption and hence improved the economic and environmental impacts. Similarly, although our study assumes that land allocation remains unchanged, the amount of land owned by households might determine the ability of the household to keep an additional cow. As shown in Table 3, the average farm size is about 1 ha. Furthermore, whether farmers will fully gain from improved livestock feeding and breeding will depend on their access to markets.

4. Discussion and conclusions

This study highlighted the utility of the Trade-offs Analysis Model for Multidimensional Impacts Assessment (TOA-MD) plus Ruminant model to examine the three pillars of CSA and assess the trade-offs and synergies associated with adoption of improved livestock feeding strategies across diverse farming systems in Lushoto district, Tanzania. We stratified our sample into the following: 1) households that owned local cows only; 2) households that owned improved cow breed only; and 3) households that did not own cows. Using our stratification strategy, we combined the Ruminant and Trade-offs Analysis Model for Multidimensional Impacts Assessment to assess how

ARTICLE IN PRESS

K.M. Shikuku et al. / Agricultural Systems xxx (2016) xxx-xxx

Table 6

Predicted adoption rates and simulated Impacts on net returns, food insecurity, poverty rates, and methane emissions intensity, by scenario2

Please refer to Table 2 for the definition of the scenarios. and stratum: Ruminant model scenario = Improved quality of diet plus increased quantify of feed.

Indicators Stratum (type of farm) Base system Improved systems Scenario 1 Scenario 2 Scenario 3 Scenario 4 Scenario 5 Scenario 6

Predicted adoption rate (%) 1 — Only local cattle n/a 62.17

60.58 53.00 45.30 37.77 46.52

2 — Only improved cattle n/a 75.82

90.35 88.84 87.17 85.32 87.45

3 — No cattle n/a n/a

100.00 100.00 100 99.92 99.98

Average farm income (USD/year) 1 — Only local cattle 728.00 38.66

36.93 29.51 23.10 17.68 24.06

2 — Only improved cattle 1116.00 77.64 134.02 126.37 118.87 111.52 120.06

3 — No cattle 324.77 n/a 333.16 305.43 277.70 249.98 282.15

Food insecurity (%) 1 — Only local cattle 21.66 - 31.95

- 30.79 - 25.58 - 20.73 - 16.44 - 21.51

2 — Only improved cattle 13.90 - 13.90

- 28.41 - 25.54 - 22.67 - 20.09 - 23.10

3 — No cattle 23.87 n/a

- 83.00 - 80.77 - 78.09 - 74.86 - 78.55

Poverty rate (%) 1 — Only local cattle 92.89 - 8.28

-9.85 - 6.04 - 4.54 - 3.34 - 4.77

2 — Only improved cattle 80.13 - 41.15

- 60.67 - 58.36 -56.00 - 53.57 - 56.38

3 — No cattle 92.95 n/a

- 57.04 - 52.82 - 48.28 - 43.44 - 49.03

Methane emission intensity (l CH4/l of milk/year) 1 — Only local cattle 26.71 - 32.63

- 23.77 - 21.00 -18.15 - 15.34 - 18.62

2 — Only improved cattle 23.90 - 28.34

- 62.98 - 61.52 - 59.97 - 58.27 - 60.21

1) n/a means not applicable.

2) For the scenarios (1,2,3,4,5,6), average income, poverty rate, food insecurity, and average GHG emission intensity are expressed as percentage changes compared to the baseline.

3) Negative figures show a percentage decrease in an indicator compared to the base scenario.

4) 1 TSh was equivalent to 0.0006 USD at the time of data collection.

improved livestock management options can increase productivity hence income, improve food security, and reduce greenhouse gas (GHG) emission intensity.

Our results predicted that adoption of improved quality of livestock diet would increase income for farm households both in stratum 1 and stratum 2. Our prediction showed that while the increase in income substantially reduced poverty for households with improved cow

breeds, gains in poverty reduction were modest for households that only owned local cows. Households with only local cows, however, received considerable gains in food security when they improved the quality of the livestock diets. Furthermore, results indicated that expected emissions intensity declined with adoption of improved quality feed with both local and improved breed cows. This finding has important implications for introducing incremental adaptation strategies. Economic

Table 7

Sensitivity analysis: increase in the quality of the feed by 10% (Ruminant model scenario = Improved livestock diets only).

