Effects of climate change on agricultural households’ welfare in Kenya

Abstract

Natural and artificial (e.g. agricultural) ecosystems confer benefits in the form of provisioning, regulating, cultural and habitat/supporting goods and services. Degradation of eco systems by natural and anthropogenic drivers compromises their ability to provide these goods and services. In Kenya, as in other regions worldwide, climate change and variability are driving weather pattern changes, and causing seasonal shifts. Such changing weather patterns and seasonal shifts act as stresses on agricultural ecosystems, compromising the production of agricultural goods and services, and leading to reduced farm returns, reduced household incomes, and increase in poverty levels. Using the example of Embu County in central Kenya as a case study, this study seeks to assess the impacts of climate change on household welfare (net farm returns, per capita incomes, and poverty) in current agricultural production systems. To address this objective we conducted a multi-disciplinary study involving climatologists, crop modellers and economic modellers. Primary data from 441 households were collected using a combination of stratified and multistage sampling. In climate modelling, 5 climate models were used to downscale future projected climate change scenarios for the mid-century timeframe of 2041-2070. Crop modelling for maize was done using DSSAT and APSIM crop models. Representative agricultural pathways were used to project the production of other non-modelled crops and dairy. Finally, economic analyses were done using the trade-off analysis multi-dimensional impact assessment tool. Results show that about 36 to 66% of the households in agro-ecological zones (AEZs) receiving limited rainfall are likely to lose from climate change. In addition, crop models indicate mixed results for net farm returns, per capita income and poverty levels in different AEZs, with poverty level declines being between 0.6 and 3.8% for APSIM, and between 0.7 and 11% for DSSAT. This therefore calls for adaptation, especially for households in AEZs likely to experience negative impacts from climate change.

Full text

CLIMATE RESEARCH
Clim Res
Vol. 67: 87–97, 2016
doi: 10.3354/cr01357
Published February 24
1. INTRODUCTION
Natural and artificial (e.g. agricultural) ecosystems,
confer benefits in the form of provisioning, regulat-
ing, cultural and habitat/supporting goods and serv-
ices (Fisher & Christie 2010). Degradation of eco -
systems compromises their ability to provide these
goods and services. This change in the state of the
ecosystem according to the DSPIR
1
framework is
brought about by pressures such as changes in
water flows, rainfall and evapotranspiration patterns
(Kelble et al. 2013). The pressures are themselves a
result of drivers; these have been classified as direct
and indirect drivers (e.g. Brander et al. 2010). The
IPBES
2
Framework classifies direct drivers as either
© Inter-Research 2016 · www.int-res.com*Corresponding author: richard.mulwa@uonbi.ac.ke
Impacts of climate change on agricultural
household welfare in Kenya
Richard Mulwa
1,
*
, Karuturi P. C. Rao
2
, Sridhar Gummadi
2
, Mary Kilavi
3
1
Centre for Advanced Studies in Environmental Law and policy (CASELAP), University of Nairobi, Kenya
2
International Centre for Research in the Arid and Semi-Arid Tropics (ICRISAT), Addis Ababa, Ethiopia
3
Kenya Meteorological Department, Nairobi, Kenya
ABSTRACT: Natural and artificial (e.g. agricultural) ecosystems confer benefits in the form of pro-
visioning, regulating, cultural and habitat/supporting goods and services. Degradation of eco -
systems by natural and anthropogenic drivers compromises their ability to provide these goods
and services. In Kenya, as in other regions worldwide, climate change and variability are driving
weather pattern changes, and causing seasonal shifts. Such changing weather patterns and sea-
sonal shifts act as stresses on agricultural ecosystems, compromising the production of agricultural
goods and services, and leading to reduced farm returns, reduced household incomes, and
increase in poverty levels. Using the example of Embu County in central Kenya as a case study,
this study seeks to assess the impacts of climate change on household welfare (net farm returns,
per capita incomes, and poverty) in current agricultural production systems. To address this objec-
tive we conducted a multi-disciplinary study involving climatologists, crop modellers and eco-
nomic modellers. Primary data from 441 households were collected using a combination of strati-
fied and multistage sampling. In climate modelling, 5 climate models were used to downscale
future projected climate change scenarios for the mid-century timeframe of 2041−2070. Crop
modelling for maize was done using DSSAT and APSIM crop models. Representative agricultural
pathways were used to project the production of other non-modelled crops and dairy. Finally, eco-
nomic analyses were done using the trade-off analysis multi-dimensional impact assessment tool.
Results show that about 36 to 66% of the households in agro-ecological zones (AEZs) receiving
limited rainfall are likely to lose from climate change. In addition, crop models indicate mixed
results for net farm returns, per capita income and poverty levels in different AEZs, with poverty
level declines being between 0.6 and 3.8% for APSIM, and between 0.7 and 11% for DSSAT. This
therefore calls for adaptation, especially for households in AEZs likely to experience negative
impacts from climate change.
KEY WORDS: Crop modelling · Economic modelling · Household welfare
Resale or republication not permitted without written consent of the publisher
1
Driver, pressure, state, impact, and response (DPSIR) model
depicts how human society affects ecosystem state
This authors' personal copy may not be publicly or systematically copied or distributed, or posted on the Open Web,
except with written permission of the copyright holder(s). It may be distributed to interested individuals on request.

Clim Res 67: 87–97, 2016
natural or anthropogenic, and indirect drivers in -
clude demographic factors, technological changes,
economic changes etc. Climate change, which is the
basis of this article, can be categorised under both
natural and anthropogenic direct drivers (Diaz et al.
2015). However, anthropogenic drivers outweigh the
natural drivers in contributing to climate change.
