ArticlePDF Available

Abstract and Figures

Coffee farming provides livelihoods for around 15 million farmers in Ethiopia and generates a quarter of the country's export earnings. Against a backdrop of rapidly increasing temperatures and decreasing rainfall, there is an urgent need to understand the influence of climate change on coffee production. Using a modelling approach in combination with remote sensing, supported by rigorous ground-truthing, we project changes in suitability for coffee farming under various climate change scenarios, specifically by assessing the exposure of coffee farming to future climatic shifts. We show that 39–59% of the current growing area could experience climatic changes that are large enough to render them unsuitable for coffee farming, in the absence of significant interventions or major influencing factors. Conversely, relocation of coffee areas, in combination with forest conservation or re-establishment, could see at least a fourfold (>400%) increase in suitable coffee farming area. We identify key coffee-growing areas that are susceptible to climate change, as well as those that are climatically resilient.
No caption available
Content may be subject to copyright.
Resilience potential of the Ethiopian coffee sector
under climate change
Justin Moat1,2*, Jenny Williams1, Susana Baena1,2, Timothy Wilkinson1, Tadesse W. Gole3,
Zeleke K. Challa3, Sebsebe Demissew1,4 and Aaron P. Davis1*
Coffee farming provides livelihoods for around 15 million farmers in Ethiopia and generates a quarter of the countrys
export earnings. Against a backdrop of rapidly increasing temperatures and decreasing rainfall, there is an urgent need to
understand the inuence of climate change on coffee production. Using a modelling approach in combination with remote
sensing, supported by rigorous ground-truthing, we project changes in suitability for coffee farming under various climate
change scenarios, specically by assessing the exposure of coffee farming to future climatic shifts. We show that 3959%
of the current growing area could experience climatic changes that are large enough to render them unsuitable for coffee
farming, in the absence of signicant interventions or major inuencing factors. Conversely, relocation of coffee areas, in
combination with forest conservation or re-establishment, could see at least a fourfold (>400%) increase in suitable
coffee farming area. We identify key coffee-growing areas that are susceptible to climate change, as well as those that are
climatically resilient.
Arabica coffee (Coffea arabica) provides Ethiopia with its most
important agricultural commodity, contributing around
one quarter of its total export earnings1. In 2015/16, Ethiopia
exported around 180,000 metric tonnes of coffee2at a value in
excess of 800 million USD, making it Africas largest coffee producer
and the worldsfth largest coffee exporter2, despite the fact that
around half of the coffee produced each year is consumed in-
country3. Coffee farming provides a livelihood income for around
15 million Ethiopians (16% of the population), based on four million
small-holder farms2,4.
In Ethiopia, coffee is produced within specic agro-ecological
zones, over numerous geographical and political boundaries. At
least 80% of Ethiopias coffee comes from forests, forest-like habi-
tats, or farms with shade (canopy) cover, representing land coverage
of around 19,000 km
(Fig. 1) with around another 4,000 km
(c. 20%) grown in small plots in partial shade or full sun. Most of
the coffee is grown in areas that are covered, or were previously
covered5, with humid evergreen forest: Moist Afromontane Forest
(MAF) and Transitional Rain Forest (TRF)6. MAF and TRF are
found at 6502,600 m (4503,000 m including extremes), although
coffee is mostly conned to 1,2002,200 m. Coffee farming is also
undertaken in association with a drier type of vegetation, classied
as Dry Afromontane Forest6, such as that found in the Harar Zone
(Fig. 1). The main coffee-growing areas of Ethiopia are found
within the south-west and south-east (Oromia Region and
Southern Nations, Nationalities and PeoplesRegion), with
modest and minor production in the north (Amhara Region and
BenishangulGumuz Region, respectively) (Fig. 1).
Coffee farmers and other coffee sector stakeholders in Ethiopia
and East Africa report that coffee production has been negatively
inuenced by changes in climate. These changes include: an
increase in the uncertainty of yearly weather patterns, particularly
in precipitation variability and timing of the wet season; an exten-
sion of the dry season (shortening of the wet season); a more
extreme (drier and hotter) end to the main dry season; more intense
(extreme) weather (heavier rain, hotter days); and warmer nights7,8.
Historical climate data shows that the mean annual temperature
of Ethiopia has increased by 1.3 °C between 1960 and 2006, at an
average rate of 0.28 °C per decade9and by 0.3 °C per decade in
the south-west10 and Amhara in the north11. These temperature
increases have been most rapid in the main wet season (July to
September) at a rate of 0.32 °C per decade9. Analyses of extreme
temperature changes in various coffee-growing areas indicate posi-
tive trends for maximum temperature, warm days, warm nights and
warm spell duration; and negative trends for cool days, cool nights,
and cold spell duration across different eco/agricultural environ-
ments (pastoral, agro-pastoral and highland), although some of
the trends are not statistically signicant12. The strong variability
within Ethiopias annual and decadal rainfall makes it difcult to
detect long-term, country-wide trends13. Despite these limitations,
studies show: that February to May14 and June to September rains
have declined13,15,16;a1520% reduction in rainfall since the mid-
1970s17 and late 2000s in southern, south-western and south-
eastern Ethiopia, particularly in the period of the initial early
(February to March) rains in the south-east and east18,19, with an
increase in drought frequency in all parts of Ethiopia during the
last 1015 years16; a decrease in June to September precipitation
in the Greater Horn of Africa by approximately 1 s.d. during the
period 19501989, corresponding to decreases of over 30 mm per
decade throughout much of the Ethiopian Highlands20; and a down-
ward trend in rainfall of 0.4 mm per month per year over the south-
western region in the period 1948200610.
The mean annual temperature of Ethiopia is projected to
increase by 1.13.1 °C by the 2060s, and 1.55.1 °C by the 2090s,
depending on the emission scenario9. Climate model projections
under the emission scenarios A2 and B1 show Ethiopia warming
in all four seasons across the country21. Projections from different
General Circulation Models (GCMs) are broadly consistent in
1Royal Botanic Gardens, Kew, Richmond, Surrey TW9 3AE, UK. 2School of Geography, University of Nottingham, Nottingham NG7 2RD, UK.
3Environment and Coffee Forest Forum (ECFF), PO Box 28513, Addis Ababa, Ethiopia. 4The National Herbarium, Department of Plant Biology
and Biodiversity Management, College of Natural Sciences, Addis Ababa University, PO Box 3434, Addis Ababa, Ethiopia.
NATURE PLANTS 3, 17081 (2017) | DOI: 10.1038/nplants.2017.81 | 1
© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
indicating slight increases in annual rainfall in Ethiopia22,23.
However, it should be made very clear that precipitation projections
vary considerably in their agreement; some increases are negligible;
the variability for GCM change is extremely high24; and for some
regions of Ethiopia GCMs do not agree on the direction of change21.
It is imperative to better understand the inuence of climate
change on coffee farming in Ethiopia, given the importance of
coffee production and consumption for this country. The impacts
of observed and projected changes in climate on coffee farming in
Ethiopia are largely unknown. The only available climate change
study for Ethiopian coffee is for native populations of Arabica
coffee, using a single species distribution model (SDM),
forest-only data and one GCM25.
In order to quantify the inuence of climate change on coffee
farming in Ethiopia, we collated and collected high-quality
ground-point data, from across the Ethiopian coffee landscape and
beyond, then modelled using an ensemble SDM approach, which
was overlaid with Landsat 8 satellite imagery26 to accurately
delimit the main coffee production areas (humid forest). This
approach, in conjunction with the use of multiple migration scen-
arios, ne-level suitability thresholding, multiple GCM appraisal
and rigorous ground-truthing (of models and remote sensing
outputs), constitutes a signicant advancement over previous
coffee climate change analyses generally25,2733. Nine environmental
variables (from BIOCLIM34) were used to understand the current
climate conditions for coffee farming, and to project suitable
growing areas for the future. Using these data and methods we quan-
tify the potential outcomes of climate change on coffee production
across the coffee landscape of Ethiopia until 2099. Our analyses are
primarily focused on exposure (based on climate scenario projec-
tions from GCMs). The two other aspects of climate change vulner-
ability, sensitivity and adaptive capacity35,36, have not been directly
addressed, although it should be noted that Arabica coffee is ident-
ied as a climate-sensitive species with a low adaptive capacity25.
The potentially benecial inuence of elevated CO
and compounding
negative inuences (for example, pests, diseases and deforestation)
were not included in our analyses, but are discussed.
Model projections overview. We modelled Arabica coffee suitability
for Ethiopia using the World Climate Research Programmes
(WCRPs) Coupled Model Intercomparison Project phase 3
(CMIP3)37, across four time intervals (19601990, 20102039,
20402069 and 20702099; Supplementary Fig. 1), using
South west
South east
S. Sudan
Dire Dawa
Bahir Dar
Addis Ababa
Debre Birhan
Shire (Inda Silase)
Arba Minch
Asbe Teferi
Mizan Teferi
0 50 100 200 300
Jimma-Limu Rift North
Bench Maji
West Hararge
(incl. Yirgachee)
Rift South
East Hararge
Central Eastern
Figure 1 | The main coffee growing zones and areas of Ethiopia. The coffee zones represented by coloured polygons: red/pink, North Zone
(coffee areas: Amhara and Benishangul Gumuz); light blue, South West Zone (coffee areas: Wellega, Illubabor, Jimma-Limu, Kaffa, Tepi and Bench Maji);
light green, Rift Zone (coffee areas: Rift North and Rift South); dark blue, South East Zone (coffee areas: Sidamo, Yirgacheffe, Bale and Central Eastern
Highlands); dark green, Harar Zone (coffee areas: Arsi, West Hararge and East Hararge).
NATURE PLANTS 3, 17081 (2017) | DOI: 10.1038/nplants.2017.81 |
© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
two emission scenarios (A1B and A2), six migration (coffee area
movement) scenarios (Box 1; Supplementary Fig. 5), two
biological/climatic coffee groups/areas and 23 GCMs, selecting
three GCMs for our nal analysis (see Methods). Coffee growing
suitability was divided into ve categories, based on statistical
thresholding: Unsuitable, Marginal, Fair, Good and Excellent (see
Methods; Box 2; Supplementary Fig. 2). We restrict our narrative
here to two migration scenarios: A and D (Box 1), which we refer
to as Full Migration (A)and No Migration (D). The Full
Migration (A) scenario represents the (maximum) potential niche
(suitability) for coffee in Ethiopia, invoking an assisted migration
scenario, where coffee areas can move to newly available niches,
regardless of whether there is forest or not (Fig. 2). This scenario
would require considerable intervention, including the provision of
canopy cover (tree planting) in many places. The No Migration (D)
scenario represents the actual (near-minimum; where scenario F
is the minimum) coffee-growing surface area, based on the
recent past (19601990) (Fig. 3). It assumes non-migration of
coffee areas, either natural25 or assisted (by planting coffee and
shade cover) and also that there is zero forest loss or gain, but
allows for climate change within the niche. Summary data for
other migration scenarios are supplied in Table 1 and
Box 1 | Migration scenarios.
Six migration scenarios used in the modelling (see
Supplementary Fig. 5 for further details, and Methods for proces-
sing details).
(A) Full Migration. Plants can grow in any suitable niche
(can move anywhere).
(B) Plants can only grow within known niche (can only move
within presently predicted niche).
(C) Plants can only grow within suitable forest cover, within
any suitable niche (can move within suitable forest).
(D) No Migration. Plants can only grow within suitable forest
cover and only in suitable known niche (restricted to
present-day forest cover and suitable niches).
(E) Plants can only grow within suitable niche but only if
niche does not drop outside of suitability during any
30-year time period.
(F) Plants can only grow within suitable forest cover and suit-
able niche, but only if niche does not drop outside of suit-
ability during any 30-year time period.
Table 1 | Total suitable niche area change for three selected GCMs.
(migration emission)
Area change for low: middle: high outcomes (km
with percentage area remaining in parentheses)
19601990 20102039 20402069 20702099
AA1B 44,820: 44,820: 44,820
(100: 100: 100)
60,996: 66,158: 74,873
(136: 148: 167)
50,387: 58,036: 72,033
(112: 129: 161)
42,247: 51,280: 61,230
(94: 114: 137)
BA1B 44,820: 44,820: 44,820
(100: 100: 100)
29,192: 35,270: 41,195
(65: 79: 92)
27,110: 29,838: 35,597
(60: 67: 79)
19,553: 24,379: 26,809
(44: 54: 60)
EA1B 44,820: 44,820: 44,820
(100: 100: 100)
29,192: 35,270: 41,195
(65: 79: 92)
26,287: 29,079: 34,923
(59: 65: 78)
18,427: 23,389: 26,110
(41: 52: 58)
AA2 44,820: 44,820: 44,820
(100: 100: 100)
62,774: 74,272: 90,527
(140: 166: 202)
54,722: 57,118: 68,728
(122: 127: 153)
42,004: 50,498: 56,371
(94: 113: 126)
BA2 44,820: 44,820: 44,820
(100: 100: 100)
31,245: 34,904: 42,046
(70: 78: 94)
28,054: 32,138: 33,438
(63: 72: 75)
19,101: 22,468: 27,647
(43: 50: 62)
EA2 44,820: 44,820: 44,820
(100: 100: 100)
31,245: 34,904: 42,046
(70: 78: 94)
27,617: 31,717: 33,091
(62: 71: 74)
17,648: 21,415: 26,577
(39: 48: 59)
CA1B 19,142: 19,142: 19,142
(100: 100: 100)
17,451: 20,575: 23,855
(91: 107: 125)
17,330: 19,588: 22,082
(91: 102: 115)
14,479: 17,139: 20,378
(76: 90: 106)
DA1B 19,142: 19,142: 19,142
(100: 100: 100)
11,370: 14,319: 17,130
(59: 75: 89)
12,131: 12,897: 14,335
(63: 67: 75)
8,604: 11,256: 12,311
(45: 59: 64)
FA1B 19,142: 19,142: 19,142
(100: 100: 100)
11,370: 14,319: 17,130
(59: 75: 89)
11,817: 12,556: 14,100
(62: 66: 74)
8,377: 10,464: 12,058
(44: 55: 63)
CA2 19,142: 19,142: 19,142
(100: 100: 100)
19,111: 21,072: 25,040
(100: 110: 131)
18,092: 20,264: 20,742
(95: 106: 108)
14,189: 17,781: 19,225
(74: 93: 100)
DA2 19,142: 19,142: 19,142
(100: 100: 100)
13,257: 14,177: 16,319
(69: 74: 85)
11,295: 12,983: 14,964
(59: 68: 78)
8,584: 9,933: 13,013
(45: 52: 68)
FA2 19,142: 19,142: 19,142
(100: 100: 100)
13,257: 14,177: 16,319
(69: 74: 85)
11,105: 12,789: 14,858
(58: 67: 78)
7,910: 9,502: 12,622
(41: 50: 66)
Percentage and km
change (that remaining) in surface areafor coffee niche suitability, against 19601990, for low, middleand high outcomes (see Methods for explanation) for migrationscenarios (AF) and
emission scenarios A1B and A2. GCMs: GFDL-CM2.1,CSIRO-MK3.5 and BCCR-BCM2.0. The table issplit: the top section containsniche-only scenarios(A, B, E), while the bottom section consists offorest and
niche scenarios (C, D, F).