Indicators Stratum (type of farm) Base system Improved systems Scenario 1 Scenario 2 Scenario 3 Scenario 4 Scenario 5 Scenario 6

Predicted adoption rate (%) 1 Only local cattle n/a 58.20 65.52 57.93 50.03 42.12 51.30

2 Only improved cattle n/a 66.44 86.65 83.87 80.75 77.28 81.28

3 No cattle n/a n/a 99.91 99.65 98.90 97.01 99.08

Average farm income (USD/year) 1 Only local cattle 728.00 33.48 41.20 33.15 26.11 20.09 27.17

2 Only improved cattle 1116.00 45.58 83.05 75.79 68.78 62.05 69.89

3 No cattle 324.77 n/a 210.70 183.03 155.49 128.30 159.89

Food insecurity (%) 1 Only local cattle 21.66 - 30.56 - 27.11 -20.22 -14.12 - 9.01 - 15.04

2 Only improved cattle 13.90 - 13.90 - 30.23 -26.79 - 23.64 - 20.63 - 24.07

3 No cattle 23.87 n/a - 73.19 -68.62 - 63.05 - 56.35 - 64.01

Poverty rate (%) 1 Only local cattle 92.89 - 6.47 - 8.00 - 6.36 - 4.76 - 3.50 - 4.98

2 Only improved cattle 80.13 - 27.85 - 46.42 -43.16 - 39.85 - 36.55 - 40.39

3 No cattle 92.95 n/a - 35.72 -30.18 - 22.92 - 19.34 - 25.54

Methane emission intensity (l CH4/l of milk/year) 1 Only local cattle 27.77 - 24.08 - 13.36 -12.21 - 10.91 - 9.51 -11.13

2 Only improved cattle 24.70 - 21.53 - 52.25 -50.20 - 47.87 - 45.39 - 48.28

Notes:

1) n/a means not applicable.

2) For the scenarios (1, 2, 3, 4, 5, 6), average income, poverty rate, food insecurity, and average GHG emission intensity are expressed as percentage changes compared to the baseline.

3) Negative figures show a percentage decrease in an indicator compared to the base scenario.

4) 1 TSh was equivalent to 0.0006 USD at the time of data collection.

ARTICLE IN PRESS

K.M. Shikuku et al. / Agricultural Systems xxx (2016) xxx-xxx 11

Table 8

Sensitivity analysis: decrease in the quality of the feed by 10% (Ruminant model scenario = Improved livestock diets only).

Indicators Stratum (type of farm) Base system Improved systems Scenario 1 Scenario 2 Scenario 3 Scenario 4 Scenario 5 Scenario 6

Predicted adoption rate (%) 1 - Only local cattle n/a 57.88 65.31 57.70 49.80 41.90 51.07

2 - - Only improved cattle n/a 67.92 87.43 84.92 82.09 78.94 82.56

3 - - No cattle n/a n/a 99.84 99.51 98.68 96.88 98.87

Average farm income (USD/year) 1 - Only local cattle 728.00 33.10 40.97 32.94 25.92 19.94 26.97

2 - Only improved cattle 1116.00 50.09 90.00 82.66 75.55 68.70 76.68

3 - No cattle 324.77 n/a 226.83 199.17 171.67 144.52 176.06

Food insecurity (%) 1 - Only local cattle 21.66 -30.42 - 36.01 - 30.29 - 24.88 -19.94 -25.76

2 - Only improved cattle 13.90 -13.32 - 30.09 - 26.65 - 23.64 -20.63 - 24.07

3 - No cattle 23.87 n/a -72.14 - 67.66 - 62.21 -55.76 -63.13

Poverty rate (%) 1 Only local cattle 92.89 -6.35 -8.26 -6.31 - 4.72 - 3.43 - 4.94

2 Only improved cattle 80.13 -29.93 - 48.85 - 45.73 - 42.56 -39.35 -43.07

3 No cattle 92.95 n/a -39.19 - 33.98 - 28.73 -23.55 -29.56

Methane emission intensity (l CH4/l of milk/year) 1 Only local cattle 27.77 -22.98 - 12.40 -11.34 -10.15 - 8.84 -10.33

2 Only improved cattle 24.70 -10.38 - 53.56 -51.63 - 49.48 -47.14 - 49.86

1) n/a means not applicable.