In Kenya (as with regions worldwide) climate
change and variability drives changes in weather
patterns and causes seasonal shifts (Republic of
Kenya 2010). Changing weather patterns and sea-
sonal shifts act as stresses on the agricultural eco-
systems, compromising the production of agricul-
tural goods and services. In Kenya, close to 70% of
the population depends directly on agriculture for
their livelihoods, and therefore any interference or
disturbance to agricultural ecosystems is likely to
have adverse impacts on the rural areas. Such im -
pacts could include reduced farm returns, reduced
house hold incomes, and in creases in poverty le -
vels. For instance, the country experienced in 2011
one of the worst droughts in the past 50 years,
where poor rains greatly undermined the food se -
curity situation, leaving about 3.5 million people in
need of food assistance (WFP 2012). Policy inter-
ventions from the government and other responses
by individuals and organizations would be required
to mitigate these impacts in the future.
Given this background, it is important to understand
the potential impacts of climate change on current
and future agricultural systems in the Kenya, and to
propose plausible policy interventions. We measured
the impacts of climate change using 4 welfare indic-
tors of: (1) changes in the percentage of gainers or los-
ers from climate change; (2) changes in net farm re-
turns; (3) changes in per capita incomes; and (4)
changes in poverty levels. We used a case study of
Embu County in Central Kenya, and our hypothesis
was that increased temperature and de pressed rain-
fall, which are our indicators of climate change, would
have a negative impact on all 4 welfare indicators.
2. METHODS
2.1. Study area
Embu County was chosen as a case study because
of its diverse agro-ecological zones (AEZs) — totaling
11 in all. The county lies on the south-eastern slopes
of Mount Kenya (Fig. 1) and covers the typical agro-
ecological profile of the country, from cold and wet
high altitude areas to the hot and dry low altitude
areas (Jaetzold et al. 2007). The AEZs in the county
are representative of most of the climatic conditions in
Kenya, and it is therefore possible to infer produc tivity
of most other regions in the country. The average an-
nual rainfall varies from >2200 mm at an altitude of
2500 m to <600 mm near the Tana River at 700 m. The
Upper Highlands (UH0) and Lower Highlands (LH0)
are so wet and steep that forest is the best land use. In
the Lower Highlands Zone (LH1) and Upper Midland
Zone (UM1) precipitation is still 1800 mm or more,
and average annual temperatures are <18°C, with the
predominant cropping systems being tea and coffee.
Compared to the Lower Mid land (LM5) and Inner
Midland (IL5) zones with 600–800 mm of rainfall, the
productivity potential of LH1 or UM1 is more than ten
times; and if the poor soils in LM5 and IL5 are consid-
ered, then the productivity potential of LH1 or UM1 is
much higher than that of LM5 and IL5. The reason for
this low potential in the low rainfall regions is the rap-
idly decreasing rainfall during the agro-humid periods
(i.e. the growing periods for annual crops).
Therefore, to improve farm returns, household
in comes and to reduce poverty rates in the face of cli-
mate change, households in the county have adopted
various management practices, such as planting
drought tolerant crop varieties, early planting and ap-
plication of manure and fertilizers. The most popular
improved maize varieties are the ‘Duma’ and ‘DK’ va-
rieties which are planted by 32 and 27% of the house-
holds, respectively. About 26% of households also
cultivate local maize varieties. In addition to improved
varieties, households are also using inorganic fertiliz-
ers and manure. The main fertilizers used in the
region are di-ammonium phosphate (DAP), nitrogen,
phosphorus and potassium (NPK) and calcium ammo-
nium nitrate (CAN). Apart from agriculture, other
sources of income in the region vary from formal em-
ployment, businesses, and re mittances from family
members. The key source of non-agricultural income
is from informal businesses followed by off-farm
labour. Therefore, there are marked differences be-
tween the AEZs in the county with regard to crop va-
rieties, management practices, and sources of income.
2.2. Conceptual framework and data
The conceptual framework employed in this study
is based on the Agricultural Model Intercomparison
and Improvement Project (AgMIP) regional
integrated assessment framework (Rosenzweig et al.
88
2
Intergovernmental platform on biodiversity and ecosystem
services
Author copy

Mulwa et al.: Agricultural household welfare in Kenya
89
2013). This framework requires experiments, surveys
and expert data for use in crop and economic model-
ling, including crop and livestock production data,
costs of production of different enterprises, and
household characteristics data, e.g. household size
and non-farm income data. In crop modelling, crops
and livestock production data are used to estimate
the relative yield distributions under climate change.
The crop modelling module uses downscaled climate
data. For crops and livestock enterpri ses which could
not be modelled, re presentative agri cultural path-
ways (RAPs) are used to estimate these re lative
yields. The estimates from cli mate mod-
elling are passed into an economic
model and combined with prices and
costs of different crops and livestock en-
terprises to estimate economic impacts
of climate change, as shown in Fig. 2.
To collect the data required for the
analysis, household surveys a from total
of 441 households were collected using
a combination of stratified and multi-
stage sampling in 2013. The strata for
the household survey were the AEZs in
the county. Out of the 11 AEZs, 5 key
AEZs were targeted for data collection.
These were (1) Upper Midland 2 (UM2),
which is the main coffee zone; (2) Upper
Midland 3 (UM3), the marginal coffee
zone; (3) Lower Midland 3 (LM3), the cotton zone; (4)
Lower Midland 4 (LM4), the marginal cotton zone;
and (5) Lower Midland 5 (LM5), the sorghum millet
zone. In each AEZ, administrative regions (division,
location and sub-location) were chosen, and one sub-
location representing each AEZ was chosen for sam-
pling. At the sub-location level, data collection from
individual households was by simple random sam-
pling. The data collected in these AEZs included: pro-
duction of different varieties of crop and livestock,
output prices and variable production costs of the
different enterprises, non-farm incomes, and other
Fig. 1. Map display-
ing target agro-eco-
logical zones (AEZs)
of Embu County in
Kenya. See Section
2.1 for descriptions
of the AEZs
Fig. 2. The AgMIP regional integrated assessment framework. Source: Rosen-
zweig et al. (2013). RCP = representative concentration pathway; SSP = shared
socio-economic pathway; RAP = representative agriculturalpathway;TOA-MD=
trade-off analysis model for multi-dimensional impact assessment
Author copy

Clim Res 67: 87–97, 2016
90
household charac teristics. All AEZs had maize and
dairy activities but UM2 and UM3 also had coffee.