Box 2 | Niche classications.
Environmental niche (coffee suitability) categories (see
Supplementary Fig. 2, Methods and Supplementary
Information for category threshold visualization and statistics).
(1) Unsuitable. Coffee farming is barely possible to impossible.
(2) Marginal. Coffee farming is possible but often proble-
matic, with poor and very inconsistent yields.
(3) Fair. Coffee farming possible and sometimes good but may
require additional inputs (for example, irrigation, good
shade and pruning). Yields poor in years with low rainfall.
(4) Good. Coffee farming good, with the potential for con-
sistent, high-quality yields. Yield and quality may be
negatively affected in some years (during adverse
weather conditions, especially low rainfall and high
(5) Excellent. Coffee farming good to excellent, with the
potential for consistent, high-quality yields. Yields and
quality can be negatively affected by adverse weather con-
ditions, but less likely.
NATURE PLANTS 3, 17081 (2017) | DOI: 10.1038/nplants.2017.81 | 3
© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
Supplementary Figs 7 and 8. In our modelling, there were no
considerable differences between A1B and A2 (CMIP3 (ref. 37)),
and so we restrict our main narrative to scenario A1B, which is
the lower emission scenario of the two; results from the A2
analysis and migration scenarios B, C, E and F are supplied
(Table 1; Supplementary Figs 7 and 8). We also review the more
2040−2069 2070−2099
SDM agreement
Very strong
Ground control
0 125 250
Bahir Dar
Debre Birhan
Asbe Teferi
Addis Ababa
Jimma Hosaina
Awassa Goba
Mizan Teferi
Arba Minch
Figure 2 | Future projections for coffee suitability under Full Migration (A) and emission scenario A1B. Median of the three GCMs was used to colour
niche suitability. SDM modelling agreement: very strong (average of 86% across study area, for all time periods), solid colour with no overlay; strong (13%),
dots overlaid; weaker (1%), diagonal hatching overlaid. The 19601990 map shows the ground control data points used in this study and major place names
(removed for clarity in subsequent maps). All maps show major roads (as dark lines) for navigation. Diagonal lines on 19601990 and 20102039 maps
show the location of the elevation prole given in Fig. 6.
NATURE PLANTS 3, 17081 (2017) | DOI: 10.1038/nplants.2017.81 |
© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
recent Coupled Model Intercomparison Project phase 5 (CMIP5)38
Representative Concentration Pathway (RCP) Scenario 8.5 (RCP8.5)
emission scenarios, following Knutti & Sedláček39 (see Methods and
Discussion). In the following two sections, we provide percentages
for the middle range projections and their area (coffee niche
suitability in km
), based on the three selected GCMs (Bjerknes
Centre for Climate Research, Bergen Climate Model, Version 2
(BCCR-BCM2.0); Centre for Australian Weather and Climate
Research Mark, 3.5 (CSIRO-MK3.5); and Geophysical Fluid
Dynamics Laboratory, USA, Coupled Climate Model 2.1 (GFDL-
CM2.1)). The range of values for low and high range outcomes,
which are also based on the above three GCMs, are given in Table 1.
1960−1990 2010−2039
2040−2069 2070−2099
SDM agreement
Very strong
0 125 250
Figure 3 | Future projections for coffee suitability under the No Migration scenario (D) and emission scenario A1B. Median of the three GCMs used. All
maps show major roads (as dark lines) for navigation. SDM agreement symbolism as in Fig. 2.
NATURE PLANTS 3, 17081 (2017) | DOI: 10.1038/nplants.2017.81 | 5
© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
Projections 19601990 to 20102039. We see a large increase
(Figs 2, 4 and 5) in coffee area suitability under the Full Migration
(A) scenario within the present/near future (20102039), compared
to the recent past (19601990). This is due to a dramatic increase
in the area available (and suitable) for coffee at higher elevation, as
temperatures increase, particularly in the South West Zone and to
a lesser extent the North Zone (see Fig. 1 for zone localities). The
overall shift for each time period is shown in Fig. 6. If we look at
the south-western plateau (South West Zone) there is a vast area
(in excess of 15,000 km
) that comes into suitability in 20102039,
1960−1990 2010−2039 2040−2069 2070−2099
Area (km2)
A1B–Full Migration (A)
Fair Good Excellent
1960−1990 2010−2039 2040−2069 2070−2099
Area (km2)
A2–Full Migration (A)
Fair Good Excellent
1960−1990 2010−2039 2040−2069 2070−2099
Area (km2)
A1B–No Migration (D)
Fair Good Excellent
1960−1990 2010−2039 2040−2069 2070−2099
Area (km2)
A2–No Migration (D)
Fair Good Excellent
Figure 4 | Future projections for coffee suitability under the scenarios of Full Migration (A) and No Migration (D) across emission scenarios A1B and A2.
Error bars represent the SDM area (km
) model variability (low and high model outcome) within the three GCMs (GFDL-CM2.1, CSIRO-MK3.5 and BCCR-
BCM2.0). See Methods (section Selection of GCMs) and Table 1 for further details. A1BA is the scenario used in the Results and Discussion.
0 50 100 km
Nekemte Ambo
Nekemte Ambo
Nekemte Ambo
Nekemte Ambo
1960 −1990 2010−2039 2040−206 9 2070−2099
Full Migration (A)No Migration (D)
Good Excellent
6,294 km2
44,820 km2
19,142 km214,319 km212,897 km211,256 km2
66,158 km256,036 km251,280 km2
6,123 km22,219 km21,115 km2
2,891 km2
4,700 km2
5,306 km2
5,976 km2
9,536 km2
4,284 km2
5,858 km2
Nekemte Ambo
Nekemte Ambo
Nekemte Ambo
Nekemte Ambo
3,312 km2
Figure 5 | Availability of suitable coffee niche in km
for migration scenarios of Full Migration (A) and No Migration (D). Total suitable area is given in
the top right of each pie chart box. See Table 1 and Supplementary Table 1 for all metrics. Scenario A: coffee plants can grow in any suitable niche (can be
moved anywhere); scenario D: coffee plants can grow only within suitable forest cover and only in a suitable known niche (restricted to movement within
present-day forest cover and suitable niches). See Supplementary Fig. 5 for further details. The background is part of south-eastern Ethiopia, and isgivento
provide a scale for the ground area.
NATURE PLANTS 3, 17081 (2017) | DOI: 10.1038/nplants.2017.81 |
© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
but which declines as the century progresses (Fig. 6). We originally
considered the possibility that this expansion in suitability might be
due to the East African Climate Paradox14, where forecasts (from
GCMs) show an increase in rainfall for East Africa compared to
actual declines in rainfall (from the 1980s to present day)14,40.
We do not see this in our data or analyses; the increase in suitable
area for 20102039 is as a result of the interaction between
climate change and the topology of Ethiopia (see Discussion).
Under the No Migration (D) scenario (Figs 35) and the middle
range projection, from 19601990 to 20102039 there is a decrease
of 25% (19,142 km
to 14,319 km
) across the Fair to Excellent
coffee suitability niche categories. Over the same period there is
an increase of 85% (from 3,312 km
to 6,123 km
) in the Excellent
category, which accounts for the corresponding decreases in the
Fair and Good categories (Figs 4 and 5; Supplementary Figs 7 and 8).
Under Full Migration (A) there is a 48% increase (44,820 km
66,158 km
) in the Fair to Excellent categories over this period
(Box 2; Table 1; Figs 4 and 5). A projection with profound impli-
cations is the early loss (20102039) of almost the entire Bale
coffee area and eastern parts of the Sidamo coffee area (Figs 2
and 3). This outcome is consistent across all three GCMs, and all
migration and emission scenarios (Figs 2 and 3). This result
appears dramatic given that much of this area currently falls
within the Excellent coffee niche suitability category. In the
Discussion, we offer a possible explanation for this critical nding.
Projections 20102039 to 20702099. Over the rest of this century
(until 2099), the models forecast a general reduction (km
) in coffee
niche suitability, but the degree of reduction is highly dependent on
the migration scenario (Box 1; Table 1; Figs 4 and 5; Supplementary
Figs 5, 7 and 8). Under the No Migration (D) scenario and the
middle range projection (Fig. 5) there is a 21% decrease (from
14,319 km
to 11,259 km
) in the suitable coffee niche, from
20102039 to 20702099. We also see a similar reduction under
the Full Migration (A) scenario, where there is a 22% decrease
(from 66,158 km
to 51,280 km
) over the same period (Fig. 5).
For both No Migration (D) and Full Migration (A), the most
obvious difference is the substantial decrease in the Excellent
category: 82% (from 6,123 km
to 1,115 km
) for No Migration (D),
and 72% (from 14,585 km
to 4,052 km
) for Full Migration (A)
(Table 1; Figs 4 and 5), from 20102039 to 20702099. During
20402069, coffee suitability holds quite well, due to niche
potential at higher elevations (as the higher ground provides
suitability for coffee), for Full Migration (A) and No Migration (D)
and most of the other migration scenarios (Figs 25). Most of the
increases and stabilities in suitability are distributed in the main
South West Zone (Figs 13 and 6). The Arsi, East Hararge and
West Hararge coffee areas (Harar Zone) do not show any niche
suitability, for any of the modelling, for any future time period,
and this result receives strong model agreement. Environmentally,
they are ill suited to coffee growing now and do not improve in the
future (Figs 13 and 6). Under the Full Migration (A) scenario,
new areas of suitability develop, especially in the Amhara,
BenishangulGumuz and Rift South coffee areas (Figs 13),
mostly in the Fair and Good categories, but with lower model
agreement (Figs 2 and 3).
Area (km2)
Elevation range (m)
A1B–Full Migration (A)
0 50 100 150 200
Elevation (m)
Distance (km)
Figure 6 | Histogram and prole for elevation shifts. a, Histogram of elevation range for Full Migration (A) scenario, under emission scenario A1B. The
change in ranges shows a potential shift of 32 m in elevation per decade (see Methods). b, Elevation prole of the South West Zone with coffee suitability
for 19601990 and 20102039. Location of prole across the south-western plateau, running from southwest corner (SW) to northeast corner (NE), as
shown in Fig. 2.
NATURE PLANTS 3, 17081 (2017) | DOI: 10.1038/nplants.2017.81 | 7
© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
Full Migration versus No Migration. Comparing Full
Migration (A) to No Migration (D) over the three future time
periods, the percentage differences in the potential coffee niche area
are substantial: 20102039, 462% (44,820 km
to 14,319 km
); 2040
2069, 450% (58,036 km
to 12,897 km
); and 20702099, 455%
(51,280 km
to 11,256 km
) (Table 1; Figs 26; Supplementary
Table 1). This also demonstrates that, in Ethiopia, considerably
more land could be used to grow coffee than is currently the case.
Modelling considerations. The outcomes of our projections should
take into account the assumptions and uncertainties associated with
SDMs generally41, within the regional context25, and the projection
of SDMs with climate change41. We have done our utmost to
represent assumptions and uncertainties. Predicting future climate
with any great certainty is not possible, but by using multiple
climate models it is possible to make future projections assuming
certain pathways of greenhouse gas emission. These are highly
uncertain and cannot be validated, since we have no analogue for
future climate. However, climate model projections provide useful
estimates of potential future conditions41, which we can use to
explore the sensitivity of coffee production. By comparing different
climate models and exploring reasons for the projected change in
the coffee farming suitability niche, we can start to understand the
vulnerabilities of coffee growing, at least in terms of exposure35,36,
and the level of condence in the projected change of the niche.
Projections 19601990 to 20102039. Projections from the recent
past (19601990) to present/near future (20102039) under the No
Migration (D) scenario (Fig. 3) agree with our on-the-ground
observations for climatic coffee farming suitability, the overall
productivity of specic areas42 and climate station data.
