Page 1
1
Long-term soil organic carbon and crop yield feedbacks differ between 16 soil-crop models
1
in sub-Saharan Africa
2
Antoine Couëdel
1
, Gatien N. Falconnier
1,2,6
, Myriam Adam
4,5
, Rémi Cardinael
1,6
, Kenneth
3
Boote
7
, Eric Justes
8
, Ward N. Smith
9
, Anthony M. Whitbread
36
, François Affholder
1,10
, Juraj
4
Balkovic
11
, Bruno Basso
12
, Arti Bhatia
13
, Bidisha Chakrabarti
13
, Regis Chikowo
6,14
, Mathias
5
Christina
1,15
, Babacar Faye
16
, Fabien Ferchaud
17,18
, Christian Folberth
11
, Folorunso M.
6
Akinseye
19,20
, Thomas Gaiser
21
, Marcelo Galdos
22
, Sebastian Gayler
23
, Aram Gorooei
24
, Brian
7
Grant
9
, Hervé Guibert
1
, Gerrit Hoogenboom
7
, Bahareh Kamali
24
, Moritz Laub
25
, Fidel
8
Maureira
12
, Fasil Mequanint
23
, Claas Nendel
26,27,28
, Cheryl H. Porter
7
, Dominique Ripoche
29
,
9
Alex C. Ruane
30
, Leonard Rusinamhodzi
31
,Shikha Sharma
13
, Upendra Singh
32
, Johan Six
25
,
10
Amit Srivastava
21
, Bernard Vanlauwe
3
, Antoine Versini
33
, Murilo Vianna
24
, Heidi Webber
26,34
,
11
Tobias Weber
23,35
, Congmu Zhang
7
, Marc Corbeels
1,3
12
1
AIDA, Univ Montpellier, CIRAD, Montpellier, France
2
International Maize and Wheat Improvement Centre (CIMMYT)-Zimbabwe, 12.5 km Peg Mazwe Road, Harare, Zimbabwe
3
International Institute of Tropical Agriculture (IITA), c/o ICIPE Compound P.O. Box 30772-00100, Nairobi, Kenya
4
AGAP Institut, Univ. Montpellier, INRAE, CIRAD, Inst. Agro. F-34398 Montpellier, France
5
Faculty of Agriculture and Food Processing, National University of Battambang, Battambang Province, 021402, Kingdom of
Cambodia
6
Department of Plant Production Sciences and Technologies, University of Zimbabwe, Harare, Zimbabwe
7
University of Florida, Gainesville, FL, USA
8
PERSYST, Univ Montpellier, CIRAD, Montpellier, France
9
Ottawa Research and Development Centre, Agriculture and Agri-Food Canada, Ottawa, ON, Canada
10
CIRAD, UPR AIDA, Maputo, Mozambique
11
Biodiversity and Natural Resources Program, International Institute for Applied Systems Analysis, Laxenburg, Austria
12
Department of Earth and Environmental Sciences, Michigan State University, East Lansing, MI, USA
13
Indian Agricultural Research Institute, New Delhi, India
14
Plant, Soil and Microbial Sciences Department, Michigan State University, East Lansing, MI 48824, USA
15
CIRAD, UPR AIDA, F-97743, Saint-Denis, La Réunion, France
16
Universit Université du Sine Saloum El Hadj Ibrahima Niass (USSEIN), Kaolack, Senegal
17
BioEcoAgro Joint Research Unit, INRAE, Université de Liège, Université de Lille, Université de Picardie Jules Verne, 02000,
Barenton-Bugny, France
18
UMR Eco&Sols, Univ Montpellier, CIRAD, INRAE, IRD, InstitutAgro Montpellier, Montpellier, France
19
International Crop Research Institute for the Semi-Arid Tropics (ICRISAT), Senegal
20
Centre d’Etude Régional pour l’Amélioration de l’Adaptation à la Sécheresse (CERAAS), Thiès Escale, Senegal
21
Institute of Crop Science and Resource Conservation, University of Bonn, Katzenburgweg 5, D-53115, Bonn, Germany
22
Sustainable Soils & Crops, Rothamsted Research, Harpenden, UK
23
Biogeophysics, Institute of Soil Science and Land Evaluation, Hohenheim University, 70599, Stuttgart, Germany
24
Institute for Crop Science and Resource Conservation (INRES), University of Bonn, Katzenburgweg 5, 53115 Bonn, Germany
25
Department of Environmental Systems Science, ETH Zurich, 8092 Zurich, Switzerland
26
Leibniz Centre for Agricultural Landscape Research (ZALF), Müncheberg, Germany
27
Global Change Research Institute CAS, Brno,
Czech Republic
28
Institute of Biochemistry and Biology, University of Potsdam, Potsdam, Germany
29
AgroClim, INRAE, Avignon, France
30
NASA Goddard Institute for Space Studies, New York, NY 10025, USA
31
International Institute of Tropical Agriculture (IITA), PMB L56 Legon-Accra, Ghana
32
International Center for Soil Fertility and Agricultural Development, Muscle Shoals, AL, USA
© 2024 published by Elsevier. This manuscript is made available under the Elsevier user license
https://www.elsevier.com/open-access/userlicense/1.0/
Version of Record:
https://www.sciencedirect.com/science/article/pii/S1161030124000303
Manuscript_78b8588c61543c142daf335ae516a96e
Page 2
2
33
CIRAD, UPR Recyclage & Risque, F-97743, Saint-Denis, La Réunion, France
34
Brandenburg University of Technology, Faculty of Environment and Natural Sciences, Cottbus, Germany
35
Soil Science Section, Faculty of Organic Agricultural Sciences, University of Kassel, Kassel, Germany
36
International Livestock Research Institute, C/o IITA East Africa Hub, Dar es Salaam, Tanzania
13
Abstract
14
Food insecurity in sub-Saharan Africa is partly due to low staple crop yields, resulting from poor soil
15
fertility and low nutrient inputs. Integrated soil fertility management (ISFM), which
includes the
16
combined use of mineral and organic fertilizers, can contribute to increasing yields and sustaining soil
17
organic carbon (SOC) in the long term. Soil-crop simulation models can help assess the performance
18
and trade-offs of a range of crop management practices including ISFM, under current and future
19
climate. Yet, uncertainty in model simulations can be high, resulting from poor model calibration
20
and/or inadequate model structure. Multi-model simulations have been shown to be more robust
21
than those with single models and help understand and reduce modelling uncertainty. In this study,
22
we aim to perform the first multi-model comparison for long-term simulations of crop yield and SOC
23
and their feedbacks in SSA. We evaluated the performance of 16 soil-crop models using data from
24
four long-term maize experiments at sites in SSA with contrasting climates and soils. Each experiment
25
had four treatments: i) no exogenous inputs, ii) addition of mineral nitrogen (N) fertilizer, iii) use of
26
organic amendments, and iv) combined use of mineral and organic inputs. We assessed model
27
performance in two steps: through blind calibration involving a minimum level of experimental data
28
provided to the modeling teams, and subsequently through full calibration, which included a more
29
extensive set of observational data. Model ensemble accuracy was greater with full calibration than
30
blind calibration. Improvement in model accuracy was larger for maize yields (nRMSE 48 vs 18%)
31
than for topsoil SOC (nRMSE 22 vs 14%). Model ensemble uncertainty (defined as the coefficient of
32
variation across the 16 models) increased over the duration of the long-term experiments.
33
Uncertainty of SOC simulations increased when organic amendments were used, whilst uncertainty
34
of yield predictions was largest when no inputs were applied. Our study revealed large discrepancies
35
among the models in simulating i) crop-to-soil feedbacks due to uncertainties in simulated carbon
36
Page 3
3
coming from roots, and ii) soil-to-crop feedbacks due to large uncertainties in simulated crop N
37
supply from soil organic matter decomposition. These discrepancies were largest when organic
38
amendments were applied. The results highlight the need for long-term experiments in which root
39
and soil N dynamics are monitored. This will provide the corresponding data to improve and
40
calibrate soil-crop models, which will lead to more robust and reliable simulations of SOC and crop
41
productivity, and their interactions.
42
43
Keywords:
soil-crop simulation, soil organic matter, soil-crop feedback, ensemble modelling, model
44
intercomparison, long-term experiments
45
46
1.
Introduction
47
48
Food insecurity in sub-Saharan Africa (SSA) can be largely attributed to low cereal crop yields often
49
caused by low soil fertility and low external input use (Sanchez, 2002). In fact, the large yield gaps of
50
cereal crops in SSA can mostly be explained by soil nutrient limitations rather than climate factors
51
(Van Ittersum et al., 2016). Soil fertility has declined in many regions of SSA as a result of the low use
52
of mineral and organic fertilizers, the shortening of fallow periods and soil erosion (Vanlauwe et al.,
53
2015b; Vanlauwe and Giller, 2006). Sustainable intensification schemes, such as Integrated soil
54
fertility management (ISFM), aiming at maximizing the agronomic use efficiency of nutrients through
55
the combined use of mineral and organic fertilizers, are promising entry points to increase both crop
56
yield and soil organic carbon (SOC), with potential positive impacts on food security and adaptation
57
to climate change (Gram et al., 2020; Vanlauwe et al., 2010).
58
Beside the short-term effects brought by the immediate crop response to mineral and organic
59
fertilizers, the performance of ISFM depends on long-term soil-crop feedbacks. Soil-crop feedbacks
60
describe ecological processes whereby a crop leaves biotic and/or abiotic soil legacies and vice versa,
61
which impact the growth of subsequent crops (Bever, 1994; Pernilla Brinkman et al., 2010).
62
Page 4
4
Indeed, SOC buildup through ISFM can i) increase or maintain nutrient supply in the long term
63
through the decomposition of greater amounts of soil organic matter (SOM), and ii) enhance or
64
maintain soil aggregation and improve soil structure which can favor root development (Adams et al.,
65
2020). In turn, higher crop productivity with associated larger amounts of crop residues (roots and
66
shoots) increases carbon (C) inputs into the soil. Other crop-to-soil feedbacks include nutrient cycling
67
(e.g. direct nitrogen (N) carry over from legumes to cereals), carry-over of pests and diseases (e.g.
68
pathogens) and mutualist biota (e.g., symbiotic N-fixing bacteria, mycorrhizal fungi) (Mariotte et al.,
69
2018; Smith-Ramesh and Reynolds, 2017). However, mechanisms involved in soil-crop feedbacks are
70
complex. For example, root growth is also known to stimulate SOM decomposition through the
71
rhizosphere priming effect (Cheng and Kuzyakov, 2005; Dijkstra et al., 2021; Kuzyakov, 2002), and
72
fertilizers do not always increase yields and C inputs into the soil, e.g. in the case of non-responsive
73
soils (Vanlauwe et al., 2015a, 2010), but on the other hand may stimulate SOM decomposition (Feng
74
et al., 2022; Zhou et al., 2014).
75
Long-term experiments on ISFM can provide valuable insights into soil-crop feedback mechanisms
76
(Bationo et al., 2012; Cardinael et al., 2022; Laub et al., 2023b). However, such experiments are
77
costly, and cannot capture the entire variability of soils, climatic conditions and management
78
practices that exist in farmers’ context (Körschens, 2021; Mirtl et al., 2018). Alternatively, process-
79
based soil-crop models that simulate daily soil processes and plant growth, can help explore long-
80
term effects of ISFM on crop growth and soil fertility for different crop genotypes, soil types, and
81
management practices. These models extend beyond the confines of experimental settings and allow
82
for the exploration of the interactions between these factors (G x E x M). More specifically, the
83
coupled soil-plant models can account for some of the complex feedback mechanisms between SOC
84
dynamics and crop growth over the long term (Basso et al., 2018; Smith et al., 2020a). In fact, the
85
existing models incorporate crop-to-soil feedbacks primarily by considering C inputs derived from
86
roots and/or shoots of the cultivated crops. As for the soil-to-crop feedbacks, they principally
87
consider N supply originating from the decomposition of crop residues and SOM.
88
Page 5
5
Some studies have assessed the capability of process-based soil-crop models to reproduce the effects
89
of drought on crop yield (Kamali et al., 2022) across SSA or those of mineral and organic fertilizers on
90
SOC content (Kamoni et al., 2007) in Kenya. However, feedbacks between SOC and crop biomass
91
have received little scrutiny in these studies. DayCent (Del Grosso et al., 2001) and LPJ-GUESS (Smith
92
et al., 2014), two process-based soil-plant models, simulated reasonably well average maize grain
93
yield and SOC responses to ISFM in a long-term experiment in Kenya , but interannual variability in
94
yield was not well captured (Ma et al., 2022; Nyawira et al., 2021). Although these studies highlighted
95
the strengths and weaknesses of these two models, they did not capture the full diversity of existing
96
soil-crop models and considered a limited number of climate and soil conditions that prevail in SSA.
97
Multi-model ensembles can help reduce the uncertainties in the simulation of soil and crop variables
98
as a result of compensation of individual model errors and wider integration of model formalisms
99
(Martre et al., 2015; Wallach et al., 2018)
.