2) For the scenarios (1,2,3,4,5,6), average income, poverty rate, food insecurity, and average GHG emission intensity are expressed as percentage changes compared to the baseline.

3) Negative figures show a percentage decrease in an indicator compared to the base scenario.

4) 1 TSh was equivalent to 0.0006 USD at the time of data collection.

gains, in terms of income, poverty, and food security were higher when households improved both the quality and quantity of the feed compared to when only quality was improved. However, no significant difference in emissions intensity was observed between improved quality and improved quality plus increased quantity of feed. We also did not observe significant difference in methane emissions between stratum 1 and stratum 2 with improved quality or quality plus quantity. Providing an improved cow breed had greater impacts on methane emissions reduction when farm households improved the quality of livestock diets and increased the amount fed to the livestock. Although there were gains in income, food security, poverty reduction, and methane emission, when households acquired an improved cow breed, impacts were lower when households paid for the cow. Providing access to credit, however, reduced the financial constraint.

Our results have several important policy implications. First, the evidence presented here suggests the need to promote improved feeding strategies and introduction of more efficient breeds of livestock in order to achieve improved food security, increased productivity, reduced poverty, and reduced methane emissions intensities. Specifically, policies

targeting to increase income and improve food security can be beneficial to farm households that own local cows if such households are encouraged to adopt improved livestock diets while those targeting to further reduce poverty should aim to promote adoption of improved cow breeds. Secondly, promotion of improved livestock feeding strategies needs to be accompanied by policies or programs that facilitate the adoption of improved breeds. Policies that focus on providing access to credit markets are required in order to ease liquidity constraints that often limit smallholder farmers from adopting improved technologies. However, this does not necessarily imply a complete shift to improved breeds as evidence from literature shows that local breeds are better adapted to drought conditions (Scarpa et al., 2003). Thirdly, for farm households that do not own cattle, there are trade-offs involved because shifting non-cattle producers to dairy production implies a net increase in GHG emissions. At the same time, introducing improved breeds to non-cattle producers has considerable positive impacts on income. It seems, therefore, that while shifting non-cattle producers to dairy production might not present a triple win, it is a path to reach the triple win if accompanied by other interventions.

Table 9

Sensitivity analysis: increase in the quality of the feed by 10% (Ruminant model scenario = Improved livestock diets plus increased feed amounts).

Indicators Stratum (type of farm) Base system Improved systems Scenario 1 Scenario 2 Scenario 3 Scenario 4 Scenario 5 Scenario 6

Predicted adoption rate (%) 1 Only local cattle n/a 62.17 71.60 64.72 57.32 49.66 58.53

2 Only improved cattle n/a 75.82 89.95 88.29 86.45 84.42 86.76

3 No cattle n/a n/a 100.00 99.98 99.94 99.81 99.95

Average farm income (USD/year) 1 Only local cattle 728.00 38.66 50.47 41.56 33.59 26.61 34.81

2 Only improved cattle 1116.00 77.64 125.23 117.62 110.17 102.89 111.36

3 No cattle 324.77 n/a 310.67 282.94 255.21 227.51 259.66

Food insecurity (%) 1 Only local cattle 21.66 -31.95 - 39.47 - 33.89 - 28.49 - 23.41 -29.32

2 Only improved cattle 13.90 -13.90 - 28.28 - 25.82 - 22.30 -19.66 -23.39

3 No cattle 23.87 n/a - 81.27 - 78.72 - 75.62 -71.89 -76.16

Poverty rate (%) 1 Only local cattle 92.89 -8.28 - 11.35 - 9.01 - 7.02 - 5.35 - 7.32

2 Only improved cattle 80.13 -41.15 - 58.66 - 56.22 - 53.71 -51.12 -54.12

3 No cattle 92.95 n/a - 53.65 -49.17 - 44.38 -39.35 -45.16

Methane emission intensity (l CH4/l of milk/year) 1 Only local cattle 26.71 -32.63 - 27.71 - 25.31 - 22.65 -19.81 -23.08

2 Only improved cattle 23.90 - 28.43 - 61.61 - 60.09 - 58.40 -56.60 -58.68

Notes:

1) n/a means not applicable.

2) For the scenarios (1,2,3,4,5,6), average income, poverty rate, food insecurity, and average GHG emission intensity are expressed as percentage changes compared to the baseline.