Specically, areas categorized by us as Unsuitable or Marginal
generally have poor and inconsistent harvests, produce lower
quality coffee and in extreme circumstances (for example, drought
episodes) coffee trees experience severe stress or death. We present
three examples to illustrate this. (1) In the Zege Peninsula, near
Bahir Dar (on the shores of Lake Tana), coffee farming is
problematic, with good harvests only every ve years. The farmers
report (personal communication) that there has been a long-term
decline in productivity, from good harvests every year or two in the
1930s, to good harvests every ve years (since the late 1990s). We
assess the suitability of the Zege coffee locality as Marginal (1960
1990) to Fair (20102039), and then Unsuitable (20402099), with
very strong SDM model agreement (Figs 2 and 3). The main
climate constraint for Zege is not the amount of total annual rain,
but rather the long dry season (Supplementary Fig. 4). (2) In the
central and eastern parts of Wellega coffee area, where the SDM
(19601990) categorizes suitability as Marginal, farmers and
intervention workers report (personal communication)
increasingly inconsistent harvests (a good harvest every three years
or so), which they attribute to a shift in less rain from the late
1970s and mid-1980s. (3) For the Harar Zone, the niche is
categorized as Unsuitable (Figs 13), which is a fair representation
of conditions there at the present time: coffee farming is mostly
suboptimal unless improved agronomy is practised (for example,
irrigation, terracing), where the local or micro-climate is more
suitable, or where the water table brings sufcient moisture to the
soil. During the course of our eld survey work we received
numerous anecdotal reports (from farmers and coffee sector
representatives) of a long-term decrease in rainfall over the Harar
Zone and a steady, long-term decline in coffee farming. It seems
clear, however, that the conversion to the narcotic khat
(Catha edulis) has also had a considerable inuence on declining
coffee production in the Harar Zone. Since at least the 1960s,
farmers have moved to growing this more lucrative crop at the
expense of coffee farming and best-practice coffee agronomy.
Given that there are also no gains in niche suitability in the Harar
Zone across any of the modelling to 2099 (Figs 13 and 6) and
that the modelling is in strong agreement (Figs 2 and 3), it seems
that Harar coffee production will continue to decline unless
considerable interventions are made. Overall, our modelling
projects low-value categorizations (Unsuitable, Marginal, Fair) for
the 19601990 to 20102039 time periods, for areas that are
currently suboptimal for coffee farming. One unexpected exception
is the Bale coffee area and the eastern part of the Sidamo coffee
area (Figs 13 and 6), which is projected to be no longer suitable
for coffee by 2039. Most of these two areas currently fall within the
Excellent niche category. This result has strong to very strong
model agreement (Figs 2 and 3). Our explanation for the loss of
the Bale area and the climate mechanisms behind declines and
increases in coffee suitability are discussed below.
Projections 20102039 to 20702099. As the century progresses
(20102039 to 20402069 to 20702099) the projected outcomes
are profound, depending on the migration scenario (Fig. 5). By
the end of this century the current coffee-growing niche of
Ethiopia could decrease by 3959% (of the original area; 4161%
suitable area remaining), under our most restrictive migration and
emission scenarios (the low-range (area in km
) outcome, under
migration scenario E and emission scenario A2) if no
interventions are made (Table 1). The scenario of No Migration
(D) (A1B = 4564% remaining) most closely resembles a do
nothingscenario, although forest and tree canopy cover would
have to remain stable (no deforestation or afforestation). At the
other extreme, under the Full Migration (A) scenario (emission
scenario A1B), there would be substantial gains with the potential
for a more than fourfold increase in coffee farming area compared
to the No Migration (D) scenario.
The main reason for the substantial increase in the area for coffee
suitability in 20102039, under the Full Migration (A) scenario, is
due to higher elevation areas coming into suitability. This is seen
predominantly in the South West Zone (Fig. 1), where a huge
area (in excess of 15,000 km
) within the south-western highlands
comes into suitability in 20102039. After 20102039, an elevation
ceiling (literally) is reached, resulting in a decline from there
onwards. There are of course interactions with rainfall (adding
more nuanced changes), but the dominant climatic factor for the
expansion is temperature, which is manifest as an upward eleva-
tional shift. An elevation prole across the South West Zone
(Fig. 6b) illustrates this topographical expansion.
Particularly vulnerable coffee areas over these two time periods,
under any migration scenario, include large parts of Wellega, the
northern part of Illubabor and Bench-Maji (Figs 13 and 6), with
variable but mostly strong SDM model agreement for these projec-
tions (Figs 2 and 3). All of the ve circumscribed coffee zones
(Fig. 1) could be negatively inuenced to some extent (Figs 2
and 3), particularly at lower altitudes (Fig. 6). The main
South West Zone has the most inherent resilience, mainly due to
the potential for the coffee niche to improve in suitability at
higher elevations (Figs 6 and 7). Across Ethiopia, the most climati-
cally suitable part of the overall niche (the Excellent category; Box 2)
is projected to drastically decline in the long-term (Figs 25).
Identifying the drivers of change. There has been considerable
focus on the inuence of increasing temperatures and supra-
optimal temperatures on coffee production8,43. It is clear from our
study, however, that rainfall is also a major contributing factor; as
well as this, eld observations, climate station data and our weather
station recordings show that temperatures rarely reach the extreme
limits for Arabica coffee43 in Ethiopia. Analysis of the BIOCLIM34
layers reveals that a combination of rainfall, temperature seasonality
NATURE PLANTS 3, 17081 (2017) | DOI: 10.1038/nplants.2017.81 |
© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
and annual temperature (east and west of the Rift Valley;
Supplementary Tables 2 and 3) were contributing most to the
SDMs. By comparing these bioclimatic variables against each of
our GCM projections (for emission scenario A1B), we were able to
identify the main drivers in our models for change for each of the
major coffee zones (Figs 13, for context; Supplementary Figs 912
show the trends for each zone). Across the entirety of the coffee
landscape, the temperature BIOCLIM (BIO1) shows projected
temperature increases of 2.73.2 °C (under emission scenario A1B)
from 19601990 to 207099 (Supplementary Fig. 9), which is in
good agreement with historical records (1960s onwards) and
temperature projections from other sources (see Introduction).
Temperature seasonality (BIO4) generally increases (Supplementary
Fig. 10). Critically, the rainfall BIOCLIMs (BIO12 and BIO18)
show little change across this century (Supplementary Figs 11 and 12).
Thus, the total rainfall for each area experienced in the present day is
key to understanding future suitability, as any increases and
decreases in rainfall are mostly projected to be negligible to slight.
Harar does show some decline in rainfall towards the end of the
century, but the GCMs are in poor agreement (varying in decline of
250, 100 and 0 mm to the end of the century). Any decreases in
rainfall for Harar would compound the already unsuitable coffee
niche in this area, particularly if temperature increases continue to
trackfutureprojections.Tosummarize,thekeyinuencing factors
can be viewed as a relatively simple interaction between rainfall and
temperature: coffee areas that start with low rainfall cannot tolerate
an increase in temperature, especially as rainfall is not projected to
change by any considerable amount; temperatures are projected to
increase steadily and substantially. If Ethiopian rainfall declines
continue (see Introduction), in contrast to the negligible changes in
precipitation observed in our projections (Supplementary Figs 11
and 12), they would further amplify the temperature-driven changes
identied here. The Harar Zone and some locations within the South
East Zone (the latter includes the coffee areas of Bale and Sidamo)
have the lowest total rainfall at the present time (<1,300 mm), giving
these areas little potential to deal with any temperature increases. In
Harar and the South East zones, air temperatures tend to reach a
seasonal maxima between January and May, that is, just before or
during the onset of the spring rains (and coffee owering time).
These areas would be particularly sensitive to warming and
evapotranspiration increases at the end of the dry season, for
example, between November and March. This is consistent with
projections indicating that increases in potential evapotranspiration
may reduce the effective length of the African growing seasons44.
025 50 100 150
Protected areas
Coee zones
Migration scenarios
Full Migration (A)
No Migration (D)
Other forest cover
East Hararge
Central Eastern
Rift North
Jimma-LimuSouth west
South east
Tepi Kaa
Bench Maji Rift South
(incl. Yirgachee)
Figure 7 | Projections for coffee suitability 20702099 (emission scenario A1B) for scenarios Full Migration (A) and No Migration (D), with main coffee
areas (black lines; see Fig. 1) and protected area boundaries53 (red lines). Migration scenario A (Full Migration (A); green-blue) represents new coffee
suitability niche for 20702099. Migration scenario B (light green) represents 19601990 niches that persist into 20702099. Migration scenario C (mid-
green) represents new coffee suitability niche for 20702099 (as in migration scenario A), but with present-day forest cover. Migration scenario D (No
Migration (D); dark green), 19601990 niches that still persist into 20702099 (as in migration scenario B), but with present-day forest cover. Grey
represents other forest cover (with no end of the century coffee niche predicted). Figure 7 should be reviewed with model variability (Fig. 2.). Note,
migration scenarios E and F are not represented on this map as they are too small to be visible at this scale.
NATURE PLANTS 3, 17081 (2017) | DOI: 10.1038/nplants.2017.81 | 9
© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
The Rift and South West zones have greater rainfall (1,320
1,690 mm) and therefore greater scope to tolerate temperature
increases. Areas that are presently too cold (the North Zone and
parts of the South West Zone, at higher altitudes), but have enough
rainfall, will start to come into suitability as temperatures increase.
A better understanding of the critical relationship between low
rainfall (and a corresponding soil moisture decit) and high temp-
eratures is starting to emerge from our in-country observations of
plant stress, in conjunction with the use of climate data logging
equipment. At the beginning of the year (January to March), we
observe low rainfall and steadily increasing temperatures, resulting
in low soil moisture levels, which in some cases reach or exceed
the permanent wilting point. For example, in the Bench Maji
coffee area we recorded severe plant stress in March (2015), after
two monthstotal rainfall of less than 10 mm per month, an
average temperature (based on 1 reading per hour) of 24 °C
(minimum 16.4 °C; maximum 34.5 °C) and a volumetric soil moist-
ure content of c. 20%, and soil water potential values of 400 kPa to
1000 kPa (both recorded at 40 cm soil depth). Arabica coffee will
perform well in areas with higher temperatures than Ethiopia, but
only if there is sufcient rainfall, especially if there are few, short
or no periods of the year with a soil water decit, for example as
found in Panama and Costa Rica45,46. However, it should be clear
that none of our projections would lead to a comparable environ-
ment for Ethiopia, given that rainfall is not projected to increase
by any considerable amount.
CMIP5 versus CMIP3. In their overview of CMIP5, Knutti
& Sedláček39 compare CMIP5 and CMIP3 and state: The spatial
patterns of temperature and precipitation change are also very
consistent. Interestingly, the local model spread has not changed
much despite substantial model development and a massive
increase in computational capacity.Our SDM results show no
considerable differences between A2 and A1B, so if we see scenarios
that fall within our present A2 and A1B we would expect similar
results. Following Knutti & Sedláček39 and comparing RCP8.5 with
A2, we are provided with an indication of likely differences. The
only substantial difference seen is an increase in annual mean
temperature (BIO1), with an average additional increase of 0.9 °C
(Supplementary Fig. 9), and an increase in temperature range
between GCMs. We see some difference in rainfall between A2 and
RCP8.5, with projections for increased rainfall (North and Harar
zones), but with large disagreement between GCMs. These increases
are unlikely to be sufcient to compensate or override projected
severe temperature increases, although in the North Zone it is
possible that some areas could move to higher suitability under this
pathway (Supplementary Figs 1114 and 1617). The larger
increase in temperature observed for RCP8.5 would indicate that
both the A2 and A1B scenarios could be more conservative,
compared to RCP8.5. It is also possible that the North Zone, which
comes in as a new area in our CMIP3 future predictions, could
come into better conditions under RCP8.5, but the increase
in rainfall is slight and highly variable (Supplementary Figs 11
and 12). It should be reiterated that RCP 8.5 is the most extreme
emissions scenario of the new CMIP5 pathways. Reviewing all of
the RCPs for CMIP5 (which for the rst time includes a low-
emission mitigation scenario), these pathways are likely to deliver a
broader range of outcomes. Overall, and as demonstrated by Knutti
&Sedláček39, CMIP5 scenarios give higher warming compared to
CMIP3. If temperature change trends track according to (CMIP5)
RCP8.5, we would see an amplicationofthenegativechanges
identied here using CMIP3.
Other inuences. Our modelling does not include the potential
inuence of elevated CO
concentrations in the atmosphere, even
though it has been shown that this may have a benecial
physiological inuence on coffee, due to improved (leaf) water-
use efciency, which may mitigate high temperatures, at least
where there is unrestricted water supply and high relative
humidity43. Free air CO
enrichment experiments show that trees
are more responsive than herbaceous plants to elevated CO
and that elevated CO
benets plant species under drought
conditions48,49, although research indicates that the inuence of
enrichment for crop production may be over-emphasized50
If projected increases in CO
prove to have a substantial role in
mitigating the impact of drier, warmer conditions on coffee
production, then our projections of climate exposure could be
over-estimated. However, in Ethiopia we observe that drought
(water stress) is one of the major causes of crop failure for coffee,
and severe drought (lethal water stress) the most signicant
climate-related cause of plant fatality. We argue, therefore, that
while there could be benets from elevated atmospheric CO
drought (in its duration, timing and severity) may far outweigh
these gains in the longer term. One might also contend that the
forest will change as the climate changes, including responses to
elevated CO
. In this study, we review changes in niche and not
forest, although our scenario of Full Migration (A) would
encompass a broad spectrum of vegetation change (species and
forest composition) within the humid forest vegetation class
(MAF and TRF6). As well as increasing drought stress, high
deforestation rates5and particularly their inuence on local and
regional climate are likely to override the benets of atmospheric
enrichment for forest performance and survival and the
physiological behaviour of Arabica coffee. We also have little
knowledge about the potential inuences of pests and diseases,
although these are likely to have a compounding negative
inuence under a warming climate25,51.