Furthermore, model intercomparisons can help
100
understand to what extent models differ in their design and skills to accurately predict soil and crop
101
variables, which in turn can guide the improvement of these models. Yet, model intercomparisons
102
have mainly focused on the impact of climate change on crop yields, particularly for wheat, maize
103
and rice (e.g. Asseng et al., 2013; Bassu et al., 2014; Kothari et al., 2022; Li et al., 2015). Few studies
104
have focused on simulating SOC dynamics (Bruni et al., 2022; Farina et al., 2020; Smith et al., 1997,
105
2012), and even fewer have simulated conjointly crop yield and SOC dynamics (Basso et al., 2018; He
106
et al., 2021; Sándor et al., 2020). In fact, multi-model comparisons regarding SOC dynamics have
107
either been done in the absence of observations for model calibration (Basso et al., 2018), or were
108
conducted for bare soil conditions only, i.e. without plant growth (Bruni et al., 2022; Farina et al.,
109
2020) or for temperate climates only (He et al., 2021; Sándor et al., 2023, 2020; Smith et al., 2012).
110
For example, Sándor et al. (2023) found that
some models
overestimated observed SOC contents,
111
while others underestimated them, and that the model ensemble median outperformed individual
112
models in 92% of cases. Inconsistencies between model simulations and observed SOC data can be
113
caused by inaccuracies in the simulation of both soil and crop processes. Accurate model comparison
114
Page 6
6
is challenging due to the complexities arising from the contribution of individual components (such
115
as the soil and crop modules) to the overall model inaccuracy (Ehrmann and Ritz, 2014).
116
When dealing with ISFM in SSA, two additional challenges need to be considered: i) water and N
117
stresses need to be well accounted for in the model to accurately simulate crop yields, especially in
118
low-input systems (Falconnier et al., 2020), and ii) organic fertilizers, that are commonly used by
119
smallholder farmers, have to be well represented in the model and their impact on SOC dynamics
120
and crop growth need to be accurately simulated (Cavalli et al., 2019; Levavasseur et al., 2021).
121
Whilst Falconnier et al. (2020) showed the potential of using multi-model ensembles for accurate
122
maize yield simulations in SSA, we are not aware of any model intercomparison that evaluates the
123
accuracy of both crop yield and SOC simulations across a range of experimental sites over the long
124
term (without model reinitialization each year). Indeed, most model intercomparisons use short-
125
term field experiments for model calibration and do not account for carry-over effects of soil
126
variables from one year to another (i.e. apply model reinitialization every year). Yet, Basso et al.
127
(2018) and Smith et al. (2020) found a large effect on simulated crop yields when including soil-crop
128
feedbacks from soil C and N processes.
129
This study aims to assess how current soil-crop models simulate long-term crop-soil feedbacks. More
130
specifically, we aimed to assess the performance of soil-crop models in simulating long-term trends
131
of rainfed maize yield and SOC for contrasting environments (climate, soil texture, initial SOC level)
132
and fertilization treatments (no input, mineral and/or organic fertilizer) in SSA. We tested five
133
hypotheses: i) local calibration using crop yield and SOC observations is needed to improve model
134
accuracy; ii) the median value of an ensemble of models outperforms single model results for
135
combined yield and SOC simulations; iii) model errors of SOC and crop yield predictions accumulate
136
over time, leading to poor predictions at the end of the long-term simulations; iv) model accuracy is
137
lower in treatments where the number of simulated soil and crop processes and their feedbacks is
138
higher (e.g. decomposition of organic inputs in ISFM, crop N stress when no mineral fertilizers are
139
Page 7
7
applied); and v) more complex soil modules, i.e. with a higher number and more types of (S)OM
140
pools, have a higher model accuracy in simulating SOC dynamics.
141
142
Methods
143
2.1 Experimental data for model calibration
144
2.1.1. Experimental design
145
We calibrated 16 soil-crop models using data from long-term rainfed maize experiments at four sites
146
across SSA. The sites span a range of climates (from warm to cool and from semi-arid to humid), soil
147
textures (from sandy to clayey) and soil fertility levels (varying initial SOC contents, 0.3-3%) (Table 1).
148
The selected long-term experiments (ranging from 9 to 18 years in duration depending on the site)
149
were conducted on station or on farm, but managed by researchers. In three sites (Côte d’Ivoire-
150
Gagnoa, Kenya-Embu and Kenya-Machanga), there were two crop growing seasons per year. The
151
crop grown was maize, except in the second growing season for the site in Côte d’Ivoire where
152
soybean was sown (in two years of the experiment) the soil was left bare (in five years of the
153
experiment) or maize was sown (in five years of the experiment). Experiments were rainfed only, and
154
included four treatments: i) a control without external C and N inputs (-C-N), ii) the application of
155
mineral N fertilizer (between 100 and 160 kg N ha
-1
per growing season depending on the site) (-
156
C+N), iii) the application of organic fertilizer (manure, or compost from crop residues, between 3 and
157
4 t C ha
-1
per year depending on site) (+C-N), and iv) the combined application of organic and mineral
158
N fertilizer (+C+N). The experimental treatments provided proxies for water-limited (+C and +N
159
treatments) and water plus nutrient-limited conditions (-C-N treatment) for crop growth. Weeds,
160
diseases and pests were controlled using appropriate management to avoid biotic stresses,
161
eliminating them as a source of uncertainty in the model predictions in this study. Crop aboveground
162
biomass residues were all removed from the fields in each site, thus the only crop residue inputs
163
were from roots.
164
Page 8
8
The experiments were extensively described in earlier publications, for Côte d’Ivoire-Gagnoa by
165
Pichot et al. (1977), Guibert (1999) and Cardinael et al. (2022), for the two sites in Kenya by Chivenge
166
et al. (2009), Gentile et al. (2009, 2011) and Laub et al. (2023a, 2023b) and for the site in Zimbabwe
167
by Zingore et al. (2007) and Rusinamhodzi et al. (2013).
168
2.1.2. Meteorological data
169
Weather data (solar radiation, maximum and minimum air temperature, precipitation and relative
170
humidity) over the duration of each experiment were collected from different sources depending on
171
the site. Daily rainfall was recorded using graduated cylindrical or tipping bucket rain gauges in all
172
sites with some missing years only in Kenya-Embu. Daily temperature (maximum and minimum) was
173
recorded with electronic sensing thermometers in Côte d’Ivoire-Gagnoa and Kenya-Machanga in all
174
experimental years, but with some missing years in Kenya-Embu; no records were available for
175
Zimbabwe-Murewa. Daily incident solar radiation and relative humidity were retrieved from the
176
NASA POWER (Prediction Of Worldwide Energy Resource) database for all years and sites
177
(https://power.larc.nasa.gov/). Missing daily temperature and rainfall data (missing years) were gap
178
filled with bias-corrected NASA-POWER data. Briefly, for correction of each variable, an equation for
179
a linear regression was created using NASA data as the independent variable (x) and observed data
180
as the dependent variable (y) at each site. The slope and intercept values obtained from the
181
regression equation, y=ax+b, were utilized to generate bias-corrected estimates for the missing
182
temperature/rainfall data.
183
2.1.3. Observations and experimental data
184
Data on the main soil characteristics at the start of the experiments and on crop management were
185
collected in the experiments. Data on crop phenology were available for two sites (Côte d’Ivoire-
186
Gagnoa and Kenya-Embu) while only general information on the maize cultivar planted was provided
187
for Zimbabwe-Murewa and Kenya-Machanga (Table 1). Observed data of grain yield was available for
188
all sites while data of total aboveground biomass was available for all sites except Zimbabwe-
189
Page 9
9
Murewa. Data from SOC measurements were available in different years over the duration of the
190
experiments (for respectively a total of 4, 5, 8 and 13 years in Zimbabwe-Murewa, Côte d’Ivoire-
191
Gagnoa, Kenya-Machanga and Kenya-Embu). SOC was determined from topsoil samples (0-15 cm in
192
Kenya-Machanga and Kenya-Embu and 0-20 cm in Zimbabwe-Murewa and Côte d’Ivoire-Gagnoa).
193
The experiment in Côte d’Ivoire contained three measurement of soil mineral N (0-20 cm) that were
194
done before the start of the first growing season in three different years.
195
2.2 Overview of the models
196
An ensemble of 16 process-based soil-crop models was used in this study (Table 2). Models differed
197
in the way they simulate soil and crop processes. The simulation of SOC dynamics is principally based
198
on first-order decay kinetics for mass loss of SOM pools, and models differed in the number and
199
nature of these pools. Some models have multiple SOM pools (e.g., CENTURY in DSSAT differentiate
200
between an active, slow and passive SOM pool), while other models have a single SOM pool (e.g.,
201
CELCIUS). Ten of the 16 models include a soil microbial biomass pool that follow first-order kinetic
202
decay. The representation of fresh OM pools (from above- and belowground crop residues, organic
203
amendments, such as manure and compost) is the most complex in SIMPLACE (based on the SoilCN
204
model, Corbeels et al. (2005), Molina et al. (1983)), in which four pools are represented (subdivided
205
in a metabolic, holo-cellulosic, ligno-cellulosic and woody litter pool), while half of the other models
206
have only a single fresh OM pool. All models simulate soil N cycling coupled to the (S)OM pool(s).
207
Multi-pool SOC models usually have pool-specific decay coefficients that govern the decomposition
208
rate of each individual pool, leading to a non-linear behavior of the entire system
(e.g. Caruso et al.,
209
2018)
.
The soil-crop models also differ in the representation of other soil processes (e.g., soil water
210
dynamics, nitrate leaching) and of crop processes (e.g. leaf area development, light interception,
211
above- and belowground biomass accumulation and grain formation) (see Table S1 for more details
212
on the differences in the simulation of crop processes between the models).
213
2.3 Model calibration procedure
214
Page 10
10
The models were calibrated in two phases, namely a blind and a full calibration phase, following a
215
similar approach as other multi-model comparisons within the Agricultural Model Intercomparison
216
and Improvement Project (AgMIP) (Rosenzweig et al., 2013). For the blind calibration, values
217
(measured or estimated) for a minimum set of input variables required to execute the model were
218
provided, i.e. general soil characteristics, initial soil conditions, weather, soil and crop management,
219
and observed dates of emergence, flowering and physiological maturity of the maize crop (Table 1).
220
Soil depth in the model runs was set at 1.5 m for all sites. Model parameter adjustment was limited
221
to fitting maize development stages to observed emergence, anthesis and maturity. The modelling
222
teams could use their own method for initialization of the SOM pools (use initial SOC value,
223
combined with spin-up runs - i.e. running the model iteratively for thousands of years to approximate
224
the steady state solution - or with empirical functions between SOM pools and soil texture). For the
225
full calibration, values for additional observed crop and soil variables were provided to the modelling
226
teams, i.e. grain yield, total aboveground biomass at harvest when available, and SOC values (0- 15
227
cm for Kenya-Embu and Kenya-Machanga and 0-20 cm for Côte d’Ivoire-Gagnoa and Zimbabwe-
228
Murewa) at various points in time, depending on the site (Table 1). The modeling teams were free to
229
adjust any model parameter they found relevant to improve the match between observed data and
230
simulations, using their usual calibration methods. The teams calibrated crop and soil-related
231
parameters (e.g. parameters governing SOM and residue decomposition) to improve the accuracy in
232
crop growth and SOC simulations. Simulations were performed by a single modeling team of
233
experimented users for each model. Different options of the same model (DSSAT, Expert-N and
234
SIMPLACE, see Table 2) were simulated by a single modelling team.
235
2.4 Data analysis
236
2.4.1 Model performance assessment
237
We assessed the accuracy of each individual model and the multi-model ensemble in reproducing
238
crop yield and SOC stocks by using the root mean square error (RMSE) and normalized RMSE
239
(nRMSE):
240
Page 11
11
=
∑
−
,
241
=
Ō
x 100 ,
242
where O
i
and S
i,m
are the observed and simulated values (for model m) for the n
th
measurement, n is
243
the number of observations and Ō is the mean of the observed values.
244
The Nash–Sutcliffe modeling efficiency (EF) was also used to assess the accuracy of the multi-model
245
ensemble in reproducing crop yield and SOC stocks as follows:
246
=
∑
Ō
∑
!
"#
!
"#
∑
Ō
!
"#
,
247
where O
i
is the observed value of the n
th
measurement, Ō is the mean of the observed values and S
i
is
248
the simulated value corresponding to Oi.
249
Uncertainty in the multi-model ensemble simulations was assessed using the coefficient of variation
250
(CV) of the 16 model simulations for yield and SOC:
251
$%
&
% =
(
)
)
× 100 ,
252
where
σ
+
is the standard deviation of the simulated values among models at site s and S is the mean
253
of the simulated values at site s.