3) Negative figures show a percentage decrease in an indicator compared to the base scenario.

4) 1 TSh was equivalent to 0.0006 USD at the time of data collection.

ARTICLE IN PRESS

12 K.M. Shikuku et al. / Agricultural Systems xxx (2016) xxx-xxx

Table 10

Sensitivity analysis: decrease in the quality of the feed by 10% (Ruminant model scenario = Improved livestock diets plus increased feed amounts).

Indicators Stratum (type of farm) Base system Improved systems Scenario 1 Scenario 2 Scenario 3 Scenario 4 Scenario 5 Scenario 6

Predicted adoption rate (%) 1 Only local cattle n/a 62.17 71.41 64.52 57.11 49.43 58.32

2 Only improved cattle n/a 74.66 90.58 89.15 87.57 85.84 87.84

3 - No cattle n/a n/a 100.00 100.00 99.98 99.93 99.98

Average farm income (USD/year) 1 Only local cattle 728.00 38.66 50.20 41.32 33.38 26.43 34.59

2 Only improved cattle 1116.00 82.25 139.58 131.91 124.38 117.00 125.58

3 No cattle 324.77 n/a 336.73 308.96 281.23 253.50 285.67

Food insecurity (%) 1 Only local cattle 21.66 -31.95 - 39.34 - 33.70 - 28.35 - 23.27 - 29.18

2 Only improved cattle 13.90 -9.06 - 28.16 - 25.29 - 22.41 - 19.83 - 22.84

3 No cattle 23.87 n/a - 83.20 - 81.02 - 78.38 - 75.24 - 76.84

Poverty rate (%) 1 Only local cattle 92.89 - 8.28 - 7.85 - 8.96 -6.97 - 5.31 - 7.27

2 Only improved cattle 80.13 - 42.38 - 60.67 - 59.65 - 57.38 - 55.04 - 57.76

3 No cattle 92.95 n/a - 57.04 -53.37 - 48.88 - 44.09 - 49.62

Methane emission intensity (l CH4/l of milk/year) 1 Only local cattle 26.71 - 32.48 - 23.77 - 24.61 - 22.00 - 19.28 - 22.44

2 Only improved cattle 23.90 - 28.29 - 62.98 - 62.23 - 60.74 - 59.09 - 60.99

1) n/a means not applicable.

2) For the scenarios (1,2,3,4,5,6), average income, poverty rate, food insecurity, and average GHG emission intensity are expressed as percentage changes compared to the baseline.

3) Negative figures show a percentage decrease in an indicator compared to the base scenario.

4) 1 TSh was equivalent to 0.0006 USD at the time of data collection.

Furthermore, there are other barriers to adoption, like farm size. Although our analysis did not account for issues of carrying capacity, we know that small farms may have issues with increasing herd sizes due to the limited land available. It might not be possible for farms to increase their herd size due to lack of land. Future research should, therefore, incorporate the effects of changing land allocation as a key aspect when simulating impacts of improved feeding and breeding strategies.

Our study provides a first attempt to combine economic and livestock simulation methods in prioritizing climate smart feeding strategies using a minimum data approach. A few points are, however, worth pointing out. First, we acknowledge that although parsimonious, accurate data to fully characterize farming systems is important in estimating ex-ante impacts using TOA-MD. Specifically, capturing heterogeneity in net-returns remains an important requirement of the model. Future data collection efforts and research should work towards an improved approach to capture the heterogeneity of milk production in a population of farms to allow estimation of the distribution of greenhouse gas produced in the same population and, therefore, estimate the economic versus environmental tradeoffs more accurately. Secondly, our study makes several assumptions in simulating impacts of improved livestock feeding strategies. It is, therefore, important to interpret our results within the assumptions that we make. We, acknowledge, for example, that smallholder farmers might not always behave to maximize economic returns and that crop-livestock interactions are an important component of farming systems in SSA. In conclusion, we believe this study can help to assess the multiple trade-offs of selected CSA practices and sustainable intensification options to identify triple win situations.