The other AEZs had pigeon peas and sorghum/ millet
enterprises which were not in UM2 and UM3. In total
there were 5 crop activities and dairy. There are
marked differences in the house hold characteristics
among the AEZs. For instance, UM2 has the smallest
geographical area and lowest mean
household size. Other AEZ house-
hold characteristics are as shown
in Table 1.
In UM2, the key crops are maize
and beans, but households also
grow coffee as a cash crop. They
also plant bananas, vegetables,
sweet potatoes, etc. for private con-
sumption and sale to markets. UM3
has similar crops to UM2, but the
mean farm size is larger (as shown
in Table 1). Households in LM3
plant the same crops as in UM2
and UM3 (except for coffee), plus
sorghum and millet in some house-
holds. Households in LM4 and
LM5 plant the same crops as
grown in LM3. All AEZs produce
milk but at varying quantities, with
the highest production in UM2
(Table 2). The prices of farmers’
produce are fairly similar in all the
AEZs, as shown in Table 3. The
mean crop and livestock outputs
and mean prices are indicators of
mean gross revenues in the AEZs.
The net farm returns in different
farms in the AEZs are determined
by the costs of production. Table 4
shows the mean costs of produc-
tion for different enterprises.
These costs vary across the differ-
ent enterprises and across the
AEZs. Maize has the highest cost
of production in all AEZs except in
UM2, where dairy and coffee pro-
duction recorded higher costs of
production.
2.3. Climate, crop and economic
modeling
This being a multi-disciplinary
study, it involved climate, crop and
economic modelling. Five climate models — CCSM4,
GFDL-ESM, HadGEM2-ESM, MIROC-5 and MPI-
ESM — were used to downscale future projected cli-
mate change scenarios for the mid-century time-
frame of 2041−2070. Crop modelling for maize was
carried out using the Decision Support System for
Agrotechnology Transfer (DSSAT) and Agricultural
AEZ Geographical Mean Mean annual non- Mean farm Mean dairy
area (ha) HH size agric. income (Ksh) size (ha) herd size
UM2 4370 4.3 114 411 0.9 2.29
UM3 8400 5.7 67 902 2.21 1.79
LM3 18 020 5.8 148 525 1.86 1.83
LM4 60 420 6.5 146 671 2.44 2.20
LM5 105 500 6.9 118877 1.74 1.88
Table 1. Characteristics of AEZs and households (HHs; HH size is no. of people).
Ksh: Kenyan shillings. Source: Jaetzold et al. (2007) and household survey data
(this study)
AEZ Maize Beans Coffee Pigeon Sorghum Milk
pea and millet
UM2 2191.20 1113.09 1836.80 0.00 0.00 1938.39
UM3 2273.20 1275.41 1850.93 0.00 0.00 679.52
LM3 1935.09 1481.33 0.00 890.47 643.49 1019.95
LM4 1675.40 1254.99 0.00 1154.20 921.76 1153.30
LM5 877.04 645.61 0.00 674.45 626.75 781.29
Table 2. Mean yield of different crops (kg ha
−1
yr
−1
) and dairy (kg farm
−1
yr
−1
) in
AEZs. Source: household survey data (this study)
AEZ Maize Beans Coffee Pigeon Sorghum Dairy
pea and millet (milk)
UM2 32.00 58.89 28.70 N/A N/A 30.20
UM3 30.70 55.35 22.67 N/A N/A 31.20
LM3 31.40 60.82 N/A 58.94 43.15 38.40
LM4 31.40 54.24 N/A 58.29 45.38 42.50
LM5 31.30 80.00 N/A 59.64 43.89 38.20
Table 3. Mean prices (in Kenyan shillings, Ksh) of different crops (Ksh kg
−1
) and dairy
(kg l
−1
) in AEZs. N/A: not applicable. Source: household survey data (this study)
AEZ Maize Beans Coffee Pigeon Sorghum Dairy
pea and millet (milk)
UM2 6952.06 2718.13 6977.03 N/A N/A 18 110.24
UM3 19 255.25 4098.46 17996.13 N/A N/A 9436.22
LM3 20 980.85 3079.52 N/A 1232.66 725.01 8407.54
LM4 14 046.51 3733.04 N/A 1182.81 1160.19 7731.88
LM5 6506.86 2147.13 N/A 776.86 1301.59 982.50
Table 4. Mean cost of production (in Kenyan shillings, Ksh) for different crops and
dairy (Ksh farm
−1
yr
−1
) in AEZs. N/A: not applicable. Source: household survey
data (this study)
Author copy

Mulwa et al.: Agricultural household welfare in Kenya
Production Systems sIMulator (APSIM) crop models.
RAPs were used to project the impacts of, and adap-
tations to, climate change for other farm activities,
which could not be modelled using crop models.
These were beans, coffee, pigeon peas, sorghum and
millet, and dairy. Economic analyses were performed
using the Trade-off Ana lysis Multi-Dimensional
(TOA-MD) impact as sess ment tool (Antle & Valdivia
2010, 2011, Antle 2011, Claessens et al. 2012).
2.4. Climate modeling
Since it is not practical to assess impacts of climate
change on agricultural systems at the local scale with
coarse data from coupled atmosphere−ocean general
circulation models (AOGCMs), location-specific cli-
mate change scenarios were developed using the
delta
3
method. Using this method, monthly changes
in temperature and precipitation from AOGCMs, cal-
culated at the grid scale, were added to the corre-
sponding observed station data. Climate change sce-
narios for mid-century (2041−2070) and end-century
(2071− 2100) periods were developed for 20
AOGCMs from the Coupled Model Inter-comparison
Project phase 5 (CMIP5) for 2 Representative Con-
centration Pathways (RCPs), 4.5 and 8.5. The climate
change scenarios were developed and analyzed for
all the stations used in this assessment. Future pro-
jected climate change scenarios were downscaled
with the delta method for 20 CMIP5 climate models.