254
255
2.4.2 Evaluation of soil-crop feedbacks
256
The relationships between i) maize yield or SOC versus time since the start of the experiment, ii) N
257
mineralized and SOC, and (ii) nRMSE of SOC simulations and the number of SOM pools, were
258
evaluated with Pearson correlation coefficients, using the cor.test function of the R Stats package (R
259
core team, 2023). For all statistical tests, the significance level, α, was set to 0.05.
260
261
Results
262
3.1
Accuracy of multi-model ensemble: blind vs full calibration
263
Page 12
12
In the blind calibration phase, the multi-model ensemble strongly overestimated average maize
264
yields across all years and treatments of the four experiments (median ensemble nRMSE of 48%,
265
Figure 1.A), whilst final topsoil SOC stocks of the treatments in the four experiments were simulated
266
with acceptable accuracy (median ensemble nRMSE of 22%, Figure 1.B). Full calibration substantially
267
improved the ability of the multi-model ensemble to reproduce average maize yields and final topsoil
268
SOC stocks of the different treatments in the four sites (Figure 1). The increase in model accuracy
269
following full calibration was stronger for the simulations of average maize yields than for the SOC
270
simulations, with a reduction of nRMSE by 30 and 8%, respectively (Figure 1). The multi-model
271
ensemble failed, however, to adequately reproduce interannual variability in maize yields (Figure 2,
272
Figure S1, Figure S2) in the blind (R
2
=0.35) and full (R
2
=0.47) calibration phases. The changes in
273
topsoil SOC stocks over time were generally well reproduced for the experimental treatments of the
274
four experiments in both the blind (R
2
=0.96) and full (R
2
=0.98) calibration phases (Figure 3, Figure S1,
275
Figure S3). The accuracy of the multi-model ensemble in simulating interannual yield variability was
276
highest in Côte d’Ivoire-Gagnoa where both low and high yields were accurately reproduced for the
277
four experimental treatments in the blind and full calibration phases (Figure 2, Figure S1, Figure S2).
278
3.2 Accuracy of individual models vs multi-model ensemble
279
Larger differences in accuracy between individual models occurred in the blind calibration phase
280
(nRMSE ranging from 42 to 150%) compared to the full calibration phase (nRMSE ranging from 17 to
281
76%) for the simulations of average maize yield across all sites and treatments (Figure 4). Individual
282
models also simulated final topsoil SOC stocks with larger differences in accuracy in the blind
283
calibration phase (nRMSE ranging from 17 to 90%) compared to the full calibration phase (nRMSE
284
ranging from 9 to 39%, with exclusion of one model whose accuracy did not change between the
285
blind and the full calibration) (Figure 4). Only DayCent was more accurate than the median multi-
286
model ensemble for the simulation of average yields after full calibration (Figure 4). Six models (ST,
287
CE, DN, DS1, DC and EI) were more accurate than the median multi-model ensemble in simulating
288
the final topsoil SOC stocks after full calibration (Figure 4). Only DayCent had better accuracy than
289
Page 13
13
the median multi-model ensemble for both simulated average maize yields and final topsoil SOC
290
stocks.
291
The ranking of the individual models differed between the blind and full calibration phases (Figure 4).
292
In the blind calibration phase, model ranking for average maize yield and final topsoil SOC stocks
293
were different; some models were among the best to reproduce final SOC stocks but among the
294
worst to reproduce average yields (e.g., EI and EC). Conversely, some models were among the best to
295
reproduce average yields but among the worst to reproduce final SOC stocks (e.g., DS2 and DN).
296
However, with full calibration, model ranking for averaged yield and final topsoil SOC stocks were
297
similar. In other words, models that performed best to reproduce final SOC stocks were also
298
performing best for average yields (e.g., ST, CE, DN).
299
3.3 Multi-model ensemble drift over time
300
With full calibration, the accuracy of the multi-model ensemble (expressed as RMSE) decreased over
301
time for the SOC simulations, and at the same time uncertainty (expressed as CV across models)
302
increased (Figure 5.C and 5.D). On the other hand, no clear trends could be observed for the maize
303
yield simulations (Figure 5.A and Figure 5.B). In fact, the change in model accuracy over time was not
304
treatment-specific for yield simulations, except for the -C+N treatment in Kenya-Embu where yields
305
were increasingly overestimated towards the end of the experiment (maximum RMSE of 3.7 t/ha,
306
Figure 6.A). Yields were simulated with greater accuracy in the no input treatment (-C-N), but with
307
greater uncertainty, than in any other treatments in all sites, except in Côte d’Ivoire-Gagnoa. The
308
change in model accuracy over time was also not treatment-specific for the SOC simulations (Figure
309
6.C). However, all treatments showed an increase in uncertainty of the SOC simulations over time, in
310
all sites, and this increase was larger in treatments with addition of C inputs (Figure 6.D).
311
3.4 Causes of uncertainty in soil-crop feedbacks
312
SOC-yield feedbacks
313
In the no input treatments (-C-N), the observed significant decline in SOC and maize yield over the
314
experimental period in Côte d’Ivoire-Gagnoa and Kenya-Machanga, indicating significant negative
315
Page 14
14
soil-crop feedbacks, were reproduced by seven out of the 16 models and eight out of the 16 models,
316
respectively (Figure 7). On the other hand, the absence of significant conjoint SOC and maize yield
317
decline over the experimental period in Kenya-Embu and Zimbabwe-Murewa were reproduced by 15
318
out of the 16 models. Some models simulated a sharp SOC decline without yield loss in Côte d’Ivoire-
319
Gagnoa (IN, EG, EC) and Kenya-Machanga (ES, EG), while others simulated a large increase in yield
320
without SOC increase in Kenya-Embu (SI1) and Zimbabwe-Murewa (ES, SI1). None of the models
321
simulated a significant negative correlation between declines in SOC and yield.
322
Soil-to-crop feedbacks (via N mineralization)
323
In treatments without organic amendments (-C treatments), where N supply for crops originates
324
from SOM decomposition, the simulated average amounts of N mineralized per year showed large
325
differences between models in all sites, ranging from 10 to 100 kg N ha
-1
year
-1
in Côte d’Ivoire, from
326
10 to 260 kg N ha
-1
year
-1
in Kenya-Embu, from 10 to 80 kg N ha
-1
year
-1
in Kenya-Machanga, and from
327
5 to 70 kg N ha
-1
year
-1
in Zimbabwe-Murewa (Figure 8). Consequently, models simulated largely
328
different annual N mineralization rates, from 0.5% to 11% of soil total N in Côte d’Ivoire, 0.5% to 6%
329
in Kenya-Embu, 0.5% to 5.5% in Kenya-Machanga, and 0.6% to 4% in Zimbabwe-Murewa (Figure S4).
330
Models differed in their simulation of changes in annual N mineralization over time in response to
331
changes in SOC (Figure 8). Some models simulated N mineralization remaining constant with an
332
increase in SOC (AP, MO), while others simulated an increase in N mineralization with SOC remaining
333
constant (SI2). Among the models that simulated a significant relationship between SOC and
334
mineralized N, large differences were found in the slope of the relationship (resulting in varying N
335
mineralization rates). The largest N mineralization rates per ton SOC were simulated by CE (10.6 kg N
336
t
-1
C) and DN (8.6 kg N t
-1
C) in all sites, compared to other models that simulated N mineralization
337
rates of about 1.2 to 2.5 kg N t
-1
C (ST, EG, ES, EC). Interestingly, the models that simulated SOC loss
338
without a decrease in yield (Figure 7) were also those that simulated a weak relationship between
339
mineralized N and SOC (IN, EG, EC) (Figure 8).
340
Page 15
15
With addition of organic amendments (+C treatments), simulated amounts of mineralized N were
341
higher, with much greater discrepancies between models than without added C (Figure S5),
342
indicating an important but highly uncertain effect of organic amendments on soil-to-crop feedbacks
343
in the model simulations.
344
Crop-to-soil feedbacks (via root inputs)
345
Simulated maize root biomass increased with N fertilization for most models (Figure 9, Figure S6). As
346
a result, in treatments without organic amendments (C input coming only from the roots), most
347
models simulated a decrease in SOC with a decrease in cumulative roots biomass over the duration
348
of the experiment for all sites. Yet, some other models did not consistently reproduce this effect (AP,
349
CE, MO, IN, SI1) (Figure 9). Interestingly, models that simulated yield gain without SOC increase
350
(Figure 7) were those that did not simulate a strong relationship between cumulative root biomass
351
and SOC (SI1, ES) (Figure S6).
352
Simulated average amounts of maize root biomass per year showed large differences between
353
models. In the no input treatment (-C-N), simulated root biomass varied from 0.2 to 1.5 t ha
-1
per
354
year, while in the +N+C treatments it varied from 0.2 to 3.8 t ha
-1
per year (Figure S6). The simulated
355
maize root:shoot ratios differed largely between models, which explains to a large extent the
356
uncertainty in the simulated root biomass (Figure S7.A). Uncertainties in the simulated maize
357
root:shoot ratios were not larger in +N and +C treatments compared to no input treatments (Figure
358
S7.B).
359
3.5 Impact of model structure on model accuracy
360
The type and the number of (S)OM pools had no significant effect on model accuracy for simulations
361
of topsoil SOC stocks towards the end of the experiments (Figure S8). In particular, the number of
362
SOM pools had no significant effect on model accuracy in simulating SOC in the -C treatments, while
363
the number of residue OM pools did not lead to improved model accuracy in simulating SOC in the
364
+C treatments (Figure S8). Moreover, the presence of a microbial SOM pool in the models had no
365
significant effect on model accuracy either (Figure S9).
366
Page 16
16
Discussion
367
4.1 Importance of local model calibration with detailed observations
368
Our study revealed that full model calibration with observed crop yield and SOC data over time
369
improves accuracy and reduces uncertainty of multi-model long-term simulations of average crop
370
yields and final topsoil SOC stocks (Figure 1, 2, 3). This has previously been shown in other crop
371
modeling studies using multiple models, although for short-term model simulations (e.g. Asseng et
372
al., 2013; Bassu et al., 2014; Falconnier et al., 2020). The multi-model ensemble failed however to
373
reproduce the observed interannual yield variability throughout the four long-term experiments
374
(Figure S1), as was also reported by other studies in SSA that, however, used, a single soil-crop
375
model (Ma et al., 2022; Nyawira et al., 2021). In contrast, in some other AgMIP crop modelling
376
studies (Bassu et al., 2014; Falconnier et al., 2020; Li et al., 2015), multi-model ensembles simulated
377
accurately interannual crop yield variability in short-term experiments. This was, however, possible
378
because the full model calibration was done using more detailed data of dynamics of leaf area index
379
(LAI), soil water, soil mineral N and plant N over the growing seasons of the experiments. In
380
particular, observed LAI and soil moisture data is extremely valuable when calibrating models, as it
381
enables the model to accurately reproduce yield variations resulting from changes in seasonal rainfall
382
patterns. Therefore, in long-term experiments, model calibration based on only a few experimental
383
years with detailed observation would help improve model accuracy, compared with calibration on
384
many years with a low level of information. Surprisingly, complex crop models (e.g. DSSAT, STICS,
385
APSIM) did not perform better than simpler ones in simulating average yield and interannual
386
variability in yield (e.g. DayCent, see table S1 for a description of crop model complexity). Possibly,
387
this was due to the lack of in-season observations of crop growth (e.g. LAI) for accurate model
388
calibration. We did not observe any impact of model structure on model accuracy. Therefore, we
389
could not formulate any specific recommendation for model improvement based on existing model
390
structures for SOC simulations.
391
Page 17
17
Thus, the full model calibration conducted in our study, based on a limited set of observed data
392
(phenological stages, grain yield, aboveground biomass and SOC), does not allow for the use of the
393
multi-model ensemble in assessing crop production risk associated with weather conditions (i.e.
394
occurrence of low yields, crop failure). Yet, the applied model calibration proved valuable for
395
exploring and better understanding the effect of crop management, such as ISFM, on overall crop
396
productivity and soil fertility, including the long-term soil-crop feedbacks at play.
397
398
4.2
The more soil and crop processes to account for, the higher the model uncertainty over time
399
In our study the accuracy of the multi-model ensemble in simulating SOC decreased over time, but
400
that of average crop yield simulations remained constant (Figure 5). This outcome contradicts our
401
initial hypothesis, which stated that yield simulations would become less accurate towards the end of
402
the experiment due to accumulation of errors over time (including those associated with soil-crop
403
feedbacks). Modelling groups performed their model calibration against observed crop yields
404
throughout all years of the experiment (and not only using yields from e.g. year 1). This may have
405
masked accuracy drifts over time.
406
The model ensemble uncertainty in simulating SOC increased over time, especially in the treatments
407
with organic amendments (Figure 6). This can be attributed to the uncertainties in simulating N
408
immobilization-mineralization processes during decomposition of the organic amendments. In
409
relation to this, we found no impact of the number of residue or SOM pools on model accuracy in our
410
study. Cavalli et al. (2019) showed that only a few model parameters, such as those related to
411
substrate C use efficiency, explained most of the accuracy of the models (i.e. APSIM, EPIC, FASSET
412
and STICS) in simulating the effect of manure on soil C and N dynamics. Levavasseur et al. (2021)
413
compared two STICS versions that included either one or two residue pools, and showed that the
414
one-pool model simulated soil C and N dynamics as accurately as the two-pool model
.