Acknowledgements

This research was conducted under the CGIAR Research Program on Climate Change, Agriculture and Food Security (CCAFS) with funding from the International Fund for Agricultural Development (IFAD) Grant number 2000000176, project titled, "Increasing food security and farming system resilience in East Africa through wide-scale adoption of climate-smart agricultural practices". We acknowledge the Agricultural Model Inter-comparison and Improvement Project (AGMIP, www.agmip.org) for the contribution on the methodology. A special thanks also to Selian Agricultural Research Institute (SARI), specifically Dr. Charles Lyamchai and George Sayula for facilitating field visits to Lushoto and the CCAFS East Africa office including James Kinyangi, John Recha, and Maren Radeny for their support.

1) A complete description of the scenarios is provided in Table 2.

2) Opportunity cost is defined here as the difference in net farm returns between the base system (current technology) and the modified system (improved technology). Adoption rate is indicated by the point where the curves cut the x-axis.

3) 1 TSh was equivalent to 0.0006 USD at the time of data collection.

1) A complete description of the scenarios is provided in Table 2.

2) Opportunity cost is defined here as the difference in net farm returns between the base system (current technology) and the modified system (improved technology). Adoption rate is indicated by the point where the curves cut the x-axis.

3) 1 TSh was equivalent yo 0.0006 USD at the time of data collection.

Fig. 4. Sensitivity analysis by scenario. Note: Base runs of baseline and scenario diets were repeated with a +10% and —10% variation to test sensitivity of the approach.

ARTICLE IN PRESS

K.M. Shikuku et al. / Agricultural Systems xxx (2016) xxx-xxx 13

References

Antle, J.M., Valdivia, R.O., 2006. Modelling the supply of ecosystem services from agriculture: a minimum-data approach. Aust J. Agric. Resour. Econ. 50 (1), 1-15.

Antle, J.M., Valdivia, R.O., 2011. TOA-MD 5.0: Trade-off Analysis Model for Multi-Dimensional Impact Assessment. http://trade-offs.oregonstate.edu.

Antle, J.M., Diagana, B., Stoorvogel, J.J., Valdivia, R.O., 2010. Minimum-data analysis of ecosystem service supply in semi-subsistence agricultural systems. Aust. J. Agric. Resour. Econ. 54 (4), 601-617.

Antle, J.M., Stoorvogel, J.J., Valdivia, R.O., 2014. New parsimonious simulation methods and tools to assess future food and environmental security of farm populations. Philos. Trans. R Soc. Lond. Ser. B Biol. Sci. 369.

Antle, J., Adhikari, S., Price, S., 2015a. An income-based food security indicator for agricultural technology impact assessment. In: Schmitz, A., Kennedy, P.L., Schmitz, T.G. (Eds.), Food Security in an Uncertain World (Frontiers of Economics and Globalization. Emerald Group Publishing Limited vol. 15, pp. 81-96.

Antle, J., Valdivia, R.O., Boote, K., Hatfield, J., Janssen, S., Jones, J., Porter, C., Rosenzweig, C., Ruane, A., Thorburn, P., 2015b. AgMIP's trans-disciplinary approach to regional integrated assessment of climate impact, vulnerability and adaptation of agricultural systems. In: Rosenzweig, C., Hillel, D. (Eds.), Handbook of Climate Change and Agroecosystems: The Agricultural Model Intercomparison and Improvement Project (AgMIP)ICP Series on Climate Change Impacts, Adaptation, and Mitigation vol. 3. Imperial College Press.

Battisti, D.S., Naylor, R.L., 2009. Historical Warnings of Future Food Insecurity with Unprecedented Seasonal Heat. Science 5911, 240-244 (323).

Bryan, E., Ringler, C., Okoba, B., Koo, J., Herrero, M., Silvestri, S., 2013. Can agriculture support climate change adaptation, greenhouse gas mitigation and rural livelihoods? Insights from Kenya. Clim Chang. 118,151-165.

Claessens, L., Antle, J.M., Stoorvogel, J.J., Valdivia, R.O., Thornton, P.K., Herrero, M., 2012. A method for evaluating climate change adaptation strategies for small-scale farmers using survey, experimental and modeled data. Agr. Syst 111,85-95.

Critical Ecosystem Partnership Fund., 2005. Eastern Arc Mountains and Coastal Forests of Tanzania and Kenya: Ecosystem Profile Elsevier, Cape Town (162 pgs).