However, note that in this study we are considering 5
GCMs (CCSM4, HadGEM2ES, MIROC5, MPI ESM
and GFDL); the mid-century (2041−2070) scenario;
and RCP 8.5 presents a higher carbon dioxide con-
centration compared to RCP 4.5.
2.5. Crop modeling
Long-term historical climate data for the baseline
period 1980−2010 for 4 locations in the target county
were collected from the archives of the Kenya
Meteo rological Department (KMD). Efforts were
made to collect daily observations on all parame-
ters — rainfall, maximum temperature, minimum
temperature and solar radiatio — that are required to
operationalize the crop models. Two plot-specific
crop simulation models, DSSAT and APSIM were
selected for the study. The data on maize
4
(H511,
H513 and ‘Katumani’ varieties) crop phenology, bio-
mass, yields and management practices for the long-
and short-range seasons over 2 years (2000 and 2001)
were obtained from Kenya Agricultural and Live-
stock Research Organization (KALRO). Yield experi-
mental data were selected that represented the 3
major maize varieties and the major crop growing
seasons. In addition to crop experimental data, his-
torical daily weather data (rainfall, maximum and
minimum temperatures, solar radiation and relative
humidity) for the period 1980 to 2012, and soil data
(including albedo, surface runoff curve number, tex-
ture; water-holding capacity at drained lower and
upper limits and at saturation; bulk density; pH; and
organic carbon for 4 to 5 layers of soil) and site infor-
mation were collected. Varieties were calibrated for
4 main parameters — days to flowering, days to
maturity, and grain and biomass yields at harvest.
For some varieties such as ‘Katumani’, default para -
meters that are available in APSIM and DSSAT mod-
els needed no further adjustments. For other vari-
eties, parameters were derived by manipulating the
thermal time required to complete various growth
stages until the simulated phenology matched the
observed pheno logy. Simulations with a final set of
parameters by both models indicated a good rela-
tionship be tween observed and simulated days to
flowering and days to maturity. The model-simulated
biomass and grain yield were closely related to the
observed data.
2.6. Economic modeling
For the analysis of impacts of climate change on
current agricultural systems we used the TOA-MD.
The TOA-MD model is a parsimonious, generic
model for analysis of technology adoption and impact
assessment, and ecosystem services analysis (Antle
2011). In this model, households are presented with a
simple binary choice: they can operate with a current
or base production System 1 (current climate, current
technology), or they can switch to an alternative sys-
tem (Claessens et al. 2012). The empirical model is
therefore based on the random utility theory where
a household is assumed to maximize a welfare-
91
3
The delta method assumes that future model biases for both
mean and variability will be the same as those in present
day simulations (Mote & Salathé 2009)
4
Note that only maize was simulated using APSIM and
DSSAT. For other crops such as beans, coffee, pigeon peas,
sorghum/millet, future production, price and variable costs
of production were estimated using RAPs based on expert
opinion
Author copy

Clim Res 67: 87–97, 2016
enhancing factor such as utility (McFadden 1973).
The random utility model theory posits that the utility
U that an individual or household i gains from partic-
ipating in system h is made up of an observable
deterministic component V (of observable attributes)
and a random component ε . Thus, the random utility
function is represented as (Greene 2003):
U
ih
= V
ih
+ ε
ih
(1a)
Therefore, an individual or a household is assumed
to maximize utility from a given system if the utility
derived from participating in the system (U
h
) is
greater than the individual/household utility before
participation in this system. This equation can be
presented as (Maddala 2001):
U
ih
= V(X
h
) + ε
ih
(1b)
where, for any individual or household i, a given
level of utility U will be associated with participation
in a certain system h. Thus, system h will be chosen
over some other system k if U
h
> U
k
. The utility
derived from participation in a project is assumed to
depend on the attributes of the system X
h
and the
attributes of the individual Z
i
(Maddala 2001).
This theoretical foundation can be empirically
translated into the TOA-MD, which has been used
for the analysis of technology adoption (Claessens et
al. 2008, 2010); payments for environmental services
(Antle & Valdivia 2006, Immerzeel et al. 2008, Antle
et al. 2010); and evaluating climate change adapta-
tions (Claessens et al. 2012). To motivate the TOA-
MD model, assume a household at a site s using a
production system h which is defined as a combina-
tion of technology, climate and a certain RAP. The
returns per hectare (V) in a certain period t — usually
a year — for this household is equal to v
t
= v
t
(s, h) . If
this household was to earn this annuity over T time
periods, the discounted net return for this system h
would be given as:
(2)
where, NR(s,h) is the discounted net return for sys-
tem h over time T; δ
t
is the discount factor; and r is the
interest rate. If there is a change in technology or cli-
mate or both, then the production in system h is also
expected to change and so are the expected returns
of the system. This change ushers in a new system of
production which we christen system k. The effect of
changing from system h to the new system k on
household’s returns can be expressed as:
ω(s,h,k) = NR(s,h) – NR(s,k) (3)
where NR are the net farm returns. If ω(s,h,k) is pos-
itive it means that the net farm returns in the old sys-
tem are higher than in the new system and hence a
loss or opportunity cost is associated with switching
from system h to system k. If the figure is negative, it
represents the gain in switching to system k. If we
were to define the spatial distribution of gains or
losses in the population of farms indexed by s using
the density function ϕ(ω|h,k); then the percent of
farms with ω(s,h,k) < a (where a is an amount in dol-
lars per ha or acre) is given by:
(4)
If we assume that climate changes and households
do not adapt, then their only option is to use the same
technology with the new climate (say System 2). The
task, therefore, is to compare the performance of
farm-level indicators under System 1 with current
technology and current climate, and System 2 where
climate has changed but households retain the same
technology (i.e. not adapted to climate change). In
this case, Eq. (4) above is interpreted as showing the
proportion of farms with losses less than a, i.e.