Our study
415
confirms that
when detailed data on soil C and N dynamics is scarce, complex models with multiple
416
conceptual (S)OM pools do not perform better than simple models with fewer pools and parameters
417
Page 18
18
(Castañeda-Vera et al., 2015)
.
Besides, the limited availability of data for calibrating SOM pool sizes
418
diminishes the effectiveness of using more complex models that encompass multiple SOM pools.
419
The uncertainty of the multi-model ensemble in simulating maize grain yield was larger for
420
treatments without mineral and/or organic fertilizer inputs (Figure 6). This result can be attributed to
421
the larger influence of soil-crop feedback processes and the associated errors in simulating them, in
422
these treatments compared to treatments with external inputs. As further discussed in the next
423
section, model differences in simulating SOC-to-crop feedbacks in no input treatments result from
424
large differences in simulating the amounts of N mineralized from SOM. This was especially true for
425
the Kenya-Machanga and Zimbabwe-Murewa sites, that have the lowest initial soil fertility (SOC
426
levels) and where N available for crop uptake that originates from SOM decomposition is limited.
427
Our study clearly shows that simulated impacts of experimental ISFM treatments on SOC dynamics
428
and overall crop yield can differ depending on the model chosen for a study (Figure 2 and 3). This
429
raises challenges in assessing the value of ISFM for different farming contexts (soils, climate). It is
430
therefore essential to exercise caution with crop modeling studies that aim to determine the
431
potential impacts of sustainable intensification practices on crop productivity and soil fertility. We
432
believe that models that correctly reproduce soil-crop feedbacks are suitable candidates to be used
433
in model simulation studies dealing with ISFM. Methods are available to determine the optimal
434
composition and size of the model ensemble for robust estimates of crop yields (Li et al., 2023;
435
Rodríguez et al., 2019).
436
4.3 More detailed experiments are required in SSA to reduce sources of uncertainties in soil-crop
437
feedbacks
438
Our study revealed large discrepancies in the simulated soil-crop feedbacks among models due to
439
differences in simulating N mineralization and root C inputs (Figure 8 and 9). These discrepancies
440
were more pronounced in the treatments with organic amendments. The models that reproduced
441
significant crop-to-SOC feedbacks (a decrease in SOC with a decrease in root biomass) and SOC-to-
442
Page 19
19
crop feedbacks (a decrease in crop N supply with a decrease in SOC) exhibited the best capability to
443
reproduce the observed SOC and yield trends over the duration of the experiments (Figure 4 and 7).
444
Yet, soil C and N feedbacks on crops have often been overlooked in long-term modelling studies
445
where soil properties were typically reinitialized in the models each year (e.g. Bassu et al., 2014;
446
Falconnier et al., 2020).
447
The simulated N mineralization rates varied considerably among the models (from about 1 to 10 fold)
448
(Figure S4). Some models simulated annual mineralization rates that approached or exceeded the
449
maximum value of 5% of soil total nitrogen reported in the literature
(Bationo et al., 2007; Masvaya
450
et al., 2017; Wetselaar and Ganry, 1982). The high uncertainty in simulated amounts of N mineralized
451
leads to uncertainties in simulated soil mineral N, nitrate leaching and plant N uptake. For example,
452
in our study most models simulated a strong accumulation of soil mineral N over the years in the
453
fertilized treatments (data not shown). This could have masked feedbacks between SOC and yields
454
(high soil mineral N leads to a weak influence of N mineralization on crop growth and yield). Yet, an
455
accurate prediction of N mineralization rates and a better understanding of the reasons behind the
456
weak correlation between changes in SOC and the amount of N mineralized in some models are a
457
prerequisite for improved model simulations, especially in low-input cropping systems. From our
458
study, it is evident that more soil mineral N data need to be collected in field experiments in SSA (and
459
elsewhere) for model calibration. A thorough calibration and testing of models for soil N dynamics
460
would be greatly beneficial for better reproducing yield responses to mineral and/or organic
461
fertilizers and better understanding long-term soil-crop feedback mechanisms.
462
In low-input cropping systems, the carry-over of root residues from one cropping season to the other
463
is important to quantify because of its critical influence on SOC and soil mineral N dynamics. The
464
maize root biomass simulated by the models was also subject to large uncertainties (Figure 9 and S6).
465
Some models simulated low root biomass production resulting in very low root:shoot ratios (Figure
466
S7), i.e. below the likely values of 0.08 to 0.43 for maize reported in the literature (Anderson, 1988;
467
Page 20
20
Eghball and Maranville, 1993; Lopez et al., 2023; Ordóñez et al., 2020).
Moreover, model differences
468
in simulating maize root biomass increased with mineral and/or organic fertilizer application (Figure
469
S6). Yet, the effects of fertilizers on crop root biomass are complex, and probably crop- and site-
470
specific. Experimental research found either an increase (Eghball and Maranville, 1993), no clear
471
effects (Anderson, 1988, 1987) or a decrease (Ordóñez et al., 2021) in crop root biomass with mineral
472
fertilization. Yet, most studies agreed that fertilization tends to cause a decline in maize root:shoot
473
ratio (Anderson, 1988, 1987; Eghball and Maranville, 1993; Lopez et al., 2023), which was, however,
474
not reproduced by the models (Figure S7). This outcome calls for more experiments aimed at better
475
understanding the effects of fertilization on root dynamics in SSA. Given the large discrepancies
476
between models, such root biomass measurements, would prove valuable for improving model
477
formalisms that describe the processes and factors of root development. This would in turn allow for
478
a better assessment of root C inputs and their contribution to SOC formation (e.g Whitbread et al.,
479
2003). Besides, more experimental research is needed to quantify the effects of roots on SOC
480
dynamics, as roots can stabilize or destabilize SOC (Dijkstra et al., 2021).
481
Specific experiments in SSA are also needed to unravel the effects of organic amendment from the
482
effects of in-situ crop residues and SOM turnover on soil-crop feedbacks. Bare soil experiments
483
without C inputs, that measure soil C and N, are valuable for calibrating SOM turnover in models.
484
Indeed, with bare soils, it is possible to evaluate model performance without crop C inputs and
485
without interferences arising from crop water use and N uptake
.
Such experiments have for example
486
been conducted in Australia and enabled to calibrate the SoilN module of APSIM (Dalal et al., 1995;
487
Dimes, 1996; Probert et al., 1998). On the other hand, long-term experiments with organic inputs on
488
bare soil can help to calibrate the residue pools associated with the organic amendment, if they are
489
established in a way that avoids erosion. Such long-term experiments have for example been
490
conducted in Sweden and were used to calibrate the CENTURY model (Gerzabek et al., 1997;
491
Paustian et al., 1992). Comparisons of soil C data between bare and cropped soils, along with model
492
simulations, can then further help to assess the effects of crops on SOC dynamics.
493
Page 21
21
Besides the data needed for calibrating SOM turnover formalisms already present in models, model
494
improvements are needed to better reproduce soil-crop feedbacks. Soil C and N fluxes remain
495
difficult to simulate due to incomplete representation of key mechanisms in the models. Campbell
496
and Paustian (2015) highlighted the need for advances in SOC modelling, such as the implementation
497
of microbial pool that plays the central role in SOM stabilization, the modelling of SOM saturation
498
kinetics, and the representation of SOM dynamics in deep soil layers. Besides, the impacts of soil
499
structure on long-term SOC dynamics are also overlooked in current soil-crop models. Soil structure
500
affects root development, water dynamics and SOC stabilization (e.g. inside aggregates) and
501
decomposition while in turn, SOC also affects soil physical and hydraulic properties (Rabot et al.,
502
2018). These complex soil structure-SOC feedbacks are now being considered in the development of
503
new SOC models (Meurer et al., 2020). Other processes such as rhizosphere-SOM interactions, such
504
as rhizodeposition and rhizosphere priming effects, are mostly not represented in models (Cavalli et
505
al., 2019) while it has been reported that rhizosphere priming effects can offset the SOC increase
506
caused by roots (Dijkstra et al., 2021). Moreover, soils can be unresponsive to fertilization due to
507
other crop growth limiting factors, such as nutrient limitations other than N, P and K, or soil
508
compaction(Vanlauwe et al., 2015a, 2010). Those processes are not represented in most models.
509
More specifically, effects of chemical fertilizers on pH are poorly represented in models. It could, for
510
example, explain the poor model inaccuracy in simulating the +N treatments in Embu where maize
511
yields were adversely affected by a decrease in pH over the years (Laub et al., 2023a).
512
Conclusion
513
We conducted the first multi-model comparison for long-term simulations of crop (maize) yields and
514
topsoil SOC trends, and their feedbacks, in SSA. The models differed considerably in their ability to
515
reproduce these trends. Models that accounted for the soil-crop feedbacks showed the best
516
accuracy. For maize yield simulations, the largest model uncertainty was observed in treatments
517
without fertilizer inputs, while for SOC simulations uncertainty was highest when organic
518
amendments were used. Models exhibited large differences in simulating i) soil-to-crop feedbacks as
519
Page 22
22
indicated by the large variations in simulated amounts of crop N supply from SOM decomposition,
520
and ii) crop-to-soil feedbacks as reflected in varying simulated amounts of root biomass residues for
521
SOC formation. As a result, we advocate for an urgent need for detailed experimental data on soil N
522
dynamics and roots in SSA. Such experimental efforts are critical to guarantee the accurate
523
integration of soil-crop feedbacks into models, ensuring their suitability for application in SSA.
524
Acknowledgments
525
We are grateful to the support provided by the AgMIP leadership team. Part of the study was funded
526
by the DSCATT project “Agricultural Intensification and Dynamics of Soil Carbon Sequestration in
527
Tropical and Temperate Farming Systems” (N° AF 1802-001, N° FT C002181), supported by the
528
Agropolis Foundation (“Programme d′ Investissement d′ Avenir” Labex Agro, ANR-10-LABX0001-01)
529
and by the TOTAL Foundation within a patronage agreement.
530
Competing interests
531
Authors have no competing interests to declare
532
Data Availability
533
Data will be made available on request.
534
535
References
536
Adams, A.M., Gillespie, A.W., Dhillon, G.S., Kar, G., Minielly, C., Koala, S., Ouattara, B., Kimaro, A.A.,
537
Bationo, A., Schoenau, J.J., Peak, D., 2020. Long-term effects of integrated soil fertility
538
management practices on soil chemical properties in the Sahel. Geoderma 366, 114207.
539
https://doi.org/10.1016/j.geoderma.2020.114207
540
Aggarwal, P.K., Kalra, N., Chander, S., Pathak, H., 2006. InfoCrop: A dynamic simulation model for the
541
assessment of crop yields, losses due to pests, and environmental impact of agro-ecosystems in
542
tropical environments. I. Model description. Agric. Syst. 89, 1–25.
543
https://doi.org/10.1016/j.agsy.2005.08.001
544
Anderson, E.L., 1988. Tillage and N fertilization effects on maize root growth and root:shoot ratio.
545
Plant Soil 108, 245–251. https://doi.org/10.1007/BF02375655
546
Anderson, E.L., 1987. Corn Root Growth and Distribution as Influenced by Tillage and Nitrogen
547
Page 23
23
Fertilization. Agron. J. 79, 544–549.
548
Asseng, S., Ewert, F., Rosenzweig, C., Jones, J.W., Hatfield, J.L., Ruane, A.C., Boote, K.J., Thorburn, P.J.,
549
Rötter, R.P., Cammarano, D., Brisson, N., Basso, B., Martre, P., Aggarwal, P.K., Angulo, C.,
550
Bertuzzi, P., Biernath, C., Challinor, A.J., Doltra, J., Gayler, S., Goldberg, R., Grant, R., Heng, L.,
551
Hooker, J., Hunt, L.A., Ingwersen, J., Izaurralde, R.C., Kersebaum, K.C., Müller, C., Naresh Kumar,
552
S., Nendel, C., O’Leary, G., Olesen, J.E., Osborne, T.M., Palosuo, T., Priesack, E., Ripoche, D.,
553
Semenov, M.A., Shcherbak, I., Steduto, P., Stöckle, C., Stratonovitch, P., Streck, T., Supit, I., Tao,
554
F., Travasso, M., Waha, K., Wallach, D., White, J.W., Williams, J.R., Wolf, J., 2013. Uncertainty in
555
simulating wheat yields under climate change. Nat. Clim. Chang. 3, 627–632.
556
https://doi.org/10.1038/ncliamte1916
557
Basso, B., Cammarano, D., Troccoli, A., Chen, D., Ritchie, J.T., 2010. Long-term wheat response to
558
nitrogen in a rainfed Mediterranean environment: Field data and simulation analysis. Eur. J.