Delgado, C., Rosegrant, M., Steinfeld, H., Ehui, S., Courbois, C., 1999. Livestock to 2020: The next food revolution. Food, Agriculture, and the Environment Discussion Paper 28. International Food Policy Research Institute, Washington, DC.

Easterling, W., Aggarwal, P., Batima, P., Brander, K., Erda, L., Howden, M., Kirilenko, A., Morton, J., Soussana,J.F., Schmidhuber,J., etal., 2007. Food, fibre, and forest products. In: Parry, M.L., Canziani, O.F., Palutikof, J.P., Linden, P.J., Hanson, C.E. (Eds.), Climate Change 2007: Impacts, Adaptation and Vulnerability. Cambridge University Press, Cambridge, UK, pp. 273-313.

Feder, G., 1980. Farm Size, Risk Aversion and the Adoption of New Technology under Uncertainty. Oxf. Econ. Pap. 32 (2), 263-283.

Foerch, W., Patti, K., Thornton, P., Kiplimo, J., 2011. Initial Sites in the CCAFS Regions: Eastern Africa, West Africa and Indo-Gangetic Plains, Version 2. CGIAR Research Program on Climate Change, Agriculture and Food Security (CCAFS), Copenhagen, Denmark (Available online at) www.ccafs.cgiar.org.

Henderson, B., Godde, C., Medina-Hidalgo, D., van Wijk, M., Silvestri, S., Douxchamps, S., Stephenson, E., Power, B., Rigolot, C., Cacho, O., Herrero, M., 2016. Closing system-wide yield gaps to increase food production and mitigate GHGs among mixed crop-livestock smallholders in Sub-Saharan Africa. Agric. Syst. 143,106-113.

Herrero, M., 1997. Modelling Dairy Grazing Systems: An Integrated Approach. Ph.D. Thesis. University of Edinburgh, Edinburgh, Scotland.

Herrero, M., Fawcett, R.H., Jessop, N.S., 2002. Predicting Intake and Nutrient Supply of Tropical and Temperate Diets for Ruminants Using a Simple Dynamic Model of Digestion. Bioparametrics Ruminant Nutrition Reference Laboratories Monograph. Institute of Ecology and Resource Management, University of Edinburgh, UK.

Herrero, M., Wirsenius, S., Henderson, B., Rigolot, C., Thornton, P., Havlik, P., De Boer, I., Gerber, P.J., 2015. Livestock and the environment: What have we learned in the past decade? Annu. Rev. Environ. Resour. 40,177-202.

Herrero, M., Havlik, P., Valin, H., Notenbaert, A., Rufino, M., Thornton, P.K., Obersteiner, M., 2013. Biomass Use, Production, Feed Efficiencies, and Greenhouse Gas Emissions from Global Livestock Systems — Supporting Information (PNAS).

Homann-Kee Tui, S., Valbuena, D., Masikati, P., Descheemaeker, K., Nyamangara, J., Claessens, L., Erenstein, O., Van Rooyen, A., Nkomboni, D., 2015. Economic tradeoffs of biomass use in crop-livestock systems: Exploring more sustainable options in semi-arid Zimbabwe. Agric. Syst. 13,48-60.

Homann-Kee Tui, S., Descheemaeker, K., Masikati, P., Chibwana Gama, A., Crespo, O., Claessens, L., Valdivia, R., 2016. Re-Designing Smallholder Farming Futures for Reduced Vulnerability to Climate Change in Semi-Arid Southern Africa (Submitted to European Journal of Agronomy).

IPCC, 2007. Fourth Assessment Report: Synthesis. www.ipcc.ch/ipccreports/ar4- syr.htm. IPCC.

Jambiya, G., 1998. The dynamics of population, land scarcity, agriculture and non-agricultural activities: West Usambara Mountains, Lushoto District, Tanzania. ASC Working Paper 28. Dar Es Salaam (42 pgs).

Kristjanson, P., Neufeldt, H. , Gassner, A , Mango, J., Kyazze, F.B., Desta, S., Sayula, G., Thiede, B., Forch, W., Thornton, P.K, Coe, R., 2012. Are food insecure smallholder households making changes in their farming practices? Evidence from East Africa. Food Secur. 4 (3), 318-397.