Migration scenarios and interventions. Under the Full Migration (A)
scenario, there would be substantial potential to increase coffee
farming and thus production volumes. As outlined above,
comparison between Full Migration (A) and No Migration (D)
(no migration, no forest loss) scenarios shows that the potential to
increase the coffee growing area, over the three future time
periods (20102039, 20402069 and 20702099), is at least
fourfold (>400%), although the niche peaks in 20102039
(Table 1; Figs 47; Supplementary Table 1). Most of this change
comes from up-slope migration to higher elevations. Under Full
Migration (A) (emission scenario A1B) there would be a projected
altitude shift for overall coffee suitability of 32 m per decade, from
19601990 to 20702099 (Fig. 6), with a considerable change in
the extreme upper limits to 2,8003,300 m (from 2,2002,600 m).
The emergence of coffee suitability areas at higher altitudes (to
those existing now) is projected over a large area of the south-
western Ethiopia plateau, but also in areas of the North Zone
(Amhara and BenishangulGumuz coffee areas), Rift Zone (Rift
South area) and South East Zone (Sidamo coffee area), although
with different levels of model agreement (Figs 2 and 3). For the
Full Migration (A) scenario to be realized, signicant intervention
would be required, including the planting of shade cover or the
restoration of humid forest cover for areas with little or no
present-day forest cover (Fig. 7). A concerted effort would be
needed to identify and establish new coffee areas at higher
altitudes, with particular attention placed on competing land-use
issues (for example, existing land-use ownership, nature conservation
sites), soil and micro-climate suitability and access to roads and
other infrastructure. Shifts could also occur without major
intervention, as farmers realize the potential for growing coffee in
their upland areas. Field observations made by us in 2015 show
that locations in the North Zone are already being developed as
new coffee areas (in correspondence with our projections; Fig. 2), at
altitudes in excess of 2,500 m (not previously seen for coffee
NATURE PLANTS 3, 17081 (2017) | DOI: 10.1038/nplants.2017.81 |
© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
farming in Ethiopia), indicating that assisted migration is
already underway.
Resilience built via the migration of coffee farms could be sup-
plemented with (location- and time-specic) improvements in
coffee productivity (kg per hectare; for example, by increasing soil
fertility and pruning) and quality (kg per $; for example, better
harvesting and post-harvest processing), in order to offset unit area
reductions in environmental suitability. In addition, there is scope
for investing in locally appropriate, cost effective agronomy adap-
tations (such as mulching, irrigation, improved shade management
and terracing). Irrigation would provide the single most effective
adaptation measure, especially where implementation costs are
low, for example, the diversion of nearby streams and rivers. Our
projections would serve to provide a framework for the likely
success, timing, duration and location of interventions.
Wild coffee forests. Ethiopias wild coffee forests have considerable
fundamental value as the main storehouse of C. arabica (Arabica
coffee) genetic resources25, with specic benets for the coffee
sector, locally, regionally and globally52. Most directly, indigenous
Arabica coffee plays a key role in Ethiopian coffee production as
an important source of planting material for farms via seed and
seedlings. Protection of these resources should thus be seen as a
key part of a resilience strategy. Using all migration scenarios (to
20702099), but selecting scenario E to represent the most
resilient areas (refugia), provides an indication of which protected
areas might have value as reserves for coffee genetic resources
(Fig. 7). This exercise identies protected areas in Kaffa (Kaffa
Biosphere Reserve) and northern Illubabor (Yayu, a UNESCO
registered Coffee Forest Biosphere)53, although Yayu would have
limited long-term potential unless contiguous areas at higher
altitudes are included in the delimitation of its core area. Another
important consideration is the preservation of specic coffee
origins (and their unique avour proles), some of which are
identied in the modelling as highly susceptible to negative
climate change inuences, including the coffees of Harar and Bale.
Summing up. Timely, precise, science-based decision making is
required now and over the coming decades, to ensure
sustainability and resilience for the Ethiopian coffee sector. Our
projections show an overall negative inuence, due to climate
change, across this century for the present coffee growing
landscape. Despite this, there is the potential to increase the area
where coffee can be grown (by at least fourfold) via relocation and
expansion, even if climate change tracks to our projections,
although this would have to be done in combination with forest
conservation and re-establishment. We also identify key coffee
areas that could be climatically resilient, even without major
intervention, at least until the end of this century.
Mapping the extent of humid forest in Ethiopia. A map of humid forest cover
(TRF/MAF6, and comparable agroforestry environments; Supplementary Fig. 6) was
produced from Landsat 8 imagery (,
downloaded for Decemberof 2013 to February of 2014 (
(38 images covering 24 scenes)). The images were atmospherically corrected
using the Fast Line-of-sight Atmospheric Analysis of Hypercubes algorithm
(FLAASH) in ENVI 5.1 (ref. 54) to obtain surface reectance values. To produce a
cloud-free composite, the images were processed to identify clouds and cloud
shadows using the Fmask 3.2 algorithm55. ERDAS Imagine 2013 was used to
produce Normalised Difference Vegetation Indexes56 from the surface reectance
cloudless composites. Finally, we produced a Normalised Difference Vegetation
Index composite using two thresholds to identify the humid forest.
A set of 300 training data points were used to identify the most appropriate
thresholds and the individual maps were combined in ArcGIS ver. 10.1
(Environmental Systems ResearchInstitute (Esri)). Testing of the nal map yielded an
accuracy assessment of 97%. The humid forestmap wasuploade d toGoo gle Earth for
review, as well as large-format mobile phones (Samsung Note II, GT-N7100) using
Locus Map Pro 3.1, Asamm Software, Praha, for validation during eight eldwork
trips (20142016), with an emphasis to identify and remove non-coffee production
areas (for example, tea, mango, avocado plantations and commercial forestry).
Point data and ground-truthing. We collected 3070 data points (Supplementary
Fig. 3), of which 1624 were coffee presence data points (1446 non-coffee points),
comprising 824 historical (19412001) and 800 contemporary (20132015) points
(with coffee production type, forest cover and condition, and global positioning
system (GPS) data recorded). To reduce sampling bias and sampling errors, the
ground data was ltered: historical data points collected before 2001 (early-
generation GPS, herbarium specimens and literature records) with an accuracy lower
than 1 km were rejected; multiple points falling within a 1 km
grid were reduced to
a single point; any points covering locations with signicant interventions (all
irrigated and high water table areas) were removed. This left 381 points to build the
nal SDM. Many of the data points appear to lie close to roads, although a
considerable amount of data was collected on foot (c. 40%). There were no
indications of a road-like bias in our SDMs. In well-sampled studies, sampling along
roads should not represent a bias issue57, especially where roads cut across numerous
environmental gradients, as in Ethiopia. Moreover, the niche and landscape were
very well sampled (Supplementary Fig. 3). We double-checked to see if our sampling
would show any signicant sampling bias by comparing histograms of elevation
versus percentage sampling. The range of elevations sampled by the 3,070 data
points was 4263,877 m, which is much greater than the range (9912,263 m) for
the 381 points used for the SDMs. Both histograms showed consistent sampling
across 250 m elevation bands.
In addition to the 3070 original points, an additional 1028 points for coffee
presence and absence were collected after the initial modelling (2015). These points
were used to informally validate the modelling and remote sensing, although no
mapping or analyses were undertaken with these data. The 800 contemporary data
points (20132015) and the additional 1028 data points (20152016) were
collected during ten eld missions, covering c. 30,000 km by road and on foot (plots
and transects), across almost the entire coffee landscape of Ethiopia. During our
eldwork, we also informally interviewed farmers for perceived changes in coffee
farming, on a year-by-year and generation-by-generation basis.
SDMthe niche. There has been much discussion about the actual niche
represented by an SDM and particularly the question over realised niche versus
fundamental niche. Most SDM studies identify the realised niche58; the closer the
realised niche gets to the fundamental niche, the better the projections are likely to
be. When using bioclimatic variables, the results will tend towards the fundamental
niche, especially when extrapolating to new areas (for example, introduced ranges)
or future climates58,59. In this study, the niche-only scenarios (A, B and E), combined
with exhaustive ground validation, show that our SDMs are very close to
representing the fundamental niche, that is, where coffee grows and is farmed. Our
niche and forest (migration) scenarios (C, D and F; Supplementary Fig. 5) preserve
the relationship to the fundamental niche; applying the humid forest mask also
means that we are closer to the realised niche than using an SDM alone.
SDMdivision into climatic groups. After running preliminary models for the
present day, investigating climate proles, and considering the biology of Arabica
coffee in Ethiopia, it became clear that we should treat east and west of the Rift Valley
(Fig. 1) separately. The rainfall pattern east of the Rift Valley is bimodal, and in the
west it is unimodal (see Supplementary Fig. 4). Molecular6062, morphological63 and
physiological data64 support Arabica coffee differentiation either side of the Rift
Valley. Thus, for our modelling, we treated east and west independently, running the
models for each and combining only at the end of the procedure to provide nal
modelling outputs.
SDMselection of environmental variables. We examined the full BIOCLIM
dataset34 of 19 variables, independently, east and west of the Rift Valley. Paired
dataset correlations and boxed plots were reviewed and we examined which variables
had signicant inuence on the Generalised Linear Models (GLM) and Generalized
Additive Models (see below). For each BIOCLIM variable, we ran an analysis of
variance (ANOVA) test by examining output model results for our validation points
and comparing these to the pseudo-absence points. We also ran models using
random pairs of variables to investigate the best variables for model t (see below).
Finally, we examined the resultant BIOCLIM variables and reduced the selection to
those that were signicant in the above tests, not highly correlated (as pairs) and
which represented the ecological requirements of Arabica coffee. By comparing the
results independently for east and west of the Rift Valley, we reduced the
19 BIOCLIM variables to nine (Supplementary Table 4), for use in our modelling.
The same variables were used for the east and west split, to allow our models to
be comparable.
SDMprocessing. Using the selected environmental variables (see above), we ran
the data (381 ground points) in the Biomod2 R package, version 3.164 (ref. 65),
using six modelling methods (GLM, Generalized Boosted Regression Models,
Generalized Additive Models, Multiple Adaptive Regression Splines (MARS),
Random Forest and Maximum Entropy (Maxent) software version 3.3.3 (refs 6568)
to produce an ensemble model for either side of the Rift Valley. Each of the models
NATURE PLANTS 3, 17081 (2017) | DOI: 10.1038/nplants.2017.81 | 11
© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
used the nine selected environment variables to maximise its t to the output
prediction. We largely followed established methodologies65,6971 for model
evaluation, runs, replications and combination. We used the default setting in
Biomod2, except for increasing the number of iterations for GLM and Maxent, to
reach convergence. Three sets of background or pseudo-absence data were
generated. Ground control data, both presence and background for each of the three
sets, was split randomly ten times with a ratio of 70:30 to build training and testing
data, respectively72. Model outputs from Biomod2 were evaluated by viewing the
actual results and using sensitivity, specicity, True Skills Statistics (TSS), Cohens
Kappa statistic (KAPPA) and relative operating characteristic curve foreast and west
of the Rift Valley models (see Supplementary Information). Following Araujo &
New70 and Thuiller et al.69,71 we took the six individual models to produce an
ensemble prediction for either side of the Rift Valley, using the suggested model
evolution threshold of 0.7 (using TSS), to remove poorly performing models
(although no models were removed during this process, as they all exceeded this
threshold). We also implemented proportional weighted overlay, to give better
tting models a higher weight. During the statistical evaluation of the six models,
expert review was used to make sure that they were in agreement with the biotic and
abiotic requirements for Arabica coffee. The testing and training data were assessed
and reviewed to give the relative importance of the variables used in each developed
model, model sensitivity and specicity, as well as overall model performance72. All
six niche model methods performed exceptionally well (over 90% for specicity and
sensitivity for each model).
There are two assumptions associated with the models for 19601990:
(1) because the 2014 TRF/MAF6forest cover (see Supplementary Fig. 6) was used
with 19601990 data, the surface area will be underestimated, as there has been
considerable deforestation since the 1960s5; and (2) we assume that the suitable
coffee niche, if forested or having sufcient shade cover, is being used for coffee
production of some description25.
SDMevaluation. Model outputs from Biomod2 (ref. 65) were evaluated by
viewing the resultant maps and using the statistics; sensitivity, specicity, TSS,
KAPPA and relative operating characteristic curve for east and west of the Rift
Valley. For our purposes, sensitivity and specicity was the most useful and easiest to
use metric, but we reviewed all statistics. All six niche model methods performed
very well (over 90% for specicity and sensitivity), but east of the Rift Valley showed
slightly lower specicity values, compared to west (this is not surprising as the
sample for the east is much smaller). GLM and MARS showed the lowest scores
(9093% for GLM and MARS for specicity), which is still very good, but suggests
they are not tting the data as well as other models. Final models were produced
from ten runs and three replicates. We checked the outputs visually to see if there
was a noticeable difference between model runs (which may suggest problems, for
example, overtting or incorrect models).
SDM and WorldClimvalidation. The ensemble SDM was loaded onto four large-
format, GPS-enabled mobile phones (Samsung Note II, GT-N7100), using Locus
Map Pro 3.1, Asamm Software, Praha, for validation during seven eldwork trips
(2015 to 2016). This activity revealed excellent correspondence between the
ensemble SDM and coffee farming areas; the SDM predicted coffee farming activity
to within a 1 km radius of each validation location.