559
Agron. 33, 132–138. https://doi.org/10.1016/j.eja.2010.04.004
560
Basso, B., Dumont, B., Maestrini, B., Shcherbak, I., Robertson, G.P., Porter, J.R., Smith, P., Paustian,
561
K., Grace, P.R., Asseng, S., Bassu, S., Biernath, C., Boote, K.J., Cammarano, D., De Sanctis, G.,
562
Durand, J.-L., Ewert, F., Gayler, S., Hyndman, D.W., Kent, J., Martre, P., Nendel, C., Priesack, E.,
563
Ripoche, D., Ruane, A.C., Sharp, J., Thorburn, P.J., Hatfield, J.L., Jones, J.W., Rosenzweig, C.,
564
2018. Soil Organic Carbon and Nitrogen Feedbacks on Crop Yields under Climate Change. Agric.
565
Environ. Lett. 3, 180026. https://doi.org/10.2134/ael2018.05.0026
566
Bassu, S., Brisson, N., Durand, J.L., Boote, K., Lizaso, J., Jones, J.W., Rosenzweig, C., Ruane, A.C.,
567
Adam, M., Baron, C., Basso, B., Biernath, C., Boogaard, H., Conijn, S., Corbeels, M., Deryng, D.,
568
De Sanctis, G., Gayler, S., Grassini, P., Hatfield, J., Hoek, S., Izaurralde, C., Jongschaap, R.,
569
Kemanian, A.R., Kersebaum, K.C., Kim, S.H., Kumar, N.S., Makowski, D., Müller, C., Nendel, C.,
570
Priesack, E., Pravia, M.V., Sau, F., Shcherbak, I., Tao, F., Teixeira, E., Timlin, D., Waha, K., 2014.
571
How do various maize crop models vary in their responses to climate change factors? Glob.
572
Chang. Biol. 20, 2301–2320. https://doi.org/10.1111/gcb.12520
573
Bationo, A., Kihara, J., Vanlauwe, B., Waswa, B., Adolwa, I., Saidou, K., 2012. Overview of long term
574
experiments in Africa, Lessons learned from Long-term Soil Fertility Management Experiments
575
in Africa. Springer Science+Business Media Dordrecht 2012. https://doi.org/10.1007/978-94-
576
007-2938-4
577
Bationo, A., Kihara, J., Vanlauwe, B., Waswa, B., Kimetu, J., 2007. Soil organic carbon dynamics,
578
functions and management in West African agro-ecosystems. Agric. Syst. 94, 13–25.
579
https://doi.org/10.1016/j.agsy.2005.08.011
580
Beaudoin, N., Lecharpentier, P., Ripoche, D., Strullu, L., Mary, B., Léonard, J., Launay, M., Justes, É.,
581
2023. STICS soil-crop model, conceptual framework, equations and uses. Versailles, Éditions
582
Quæ.
583
Bever, J.D., 1994. Feeback between Plants and Their Soil Communities in an Old Field Community.
584
Ecology 75, 1965–1977. https://doi.org/https://doi.org/10.2307/1941601
585
Biernath, C., Gayler, S., Bittner, S., Klein, C., Högy, P., Fangmeier, A., Priesack, E., 2011. Evaluating the
586
ability of four crop models to predict different environmental impacts on spring wheat grown in
587
open-top chambers. Eur. J. Agron. 35, 71–82. https://doi.org/10.1016/j.eja.2011.04.001
588
Brisson, N., Gary, C., Justes, E., Roche, R., Mary, B., Ripoche, D., Zimmer, D., Sierra, J., Bertuzzi, P.,
589
Burger, P., Bussiere, F., Cabidoche, Y.M., Cellier, P., Debaeke, P., Gaudillere, J.P., Henault, C.,
590
Maraux, F., Seguin, B., Sinoquet, H., 2003. An overview of the crop model STICS. Eur. J. Agron.
591
18, 309–332. https://doi.org/10.1016/S0223-5234(03)00100-4
592
Page 24
24
Bruni, E., Chenu, C., Abramoff, R.Z., Baldoni, G., Barkusky, D., Clivot, H., Huang, Y., Kätterer, T., Pikula,
593
D., Spiegel, H., Virto, I., Guenet, B., 2022. Multi-modelling predictions show high uncertainty of
594
required carbon input changes to reach a 4 ‰ target. Eur. J. Soil Sci. 1–26.
595
https://doi.org/10.1111/ejss.13330
596
Campbell, E.E., Paustian, K., 2015. Current developments in soil organic matter modeling and the
597
expansion of model applications: A review. Environ. Res. Lett. 10.
598
https://doi.org/10.1088/1748-9326/10/12/123004
599
Cardinael, R., Guibert, H., Brédoumy, S.T.K., Gigou, J., N’Goran, K.E., Corbeels, M., 2022. Sustaining
600
maize yields and soil carbon following land clearing in the forest–savannah transition zone of
601
West Africa: Results from a 20-year experiment. F. Crop. Res. 275, 108335.
602
https://doi.org/10.1016/j.fcr.2021.108335
603
Caruso, T., De Vries, F.T., Bardgett, R.D., Lehmann, J., 2018. Soil organic carbon dynamics matching
604
ecological equilibrium theory. Ecol. Evol. 8, 11169–11178. https://doi.org/10.1002/ece3.4586
605
Castañeda-Vera, A., Leffelaar, P.A., Álvaro-Fuentes, J., Cantero-Martínez, C., Mínguez, M.I., 2015.
606
Selecting crop models for decision making in wheat insurance. Eur. J. Agron. 68, 97–116.
607
https://doi.org/10.1016/j.eja.2015.04.008
608
Cavalli, D., Bellocchi, G., Corti, M., Marino Gallina, P., Bechini, L., 2019. Sensitivity analysis of C and N
609
modules in biogeochemical crop and grassland models following manure addition to soil. Eur. J.
610
Soil Sci. 70, 833–846. https://doi.org/10.1111/ejss.12793
611
Cheng, W., Kuzyakov, Y., 2005. Root effects on soil organic matter decomposition, in: Roots and Soil
612
Management: Inter- Actions between Roots and the Soil. American Society of Agronomy, Crop
613
Science Society of America, Soil Science Society of America, 677 S. Segoe Rd., Madison, WI
614
53711, USA, pp. 119–143. https://doi.org/10.2134/agronmonogr48.c7
615
Chivenge, P., Vanlauwe, B., Gentile, R., Wangechi, H., Mugendi, D., van Kessel, C., Six, J., 2009.
616
Organic and mineral input management to enhance crop productivity in central Kenya. Agron. J.
617
101, 1266–1275. https://doi.org/10.2134/agronj2008.0188x
618
Corbeels, M., McMurtrie, R.E., Pepper, D.A., O’Connell, A.M., 2005. A process-based model of
619
nitrogen cycling in forest plantations: Part I. Structure, calibration and analysis of the
620
decomposition model. Ecol. Modell. 187, 426–448.
621
https://doi.org/10.1016/j.ecolmodel.2004.09.005
622
Dalal, R.C., Strong, W.M., Weston, E.J., Cooper, J.E., Lehane, K.J., King, A.J., Chicken, C.J., 1995.
623
Sustaining productivity of a Vertisol at Warra, Queensland, with fertilisers, no-tillage, or
624
legumes. 1. Organic matter status. Aust. J. Exp. Agric. 35, 903–913.
625
Del Grosso, S., Parton, W., Mosier, A., Hartman, M., Brenner, J., Ojima, D., Schimel, D., 2001.
626
Simulated Interaction of Carbon Dynamics and Nitrogen Trace Gas Fluxes Using the DAYCENT
627
Model, in: Modeling Carbon and Nitrogen Dynamics for Soil Management.
628
https://doi.org/10.1201/9781420032635.ch8
629
Dijkstra, F.A., Zhu, B., Cheng, W., 2021. Root effects on soil organic carbon: a double-edged sword.
630
New Phytol. 230, 60–65. https://doi.org/10.1111/nph.17082
631
Dimes, J.., 1996. Simulation of mineral N supply to no-till crops in the semi arid tropics. Griffith
632
University, Queensland.
633
Eghball, B., Maranville, J.W., 1993. Root Development and Nitrogen Influx of Corn Genotypes Grown
634
under Combined Drought and Nitrogen Stresses. Agron. J. 85, 147–152.
635
Page 25
25
Ehrmann, J., Ritz, K., 2014. Plant: Soil interactions in temperate multi-cropping production systems.
636
Plant Soil 376, 1–29. https://doi.org/10.1007/s11104-013-1921-8
637
Enders, A., Vianna, M., Gaiser, T., Krauss, G., Webber, H., Srivastava, A.K., Seidel, S.J., Tewes, A.,
638
Rezaei, E.E., Ewert, F., 2023. SIMPLACE—a versatile modelling and simulation framework for
639
sustainable crops and agroecosystems. in silico Plants 5, 1–18.
640
https://doi.org/10.1093/insilicoplants/diad006
641
Falconnier, G.N., Corbeels, M., Boote, K.J., Affholder, F., Adam, M., MacCarthy, D.S., Ruane, A.C.,
642
Nendel, C., Whitbread, A.M., Justes, É., Ahuja, L.R., Akinseye, F.M., Alou, I.N., Amouzou, K.A.,
643
Anapalli, S.S., Baron, C., Basso, B., Baudron, F., Bertuzzi, P., Challinor, A.J., Chen, Y., Deryng, D.,
644
Elsayed, M.L., Faye, B., Gaiser, T., Galdos, M., Gayler, S., Gerardeaux, E., Giner, M., Grant, B.,
645
Hoogenboom, G., Ibrahim, E.S., Kamali, B., Kersebaum, K.C., Kim, S.H., van der Laan, M., Leroux,
646
L., Lizaso, J.I., Maestrini, B., Meier, E.A., Mequanint, F., Ndoli, A., Porter, C.H., Priesack, E.,
647
Ripoche, D., Sida, T.S., Singh, U., Smith, W.N., Srivastava, A., Sinha, S., Tao, F., Thorburn, P.J.,
648
Timlin, D., Traore, B., Twine, T., Webber, H., 2020. Modelling climate change impacts on maize
649
yields under low nitrogen input conditions in sub-Saharan Africa. Glob. Chang. Biol. 1–23.
650
https://doi.org/10.1111/gcb.15261
651
Farina, R., Sándor, R., Abdalla, M., Álvaro-Fuentes, J., Bechini, L., Bolinder, M.A., Brilli, L., Chenu, C.,
652
Clivot, H., De Antoni Migliorati, M., Di Bene, C., Dorich, C.D., Ehrhardt, F., Ferchaud, F., Fitton,
653
N., Francaviglia, R., Franko, U., Giltrap, D.L., Grant, B.B., Guenet, B., Harrison, M.T., Kirschbaum,
654
M.U.F., Kuka, K., Kulmala, L., Liski, J., McGrath, M.J., Meier, E., Menichetti, L., Moyano, F.,
655
Nendel, C., Recous, S., Reibold, N., Shepherd, A., Smith, W.N., Smith, P., Soussana, J.F., Stella, T.,
656
Taghizadeh-Toosi, A., Tsutskikh, E., Bellocchi, G., 2020. Ensemble modelling, uncertainty and
657
robust predictions of organic carbon in long-term bare-fallow soils. Glob. Chang. Biol. 1–25.
658
https://doi.org/10.1111/gcb.15441
659
Faye, B., Webber, H., Diop, M., Mbaye, M.L., Owusu-Sekyere, J.D., Naab, J.B., Gaiser, T., 2018.
660
Potential impact of climate change on peanut yield in Senegal, West Africa. F. Crop. Res. 219,
661
148–159. https://doi.org/10.1016/j.fcr.2018.01.034
662
Feng, X., Qin, S., Zhang, D., Chen, P., Hu, J., Wang, G., Liu, Y., Wei, B., Li, Q., Yang, Y., Chen, L., 2022.
663
Nitrogen input enhances microbial carbon use efficiency by altering plant–microbe–mineral
664
interactions. Glob. Chang. Biol. 28, 4845–4860.
665
https://doi.org/https://doi.org/10.1111/gcb.16229
666
Gaiser, T., Perkons, U., Küpper, P.M., Kautz, T., Uteau-Puschmann, D., Ewert, F., Enders, A., Krauss,
667
G., 2013. Modeling biopore effects on root growth and biomass production on soils with
668
pronounced sub-soil clay accumulation. Ecol. Modell. 256, 6–15.
669
https://doi.org/10.1016/j.ecolmodel.2013.02.016
670
Gentile, R., Vanlauwe, B., Six, J., 2011. Litter quality impacts short- but not long-term soil carbon
671
dynamics in soil aggregate fractions Published by : Ecological Society of America Stable URL :
672
http://www.jstor.org/stable/23021619 REFERENCES Linked references are available on JSTOR
673
for this ar. Ecol. Appl. 21, 695–703.