ω(s,1,2). Therefore, r(0,1,2) can be interpreted as the
proportion of farms that are positively impacted by
climate change and 1 – r(0,1,2) is the proportion of
farms that is negatively impacted by climate change.
Although it is possible for farmers to adapt to new
technologies, this study does not extend to the
impacts of adaptation to climate change, but restricts
itself to the impacts of climate change on farmers in
the case where climate changes but farmers fail to
adapt, or what we would call the ‘pure climate
change impact’. The comparison is therefore be -
tween a scenario of current climate with current pro-
duction technology against a future climate but still
using current technology. To capture this pure cli-
mate change impact it is assumed that the climate
will change but in terms of technology, households
will continue with ‘business as usual’. This compari-
son can be presented as: (a) System 1 = current cli-
mate, current technology; and, (b) System 2 = future
climate, current technology.
For the purposes of this study, current climate is de -
fined as what exists now, i.e. as influenced by present
direct drivers e.g. prevailing rainfall and tempera-
tures as currently experienced by farms. Current
technology is comprised of existing management
practices, crop varieties, farm sizes etc. To capture
the pure climate change impact on production, we
assume that this present technology does not change
in the different AEZs. We however assume that cli-
∫
= ϕω ω
−∞
( , , ) 100 ( , )drahk hk
a
∑
= δ⇒
−δ
⎡
⎣
⎢
⎤
⎦
⎥
=
−
(, ) (, ) (, )
1
1
NRsh v sh v sh
r
t
T
tt t
T
92
Author copy

Mulwa et al.: Agricultural household welfare in Kenya
mate does change (as indicated by relative crop
yields and as estimated by crop modellers and
RAPS). This would give rise to a future climate with
changes in rainfall and temperature which are
expected to change the levels of crop production.
Indirect drivers such as technology are held constant,
but economic factors such as prices of inputs are
expected to change. If one wanted to capture the
impact of adaptation, then both climate and technol-
ogy would change, but this is beyond the scope of
this study. For our analysis, the present production
system under current climate and current technology
(System 1) is shocked with future (2041–2070) cli-
mate (System 2) to determine how it re sponds to such
a shock.
3. RESULTS
3.1. Relative yield distributions
The relative yield distributions in Table 5 were
obtained from crop modelling, both in APSIM and
DSSAT models. They depict the expected changes in
maize production based on the different climate
models in all the 5 AEZs. For instance, in APSIM,
CCSM4 indicates a 25% decline in maize production
in LM4, and minor declines or increments in the
other AEZs. The other models yield changes in pro-
duction ranging from a 12% decline in LM4 to a 7%
increase in UM2 and LM5. In DSSAT, the increments
are much larger and range from 6 to 63% increase in
maize production while the declines are between 2
and 8% (Table 5).
Based on results from the maize simulation, histor-
ical data and expert opinion, we made certain as -
sumptions on expected changes in other crops in the
system which were not simulated using APSIM or
DSSAT. For instance with climate change, bean pro-
duction is expected to increase by 10% in UM2, UM3
and LM3, and decline by 10% in LM4 and LM5. Cof-
fee is grown in UM2 and UM3, both of which gain
from climate change, hence its production is ex -
pected to increase by 20% in both AEZs. Pigeon pea
and sorghum are drought-tolerant crops grown in
marginal areas and are not expected to be adversely
affected by climate change. In fact, the simultaneous
increment in rainfall and temperature in the region is
expected to boost pigeon pea and sorghum produc-
tion by 20 and 15%, respectively, in LM3, and
decrease production of both crops by 10% in LM4
and LM5 (Table 6). Dairy production is also expected
to increase by 10% in all AEZs.
3.2. Losers from climate change
Once again, note that the aim of our analysis is to
assess the impact of future climate on current agri-
cultural systems, i.e. to shock the current system with
climate change and observe changes in welfare indi-
cators. Therefore, output prices, both for crops and
livestock (dairy), were held constant (Table 3). How-
ever, production costs for all enterprises are expected
to change, as production changes, e.g. an increase
(decrease) in bean output is expected to increase
(decrease) variable costs of production. Other house-
hold characteristics such as farm size, herd size, non-
agricultural income, etc. are also assumed to remain
93
AEZ Observed yield Time-averaged relative yield (r = S
2
/S
1
)
(kg ha
−1
) APSIM DSSAT
CCSM4 GFDL HadGEM MIROC MPI CCSM4 GFDL HadGEM MIROC MPI
UM2 2191.20 0.97 1.06 1.07 1.06 1.03 1.12 1.17 1.07 1.11 1.23
UM3 2273.20 1.00 1.02 0.98 1.02 0.99 1.35 1.39 1.27 1.28 1.44
LM3 1935.09 1.02 1.03 1.00 1.02 1.00 1.32 1.51 1.63 1.24 1.41
LM4 1675.40 0.75 1.05 1.06 0.91 0.88 0.98 1.12 1.30 0.92 1.06
LM5 877.04 0.98 1.06 1.07 1.02 0.89 1.13 1.27 1.27 1.29 1.23
Table 5. Simulated and observed maize yields in different agro-ecological zones (AEZs). <1 and >1 indicate that climate change
has a negative or positive impact, respectively, on maize production. r = S
2
/S
1
is the relative yield, where S
2
is the future simulated
yield (System 2) and S
1
is the base simulated yield (System 1)
Crop Beans Coffee Pigeon pea Sorghum Dairy
UM2 1.1 1.2 N/A N/A 1.1
UM3 1.1 1.2 N/A N/A 1.1
LM3 1.1 N/A 1.1 1.15 1.1
LM4 0.9 N/A 0.9 0.9 1.1
LM5 0.9 N/A 0.9 0.9 1.1
Table 6. Expected relative yields (r = S
2
/S
1
) of non-model -
led crops using representative agricultural pathways
(RAPs). N/A: not applicable
Author copy

Clim Res 67: 87–97, 2016
constant. Any change between the two systems is
therefore purely the effect of climate on the current
system. For impact analysis, the household and farm
characteristics highlighted earlier, production levels
of different crops and diary, output prices, and costs
of production for the current system were keyed-in
the TOA-MD model and compared with a future sys-
tem where production levels and costs of production
are changed due to climate change. Results from
analysis indicate that if the current production sys-
tem in Embu County is subjected to climate change
shock, then LM4 and LM5 will have the highest num-
ber of losers. This is because climatic changes in
these two AEZs is expected to have adverse effects
on crop production and other ecosystems goods and
services from the farms. From APSIM simulations,
about 36−56% of the households in LM5 and 39−
66% in LM4 are expected to be worse off than they
are today if the current system was to be subjected to
climate change. The figures are lower using DSSAT
estimations (Table 7). Losses in UM3 and LM3 are
lower in both APSIM and DSSAT for all GCMs.