Lipper, L., Thornton, P., Campbell, B.M., Baedeker, T., Braimoh, A., Bwalya, M., Caron, P., Cattaneo, A., Garrity, D.P., Henry, K., Hottle, R., Jackson, L., Jarvis, A., Kossam, F., Mann, W., McCarthy, N., Meybeck, A., Neufeldt, H., Remington, T., Sen, P.T., Sessa, R., Shula, R., Tibu, A., Torquebiau, E., 2014. Nat. Clim. Chang. 4,1068-1072.

Lukuyu, B., Ravichandran, T., Maass, B., Laswai, G., Bwire, J., Duncan, A., 2015. Enhancing livestock producitivity through feed and feeding interventions in India and Tanzania. ILRI Project Report. International Livestock Research Institute (ILRI), Nairobi, Kenya.

Lyamchai, C., Yanda, P., Sayula, G., Kristjanson, P., 2011. Summary of Baseline Household Survey Results. Lushoto, Tanzania. Copenhagen.

Mangesho, W., Loina, R., Bwire, J., in collaboration with Maass, B.L., Lukuyu, B., Wassena, F.J., 2013. Feed Assessment 3 — Lushoto District, Tanga Region, the United Republic of Tanzania. Report. International Center for Tropical Agriculture (CIAT), Nairobi, Kenya; Tanzania Livestock Research Institute (TaLiRI), Tanga, Tanzania; International Livestock Research Institute (ILRI), Nairobi, Kenya. 16 pp.

Morris, J., Fraval, S., Githoro, E., Mugatha, S., Ran, Y., 2014. Participatory GIS Expert Workshop Report. ILRI, SEI, Nairobi, Kenya.

Mwongera, C., Shikuku, K.M., Winowiecki, L., Twyman, J., Läderach, P., Ampaire, E., van Asten, P., Twomlow, S., 2015. Climate-smart agriculture rapid appraisal (CSA-RA): a prioritization tool for outscaling CSA. Step-by-Step Guidelines. International Center for Tropical Agriculture (CIAT), Cali, Colombia (44 p.).

Okoba, B.O., Tenge, A.J., Sterk, G., Stroosnijder, L., 2007. Participatory Soil and Water Conservation Planning Using an Erosion Mapping Tool in the Central Highlands of Kenya. Land Degradation and Development. Vol. 319 pp. 303-319.

Pannel, D.J., 2008. Public Benefits, Private Benefits, and Policy Mechanism Choice for Land-Use Change for Environmental Benefits. Land Econ. 84 (2), 225-240.

Rosenzweig, C., Jones, J.W., Hatfield, J.L., Antle, J.M., Ruane, A.C., Boote, K.J., Thorburn, P.J., Valdivia, R.O., Porter, C.H., Janssen, S., Mutter, C.Z., 2015. Appendix 1. Guide for regional integrated assessments: handbook of methods and procedures, version 5.1. In: Rosenzweig, C., Hillel, D. (Eds.), Handbook of Climate Change and Agroecosystems: The Agricultural Model Intercomparison and Improvement Project (AgMIP)ICP Series on Climate Change Impacts, Adaptation, and Mitigation vol. 3. Imperial College Press, pp. 331-386.

Rufino, M.C., Quiros, C., Boureima, M., Desta, S., Douxchamps, S., Herrero, M., Kiplimo, J., Lamissa, D., Joash, M., Moussa, A.S., Naab, J., Ndour, Y., Sayula, G., Silvestri, S., Singh, D., Teufel, N., Wanyama, I., 2012. Developing generic tools for characterizing agricultural systems for climate and global change studies (IMPACTlite-phase 2). CGIAR Research Program on Climate Change. Agriculture and Food Security (CCAFS).

Scarpa, R., Eric, E.S.K., Kristjanson, P., Radeny, M., Drucker, A.G., Rege, J.E.O., 2003. Valuing indigenous cattle breeds in Kenya: an empirical comparison of stated and revealed preference value estimates. Ecol. Econ. 45, 409-426.

Shikuku, K.M., Mwongera, C., Winowiecki, L., Twyman, J., Atibo, C., Läderach, P., 2015. Understanding Farmers' Indicators in Climate-Smart Agriculture Prioritization in Nwoya District, Northern Uganda. International Center for Tropical Agriculture (CIAT), Cali, Colombia (46 p.).

Sidahmed, A., 2008. Livestock and climate change: coping and risk management strategies for a sustainable future. In livestock and global climate change conference proceeding, May 2008, Tunisia.