As a means of checking the WorldClim34 precipitation data, we compared
precipitation data from 27 weather stations over a 30-year period (19842014)
(National Meteorology Agency Ethiopia) against the WorldClim34 data
(19601990). We found acceptable to excellent agreement between the two data sets,
when visualized as c. 30 year averages for numerous locations. Climate proles
(30-year period; 19842014) for total monthly rainfall and average monthly
temperature (minimum, median and maximum) for an additional 20 weather
stations (National Meteorology Agency) situated within, or adjacent to, coffee
growing areas, were also found to be in good agreement with 19601990
WorldClim34 data. Six examples are provided in Supplementary Fig. 4. At Yayu
(Illubabor), Ashi (Wellega), Gesha (Bench-Maji), Yirgacheffe and Yirga Alem
(Sidamo), Bale (Bale) and Gololocha (Arsi) (see Fig. 1) we installed (20132016)
proprietary data-logging equipment (Tinytag Plus 2TGP-4500, with Stevenson
type screen (ACS-5050), both from Gemini Data Loggers), to measure air
temperature and relative humidity (RH%). At Ashi, Gesha and Bale we also installed
soil temperature probes (Tinytag Plus 2TGP-4510, Gemini Data Loggers), soil
moisture (10HS) and soil water potential sensors (MPS-6) connected to EM50
digital data loggers (all from Decagon Devices) at 20 and 40 cm soil depths and low
resolution rain gauges (ECRN-50, Decagon Devices). The logging interval was every
hour on the hour for Tinytag Plus 2 and twice daily for the Decagon equipment. The
data was not analysed for this study, but simply used to ground-truth WorldClim34
data and the coffee niche and to better understand the climatic variables causing
physiological stress in coffee.
SDMassessment of climatic variables. We examined the BIOCLIM34 data to see
which bioclimatic variables were contributing most to the SDMs, by summarizing
the contribution of each variable for each of the six niche modelling methods
(Supplementary Table 2 and 3), for east and west of the Rift Valley. For the east, the
most important were temperature seasonality (BIO4) and precipitation of the
warmest quarter (BIO18); and the west, annual mean temperature (BIO1) and
annual precipitation (BIO12).
Selection of GCMs. The evaluation and selection of GCMs for a study area and set
of circumstances is a challenging activity24,73,74. Ideally, we would have used all
GCMs to assess the full range of outcomes, but selected three in order to keep
processing time and storage to a manageable level and to reduce redundancy (for
example, where GCMs were in agreement). To make our selection, we processed all
23 GCMs available for the study area, excluding models that did not t our future
time periods (20102039, 20402069 and 20702099). We then reviewed the
projections for each of the contributing BIOCLIM layers, for eight coffee locations
either side of the Rift Valley, as a representative sample of the of the coffee farming
landscape: for the South West Zone (Yayu (Illubabor), Dembi Dollo and Ashi
(Wellega)); South East Zone (Bale (two sub-locations), Yirgacheffe and Yirga Alem
(Sidamo)); and Harar Zone (Gololocha (Arsi)); an example for one of these locations
(Ashi) is shown in Supplementary Fig. 15. On reviewing these graphs, we rejected
GCMs that were highly anomalous for temperatures and anomalously high or low
values for precipitation; an emphasis was placed on the 20102039 period, where
anomalies are more easily assessed. We selected a representative GCM for those
showing similar results across the nine BIOCLIMs, two emission scenarios and four
time intervals. This still left us with eleven GCMs from which we choose three
(BCCR-BCM2.0, CSIRO-MK3.5 and GFDL-CM2.1) to represent climatic changes
over time (for the subset of 11 GCMs; across the nine BIOCLIMs).
GCM data downscaling. The original ground resolution for the GCMs was
approximately 200 km. To make this data more useful for SDM and to see variation
at higher resolutions, we used downscaled GCMs34,75,76. We employed the
downscaled (Delta method of the Intergovernmental Panel on Climate Change
(IPCC) AR4 (ref. 76)) climate models from CGIAR Research Program on Climate
Change, Agriculture and Food Security (CCAFS) (
data_spatial_downscaling/ and The
downscaled method used by CCAFS is based on thin plate spline interpolations of
deltas (anomalies) of the original low-resolution GCM outputs. These deltas are
applied to the WorldClim34 data (the same data we used for our 19601990 SDM) at
a resolution of 30 arc seconds (approximately 1 km
cells size for Ethiopia) by
interpolating between GCM cell centroids76. The 19 bioclimatic indices (BIOCLIM)77
were calculated by CCAFS from the resultant rasters.
GCMsmodelling future projections. Our three selected GCMs were: (1) BCCR-
BCM2.0BCCR, Univ. of Bergen, Norway (BCM Version 2); (2) CSIRO-MK3.5
Centre for Australian Weather and Climate Research (Mark 3.5); and
(3) GFDL-CM2.1GFDL USA (Coupled Climate Model 2.1). The future SDM
variability of the selected GCMs is illustrated in Figs 2 and 3. We used two CO
emission scenarios, A1B and A2: A1B reaching 13.5 giga tonnes of carbon (GtC) by
2099; and A2 reaching 29.1 GtC; this is against a baseline of 9.8 GtC (in 2014/15).
The projections (areas of coffee-growing suitability given as percentages and km
produced from the three GCMs were categorised into three classes: low, middle and
high (Table 1). For the Results and Discussion, we provide percentages and area
measurements (km
) for the middle-range outcome, based on the three GCMs
(Table 1; Supplementary Table 1). The GCM CSIRO-MK3.5 predominately produced
the middle-range outcomes; BCCR-BCM2.0 tended to produce low-range outcomes
and GFDL-CM2.1 the higher ones, although these switched between low, middle and
high, depending on the suitability category and time intervals across this century.
The future projections were modelled in the Biomod2 package Ensemble
Forecasting65, which projects the ensemble recent past and current model in
space and time. The projection was performed on the nine BIOCLIM variables
as identied above. Future models were run for each GCM, for east and west
of the Rift Valley and over three time periods (20102039, 20402069 and
20702099; Supplementary Fig. 1), giving 18 future projections
(two regions, over three dates and three GCMs) as a nal output.
Combining SDMs and GCMsanalysis and mapping. The outputs from the four
time periods (19601990, 20102039, 20402069 and 20702099) were imported
into ArcGIS ver. 10.1 (Esri), with the spatial analysis extension, for further analysis
and visualization. The Rift Valley separation, from each modelling stage, was
combined using the raster calculator tool to give the maximum suitability value for
each pixel, giving 19 models for the nal analysis (one time period (19601990) plus
three GCMs × two emission scenarios, across three further time periods). To
visualise the SDM variability, the projections were processed to show the range of
SDM values and these ranges were categorised into three classes: (1) Very Strong,
model variability 0 to 200; (2) Strong, 200 to 400; and (3) Weaker, 600 or higher.
Very strong (solid colours, with no overlay) shows where the model class (Excellent,
Good, Fair, Marginal and Unsuitable) is unlikely to change, except at the edge of the
ranges. Strong (overlaid with dots) shows where a class could change up or down by
one class (for example, Good could range from Excellent to Fair). The Weaker class
(diagonal lines) is where classes could change by more than one class; the largest
range is 610, indicating a movement of up to one and a half classes. We adopted the
same symbology as used in a widely read IPCC report22, although it should be noted
that we used it for SDM variation not GCM variation (as in the IPCC report). We
NATURE PLANTS 3, 17081 (2017) | DOI: 10.1038/nplants.2017.81 |
© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
also calculated the median for each model, which was used only for visualization of
the results (Figs 2 and 3).
Categorisation of projection maps. To provide area measurements (in km
) for
each category (Unsuitable, Marginal, Fair, Good, Excellent), we needed to threshold
the continuous SDM values (those between 0 and 1,000). To achieve this, we
examined the SDM values for our ground point data (ground control, absence and
pseudo absences data), by comparing TSS and the cumulative percentage of points
to give us our initial threshold. This was further subdivided to give a total of
ve categories (Supplementary Fig. 2; Supplementary Information).
Migration scenarios for climate change projections. The six migration scenarios
(Box 1; Supplementary Fig. 5) were processed in ArcGIS, using simple map algebra,
summarized, with the following processing: (A) no processing; (B) all models
cookie-cut with 19601990 suitable niche (all areas outside of the 19601990 niche
were clipped for all dates); (C) as A but cookie-cut with forest cover; (D) as B but
cookie-cut with forest cover; (E) as B but 20402069 cookie-cut with 20102039
suitable niche and 20702099 cookie-cut with 20102039 and 20402069 suitable
niche; (F) as E but additionally cookie-cut with forest cover. The forest cover niche
models were resampled at 30 m resolution using a simple nearest-neighbour
resampling (to preserve the original model resolution of 1 km
Analysis of regional trends and drivers. To understand the trends and the main
drivers for climate change outcomes we produced graphs of change for the most
inuential BIOCLIMs. Supplementary Figs 912 show the trends for each of the ve
coffee zones (Fig. 1), from 19601990 to 20702099. The most inuential
BIOCLIMs were: temperature seasonality (BIO4), precipitation of the warmest
quarter (BIO18), annual mean temperature (BIO1) and annual precipitation
(BIO12). We calculated the mean, max, min and range for each zone, for CMIP3
scenarios A1B and A2, the three GCMs and all four date periods.
CMIP3 versus CMIP5. Our projections are based on the WCRPs CMIP3 multi-
model dataset37. The newer and more developed WCRP CMIP5 multi-model
ensemble38 was released as a downscaled dataset after the project research period.
A direct comparison between CMIP3 and CMIP5 is not straightforward, as CMIP3
uses emission scenarios, whereas CMIP5 uses RCPs39. Following the methods of
Knutti & Sedláček39, we compared CMIP5 RCP8.5 against CMIP3 SRES scenario
A2, which represents the two scenarios with the highest temperature changes.
Supplementary Figs 15 and 16 show a comparison of the GCM robustness across
Africa for CMIP5 (RCP8.5) vs CMIP3 (SRES A2) based on Knutti & Sedláček39.For
further comparison, we included CMIP5 RCP8.5 in the same analysis that we used to
represent the ve coffee zones (Fig. 1) for CMIP3 A1B and A2 (Supplementary
Figs 912). Both the GCM and the downscaling methods (Delta method76 IPCC
AR478 to Delta method IPCC AR5 (ref. 79)) were updated when moving from
CMIP3 to CMIP5. Where possible, we used the most up to date version of each
GCM: GFDC-CM2.1 and CSIRO-MK3.5 for CMIP3 scenarios A2 and A1B were
updated to GFDL-CM3 and CSIRO-MK3.5 for the CMIP5 scenarios, respectively.
There was no updated GCM for BCCR-BCM2.0, so we used BCCR-CSM1.1m to
give us an output level for comparative purposes. The period 20102039 was not
available for CMIP5, so we substituted the period 20202049.
Data availability. The data that supports the ndings of this study are available from
the corresponding author upon request. The supply of ground point data is
restricted, as ownership is shared across the authorship and externally.
Received 12 February 2017; accepted 2 May 2017;
published 19 June 2017
1. Minten, B., Tamru, S., Kuma, T. & Nyarko, Y. Structure and Performance of
Ethiopias Coffee Export Sector. Working paper 66 (EDRI/IFPRI, 2014).
2. Historical Data on the Global Coffee Trade (ICO, 2016);
new_historical.asp (2016).
3. Chemonics International. Ethiopia Coffee Industry Value Chain Analysis.
Proling the Actors, their Interactions, Costs, Constraints and Opportunities
(USAID, 2010);
4. Tefera, A. Ethiopia: Coffee Annual Report. GAIN Report Number ET1514
(USDA Foreign Agricultural Service, 2015).
5. Hailu, B., Maeda, E. E., Heiskanen, J. & Pellikka, P. Reconstructing pre-
agricultural expansion vegetation cover of Ethiopia. Appl. Geogr. 62,
357365 (2015).
6. Friis, I., Demissew, S. & Breugel, P. V. Atlas of the potential vegetation of
Ethiopia. Biol. Skrif. 58, 1307 (2010).
7. Dudu, V. P. Impacts of Climate Change on Coffee Farming in Ethiopia
(LAP Lambert Academic, 2012).
8. Craparo, A. C. W., Van Asten, P. J. A., Läderach, A., Jassogneb, L. T. P.
& Graba, S. W. Coffea arabica yields decline in Tanzania due to climate change:
global implications. Agr. Forest Meteorol. 207, 110 (2015).
9. McSweeney, C., New, M. & Lizcano, G. UNDP Climate Change Country Proles:
Ethiopia (UNDP, 2010);
10. Jury, M. R. & Funk, C. Climatic trends over Ethiopia: regional signals and
drivers. Int. J. Climatol. 33, 19241935 (2013).
11. Siraj, A. S. et al. Altitudinal changes in malaria incidence in highlands of
Ethiopia and Colombia. Science 243, 11541158 (2014).
12. Mekasha, A., Tesfaye, K. & Duncan, J. Trends in daily observed temperature and
precipitation over three Ethiopian eco-environments. Int. J. Climatol. 34,
19901999 (2014).
13. Williams, A. P. & Funk, C. A westward extension of the warm pool leads to a
westward extension of the walker circulation, drying Eastern Africa. Clim.
Dynam. 37, 24172435 (2011).
14. Rowell, D. P., Booth, B. B. B., Nicholson, S. E. & Good, P. Reconciling past and
future rainfall trends over east Africa. J. Climate 28, 97689788 (2015).
15. Seleshi, Y. & Camberlin, P. Recent changes in dry spell and extreme rainfall
events in Ethiopia. Theor. Appl. Climatol. 83, 181191 (2006).
16. Viste, E., Korecha, D. & Sorteberg, A. Recent drought and precipitation
tendencies in Ethiopia. Theor. Appl. Climatol. 112, 535551 (2013).