3.3. Impact of climate change on net farm returns
The impact of climate change on net farm returns is
as shown in Table 8 both for APSIM and DSSAT,
respectively. In the APSIM analysis, for all the AEZs
(except CCSM4 for LM4; and MPI-ESM for LM4 and
LM5), climate change will have a positive impact on
net farm returns. In some GCMs, declines in maize
production were recorded as reported in Table 6.
Therefore, the positive net farm returns could be
explained by net farm returns from other crops (cof-
fee, beans, pigeon peas, and sorghum), which are
expected to increase in yields in some AEZs due to
climate change. Examples of this are in UM2 for
CCSM4 and LM4 for MIROC-5, which record a
decline in maize production but increased net farm
returns. In instances where loss in maize production
and loss in other crops was recorded, net farm re -
turns also recorded a decline, e.g. CCSM4 for LM4,
and MPI-ESM for LM4 and LM5. The gains in net
returns are highest in LM3 and UM3 and lowest in
LM5. Results from the DSSAT model indicate gains
in net farm returns in all AEZs, and the trends are
similar to those of APSIM, though higher.
The total gains and losses from climate change
were expressed as a percent of net farm returns for
the different models (Fig. 3). APSIM simulations indi-
cate minimal net impact from climate change, with
GFDL recording the highest net gains of 53.2%, while
MPI-ESM recorded a net gain of 0.45% (Fig. 3a). The
net impacts under DSSAT (Fig. 3b) are almost uni-
formly spread across the different models, with the
least (CCSM4) recording 5.6% net gains, while
HADGEM recorded the highest net gains of 9.1%.
94
AEZs APSIM DSSAT
CCSM4 GFDL HadGEM MIROC-5 MPI-ESM CCSM4 GFDL HadGEM MIROC-5 MPI-ESM
UM2 31.62 28.77 27.79 28.04 29.15 26.19 24.82 28.15 28.24 22.95
UM3 33.05 31.94 33.81 32.00 33.70 21.70 20.56 23.22 27.50 20.94
LM3 29.56 39.03 31.24 29.67 30.33 21.73 19.48 18.62 24.78 20.24
LM4 56.41 44.43 38.80 45.26 51.90 43.56 34.84 26.89 47.74 38.06
LM5 40.10 37.43 39.94 36.05 56.09 36.95 33.81 37.75 34.26 35.67
Mean 43.65 39.30 38.26 37.95 50.89 36.70 32.05 31.83 37.11 34.08
Table 7. Percentage of households expected to be worse-off (‘losers’) with climate change
AEZs APSIM DSSAT
CCSM4 GFDL HadGEM MIROC-5 MPI-ESM CCSM4 GFDL HadGEM MIROC-5 MPI-ESM
UM2 82 99 106 103 97 116 124 102 122 145
UM3 104 116 98 112 99 315 335 274 276 372
LM3 120 178 108 119 115 265 357 414 223 311
LM4 -67 115 92 107 -8 35 95 165 12 68
LM5 24 29 27 33 -20 30 39 29 103 34
Mean 17 70 56 65 5 68 97 110 100 85
Table 8. Change in net farm returns (US$ farm
−1
yr
−1
) with climate change (APSIM and DSSAT)
Author copy

Mulwa et al.: Agricultural household welfare in Kenya
3.4. Impact of climate change on per capita income
The sensitivity of per capita income to climate
change is estimated by considering the difference in
per capita income (in US$ d
−1
) between System 1 and
System 2 (Table 9). The results indicate that climate
change will cause a change of varying de-
grees in per capita income in different
AEZs, with the highest gains made in
UM2, UM3 and LM3. In LM4, CCSM4 un-
der APSIM predicts a decline in per capita
income, while the gains in LM5 are very
marginal. This illustrates that climate
change will have negative impacts on
households in LM4 and LM5. The predic-
tions under DSSAT tell a similar story,
though no AEZ under DSSAT reported
negative change, and households in LM4
and LM5 gained the least from climate
change while those in UM3 and LM3
gained the most.