674
Gerzabek, M.H., PICHLMAYER, F., KIRCHMANN, H., HABERHAUER, G., 1997. The response of soil
675
organic matter to manure amendments in a long-term experiment at Ultuna, Sweden. Eur. J.
676
Soil Sci. 48, 273–282. https://doi.org/https://doi.org/10.1111/j.1365-2389.1997.tb00547.x
677
Gijsman, A.J., Hoogenboom, G., Parton, W.J., Kerridge, P.C., 2002. Modifying DSSAT Crop Models for
678
Low-Input Agricultural Systems Using a Soil Organic Matter–Residue Module from CENTURY.
679
Agron. J. 94, 462–474. https://doi.org/https://doi.org/10.2134/agronj2002.4620
680
Godwin, D.C., Jones, C.A., 1991. Modeling Plant and Soil Systems-Agronomy Monograph no.
681
Page 26
26
Gram, G., Roobroeck, D., Pypers, P., Six, J., Merckx, R., Vanlauwe, B., 2020. Combining organic and
682
mineral fertilizers as a climate-smart integrated soil fertility management practice in sub-
683
Saharan Africa: A meta-analysis, PLoS ONE. https://doi.org/10.1371/journal.pone.0239552
684
Guibert, H., 1999. Évolution de la matière organique et de la capacité d’échange cationique des
685
alfisols tropicaux cultivés. Institut National Polytechnique de Lorraine.
686
He, W., Grant, B.B., Jing, Q., Lemke, R., St. Luce, M., Jiang, R., Qian, B., Campbell, C.A., VanderZaag,
687
A., Zou, G., Smith, W.N., 2021. Measuring and modeling soil carbon sequestration under diverse
688
cropping systems in the semiarid prairies of western Canada. J. Clean. Prod. 328, 129614.
689
https://doi.org/10.1016/j.jclepro.2021.129614
690
Holzworth, D.P., Huth, N.I., deVoil, P.G., Zurcher, E.J., Herrmann, N.I., McLean, G., Chenu, K., van
691
Oosterom, E.J., Snow, V., Murphy, C., Moore, A.D., Brown, H., Whish, J.P.M., Verrall, S., Fainges,
692
J., Bell, L.W., Peake, A.S., Poulton, P.L., Hochman, Z., Thorburn, P.J., Gaydon, D.S., Dalgliesh,
693
N.P., Rodriguez, D., Cox, H., Chapman, S., Doherty, A., Teixeira, E., Sharp, J., Cichota, R., Vogeler,
694
I., Li, F.Y., Wang, E., Hammer, G.L., Robertson, M.J., Dimes, J.P., Whitbread, A.M., Hunt, J., van
695
Rees, H., McClelland, T., Carberry, P.S., Hargreaves, J.N.G., MacLeod, N., McDonald, C.,
696
Harsdorf, J., Wedgwood, S., Keating, B.A., 2014. APSIM - Evolution towards a new generation of
697
agricultural systems simulation. Environ. Model. Softw. 62, 327–350.
698
https://doi.org/10.1016/j.envsoft.2014.07.009
699
Izaurralde, R.C., Williams, J.R., McGill, W.B., Rosenberg, N.J., Jakas, M.C.Q., 2006. Simulating soil C
700
dynamics with EPIC: Model description and testing against long-term data. Ecol. Modell. 192,
701
362–384. https://doi.org/10.1016/j.ecolmodel.2005.07.010
702
Kamali, B., Jahanbakhshi, F., Dogaru, D., Dietrich, J., Nendel, C., Aghakouchak, A., 2022. Probabilistic
703
modeling of crop-yield loss risk under drought: A spatial showcase for sub-Saharan Africa.
704
Environ. Res. Lett. 17. https://doi.org/10.1088/1748-9326/ac4ec1
705
Kamoni, P.T., Gicheru, P.T., Wokabi, S.M., Easter, M., Milne, E., Coleman, K., Falloon, P., Paustian, K.,
706
Killian, K., Kihanda, F.M., 2007. Evaluation of two soil carbon models using two Kenyan long
707
term experimental datasets. Agric. Ecosyst. Environ. 122, 95–104.
708
https://doi.org/10.1016/j.agee.2007.01.011
709
Körschens, M., 2021. Long-Term Field Experiments (LTEs)—Importance, Overview, Soil Organic
710
Matter, in: Exploring and Optimizing Agricultural Landscapes. https://doi.org/10.1007/978-3-
711
030-67448-9_8
712
Kothari, K., Battisti, R., Boote, K.J., Archontoulis, S. V., Confalone, A., Constantin, J., Cuadra, S. V.,
713
Debaeke, P., Faye, B., Grant, B., Hoogenboom, G., Jing, Q., van der Laan, M., Macena da Silva,
714
F.A., Marin, F.R., Nehbandani, A., Nendel, C., Purcell, L.C., Qian, B., Ruane, A.C., Schoving, C.,
715
Silva, E.H.F.M., Smith, W., Soltani, A., Srivastava, A., Vieira, N.A., Slone, S., Salmerón, M., 2022.
716
Are soybean models ready for climate change food impact assessments? Eur. J. Agron. 135.
717
https://doi.org/10.1016/j.eja.2022.126482
718
Kuzyakov, Y., 2002. Review: Factors affecting rhizosphere priming effects. J. Plant Nutr. Soil Sci. 165,
719
382–396. https://doi.org/10.1002/1522-2624(200208)165:4<382::aid-jpln382>3.0.co;2-%23
720
Laub, M., Corbeels, M., Couëdel, A., Ndungu, S.M., Mucheru-muna, M.W., Mugendi, D., Necpalova,
721
M., Waswa, W., 2023a. Managing soil organic carbon in tropical agroecosystems : evidence
722
from four long-term experiments in Kenya. SOIL 301–323.
723
Laub, M., Corbeels, M., Mathu Ndungu, S., Mucheru-Muna, M.W., Mugendi, D., Necpalova, M., Van
724
de Broek, M., Waswa, W., Vanlauwe, B., Six, J., 2023b. Combining manure with mineral N
725
fertilizer maintains maize yields: Evidence from four long-term experiments in Kenya. F. Crop.
726
Page 27
27
Res. 291, 108788. https://doi.org/10.1016/j.fcr.2022.108788
727
Levavasseur, F., Mary, B., Houot, S., 2021. C and N dynamics with repeated organic amendments can
728
be simulated with the STICS model. Nutr. Cycl. Agroecosystems 5.
729
https://doi.org/10.1007/s10705-020-10106-5
730
Li, L., Wang, B., Feng, P., Jägermeyr, J., Asseng, S., Müller, C., Macadam, I., Liu, D.L., Waters, C.,
731
Zhang, Y., He, Q., Shi, Y., Chen, S., Guo, X., Li, Y., He, J., Feng, H., Yang, G., Tian, H., Yu, Q., 2023.
732
The optimization of model ensemble composition and size can enhance the robustness of crop
733
yield projections. Commun. Earth Environ. 4, 1–11. https://doi.org/10.1038/s43247-023-01016-
734
9
735
Li, T., Hasegawa, T., Yin, X., Zhu, Y., Boote, K., Adam, M., Bregaglio, S., Buis, S., Confalonieri, R.,
736
Fumoto, T., Gaydon, D., Marcaida, M., Nakagawa, H., Oriol, P., Ruane, A.C., Ruget, F., Singh, B.,
737
Singh, U., Tang, L., Tao, F., Wilkens, P., Yoshida, H., Zhang, Z., Bouman, B., 2015. Uncertainties in
738
predicting rice yield by current crop models under a wide range of climatic conditions. Glob.
739
Chang. Biol. 21, 1328–1341. https://doi.org/10.1111/gcb.12758
740
Lopez, G., Ahmadi, S.H., Amelung, W., Athmann, M., Ewert, F., Gaiser, T., Gocke, M.I., Kautz, T.,
741
Postma, J., Rachmilevitch, S., Schaaf, G., Schnepf, A., Stoschus, A., Watt, M., Yu, P., Seidel, S.J.,
742
2023. Nutrient deficiency effects on root architecture and root-to-shoot ratio in arable crops.
743
Front. Plant Sci. 13, 1–18. https://doi.org/10.3389/fpls.2022.1067498
744
Ma, J., Rabin, S.S., Anthoni, P., Bayer, A.D., Nyawira, S.S., Olin, S., Xia, L., Arneth, A., 2022. Assessing
745
the impacts of agricultural managements on soil carbon stocks, nitrogen loss and crop
746
production - a modelling study in Eastern Africa. Biogeosciences 2022, 1–31.
747
Mariotte, P., Mehrabi, Z., Bezemer, T.M., De Deyn, G.B., Kulmatiski, A., Drigo, B., Veen, G.F.C., van
748
der Heijden, M.G.A., Kardol, P., 2018. Plant-Soil Feedback: Bridging Natural and Agricultural
749
Sciences. Trends Ecol. Evol. 33, 129–142. https://doi.org/10.1016/j.tree.2017.11.005
750
Martre, P., Wallach, D., Asseng, S., Ewert, F., Jones, J.W., Rötter, R.P., Boote, K.J., Ruane, A.C.,
751
Thorburn, P.J., Cammarano, D., Hatfield, J.L., Rosenzweig, C., Aggarwal, P.K., Angulo, C., Basso,
752
B., Bertuzzi, P., Biernath, C., Brisson, N., Challinor, A.J., Doltra, J., Gayler, S., Goldberg, R., Grant,
753
R.F., Heng, L., Hooker, J., Hunt, L.A., Ingwersen, J., Izaurralde, R.C., Kersebaum, K.C., Müller, C.,
754
Kumar, S.N., Nendel, C., O’leary, G., Olesen, J.E., Osborne, T.M., Palosuo, T., Priesack, E.,
755
Ripoche, D., Semenov, M.A., Shcherbak, I., Steduto, P., Stöckle, C.O., Stratonovitch, P., Streck,
756
T., Supit, I., Tao, F., Travasso, M., Waha, K., White, J.W., Wolf, J., 2015. Multimodel ensembles
757
of wheat growth: Many models are better than one. Glob. Chang. Biol. 21, 911–925.
758
https://doi.org/10.1111/gcb.12768
759
Masvaya, E.N., Nyamangara, J., Descheemaeker, K., Giller, K.E., 2017. Tillage, mulch and fertiliser
760
impacts on soil nitrogen availability and maize production in semi-arid Zimbabwe. Soil Tillage
761
Res. 168, 125–132. https://doi.org/10.1016/j.still.2016.12.007
762
Meurer, K., Barron, J., Chenu, C., Coucheney, E., Fielding, M., Hallett, P., Herrmann, A.M., Keller, T.,
763
Koestel, J., Larsbo, M., Lewan, E., Or, D., Parsons, D., Parvin, N., Taylor, A., Vereecken, H., Jarvis,
764
N., 2020. A framework for modelling soil structure dynamics induced by biological activity. Glob.
765
Chang. Biol. 26, 5382–5403. https://doi.org/10.1111/gcb.15289
766
Meurer, K.H.E., Chenu, C., Coucheney, E., Herrmann, A.M., Keller, T., Kätterer, T., Svensson, D.N.,
767
Jarvis, N., 2020. Modelling dynamic interactions between soil structure and the storage and
768
turnover of soil organic matter. Biogeosciences 17, 5025–5042. https://doi.org/10.5194/bg-17-
769
5025-2020
770
Mirtl, M., T. Borer, E., Djukic, I., Forsius, M., Haubold, H., Hugo, W., Jourdan, J., Lindenmayer, D.,
771
Page 28
28
McDowell, W.H., Muraoka, H., Orenstein, D.E., Pauw, J.C., Peterseil, J., Shibata, H., Wohner, C.,
772
Yu, X., Haase, P., 2018. Genesis, goals and achievements of Long-Term Ecological Research at
773
the global scale: A critical review of ILTER and future directions. Sci. Total Environ. 626, 1439–
774
1462. https://doi.org/10.1016/j.scitotenv.2017.12.001
775
Molina, J.A.E., Clapp, C.E., Shaffer, M.J., Chichester, F.W., Larson, W.E., 1983. NCSOIL, A Model of
776
Nitrogen and Carbon Transformations in Soil: Description, Calibration, and Behavior. Soil Sci.
777
Soc. Am. J. 47, 85–91.
778
https://doi.org/https://doi.org/10.2136/sssaj1983.03615995004700010017x
779
Nendel, C., Berg, M., Kersebaum, K.C., Mirschel, W., Specka, X., Wegehenkel, M., Wenkel, K.O.,
780
Wieland, R., 2011. The MONICA model: Testing predictability for crop growth, soil moisture and
781
nitrogen dynamics. Ecol. Modell. 222, 1614–1625.
782
https://doi.org/10.1016/j.ecolmodel.2011.02.018
783
Nyawira, S.S., Hartman, M.D., Nguyen, T.H., Margenot, A.J., Kihara, J., Paul, B.K., Williams, S., Bolo, P.,
784
Sommer, R., 2021. Simulating soil organic carbon in maize-based systems under improved
785
agronomic management in Western Kenya. Soil Tillage Res. 211, 105000.
