2
Appendix S2. Analysis of aggregation
error
3
4
Systematic differences in crop yield response to climate change resulting from heterogeneity
in cropping intensity
5
A. J. Challinor, B. Parkes and J. Ramirez-Villegas
1
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
As discussed in the main paper, no aggregation effect is seen in the sensitivity analysis for
either crop. It seems prudent to ask whether the lack of signal at the domain-wide level might
be masking a signal in particular regions. Two grid cells with strong temperature and
precipitation gradients were chosen for this purpose. Cell G is a region of high groundnut
cultivation, whilst cell M is a region of high maize cultivation (see main Fig. 2). Both have
strong temperature and precipitation gradients and cell G has a lower precipitation than cell M
(Figure S1). Figure S11 shows the results of the temperature sensitivity analysis for cell G.
Model results for both crops in cell G confirm that a precipitation gradient can result in an
aggregation effect, but such effect is dependent on the direct effect of aggregation on the
weather input. An aggregation effect is detected in both groundnut and maize yield, but not in
crop suitability. Figure S12 shows the results for cell M, where there is no aggregation effect in
either yield or suitability, for either crop. The key difference between the two cells is in the
mean precipitation. In cell M, water is not limiting, and there is no aggregation effect. In cell G,
the strong precipitation gradients over a drier environment has a different effect: dryer 12km
grid cells suffer more water stress, whilst the averaging process does not reduce the water
available in the wetter cells by enough to significantly affect yield. There is some suggestion
that, as with the niche effect, lower baseline yields play a role in the aggregation effect.
However, the aggregation effect is evident when either percentage changes or absolute
differences in yield are used (Figure S13).
25
26
27
28
29
30
31
32
33
Figure S14 investigates the aggregation effect in groundnut further, by plotting the baseline
values of groundnut productivity for cell G for both the 12km and 3 degree simulations. The
coarser-scale simulation has lower yield than the higher resolution simulations. The same is
true of maize in cell G (Figure S15). Hence, as with the domain-wide niche effect, lower values
of baseline yields contribute to the aggregation effect. As with the sensitivity analysis, no
significant aggregation effect is seen in the baseline values of suitability in either of these
figures. The corresponding figure for cell M, Figure S16, shows that, as with the sensitivity
analysis, no aggregation effect is seen in this wetter environment in the baseline values of
either yield or suitability, in either crop.
34
35
36
37
38
39
The aggregation effect in the sensitivity analysis may be mediated entirely by differences in
baseline values, or it may contain a component due to the simulated temperature changes.
Figure 13 assesses the aggregation effect across the whole domain for a temperature change
of +1 K. The panels compare results with and without normalisation, i.e. percentage changes
are compared to yield differences. In both cases, yield reductions in some 3 degree simulations
are larger than those in the 12km simulations – i.e. the aggregation effect is evident in some
40
41
grid cells when either metric is used. Hence, the aggregation effect is not mediated through
baseline differences alone.
42
43
44
45
46
A. GLAM maize CYG=1
B. GLAM maize CYG=0.5
C. GLAM groundnut CYG=1
D. GLAM groundnut CYG=0.5
E. EcoCrop maize
F. EcoCrop groundnut
Figure S11. Scatter plots of crop yield and suitability response to temperature change for both
crops in grid cell G (see main Fig. 2). GLAM maize yield simulations (A, B) have two values of
CYG. Corresponding groundnut simulations are also shown (C, D). EcoCrop maize suitability
simulations for both crops are shown in panels E and F.
47
48
49
50
51
52
53
A. GLAM maize CYG=1
B. GLAM maize CYG=0.5
C. GLAM groundnut CYG=1
D. GLAM groundnut CYG=0.5
E. EcoCrop maize
F. EcoCrop groundnut
Figure S12. Scatter plots of crop yield and suitability response to temperature change for both
crops in grid cell M (see main Fig. 2). GLAM maize yield simulations with CYG = 1 (A) and CYG =
0.5 (B); corresponding GLAM groundnut simulations (C, D); and EcoCrop maize (E) and
groundnut (F) suitability simulations.
54
55
56
57
58
59
A. 3 degree differences
B. 12km differences
C. 3 degree percent
D. 12km percent
Figure S13. Comparison between differences (in kg ha-1) and relative changes (%) in yield
between the control and the +1 K simulation. Maize on the 3 degree (A) and 12 km grid
aggregated to the 3 degree grid (B); and corresponding percentage changes (C, D).
60
A. GLAM
B. GLAM
C. EcoCrop
D. EcoCrop
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
Figure S14. Effect of aggregating 12 km meteorological data to 3x3 degree for groundnut in cell
G. Line shows: normalised GLAM yield (CYG = 1) against temperature (A) and precipitation (B);
EcoCrop suitability against temperature (C) and precipitation (D). Mean temperature and
precipitation during the crop growth period in the 12km simulations are shown as bars. These
are centred on mean values of temperature or precipitation from the 12km simulations. Black
stars show the 3 degree value of productivity or suitability and red stars show the average
values from the 12km simulations. Thus the red stars are by definition at x=0, whilst the x
value of the black stars reflects differences in planting and harvest date, which affect the
averaging period. The Productivity index, defined in order to normalise the results from GLAM,
is the ratio between the mean yield for a given 0.5 degree range and the maximum yield found
in a single 12km cell.
76
77
78
79
80
81
82
A. GLAM
B. GLAM
C. EcoCrop
D. EcoCrop
Figure S15. Effect of aggregating 12 km meteorological data to 3x3 degree for maize in cell G.
Normalised GLAM yield (CYG = 1) against temperature (A) and precipitation (B); EcoCrop
suitability against temperature (C) precipitation (D). This construction of this figure is exactly as
Supplementary Fig. 16. The Productivity index, defined in order to normalise the results from
GLAM, is the ratio between the mean yield for a given 0.5 degree range and the maximum
yield found in a single 12km cell.
83
84
85
86
87
88
89
90
A. GLAM CYG=1
B. GLAM CYG=0.5
C. GLAM CYG=0.5
D. GLAM CYG=0.5
E. EcoCrop
F. Ecocrop
Figure S16. Line plots showing the effect of aggregating 12 km meteorological data to 3x3
degree for the maize in focus area (marked as ‘M’ in Fig. 1, main text). Normalised GLAM yield
with CYG = 1 against temperature (A) and precipitation (B); normalised GLAM yield with CYG =
0.5 against temperature (C) and precipitation (D); EcoCrop suitability against temperature (E)
and precipitation (F). The Productivity index, defined in order to normalise the results from
GLAM, is the ratio between the mean yield for a given 0.5 degree range and the maximum
yield found in a single 12km cell.