3.5. Impact of climate change on
poverty
Another important indicator of house-
hold welfare is poverty level. Changes in
poverty levels from the APSIM and DSSAT
models are shown in Table 10. From the APSIM esti-
mates, changes in poverty levels indicate that climate
change will reduce poverty levels in all AEZs for all
GCMs (except CCSM4 in LM4 and MPI-ESM in
LM5). However, the levels do vary, as seen from the
distribution of results across the AEZs. The highest re-
95
AEZs APSIM DSSAT
CCSM4 GFDL HadGEM MIROC-5 MPI-ESM CCSM4 GFDL HadGEM MIROC-5 MPI-ESM
UM2 0.07 0.08 0.09 0.08 0.08 0.09 0.10 0.08 0.10 0.12
UM3 0.06 0.07 0.06 0.06 0.06 0.18 0.19 0.16 0.16 0.21
LM3 0.07 0.10 0.06 0.07 0.07 0.15 0.21 0.24 0.13 0.18
LM4 –0.030 0.06 0.05 0.05 0.00 0.02 0.05 0.08 0.01 0.03
LM5 0.01 0.01 0.01 0.02 –0.010 0.01 0.02 0.01 0.05 0.02
Aggregate 0.03 0.06 0.05 0.06 0.04 0.09 0.11 0.11 0.09 0.11
Table 9. Change in per capita income (US$ d
−1
) with climate change (APSIM and DSSAT)
17.86
53.16
18.24
33.47
13.43
–16.49
–44.90
–13.65
–27.84
–12.98
1.38
8.25
4.59
5.63
0.45
–60.00
–40.00
–20.00
0.00
20.00
40.00
60.00
CCSM4
Proportion of net farm returns (%)
a
b
GFDL HadGEM MIROC-5
Gains Losses Net impact
17.37
20.23
21.63
27.63
19.05
–11.79
–12.24
–12.56
–19.45
–12.00
5.58
7.9 9
9.07
8.18
7.0 4
–10.00
–20.00
–30.00
0.00
10.00
20.00
30.00
CCSM4 GFDL HadGEM MIROC-5
MPI-ESM
MPI-ESM
Fig. 3. Gains, losses and net impact as percent of net farm returns for all
agroecological zones. (a) APSIM, (b) DSSAT
AEZs APSIM DSSAT
CCSM4 GFDL HadGEM MIROC-5 MPI-ESM CCSM4 GFDL HadGEM MIROC–5 MPI-ESM
UM2 –1.4 –1.7 –1.8 –1.8 –1.7 –2.0 –2.1 –1.8 –2.1 –2.5
UM3 –3.4 –3.8 –3.2 –3.7 –3.3 –9.6 –10.20 –8.5 –8.6 –11.10
LM3 –1.8 –1.6 –1.6 –1.8 –1.7 –3.8 –4.8 –5.4 –3.1 –4.2
LM4 1.4 –2.2 –1.6 –1.1 0.2 –0.7 –1.7 –2.9 –0.2 –1.3
LM5 –0.6 –0.8 –0.7 –0.8 0.5 –0.8 –1.01 –0.7 –2.7 –0.9
Aggregate –1.2 –2.0 –1.8 –1.8 –1.2 –3.4 –4.0 –3.9 –5.1 –4.0
Table 10. Percent decline in poverty levels (APSIM and DSSAT)
Author copy

Clim Res 67: 87–97, 2016
duction in the poverty rate is in UM3, where APSIM
projected reductions in poverty of up to 3.8% while
DSSAT projected poverty declines of >11% in UM3.
In both models the smallest poverty rate reduction
was in LM4 and LM5.
4. DISCUSSION
Our results indicate that climate change will have
mixed impacts on maize production (Table 4) in dif-
ferent AEZs based on the GCM and the crop model
used. For instance, using APSIM, climate change
will have a negative impact on maize production in
LM4 and LM5 for CCSM4 and MPI-ESM. Losses are
also recorded in LM4 for MIROC-5 and UM3 for
HadGEM. The other GCMs indicate gains from cli-
mate change under APSIM. Overall, APSIM shows a
negative impact of climate change on current pro-
duction systems for some GCMs, while DSSAT shows
an overall increase in maize production due to cli-
mate change. APSIM results show that ~36−56% of
the households in LM5 and 39−66% in LM4 are pro-
jected to be worse off than they are today were the
current system subjected to climate change. The fig-
ures are lower for DSSAT estimations. According to
the DSSAT results there are positive net farm returns
and per capita income for all GCMs, but these are
low for UM2, LM4 and LM5, and APSIM results indi-
cate very low or negative net farm returns and per
capita income with climate change for LM4 and LM5,
while the other AEZs show gains for all GCMs. This
indicates that households in LM4 and LM5 are more
vulnerable to climate change compared to the other
AEZs. There are also mixed results regarding the
impacts of climate change on poverty levels from the
APSIM and DSSAT estimates.
Climate change is therefore likely to have negative
impacts on maize production in the LM4 and LM5
AEZs of Embu County. This is expected because al-
though these 2 AEZs are not ideal for maize produc-
tion, quite a large number of households still grow it.
In the other AEZs, climate change is likely to have a
positive impact on maize production. Note that Embu
County is on the slopes of Mount Kenya, and temper-
atures in the upper AEZs (UM2, UM3 and LM3) are
sub-optimal for maize. Therefore an in crease in tem-
perature might boost maize production in the AEZs.
Increases or decreases in maize production in the dif-
ferent AEZs influence the levels of the welfare indica-
tors. The other crops, along with dairy production,
also determine the levels of welfare indicators. For in-
stance, certain AEZs report a decline in maize produc-
tion, but with positive net farm returns and per capita
income, an indication that the contribution of the
other crops outweighs the negative impact of climate
change on maize production in that particular AEZ.
This suggests that all crops and livestock in the farm-
ing systems in the County are potentially important in
determining the overall welfare of households under
climate change.
5. CONCLUSIONS AND POLICY IMPLICATIONS
This study analysed the sensitivity of current Ken -
yan agricultural production systems to climate change.
Climate change is expected to have mixed impacts in
Embu County, with some AEZs gaining and others
losing from climate change. The IPCC (2014) WGII re-
port indicates that negative impacts of climate change
on crop yields are more common than positive im-
pacts. The higher-elevation AEZs in Embu County
present an example of where climate change is ex-
pected to have positive impacts on agricultural pro-
duction. The lower-elevation AEZs, which are make
up >80% of Kenya’s farmlands, are expected to be
negatively affected by climate change. We recom-
mend that farmers both from higher- and lower-
elevation AEZs take steps to adapt to climate change.