17. Funk, C. et al. A Climate Trend Analysis of Ethiopia, USGS Fact Sheet 20123053
(USGS, 2012).
18. Funk, C. et al. Warming of the Indian Ocean threatens eastern and southern
African food security but could be mitigated by agricultural development. Proc.
Natl Acad. Sci. USA 105, 1108111086 (2008).
19. Funk, C. et al. Recent Drought Tendencies in Ethiopia and Equatorial Subtropical
Eastern Africa (FEWS NET, 2005).
20. Williams, A. P. et al. Recent summer precipitation trends in the greater horn of
Africa and the emerging role of Indian ocean sea surface temperature. Clim.
Dynam. 39, 23072328 (2012).
21. Conway, D. & Schipper, E. L. F. Adaptation to climate change in Africa:
challenges and opportunities identied from Ethiopia. Glob. Environ. Chang.
21, 227237 (2011).
22. IPCC Summary for Policymakers (eds Stocker, T. F. et al.) (Cambridge Univ.
Press, 2013).
23. IPCC Climate Change 2014: Impacts, Adaptation, and Vulnerability
(eds Barros, V. R. et al.) (Cambridge Univ. Press, 2014).
24. Ramirez-Villegas, J., Challinor, A. J., Thornton, P. K. & Jarvis, A. Implications of
regional improvement in global climate models for agricultural impact research.
Environ. Res. Lett. 8, 24018 (2013).
25. Davis, A. P., Gole, T. W., Baena, S. & Moat, J. The impact of climate change on
natural populations of Arabica coffee: predicting future trends and identifying
priorities. PLoS ONE 7, e47981 (2012).
26. Landsat Missions: Landsat 8 (USGS, accessed February 2014); http://landsat.usgs.
27. Schroth, G. et al. Towards a climate change adaptation strategy for coffee
communities and ecosystems in the Sierra Madre de Chiapas, Mexico. Mitig.
Adapt. Strateg. Glob. Change 14, 605625 (2009).
28. Baca, M., Laderach, P., Haggar, J., Schroth, G. & Ovalle, O. An integrated
framework for assessing vulnerability to climate change and developing
adaptation strategies for coffee growing families in Mesoamerica. PLoS ONE 9,
e88463 (2014).
29. Magrach, A. & Ghazoul, J. Climate and pest-driven geographic shifts in global
coffee production: implications for forest cover, biodiversity and carbon storage.
PLoS ONE 10, e0133071 (2015).
30. Bunn, C., Läderach, P., Pérez Jimenez, J. G., Montagnon, C. & Schilling, T.
Multiclass classication of agro-ecological zones for Arabica coffee: an improved
understanding of the impacts of climate change. PLoS ONE 10, e0140490 (2015).
31. Ovalle-Rivera, O., Läderach, P., Bunn, C., Obersteiner, M. & Schroth, G.
Projected shifts in coffea arabica suitability among major global producing
regions due to climate change. PLoS ONE 10, e0124155 (2015).
32. Bunn, C. P. L., Ovalle-Rivera, O. & Kirschke, D. A bitter cup: climate change prole
of global production of Arabica and robusta coffe. Clim. Change 129, 89101 (2015).
33. Chemura, A., Kutywayo, D., Chidoko, P. & Mahoya, C. Bioclimatic modelling of
current and projected climatic suitability of coffee (Coffea arabica) production in
Zimbabwe. Reg. Environ. Change 16, 473485 (2016).
34. Hijmans, R. J., Cameron, S. E., Parra, J. L., Jones, P. G. & Jarvis, A. Very high
resolution interpolated climate surfaces for global land areas. Int. J. Climatol. 25,
19651978 (2005).
35. Dawson, T. P., Jackson, S. T., House, J. I., Prentice, I. C. & Mace, G. M.
Beyond predictions: biodiversity conservation in a changing climate. Science
332, 5358 (2011).
36. Foden, W. & Young, B. E. IUCN SSC Guidelines for Assessing Species
Vulnerability to Climate Change Version 1.0. Occasional Paper of the IUCN
Species Survival Commission No. 59 (IUCN, 2016);
37. Meehl, G. A. et al. The WCRP CMIP3 multi-model dataset: a new era in climate
change research. Bull. Amer. Meteor. Soc. 88, 13831394 (2007).
38. Taylor, K. E., Stouffer, R. J. & Meehl, G. A. An overview of CMIP5 and the
experiment design. Bull. Amer. Meteor. Soc. 93, 485498 (2012).
NATURE PLANTS 3, 17081 (2017) | DOI: 10.1038/nplants.2017.81 | 13
© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
39. Knutti, R. & Sedláček, J. Robustness and uncertainties in the new CMIP5 climate
model projections. Nat. Clim. Chang. 3, 369373 (2013).
40. Tierney, J. E., Ummenhofer, C. C. & deMenocal, P. B. Past and future rainfall in
the Horn of Africa. Sci. Adv. 1, e1500682 (2015).
41. Wiens, J. A., Stralberg, D., Jongsomjit, D., Howell, C. A. & Snyder, M. A. Niches,
models, and climate change: assessing the assumptions and uncertainties. Proc.
Natl Acad. Sci. USA 106, 1972919736 (2009).
42. Boot, W. J. Ethiopian Coffee Buying Manual: Practical Guidelines for Purchasing
and Importing Ethiopian Speciality Coffee Beans. (USAID, 2011); https://
43. Rodrigues, W. P. et al. Long-term elevated air [CO2] strengthens photosynthetic
functioning and mitigates the impact of supra-optimal temperatures in tropical
Coffea arabica and C. canephora species. Glob. Change Biol. 22, 415431 (2016).
44. Cook, K. H. & Vizy, E. K. Impact of climate change on mid-twenty-rst century
growing seasons in Africa. Clim. Dynam. 39, 29372955 (2012).
45. Wrigley, G. Coffee Tropical Agriculture Series (Longman
Scientic & Technical, 1988).
46. Wellman, F. L. Coffee: Botany, Cultivation and Utilization (L. Hill;
Interscience, 1961).
47. Ainsworth, E. A. & Long, S. P. What have we learned from 15 years of free-air
enrichment (FACE)? A meta-analytic review of the responses of
photosynthesis, canopy properties and plant production to rising CO
Phytol. 165, 351372 (2005).
48. Ainsworth, E. A., Leakey, A., Ort, D. R. & Long, S. P. FACE-ing the facts:
inconsistencies and interdependence among eld, chamber and modeling
studies of elevated [CO
] impacts on crop yield and food supply. New Phytol.
179, 59 (2008).
49. Manderscheid, R., Erbs, M. & Weigel, H.-J. Interactive effects of free-air CO
enrichment and drought stress on maize growth. Eur. J. Agron. 52, 1121 (2014).
50. Long, S. P., Ainsworth, E. A., Leakey, A. D. B., Nösberger, J. & Ort, D. R. Food for
thought: lower-than-expected crop yield stimulation with rising CO
concentrations. Science 312, 19181921 (2006).
51. Jaramillo, J. et al. Some like it hot: the inuence and implications of climate
change on coffee berry borer (Hypothenemus hampei) and coffee production in
East Africa. PLoS ONE 6, e24528 (2011).
52. Hein, L. & Gatzweiler, F. The economic value of coffee (Coffea arabica) genetic
resources. Ecol. Econ. 60, 176185 (2006).
53. The World Database on Protected Areas (UNEP/WCMC, 2016); http://www.
54. Matthew, M. W. et al. Status of atmospheric correction using a MODTRAN4-
based algorithm. Proc. SPIE 4049,VI, 199207 (2000).
55. Zhu, Z., Wang, S. & Woodcock, C. E. Improvement and expansion of the Fmask
algorithm: cloud, cloud shadow, and snow detection for Landsats 47, 8, and
Sentinel 2 images. Remote Sens. Environ. 159, 269277 (2015).
56. Kriegler, F. J., Malila, W. A., Nalepka, R. F. & Richardson, W. Preprocessing
transformations and their effects on multispectral recognition. Proc. 6th Intl.
Symp. on Remote Sensing of Environment Vol. II, 97131 (1969).
57. Kadmon, R., Farber, O. & Danin, A. Effect of roadside bias on the accuracy of
predictive maps produced by bioclimatic models. Ecol. Appl. 14, 401413 (2004).
58. Franklin, J. Mapping Species Distributions 1st edn (Cambridge
Univ. Press, 2010).
59. Jiménez-Valverde, A., Lobo, J. & Hortal, J. Not as good as they seem: the
importance of concepts in species distribution modelling. Divers. Distrib. 14,
885890 (2008).
60. Anthony, F. M. et al. Genetic diversity of wild coffee (Coffea arabica L.) using
molecular markers. Euphytica 118, 5365 (2001).
61. Aga, A., Bryngelsson, T., Bekele, E. & Salomon, B. Genetic diversity of forest
arabica coffee (Coffea arabica L.) in Ethiopia as revealed by random amplied
polymorphic DNA (RAPD) analysis. Hereditas 138, 3646 (2003).
62. Lashermes, P., Trouslot, P., Anthony, F., Combes, M. C. & Charrier, A. Genetic
diversity for RAPD markers between cultivated and wild accessions of
Coffea arabica.Euphytica 87, 5964 (1996).
63. Montagnon, C. & Bouharmont, P. Multivariate analysis of phenotypic diversity
of Coffea arabica.Genet. Resour. Crop. Ev. 43, 221227 (1996).
64. Burkhardt, J., Kufa, T., Beining, A., Goldbach, H. E. & Fetene, M. in
21st International Conference on Coffee Science, 10321036 (2007).
65. Thuiller, W., Georges, D. & Engler, R. Biomod2: Ensemble platform for species
distribution modeling. R package v.3.1-64 (CRAN, 2014); http://CRAN.R-
66. Phillips, S. J., Dudik, M. & Schapire, R. E. in Proc. 21st International Conference
on Machine Learning Vol. 69, 655662 (ACM Press, 2004).
67. Phillips, S. J., Anderson, R. P. & Schapire, R. E. Maximum entropy modelling of
species geographic distributions. Ecol. Model. 190, 231259 (2006).
68. Phillips, S. J. & Dudik, M. Modeling of species distributions with
Maxent: new extensions and a comprehensive evaluation. Ecography 31,
161175 (2008).
69. Thuiller, W., Lafourcade, B., Engler, R. & Araújo, M. B. Biomoda platform
for ensemble forecasting of species distributions. Ecography 32, 369373 (2009).
70. Araujo, M. B. & New, M. Ensemble forecasting of species distributions. Trends.
Ecol. Evo. 22, 4247 (2007).
71. Thuiller, W., Lafourcade, B. & Araujo, M. B. Presentation for BIOMOD (2010);*checkout*/pkg/inst/doc/Biomod_
72. Pearson, R. G. Species distribution modelling for conservation educators and
practitioners (American Museum of Natural History, 2008);
73. Knutti, R., Furrer, R., Tebaldi, C., Cermak, J. & Meehl, G. A. Challenges in
combining projections from multiple climate models. J. Climate 23,
27392758 (2010).
74. Jury, M. R. Statistical evaluation of CMIP5 climate change model simulations for
the Ethiopian highlands. Int. J. Climatol. 35, 3744 (2014).
75. Tabor, K. & Williams, J. W. Globally downscaled climate projections for
assessing the conservation impacts of climate change. Ecol. Appl. 20,
554565 (2010).
76. Ramirez-Villegas, J. & Jarvis, A. Downscaling Global Circulation Model Outputs:
The Delta Method Decision and Policy Analysis Working Paper No. 1 (CIAT,
77. Busby, J. R. in Nature conservation: cost effective biological surveys and data
analysis (eds Margules, C. R. & Austin, M. P.) 6468 (CSIRO, 1991).
78. Statistical Downscaling (Delta Method) CMIP3 (CGIAR/CCAFS, accessed
February 2017);
79. Statistical Downscaling (Delta Method) CMIP5 (CGIAR/CCAFS, accessed
February 2017);
This study was conducted for the project Building a Climate Resilient Coffee Economy for
Ethiopia, within the Strategic Climate Institutions Programme (SCIP) Fund, nanced by the
governments of the UK (DFID), Denmark and Norway. We aregrateful to in-countryproject
partners (Ethiopian Biodiversity Institute, National Meteorology Agency, Ministry of
Environment and Forest, Ministry of Agriculture, Addis Ababa University and the Oromia
Coffee FarmersCooperative Union (OCFCU)), fund managers KPMG (Ethiopia),
Departmentfor International Development (DFiD, Ethiopia) and the EthiopianCommodity
Exchange (ECX). We thank: those individuals that assisted with eldwork, including
D. Chomen (OCFCU), R. OSullivan (RBG, Kew) and E. Sage (Speciality Coffee Association
of America); C. Schmitt (University of Freiburg) for the use of coffee plot study data;
D. Georges (LECA,CNRS) for helping with issues in R and the Biomod2package; A. Cooper
(RBG Kew) for providing assistance with image processing in ENVI; and A. Moat, S.
Bachman(RBG Kew), R. Fields and D. Boyd (University of Nottingham)for reviewing earlier
versionsof this contribution. We also acknowledge the Program for ClimateModel Diagnosis
and Intercomparison and the WCRPs Working Group on Coupled Modelling for their roles
in making available the WCRP CMIP3 and CMIP5 multi-model dataset. Support of these
datasets is provided by the Ofce of Science, US Department of Energy. We gratefully
acknowledge coffee farmers and coffee farming communities across Ethiopia for their
participation in the SCIP project, and especially for their hospitality and assistance during
eld work.