786
https://doi.org/10.1016/j.still.2021.105000
787
Ordóñez, R.A., Archontoulis, S. V., Martinez-Feria, R., Hatfield, J.L., Wright, E.E., Castellano, M.J.,
788
2020. Root to shoot and carbon to nitrogen ratios of maize and soybean crops in the US
789
Midwest. Eur. J. Agron. 120, 126130. https://doi.org/10.1016/j.eja.2020.126130
790
Ordóñez, R.A., Castellano, M.J., Danalatos, G.N., Wright, E.E., Hatfield, J.L., Burras, L., Archontoulis, S.
791
V., 2021. Insufficient and excessive N fertilizer input reduces maize root mass across soil types.
792
F. Crop. Res. 267. https://doi.org/10.1016/j.fcr.2021.108142
793
Paustian, K., Parton, W.J., Persson, J., 1992. Modeling Soil Organic Matter in Organic-Amended and
794
Nitrogen-Fertilized Long-Term Plots. Soil Sci. Soc. Am. J. 56, 476–488.
795
https://doi.org/https://doi.org/10.2136/sssaj1992.03615995005600020023x
796
Pernilla Brinkman, E., der Putten, W.H., Bakker, E.-J., Verhoeven, K.J.F., 2010. Plant–soil feedback:
797
experimental approaches, statistical analyses and ecological interpretations. J. Ecol. 98, 1063–
798
1073. https://doi.org/https://doi.org/10.1111/j.1365-2745.2010.01695.x
799
Pichot, J.-P., Al Zahawe, F., Chabalier, P.F., 1977. Evolution d’un sol ferralitique de Côte d’Ivoire après
800
la mise en culture. Effet des apports d’engrais azoté et de compost. Soil Org. matter Stud.
801
Probert, M.E., Dimes, J.P., Keating, B.A., Dalal, R.C., Strong, W.M., 1998. APSIM’s water and nitrogen
802
modules and simulation of the dynamics of water and nitrogen in fallow systems. Agric. Syst.
803
56, 1–28. https://doi.org/10.1016/S0308-521X(97)00028-0
804
R core team, 2023. R: a Language and Environment for Statistical Computing. URL. R Foundation 656
805
for Statistical Computing, Vienna, Austria. https://www.R-project. org/.
806
Rabot, E., Wiesmeier, M., Schlüter, S., Vogel, H.J., 2018. Soil structure as an indicator of soil
807
functions: A review. Geoderma 314, 122–137. https://doi.org/10.1016/j.geoderma.2017.11.009
808
Ricome, A., Affholder, F., Gérard, F., Muller, B., Poeydebat, C., Quirion, P., Sall, M., 2017. Are
809
subsidies to weather-index insurance the best use of public funds? A bio-economic farm model
810
applied to the Senegalese groundnut basin. Agric. Syst. 156, 149–176.
811
https://doi.org/10.1016/j.agsy.2017.05.015
812
Ritchie, J.T., Singh, U., Godwin, D.C., Bowen, W.T., 1998. Cereal growth, development and yield, in:
813
Understanding Options for Agricultural Production. pp. 79–98. https://doi.org/10.1007/978-94-
814
017-3624-4_5
815
Page 29
29
Rodríguez, A., Ruiz-Ramos, M., Palosuo, T., Carter, T.R., Fronzek, S., Lorite, I.J., Ferrise, R., Pirttioja, N.,
816
Bindi, M., Baranowski, P., Buis, S., Cammarano, D., Chen, Y., Dumont, B., Ewert, F., Gaiser, T.,
817
Hlavinka, P., Hoffmann, H., Höhn, J.G., Jurecka, F., Kersebaum, K.C., Krzyszczak, J., Lana, M.,
818
Mechiche-Alami, A., Minet, J., Montesino, M., Nendel, C., Porter, J.R., Ruget, F., Semenov, M.A.,
819
Steinmetz, Z., Stratonovitch, P., Supit, I., Tao, F., Trnka, M., de Wit, A., Rötter, R.P., 2019.
820
Implications of crop model ensemble size and composition for estimates of adaptation effects
821
and agreement of recommendations. Agric. For. Meteorol. 264, 351–362.
822
https://doi.org/10.1016/j.agrformet.2018.09.018
823
Rosenzweig, C., Jones, J.W., Hatfield, J.L., Ruane, A.C., Boote, K.J., Thorburn, P., Antle, J.M., Nelson,
824
G.C., Porter, C., Janssen, S., Asseng, S., Basso, B., Ewert, F., Wallach, D., Baigorria, G., Winter,
825
J.M., 2013. The Agricultural Model Intercomparison and Improvement Project (AgMIP):
826
Protocols and pilot studies. Agric. For. Meteorol. 170, 166–182.
827
https://doi.org/10.1016/j.agrformet.2012.09.011
828
Rusinamhodzi, L., Corbeels, M., Zingore, S., Nyamangara, J., Giller, K.E., 2013. Pushing the envelope?
829
Maize production intensification and the role of cattle manure in recovery of degraded soils in
830
smallholder farming areas of Zimbabwe. F. Crop. Res. 147, 40–53.
831
https://doi.org/10.1016/j.fcr.2013.03.014
832
Sanchez, P.A., 2002. Soil fertility and hunger in Africa. Science (80-. ). 295, 2019–2020.
833
https://doi.org/10.1126/science.1065256
834
Sándor, R., Ehrhardt, F., Grace, P., Recous, S., Smith, P., Snow, V., Soussana, J., Basso, B., Bhatia, A.,
835
Brilli, L., Doltra, J., Dorich, C.D., Doro, L., Fitton, N., Grant, B., Harrison, M.T., Skiba, U.,
836
Kirschbaum, M.U.F., Klumpp, K., Laville, P., Léonard, J., Martin, R., Massad, R.S., Moore, A.,
837
Myrgiotis, V., Rolinski, S., Sharp, J., Smith, W., Wu, L., Zhang, Q., Bellocchi, G., 2023. Residual
838
correlation and ensemble modelling to improve crop and grassland models. Environ. Model.
839
Softw. 105625. https://doi.org/10.1016/j.envsoft.2023.105625
840
Sándor, R., Ehrhardt, F., Grace, P., Recous, S., Smith, P., Snow, V., Soussana, J.F., Basso, B., Bhatia, A.,
841
Brilli, L., Doltra, J., Dorich, C.D., Doro, L., Fitton, N., Grant, B., Harrison, M.T., Kirschbaum,
842
M.U.F., Klumpp, K., Laville, P., Léonard, J., Martin, R., Massad, R.S., Moore, A., Myrgiotis, V.,
843
Pattey, E., Rolinski, S., Sharp, J., Skiba, U., Smith, W., Wu, L., Zhang, Q., Bellocchi, G., 2020.
844
Ensemble modelling of carbon fluxes in grasslands and croplands. F. Crop. Res. 252, 107791.
845
https://doi.org/10.1016/j.fcr.2020.107791
846
Smith-Ramesh, L.M., Reynolds, H.L., 2017. The next frontier of plant–soil feedback research:
847
unraveling context dependence across biotic and abiotic gradients. J. Veg. Sci. 28, 484–494.
848
https://doi.org/https://doi.org/10.1111/jvs.12519
849
Smith, B., Wärlind, D., Arneth, A., Hickler, T., Leadley, P., Siltberg, J., Zaehle, S., 2014. Implications of
850
incorporating N cycling and N limitations on primary production in an individual-based dynamic
851
vegetation model. Biogeosciences 11, 2027–2054. https://doi.org/10.5194/bg-11-2027-2014
852
Smith, P., Smith, J.U., Powlson, D.S., McGill, W.B., Arah, J.R.M., Chertov, O.G., Coleman, K., Franko,
853
U., Frolking, S., Jenkinson, D.S., Jensen, L.S., Kelly, R.H., Klein-Gunnewiek, H., Komarov, A.S., Li,
854
C., Molina, J.A.E., Mueller, T., Parton, W.J., Thornley, J.H.M., Whitmore, A.P., 1997. A
855
comparison of the performance of nine soil organic matter models using datasets from seven
856
long-term experiments. Geoderma 81, 153–225. https://doi.org/10.1016/S0016-
857
7061(97)00087-6
858
Smith, W., Grant, B., Qi, Z., He, W., Qian, B., Jing, Q., VanderZaag, A., Drury, C.F., St. Luce, M.,
859
Wagner-Riddle, C., 2020a. Towards an improved methodology for modelling climate change
860
impacts on cropping systems in cool climates. Sci. Total Environ. 728, 138845.
861
Page 30
30
https://doi.org/10.1016/j.scitotenv.2020.138845
862
Smith, W., Grant, B., Qi, Z., He, W., VanderZaag, A., Drury, C.F., Helmers, M., 2020b. Development of
863
the DNDC model to improve soil hydrology and incorporate mechanistic tile drainage: A
864
comparative analysis with RZWQM2. Environ. Model. Softw. 123, 104577.
865
https://doi.org/10.1016/j.envsoft.2019.104577
866
Smith, W.N., Grant, B.B., Campbell, C.A., McConkey, B.G., Desjardins, R.L., Kröbel, R., Malhi, S.S.,
867
2012. Crop residue removal effects on soil carbon: Measured and inter-model comparisons.
868
Agric. Ecosyst. Environ. 161, 27–38. https://doi.org/10.1016/j.agee.2012.07.024
869
Van Ittersum, M.K., Van Bussel, L.G.J., Wolf, J., Grassini, P., Van Wart, J., Guilpart, N., Claessens, L., De
870
Groot, H., Wiebe, K., Mason-D’Croz, D., Yang, H., Boogaard, H., Van Oort, P.A.J., Van Loon, M.P.,
871
Saito, K., Adimo, O., Adjei-Nsiah, S., Agali, A., Bala, A., Chikowo, R., Kaizzi, K., Kouressy, M.,
872
Makoi, J.H.J.R., Ouattara, K., Tesfaye, K., Cassman, K.G., 2016. Can sub-Saharan Africa feed
873
itself? Proc. Natl. Acad. Sci. U. S. A. 113, 14964–14969.
874
https://doi.org/10.1073/pnas.1610359113
875
Vanlauwe, B., Bationo, A., Chianu, J., Giller, K.E., Merckx, R., Mokwunye, U., Ohiokpehai, O., Pypers,
876
P., Tabo, R., Shepherd, K.D., Smaling, E.M.A., Woomer, P.L., Sanginga, N., 2010. Integrated soil
877
fertility management: Operational definition and consequences for implementation and
878
dissemination. Outlook Agric. 39, 17–24. https://doi.org/10.5367/000000010791169998
879
Vanlauwe, B., Descheemaeker, K., Giller, K.E., Huising, J., Merckx, R., Nziguheba, G., Wendt, J.,
880
Zingore, S., 2015a. Integrated soil fertility management in sub-Saharan Africa: Unravelling local
881
adaptation. Soil 1, 491–508. https://doi.org/10.5194/soil-1-491-2015
882
Vanlauwe, B., Giller, K.E., 2006. Popular myths around soil fertility management in sub-Saharan
883
Africa. Agric. Ecosyst. Environ. 116, 34–46. https://doi.org/10.1016/j.agee.2006.03.016
884
Vanlauwe, B., Six, J., Sanginga, N., Adesina, A.A., 2015b. Soil fertility decline at the base of rural
885
poverty in sub-Saharan Africa. Nat. Plants 1, 15101. https://doi.org/10.1038/nplants.2015.101
886
Wallach, D., Martre, P., Liu, B., Asseng, S., Ewert, F., Thorburn, P.J., van Ittersum, M., Aggarwal, P.K.,
887
Ahmed, M., Basso, B., Biernath, C., Cammarano, D., Challinor, A.J., De Sanctis, G., Dumont, B.,
888
Eyshi Rezaei, E., Fereres, E., Fitzgerald, G.J., Gao, Y., Garcia-Vila, M., Gayler, S., Girousse, C.,
889
Hoogenboom, G., Horan, H., Izaurralde, R.C., Jones, C.D., Kassie, B.T., Kersebaum, K.C., Klein, C.,
890
Koehler, A.K., Maiorano, A., Minoli, S., Müller, C., Naresh Kumar, S., Nendel, C., O’Leary, G.J.,
891
Palosuo, T., Priesack, E., Ripoche, D., Rötter, R.P., Semenov, M.A., Stöckle, C., Stratonovitch, P.,
892
Streck, T., Supit, I., Tao, F., Wolf, J., Zhang, Z., 2018. Multimodel ensembles improve predictions
893
of crop–environment–management interactions. Glob. Chang. Biol. 24, 5072–5083.