Silvestri, S., Douxchamps, S., Kristjanson, P., Förch, W., Radeny, M., Mutie, I., Quiros, F.C., Herrero, M., Ndungu, A., Ndiwa, N., Mango, J., Claessens, L., Rufino, M.C., 2015. Households and food security: lessons from food secure households in East Africa. Agric. Food Secur. 4, 23.

Smale, M., Heisey, P.W., Leathers, H.D., 1995. Maize of the ancestors and modern varieties: The microeconomics of high-yielding variety adoption in Malawi. Econ. Dev. Cult. Chang. 43 (2), 351-368.

Staal, S., Owango, M., Muriuki, H., Kenyanjui, M., Lukuyu, B., Njoroge, L., Njubi, D., Baltenweck, I., Musembi, F., Bwana, O., Muriuki, K., Gichungu, G., Omore, A., Thorpe, W., 2001. Dairy Systems Characterisation of the Greater Nairobi Milk Shed: Smallholder Dairy (R&D) Project. KARI/MoA/ILRI Collaborative Dairy Research Programme. ILRI, Nairobi, Kenya, p. 68 (http://www.cgiar.org/ilri/info/databases.cfm).

Steenwerth, et al., 2014. Climate-smart agriculture global research agenda: scientific basis for action. Agric. Food Sec. 3, 11.

Tenge, J., De Graaff, J., Hella, J.P., 2005. Financial efficiency of major soil and water conservation measures in West Usambara highlands, Tanzania. Appl. Geogr. 25, 348-366.

Thornton, P., Herrero, M., Freeman, A., Mwai, O., Rege, E., Jones, P., McDermott, J., 2007. Vulnerability, Climate change and Livestock - Research Opportunities and Challenges for Poverty.

Thornton, P.K., Van De Steeg, J., Notenbaert, A.M., Herrero, M., n.d.2009. The impacts ofcli-mate change on livestock and livestock systems in developing countries: a review of what we know and what we need to know. Agric. Syst., 101:113-127.

Tubiello, F.N., Soussana, J.F., Howden, S.M., 2007. Crop and pasture response to climate change. Proc. Natl. Acad. Sci. 104,19686-19690.

United Nations Food and Agriculture Organization (FAO), 2013. Climate-smart agriculture sourcebook. FAO, Rome.

United Nations, 2015. Department of Economic and Social Affairs, Population Division. World Population Prospects: The 2015 Revision, Key Findings and Advance Tables. Working Paper No. ESA/P/WP.241.

Vermeulen, S.J., Challinor, A.J., Thornton, P.K., Campbell, B.H., Eriyagama, N., Vervoort, J.M., Kinyangi, J., Jarvis, A., Laderach, P., Ramirez-Villegas, J., Nicklin, K.J., Hawkins, E., Smith, D.B., 2013. Addressing uncertainty in adaptation planning for agriculture. Proc. Natl. Acad. Sci. 110,8357-8362.

Waithaka, M.M., Thornton, P.K., Herrero, M., Shepherd, K.D., 2006. Bio-economic evaluation of farmers' perceptions of viable farms in western Kenya. Agric. Syst. 90,243-271.

Zingore, S., Gonzalez-Estrada, E., Delve, R.J., Herrero, M., Dimes, J.P., Giller, K.E., 2009. An integrated evaluation of strategies for enhancing productivity and profitability of resource-constrained smallholder farms in Zimbabwe. Agric. Syst. 101, 57-68.

Watson, J.R., 1972. Conservation problems, policies and the origins of the Mlalo Basin rehabilitation scheme, Usambara Mountains, Tanzania. Geogr. Ann. 54 (3/4), 221-226.

Wickama, J., Okoba, B., Sterk, G., 2014. Effectiveness of sustainable land management measures in West Usambara highlands, Tanzania. Catena 118, 91-102.

Winowiecki, L., Vagen, T.-G., Massawe, B., Jelinski, N.A., Lyamchai, C., Sayula, G., Msoka, E., 2015. Landscape-scale variability of soil health indicators: effects of cultivation on soil organic carbon in the Usambara Mountains of Tanzania. Nutr. Cycl. Agroecosyst. http://dx.doi.org/10.1007/s10705-015-9750-1.