Author contributions
J.M. and A.P.D. conceived and led the study; A.P.D. and J.M. led the project; all authors
collected data; J.M., A.P.D., J.W., S.B. and T.W. analysed and processed the data. J.M.
and A.P.D. wrote the paper with contributions from all authors.
Additional information
Supplementary information is available for this paper.
Reprints and permissions information is available at
Correspondence and requests for materials should be addressed to J.M. and A.P.D.
How to cite this article: Moat, J. et al. Resilience potential of the Ethiopian coffee sector under
climate change. Nat. Plants 3, 17081 (2017).
Publishers note: Springer Nature remains neutral with regard to jurisdictional claims in
published maps and institutional afliations.
Competing interests
The authors declare no competing nancial interests.
NATURE PLANTS 3, 17081 (2017) | DOI: 10.1038/nplants.2017.81 |
© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
... Many studies on SDM have tried to determine the realized niche; the greater the realized niche resembles the fundamental niche, the better the projections are plausible. When SDM models are based on bioclimatic factors, the projections tend to be a fundamental niche, especially for new places and future predictions [59]. In Ethiopia, very few studies have been conducted to study the impact of climate change on species distribution using SDM, such as [49,50,[59][60][61][62][63]. ...
... When SDM models are based on bioclimatic factors, the projections tend to be a fundamental niche, especially for new places and future predictions [59]. In Ethiopia, very few studies have been conducted to study the impact of climate change on species distribution using SDM, such as [49,50,[59][60][61][62][63]. Using ensemble forecasting approaches, here we present the distribution of highland bamboo in Ethiopia. ...
... Future predictions of habitat suitability by employing SDMs should consider assumptions and uncertainties associated with it since projecting future climate with high confidence is impractical. Applications and uses of SDMs for estimating range shifts and sensitivity to the extinction of a species to climate change, however, are inevitable [14,59]. ...
Highland bamboo (Oldeania alpina formerly Arundinaria alpina or Yushania alpina) is a species of significant conservation value in Afromontane ecosystems across Africa. It also plays a significant role in the livelihoods of local communities. However, global climate change is anticipated to alter its ecological niche, leading to range shifts and possible habitat contractions. This study aimed to identify potentially suitable habitats for highland bamboo in Ethiopia, determine the resilience of the species under climate change, and establish the environmental factors affecting its habitat. Species distribution modeling (SDM) was implemented in the SDM R package using 231 georeferenced presence records together with climate, topographic, and soil data. To assess climate change risks to the species, predictive models were developed assuming climate scenarios for 2061–2080 under two shared socio-economic pathways (SSPs), namely, SSP2-45 and SSP5-85. The results indicated that highland bamboo mainly grows in high elevation areas with altitudes of 2100–3100 m asl with mean annual temperatures of 11.5–19.3 �C, annual precipitation of 873–1962 mm, precipitation of the driest quarter of 36–147 mm, soil pH of 5.6, and soil CEC of 30.7 cmolc/kg. The current potentially suitable habitat for this species in Ethiopia was estimated at 61,831.58 km2, with the majority of habitats being in the southern and southwestern parts of the country. Our models predicted that the suitable habitat will shrink by 13.4% under the SSP5-85 scenario, while potential new suitable areas for this species were identified under the SSP2-45 scenario. Future vulnerable areas were mostly found in central Ethiopia. Based on the predictions, we conclude that most of the suitable habitats for highland bamboo will remain suitable between the years 2061 and 2080.
... Many studies on SDM have tried to determine the realized niche; the greater the realized niche resembles the fundamental niche, the better the projections are plausible. When SDM models are based on bioclimatic factors, the projections tend to be a fundamental niche, especially for new places and future predictions [59]. In Ethiopia, very few studies have been conducted to study the impact of climate change on species distribution using SDM, such as [49,50,[59][60][61][62][63]. ...
... When SDM models are based on bioclimatic factors, the projections tend to be a fundamental niche, especially for new places and future predictions [59]. In Ethiopia, very few studies have been conducted to study the impact of climate change on species distribution using SDM, such as [49,50,[59][60][61][62][63]. Using ensemble forecasting approaches, here we present the distribution of highland bamboo in Ethiopia. ...
... Future predictions of habitat suitability by employing SDMs should consider assumptions and uncertainties associated with it since projecting future climate with high confidence is impractical. Applications and uses of SDMs for estimating range shifts and sensitivity to the extinction of a species to climate change, however, are inevitable [14,59]. ...
Full-text available
Highland bamboo (Oldeania alpina formerly Arundinaria alpina or Yushania alpina) is a species of significant conservation value in Afromontane ecosystems across Africa. It also plays a significant role in the livelihoods of local communities. However, global climate change is anticipated to alter its ecological niche, leading to range shifts and possible habitat contractions. This study aimed to identify potentially suitable habitats for highland bamboo in Ethiopia, determine the resilience of the species under climate change, and establish the environmental factors affecting its habitat. Species distribution modeling (SDM) was implemented in the SDM R package using 231 georeferenced presence records together with climate, topographic, and soil data. To assess climate change risks to the species, predictive models were developed assuming climate scenarios for 2061–2080 under two shared socio-economic pathways (SSPs), namely, SSP2-45 and SSP5-85. The results indicated that highland bamboo mainly grows in high elevation areas with altitudes of 2100–3100 m asl with mean annual temperatures of 11.5–19.3 �C, annual precipitation of 873–1962 mm, precipitation of the driest quarter of 36–147 mm, soil pH of 5.6, and soil CEC of 30.7 cmolc/kg. The current potentially suitable habitat for this species in Ethiopia was estimated at 61,831.58 km2, with the majority of habitats being in the southern and southwestern parts of the country. Our models predicted that the suitable habitat will shrink by 13.4% under the SSP5-85 scenario, while potential new suitable areas for this species were identified under the SSP2-45 scenario. Future vulnerable areas were mostly found in central Ethiopia. Based on the predictions, we conclude that most of the suitable habitats for highland bamboo will remain suitable between the years 2061 and 2080.
... On the bases of biological diversity of the species and level of management, the coffee production systems in Ethiopia are grouped into four types: forest, semi-forest, garden, and plantation coffee (Aerts et al. 2011;Teferi 2017). Forest coffee refers to production from trees that grow wild under the shade of natural forest trees, do not have a defined owner, and have minimal human intervention (Moat et al. 2017;USDA 2021). The only management practice is access clearing to allow movement in the forest during harvesting time (Degaga 2020). ...
... In semi-forest coffee production, rigorous interventions are practised, including the removal of competing shrubs and selective thinning of the upper canopy to reduce crown closure while maintaining high crown cover (Aerts et al. 2011;Garuma et al. 2015). The farmer who prunes and weeds the forest area and plants coffee seedlings in open areas claims to be the owner of the semi-forest coffee harvest (Moat et al. 2017). Garden coffee is typically grown in the vicinity of a farmer's residence and planted at low densities under some shade. ...
... A survey of the weed flora was carried out in the major coffeegrowing areas of two administrative regions in Ethiopia, namely Oromia and the Southern Nations, Nationalities, and Peoples' Region (SNNPR) (Figure 1). These two regions account for 99% of the coffee produced in the country (Moat et al. 2017). The geographical location (latitude, longitude and altitude) of each sampled field in the study area was determined by using Global Positioning Satellite (GPS) and the overall cumulative value was then tabulated by district ( Table 1). ...
... Regional warming and increasingly erratic rainfall have already increased the frequency of poor harvests, affecting coffee productivity (DaMatta et al. 2018;Ramalho et al. 2018;Laderach et al. 2017;Craparo et al. 2015). Rising temperatures, droughts, and erratic weather patterns are predicted to reduce the overall land suitable for growing Arabica coffee in Ethiopia by 50% between 2040 and 2070 primarily in the Harar, Sidamo, and Jimma areas (Moat et al. 2017). Nearly half of the current coffee growing areas of the country would lose 20-40% climate suitability, mainly in areas of low to medium elevations (Ovalle-Rivera et al. 2015;Davis et al. 2012). ...
... There are several alternatives to reduce the negative impacts of high temperatures on coffee production and to produce high-quality beans. One of the option is moving its cultivation areas to higher elevations as suitability moves upslope to compensate for the increased temperatures (Moat et al. 2017;Ovalle-Rivera et al. 2015). The other option is via managing the canopy shade and the associated cooling of the understorey microclimate (De Frenne et al. 2019, Davis et al. 2012. ...
... We then classified farmland as of low heterogeneity (< 5% woody vegetation), medium heterogeneity (5-20% woody vegetation), and high heterogeneity (> 20% woody vegetation). We classified altitude into five ranges (< 1300 m, 1300-1500 m, 1500-2100 m, 2100-2300 m, > 2300 m), mainly based on the altitudes where coffee growing is viable, both for currently suitable ranges (Senbeta and Denich 2006, Hylander et al. 2013b, Tadesse et al. 2014, Shumi et al. 2018) and a projected future altitudinal shift until 2040 (Moat et al. 2017). Distance from the edge of the forest was used to differentiate between interior forest and edge forest, where forest beyond 150 m from the edge was classified as interior (Shumi et al. 2019b). ...
... Twenty-seven percent of forest, 17% of arable land, 5% of pasture, and 1% of farmland woody vegetation most suitable for coffee growing were converted to plantations ( Fig. 5 and Table A1.5). These conversions took place not only in the current coffee growing altitudes up to 2100 m but up to 2300 m reflecting the predicted shift in suitable areas due to climate change (Moat et al. 2017). In contrast, lower altitudes (1300-1500 m) lost coffee because of increasing climatic unsuitability. ...
... Several studies have indicated high sensitivity of coffee to weather (Craparo et al., 2015;Kath et al., 2020Kath et al., , 2021 and climate (Davis et al., 2012;Bunn et al., 2015b;Moat et al., 2017). The interactive effects of precipitation and temperature define where coffee can be grown as an economically profitable crop as well as year-to-year variability in coffee yield and quality. ...
Full-text available
Weather and climate strongly impact coffee; however, few studies have measured this impact on robusta coffee yield. This is because the yield record is not long enough, and/or the data are only available at a local farm level. A data-driven approach is developed here to 1) identify how sensitive Vietnamese robusta coffee is to weather on district and provincial levels, 2) during which key moments weather is most influential for yield, and 3) how long before harvest, yield could potentially be forecasted. Robusta coffee yield time series were available from 2000 to 2018 for the Central Highlands, where 40% of global robusta coffee is produced. Multiple linear regression has been used to assess the effect of weather on coffee yield, with regularization techniques such as PCA and leave-one-out to avoid over-fitting the regression models. The data suggest that robusta coffee in Vietnam is most sensitive to two key moments: a prolonged rainy season of the previous year favoring vegetative growth, thereby increasing the potential yield (i.e., number of fruiting nodes), while low rainfall during bean formation decreases yield. Depending on location, these moments could be used to forecast the yield anomaly with 3–6 months’ anticipation. The sensitivity of yield anomalies to weather varied substantially between provinces and even districts. In Dak Lak and some Lam Dong districts, weather explained up to 36% of the robusta coffee yield anomalies variation, while low sensitivities were identified in Dak Nong and Gia Lai districts. Our statistical model can be used as a seasonal forecasting tool for the management of coffee production. It can also be applied to climate change studies, i.e., using this statistical model in climate simulations to see the tendency of coffee in the following decades.
... This provides a general mechanism for adaptation to absolute elevation that is likely to be conserved throughout angiosperms. In addition, as this mechanism appears to not have been selected in breeding of quinoa, it may represent an untapped trait for crop improvement at unadapted altitude 26 . It also represents a component that deserves investigation in relation to plant ecological adaptation. ...
Full-text available
Flowering plants (angiosperms) can grow at extreme altitudes, and have been observed growing as high as 6,400 metres above sea level1,2; however, the molecular mechanisms that enable plant adaptation specifically to altitude are unknown. One distinguishing feature of increasing altitude is a reduction in the partial pressure of oxygen (pO2). Here we investigated the relationship between altitude and oxygen sensing in relation to chlorophyll biosynthesis—which requires molecular oxygen3—and hypoxia-related gene expression. We show that in etiolated seedlings of angiosperm species, steady-state levels of the phototoxic chlorophyll precursor protochlorophyllide are influenced by sensing of atmospheric oxygen concentration. In Arabidopsis thaliana, this is mediated by the PLANT CYSTEINE OXIDASE (PCO) N-degron pathway substrates GROUP VII ETHYLENE RESPONSE FACTOR transcription factors (ERFVIIs). ERFVIIs positively regulate expression of FLUORESCENT IN BLUE LIGHT (FLU), which represses the first committed step of chlorophyll biosynthesis, forming an inactivation complex with tetrapyrrole synthesis enzymes that are negatively regulated by ERFVIIs, thereby suppressing protochlorophyllide. In natural populations representing diverse angiosperm clades, we find oxygen-dependent altitudinal clines for steady-state levels of protochlorophyllide, expression of inactivation complex components and hypoxia-related genes. Finally, A. thaliana accessions from contrasting altitudes display altitude-dependent ERFVII activity and accumulation. We thus identify a mechanism for genetic adaptation to absolute altitude through alteration of the sensitivity of the oxygen-sensing system. Plants have adapted to grow at specific altitudes by regulating chlorophyll synthesis in response to ambient oxygen concentration, calibrated by altitude-dependent activity of GROUP VII ETHYLENE RESPONSE FACTOR.
... Climate change is predicted to negatively impact Coffea species, particularly because the rate at which change takes place might be too high for species to migrate in time or adapt via novel mutations Davis et al., 2012Davis et al., , 2019de Aquino et al., 2022;Moat et al., 2017Moat et al., , 2019Ovalle-Rivera et al., 2015). Coffee is the most important commodity crop globally, providing income to more than 125 million people, mostly in tropical nations, and having a retail market value of over 83 billion USD in 2017 alone (Voora et al., 2019). ...