Though we have shown that farmers in higher AEZs
would gain from climate change, adaptation to climate
change will boost their expected returns. Farmers in
lower AEZs need to adapt to climate change to boost
production and mitigate the potential negative im-
pacts of climate change. This adaptation can be au-
tonomous, where individual farms choose different
adaptations based on available information. However,
planned adaptation led by government or government
agencies and non-governmental organizations (NGOs)
would probably be a better strategy, as it would bene-
fit from government or donor funding and resources.
Finally, note that for the non-modelled crops, esti-
mates from expert opinions were used. Modelling
these crops using APSIM, DSSAT or other crop mod-
els would give a more complete picture of impact of
climate change on current agricultural system in the
county. It would also be interesting to model the
impact of adaptations to climate change on these
welfare indicators.
LITERATURE CITED
Antle JM (2011) Parsimonious multi-dimensional impact
assessment. Am J Agric Econ 93: 1292−1311
Antle JM, Valdivia RO (2006) Modelling the supply of eco-
96
Author copy
➤
➤
➤
➤

Mulwa et al.: Agricultural household welfare in Kenya
system services from agriculture: a minimum-data
approach. Aust J Agric Resour Econ 50: 1−15
Antle JM, Valdivia RO (2010) TOA-MD Version 4: Minimum-
Data Tradeoff Analysis Model.
http: //tradeoffs. oregonstate.
edu/climate/agmip
Antle J, Valdivia R (2011) Methods for assessing economic,
environmental and social impacts of aquaculture tech-
nologies: adoption of integrated agriculture-aquaculture
in Malawi. Oregon State University, Corvallis, OR
Antle JM, Diagana B, Stoorvogel JJ, Valdivia RO (2010)
Minimum-data analysis of ecosystem service supply in
semi-subsistence agricultural systems: evidence from
Kenya and Senegal. Aust J Agric Resour Econ 54:
601−617
Brander L, Gómez-Baggethun G, Martín-López B, Verma M
(2010) The economics of valuing ecosystem services and
biodiversity, In Kumar P (ed) TEEB: The economics of
ecosystems and biodiversity. Ecological and Economic
Foundations, Earth Scan, London. Chapter 5, p 1–133
Claessens L, Stoorvogel JJ, Antle JM (2008) Ex ante assess-
ment of dual-purpose sweet potato in the crop-livestock
system of western Kenya: a minimum-data approach.
Agric Syst 99: 13−22
Claessens L, Stoorvogel JJ, Antle JM (2010) Exploring the
impacts of field interactions on an integrated assessment
of terraced crop systems in the Peruvian Andes. J Land
Use Sci 5: 259−275
Claessens L, Antle JM, Stoorvogel JJ, Valdivia RO, Thorn-
ton PK, Herrero M (2012) A method for evaluating cli-
mate change adaptation strategies for small-scale house-
holds using survey, experimental and modeled data.
Agric Syst 111: 85−95
Diaz S, Demissew S, Carabias J, Joly C and others (2015)
The IPBES conceptual framework — connecting nature
and people. Current Opinion Env Sustainability 14: 1−16
Fisher B, Christie M (2010) Integrating the ecological and
economic dimensions in biodiversity and ecosystem
service valuation. In Kumar P (ed) TEEB: the economics
of ecosystems and biodiversity. Ecological and Economic
Foundations, Earth Scan, London. Chapter 1, p 1–40
Greene WH (2003) Econometric analysis, 5th edn. Prentice
Hall, Upper Saddle River, NJ
Immerzeel W, Stoorvogel JJ, Antle JM (2008) Can payments
for ecosystem services secure the water tower of Tibet?
Agric Syst 96: 52−63
IPCC (2014) Climate change 2014: impacts, adaptation and
vulnerability. Contribution of Working Group II to the
Fifth Assessment Report of the IPPC. Part B: Regional
aspects. Cambridge University Press, Cambridge.
Jaetzold R, Schmidt H, Hornetz B, Shisanya C (2007) Farm
management handbook of kenya: natural conditions and
farm management information, Vol. II, 2nd edn, Part C.
East Kenya: Subpart C1, Eastern Province. Ministry of
Agriculture, Nairobi.
Kelble CR, Loomis DK, Lovelace S, Nuttle WK and others
(2013) The EBM-DPSER conceptual model: integrating
ecosystem services into the DPSIR framework. PLoS
ONE 8: e70766
Maddala AS (2001) Limited dependent and quantitative
variables in econometrics. Cambridge University Press,
Cambridge
McFadden D (1973) Conditional logit analysis of qualitative
choice behaviour. In: Zarembka P (ed) Frontiers in
econometrics. Academic Press, New York, NY
Mote PW, Salathé EP Jr (2010) Future climate in the Pacific
Northwest. Clim Change 102: 29−50
Republic of Kenya (2010) National climate change response
strategy. Government Press, Nairobi
Rosenzweig C, Jones J, Hatfield J, Antle J (2013) Guide for
regional integrated assessments: handbook of methods
and procedures, Version 5. Agricultural Model Intercom-
parison and Improvement Project (AgMIP). www.
agmip.org/refbase/uploads/cynthiarosenzweig/2013/
170 _ CynthiaRosenzweig2013.pdf
WFP (World Food Programme) (2012) Countries: Kenya —
overview. https://www.wfp.org/node/3497/2608/279525
97
Editorial responsibility: Eduardo Zorita,
Geesthacht, Germany
Submitted: March 16, 2015; Accepted: November 21, 2015
Proofs received from author(s): January 26, 2016
➤
➤
➤
➤
➤
➤
➤
➤
Author copy
➤
➤
➤
➤
➤
➤
➤
➤