The assessment of population vulnerability under climate change is crucial for planning conservation as well as for ensuring food security. Coffea canephora is, in its native habitat, an understorey tree that is mainly distributed in the lowland rainforests of tropical Africa. Also known as Robusta, its commercial value constitutes a significant revenue for many human populations in tropical countries. Comparing ecological and genomic vulnerabilities within the species’ native range can provide valuable insights about habitat loss and the species’ adaptive potential, allowing to identify genotypes that may act as a resource for varietal improvement. By applying species distribution models, we assessed ecological vulnerability as the decrease in climatic suitability under future climatic conditions from 492 occurrences. We then quantified genomic vulnerability (or risk of maladaptation) as the allelic composition change required to keep pace with predicted climate change. Genomic vulnerability was estimated from genomic environmental correlations throughout the native range. Suitable habitat was predicted to diminish to half its size by 2050, with populations near coastlines and around the Congo River being the most vulnerable. Whole‐genome sequencing revealed 165 candidate SNPs associated with climatic adaptation in C. canephora, which were located in genes involved in plant response to biotic and abiotic stressors. Genomic vulnerability was higher for populations in West Africa and in the region at the border between DRC and Uganda. Despite an overall low correlation between genomic and ecological vulnerability at broad scale, these two components of vulnerability overlap spatially in ways that may become damaging. Genomic vulnerability was estimated to be 23% higher in populations where habitat will be lost in 2050 compared to regions where habitat will remain suitable. These results highlight how ecological and genomic vulnerabilities are relevant when planning on how to cope with climate change regarding an economically important species. We characterized the ecological and genomic vulnerability of the wild Robusta coffee (Coffea canephora). Using species distribution modelling, we predicted a loss of ~50% in suitable habitat area in 2050. We detected 165 genetic markers that could be involved in facilitating adaptation to future climate conditions. Yet, we estimated that genomic vulnerability was higher for populations at the West and East margins of the current distribution and ~23% higher in populations where habitat will be lost in 2050. Our study highlights how ecological and genomic vulnerabilities are relevant for conservation policies and when planning for varietal improvement.
Estimating crop biomass is critical for countries whose primary source of income is agriculture. It is a valuable indicator for evaluating crop yields and provides information to growers and managers for developing climate change adaptation strategies. The objective of the study was to model the impacts of agroclimatic indicators on the performance of aboveground biomass (AGB) in Arabica coffee trees, a critical income source for millions of Ethiopians. One hundred thirty-five coffee tree stump diameters were measured at 40 cm above ground level. The historical (1998–2010) and future (2041–2070) agroclimatic data were downloaded from the European Copernicus climate change services website. All datasets were tested for missing data, outliers, and multicollinearity and were grouped into three clusters using the K-mean clustering method. The parameter estimates (coefficients of regression) were analyzed using a generalized regression model. The performance of coffee AGB in each cluster was estimated using an artificial neural network model. The future expected change in AGB of coffee trees was compared using a paired t-test. The regression model's results reveal that the sensitivity of C. arabica trees to agroclimatic variables significantly differs based on the kind of indicator, RCP scenario, and microclimate. Under the current climatic conditions, the rise of the coldest minimum (TNn) and warmest (TXx) temperatures raises the AGB of the coffee tree, but the rise of the warmest minimum (TNx) and coldest maximum (TXn) temperatures decreases it (P <0.05). Under the RCP4.5, the rise of consecutively dry days (CDD) and TNx increased the AGB of the coffee tree, while TNx and TXx decreased it (P<0.05. Except for TXx, all indicators would significantly reduce the AGB of coffee trees under RCP8.5 (P <0.05). The average values of AGB under the current, RCP4.5, and RCP85 climate change scenarios, respectively, were 26.66, 28.79, and 24.41 kg/tree. Compared to the current climatic conditions, the predicted values of AGB under RCP4.5 and RCP8.5 will increase in the first and third clusters and decrease in the second. As a result, early warning systems and adaptive strategies will be necessary to reduce the detrimental consequences of climate change. More research into the effects of other climatic conditions on crops, such as physiologically effective degree days, cold, hot, and rainy periods, is also required.
Full-text available
Predicting climate change impacts on biodiversity is a major scientific challenge, but doing so is important for assessing extinction risk, developing conservation action plans, providing guidance for laws and regulations, and identifying the mechanisms and patterns of impact to inform climate change adaptation. In the few decades since the threat of climate change has been recognised, the conservation community has begun assessing vulnerability to climate change. There is no single ‘correct’ or established way to carry out climate change vulnerability assessments (CCVA) of species. A range of methods have been developed, and a large and burgeoning scientific literature is emerging on this subject. This document aims to ease the challenge that conservation practitioners face in interpreting and using the complex and often inconsistent CCVA literature. The intended target audiences include conservation practitioners (e.g., for CCVA of their focal species or the species in their focal area) and researchers (e.g., for carrying out CCVA to serve conservation, or to evaluate the rigorousness of others’ studies). These guidelines cover an outline of some of the terms commonly used in CCVA, and describe three dominant CCVA approaches, namely correlative (niche-based), mechanistic and trait-based approaches. This guide is structured to provide readers first with background information on definitions and metrics associated with CCVA. A discussion on identifying CCVA objectives follows, setting the stage for core guidance on selecting and applying appropriate methods. The subsequent sections focus on interpreting and communicating results, as well as suggestions for using results in Red List assessments and addressing the many sources of uncertainty in CCVAs. A final section explores future directions for CCVAs and research needs. The guide ends with ten case studies that provide essentially worked examples of CCVAs that cover the range of methods described. This guidance document has been developed by a Climate Change Vulnerability Assessment working group convened under the IUCN Species Survival Commission’s Climate Change Specialist Group. The authors’ collective experience covers a broad range of ecosystems, taxonomic groups, conservation sectors and geographic regions, and has been supplemented by an extensive literature review. No guidance on this topic can be exhaustive, but nonetheless, this document should provide a useful reference for those wishing to understand and assess climate change impacts on their focal species, at site, site network and/or at broader spatial scales.
Full-text available
The recent decline in Horn of Africa rainfall during the March-May "long rains" season has fomented drought and famine, threatening food security in an already vulnerable region. Some attribute this decline to anthropogenic forcing, whereas others maintain that it is a feature of internal climate variability. We show that the rate of drying in the Horn of Africa during the 20th century is unusual in the context of the last 2000 years, is synchronous with recent global and regional warming, and therefore may have an anthropogenic component. In contrast to 20th century drying, climate models predict that the Horn of Africa will become wetter as global temperatures rise. The projected increase in rainfall mainly occurs during the September-November "short rains" season, in response to large-scale weakening of the Walker circulation. Most of the models overestimate short rains precipitation while underestimating long rains precipitation, causing the Walker circulation response to unrealistically dominate the annual mean. Our results highlight the need for accurate simulation of the seasonal cycle and an improved understanding of the dynamics of the long rains season to predict future rainfall in the Horn of Africa.
Full-text available
Cultivation of Coffea arabica is highly sensitive to and has been shown to be negatively impacted by progressive climatic changes. Previous research contributed little to support forward-looking adaptation. Agro-ecological zoning is a common tool to identify homolo-gous environments and prioritize research. We demonstrate here a pragmatic approach to describe spatial changes in agro-climatic zones suitable for coffee under current and future climates. We defined agro-ecological zones suitable to produce arabica coffee by clustering geo-referenced coffee occurrence locations based on bio-climatic variables. We used random forest classification of climate data layers to model the spatial distribution of these agro-ecological zones. We used these zones to identify spatially explicit impact scenarios and to choose locations for the long-term evaluation of adaptation measures as climate changes. We found that in zones currently classified as hot and dry, climate change will impact arabica more than those that are better suited to it. Research in these zones should therefore focus on expanding arabica's environmental limits. Zones that currently have climates better suited for arabica will migrate upwards by about 500m in elevation. In these zones the up-slope migration will be gradual, but will likely have negative ecosystem impacts. Additionally, we identified locations that with high probability will not change their climatic characteristics and are suitable to evaluate C. arabica germplasm in the face of climate change. These locations should be used to investigate long term adaptation strategies to production systems.
Full-text available
Coffee is highly sensitive to temperature and rainfall, making its cultivation vulnerable to geographic shifts in response to a changing climate. This could lead to the establishment of coffee plantations in new areas and potential conflicts with other land covers including natural forest, with consequent implications for biodiversity and ecosystem services. We project areas suitable for future coffee cultivation based on several climate scenarios and expected responses of the coffee berry borer, a principle pest of coffee crops. We show that the global climatically-suitable area will suffer marked shifts from some current major centres of cultivation. Most areas will be suited to Robusta coffee, demand for which could be met without incurring forest encroachment. The cultivation of Arabica, which represents 70% of consumed coffee, can also be accommodated in the future, but only by incurring some natural forest loss. This has corresponding implications for carbon storage, and is likely to affect areas currently designated as priority areas for biodiversity. Where Arabica coffee does encroach on natural forests, we project average local losses of 35% of threatened vertebrate species. The interaction of climate and coffee berry borer greatly influences projected outcomes.
This latest Fifth Assessment Report of the Intergovernmental Panel on Climate Change (IPCC) again forms the standard scientific reference for all those concerned with the environmental and social consequences of climate change, including students and researchers across the natural and social sciences, professionals in medicine and law, and practitioners in environmental planning, resource management, development, disaster risk reduction, and adaptation. It provides relevant material for decision makers and stakeholders at all levels of government, NGOs, and the private sector worldwide. This assessment provides information on: • Impacts of climate changes that have already occurred and risks of future impacts; • Vulnerabilities and interactions that make some climate events devastating, while others pass with little notice; • Risks of climate change impacts for the health and security of people and ecosystems; • Risks of climate change impacts for human activities; • Prospects for adaptation, including opportunities, barriers, and financing; • Regional and sectoral key risks and vulnerabilities.
The "long rains" season of East Africa has recently experienced a series of devastating droughts, whereas the majority of climate models predict increasing rainfall for the coming decades. This has been termed the East African climate paradox and has implications for developing viable adaptation policies. A logical framework is adopted that leads to six key hypotheses that could explain this paradox. The first hypothesis that the recent observed trend is due to poor quality data is promptly rejected. An initial judgment on the second hypothesis that the projected trend is founded on poor modeling is beyond the scope of a single study. Analysis of a natural variability hypothesis suggests this is unlikely to have been the dominant driver of recent droughts, although it may have contributed. The next two hypotheses explore whether the balance between competing forcings could be changing. Regarding the possibility that the past trend could be due to changing anthropogenic aerosol emissions, the results of sensitivity experiments are highly model dependent, but some show a significant impact on the patterns of tropical SST trends, aspects of which likely caused the recent long rains droughts. Further experiments suggest land-use changes are unlikely to have caused the recent droughts. The last hypothesis that the response to CO2 emissions is nonlinear explains no more than 10% of the contrast between recent and projected trends. In conclusion, it is recommended that research priorities now focus on providing a process-based expert judgment of the reliability of East Africa projections, improving the modeling of aerosol impacts on rainfall, and better understanding the relevant natural variability.
The tropical coffee crop has been predicted to be threatened by future climate changes and global warming. However, the real biological effects of such changes remain unknown. Therefore, this work aims to link the physiological and biochemical responses of photosynthesis to elevated air [CO2 ] and temperature in cultivated genotypes of Coffea arabica L. (cv. Icatu and IPR108) and C. canephora cv. Conilon CL153. Plants were grown for 1 year at 25/20ºC (day/night) and 380 or 700 μL CO2 L(-1) , then subjected to temperature increase (0.5ºC/day) to 42/34ºC. Leaf impacts related to stomatal traits, gas exchanges, C-isotope composition, fluorescence parameters, thylakoid electron transport and enzyme activities were assessed at 25/20ºC, 31/25ºC, 37/30ºC and 42/34ºC. The results showed that 1) both species were remarkably heat tolerant up to 37/30ºC, but at 42/34ºC a threshold for irreversible non-stomatal deleterious effects was reached. Impairments were greater in C. arabica (especially in Icatu) and under normal [CO2 ]. Photosystems and thylakoid electron transport were shown to be quite heat tolerant, contrasting to the enzymes related to energy metabolism, including RuBisCO, which were the most sensitive components. 2) Significant stomatal trait modifications were promoted almost exclusively by temperature and were species dependent. Elevated [CO2 ] 3) strongly mitigated the impact of temperature on both species, particularly at 42/34ºC, modifying the response to supra-optimal temperatures, 4) promoted higher water use efficiency under moderately higher temperature (31/25 ºC), and 5) did not provoke photosynthetic down-regulation. Instead, enhancements in [CO2 ] strengthened photosynthetic photochemical efficiency, energy use and biochemical functioning at all temperatures.. Our novel findings demonstrate a relevant heat resilience of coffee species and that elevated [CO2 ] remarkably mitigated the impact of heat on coffee physiology, therefore playing a key role in this crop sustainability under future climate change scenarios. This article is protected by copyright. All rights reserved.
Open access to an unprecedented, comprehensive coordinated set of global coupled climate model experiments for twentieth and twenty-first century climate and other experiments is changing the way researchers and students analyze and learn about climate. The history of climate change modeling was first characterized in the 1980s by a number of distinct groups developing, running, and analyzing model output from their own models with little opportunity for anyone outside of those groups to have access to the model data. This was partly a consequence of relatively primitive computer networking and data transfer capabilities, along with the daunting task of collecting and storing such large amounts