894
https://doi.org/10.1111/gcb.14411
895
Wetselaar, R., Ganry, F., 1982. Nitrogen balance in tropical agrosystems, in: Dommergues, Y.R., Diem,
896
H.G. (Eds.), Microbiology of Tropical Soils and Plant Productivity. Springer Netherlands,
897
Dordrecht, pp. 1–36. https://doi.org/10.1007/978-94-009-7529-3_1
898
Whitbread, A., Blair, G., Konboon, Y., Lefroy, R., Naklang, K., 2003. Managing crop residues, fertilizers
899
and leaf litters to improve soil C, nutrient balances, and the grain yield of rice and wheat
900
cropping systems in Thailand and Australia. Agric. Ecosyst. Environ. 100, 251–263.
901
https://doi.org/10.1016/S0167-8809(03)00189-0
902
Zhou, L., Zhou, X., Zhang, B., Lu, M., Luo, Y., Liu, L., Li, B., 2014. Different responses of soil respiration
903
and its components to nitrogen addition among biomes: a meta-analysis. Glob. Chang. Biol. 20,
904
2332–2343. https://doi.org/https://doi.org/10.1111/gcb.12490
905
Page 31
31
Zingore, S., Murwira, H.K., Delve, R.J., Giller, K.E., 2007. Influence of nutrient management strategies
906
on variability of soil fertility, crop yields and nutrient balances on smallholder farms in
907
Zimbabwe. Agric. Ecosyst. Environ. 119, 112–126. https://doi.org/10.1016/j.agee.2006.06.019
908
Page 32
32
Tables and Figures
Table 1. Characteristics of the long-term experiments and the observed data available for model calibration
Site
General
information
Country
Côte d'Ivoire
Kenya
Kenya
Zimbabwe
Location
Gagnoa
Embu
Machanga
Murewa
Latitude
6.13
-0.52
-0.79
-17.81
Longitude
-5.93
37.46
37.66
31.56
Elevation (m)
213
1380
1022
1262
Climate
Average annual rainfall
(mm)*
1402
1175
795
898
Average annual temperature
(°C)*
25.5
20.1
23.7
20.9
FAO agro-ecological zone
warm-humid
cool sub-humid
warm semi-arid
cool-semi-arid
Rainfall pattern
bimodal
bimodal
bimodal
unimodal
Soil
Type (FAO)
Lixisols
Humic Nitisols
Ferric Alisols
Lixisols
Texture
sandy clay loam
clay
sandy clay loam
sandy
Initial SOC (%) (0-20cm)
1.1
2.9
0.3
0.5
Initial SON (%) (0-20cm)
0.11
0.3
0.02
0.04
Management
Experiment duration (years)
13
18
10
9
First crop cycle
maize
maize
maize
maize
Second crop cycle
maize or soybean or fallow maize
maize
none
Maize cultivar
hybrid (medium maturity,
variety H507 and IRAT83)
hybrid (early maturity
drought tolerant, variety
H513)
hybrid (early maturity
drought tolerant, variety
Katumani)
hybrid (medium maturity,
drought tolerant, variety
SC525)
Mineral P and K fertilization
(kg/ha/cycle)
100 P + 150 K
60 P + 60 K
60 P + 60 K
/
Treatments
1- No N input
0 N
0 N
0 N
0 N
Page 33
33
2- Mineral N fertilizer
(kg/ha/cycle)
160 N
120 N
120 N
100 N + 30 P
3- Organic fertilizer (t
C/ha/year)
3 t C/ha compost of crop
residues
4 t C/ha manure
4 t C/ha manure
none
4- Mineral N fertilizer
(kg/ha/cycle) + organic
fertilizer (t C/ha/year)
160 N + 100 P + 150 K +
3 t C/ha compost of crop
residues
120 N + 60 P + 60 K+ 4 t
C/ha manure
120 N + 60 P + 60 K + 4 t
C/ha manure
100 N + 3 t C/ha manure
Organic fertilizer composition 29.3% C, 2.0% N, 1.0% P,
5.9% K
28.4% C, 2.4% N, 0.4% P,
3.6% K
28.4% C, 2.4% N, 0.4% P,
3.6% K
20% C, 1.1% N, 0.2% P,
0,6% K
Available data
(blind
calibration
phase)
Emergence date
for 8 crop cycles
3 crop cycles
3 crop cycles
none
Anthesis date
for 19 crop cycles
3 crop cycles
3 crop cycles
none
Physiological maturity date
none
3 crop cycles
3 crop cycles
none
Harvest date
all crop cycles
all crop cycles
all crop cycles
all crop cycles
Available data
(full calibration
phase)
Harvested grain yield (dry
matter)
all crop cycles
all crop cycles
all crop cycles
all crop cycles
Total aboveground biomass
(dry matter)
all crop cycles
none
all crop cycles
all crop cycles
Soil organic C and N (from 4
to 8 points depending on sites)
5 years
13 years
5 years
4 years
Soil mineral N (one point
before sowing)
3 years
none
none
none
* Means calculated over the length of the long-term experiment
Page 34
34
Table 2. The 16 soil-crop process-based models used in the model intercomparison study.
Model
Abbreviation Total C
pools
Residue
pools
Microbial
pools
SOM pools
Soil sub-model Crop sub-model
Model reference
APSIM
AP
3
1
1
1
SoilN
APSIM
Holzworth et al. (2014)
CELCIUS
CE
2
1
0
1
CELCIUS
CELCIUS
Ricome et al. (2017)
DayCent
DC
5
2
0
3
CENTURY
DayCent
Del Grosso et al. (2001)
DNDC
DN
8
3
2
3
DNDC
DNDC
Smith et al. (2020)
DSSAT-CERES-Maize + Century
DS1
5
2
0
3
CENTURY
CERES maize
Ritchie et al. (1998); Gijsman et al.
(2002)
DSSAT-CERES-Maize + Ceres-
SOM
DS2
2
1
0
1
CERES
CERES maize
Godwin and Jones (1991)
EPIC
EI
5
2
0
3
EPIC
EPIC
Izaurralde et al. (2006)
Expert-N v5.1-Gecros
EG
3
1
1
1
SoilN
similar to SUCROS
Biernath et al. (2011)
Expert-N v5.1-Spass
ES
3
1
1
1
SoilN
SPASS
Biernath et al. (2011)
Expert-N v5.1-Ceres
EC
3
1
1
1
SoilN
CERES maize
Biernath et al. (2011)
InfoCrop
IN
3
3
1
1
InfoCrop
similar to SUCROS
Aggarwal et al. (2006)
MONICA v3.3.1
MO
6
2
2
2
DAISY
HERMES/AGROSIM Nendel et al. (2011)
SALUS
SA
5
2
0
3
CENTURY
SALUS
Basso et al. (2010)
SIMPLACE-Lintul + Option 1
SI1
7
4
1
2
SoilCN
Lintul
Enders et al. (2023); Gaiser et al.
(2013)
SIMPLACE-Lintul + Option 2
SI2
7
4
1
2
SoilCN
Lintul
Faye et al. (2018)
STICS v10
ST
4
1
1
2
STICS
STICS
Brisson et al. (2003); Beaudoin et al.
(2023)
Page 35
35
Figure 1. Observed and simulated average maize yield (t dry matter ha
-1
) across all years of the long-
term experiments (A) and soil organic carbon (SOC, t C ha
-1
in 0-15 cm soil depth in Kenya-Embu and
Kenya-Machanga and 0-20 cm soil depth in Zimbabwe-Murewa and Côte d’Ivoire-Gagnoa) at the end
of the long-term experiment (B) using blind and full model calibration, at four sites in sub-Saharan
Africa, for four N fertilizer and organic amendment treatments. The simulations are the median of an
ensemble of 16 soil-crop models. Treatments are without inputs (-N), with mineral N fertilizer inputs
only (-C+N), with organic amendments only (+C-N) and with combined organic and mineral N
fertilizer inputs (+C+N). nRMSE is the normalized Root Mean Square Error while EF is the Model
Efficiency.
Page 36
36
Figure 2. Observed (red dots) and simulated maize yield (t dry matter ha
-1
) using 16 fully calibrated models, in four sites in sub-Saharan Africa, for four N
fertilizer and organic treatments. Black solid lines are medians of the simulations. Dark gray areas indicate the 25th to 75th percentile range of the values
simulated. Treatments are without inputs (-C-N), with mineral N fertilizer inputs only (-C+N), with organic amendments only (+C-N) and with combined
organic and mineral N fertilizer inputs (+C+N).
Page 37
37
Figure 3 Observed (red dots) and simulated relative change in soil organic carbon (SOC) (in 0-15 cm soil depth in Kenya-Embu and Kenya-Machanga and 0-
20cm soil depth in Zimbabwe-Murewa and Côte d’Ivoire-Gagnoa) using 16 fully calibrated models, in four sites in sub-Saharan Africa, for four N fertilizer
and organic treatments. Black solid lines are medians of the simulations. Dark gray areas indicate the 25th to 75th percentile range of the values simulated.
Treatments are without inputs (-C-N), with mineral N fertilizer inputs only (-C+N), with organic amendments only (+C-N) and with combined organic and
mineral N fertilizer inputs (+C+N).
Page 38
38
Figure 4. Normalized root mean square error (nRMSE) for maize grain yield and soil organic carbon
(SOC) simulations by individual models and model ensemble mean (e.mean) and median (e.median)
for blind and full calibration. Model abbreviations: AP (APSIM), CE (CELCIUS), DS1 (DSSAT-
Century), DS2 (DSSAT-Ceres), DC (DayCent), DN (DNDC), EC (Expert-N-Ceres), EG (Expert-N-
Gecros), EI (EPIC), ES (Expert-N-Spass), IN (InfoCrop), MO (MONICA), SI1 (SIMPLACE-1), SI2
(SIMPLACE-2), SA (SALUS), ST (STICS).
Page 39
39
1
2
Figure 5. Root means square error (RMSE) and coefficient of variation (CV) for maize grain yield (A, B) and soil organic carbon (SOC) (C, D) in each
3
experimental year for fully calibrated models, in four sites in sub-Saharan Africa, across all treatments. RMSE was computed from the median of an ensemble
4
of 16 soil-crop models while CV was based on the 16 models’ simulations.
5
6
Page 40
40
7
Figure 6. Root means square error (RMSE) and coefficient of variation (CV) for maize grain yield (A, B) and soil organic carbon (SOC) (C, D) in each
8
experimental year for fully calibrated models, in four sites in sub-Saharan Africa. Treatments are without inputs (-C-N), with mineral N fertilizer inputs only (-
9
C+N), with organic amendments only (+C-N) and with combined organic and mineral N fertilizer inputs (+C+N).RMSE was computed from the median of an
10
ensemble of 16 soil-crop models while coefficient of variation was computed from the 16 models individual simulations.
11
Page 41
41
12
Figure 7. Simulated (dots) and observed (triangle) yearly change in maize grain yield and soil organic carbon (SOC) over the duration of long-term
13
experiments in four sites in sub-Saharan Africa for no input (-C-N) treatment. SOC was measured in 0-15 cm soil depth in Kenya-Embu and Kenya-Machanga
14
and in 0-20 cm soil depth in Zimbabwe-Murewa and Côte d’Ivoire-Gagnoa. Blue symbols indicate when both changes in yield and SOC are significantly (p
≤
15
0.05) different from 0. Red symbols indicate that change in yield or change in SOC was not significantly (P > 0.5) different from 0. Model abbreviations: AP
16
(APSIM), CE (CELCIUS), DS1 (DSSAT-Century), DS2 (DSSAT-Ceres), DC (DayCent), DN (DNDC), EC (Expert-N-Ceres), EG (Expert-N-Gecros), EI
17
(EPIC), ES (Expert-N-Spass), IN (InfoCrop), MO (MONICA), SI1 (SIMPLACE-1), SI2 (SIMPLACE-2), SA (SALUS), ST (STICS).
18
Page 42
42
19
20
Figure 8. Simulated soil organic carbon (SOC) and annual ammount of mineralized nitrogen (N) by individual models in four sites in sub-Saharan Africa for
21
no input treatments only. Each point corresponds to one year of simulation. Only significant regression lines at p<0.05 are displayed. Model abbreviations: AP
22
(APSIM), CE (CELCIUS), DS1 (DSSAT-Century), DN (DNDC), EC (Expert-N-Ceres), EG (Expert-N-Gecros), ES (Expert-N-Spass), IN (InfoCrop), MO
23
(MONICA), SI1 (SIMPLACE-1), SI2 (SIMPLACE-2), SA (SALUS), ST (STICS). 12 of the 16 models provided N mineralized.
24
Page 43
43
25
Figure 9. Simulated change in soil orgainc carbon (SOC) and total maize root biomass produced during each crop cycle (t dry matter ha-1) over the length of
26
the experiment in four sites in sub-Saharan Africa for treatments with no input (-C-N) and with mineral N fertilizer (-C+N). Model abbreviations: AP
27
(APSIM), CE (CELCIUS), DS1 (DSSAT-Century), DS2 (DSSAT-Ceres), DC (DayCent), DN (DNDC), EC (Expert-N-Ceres), EG (Expert-N-Gecros), EI
28
(EPIC), ES (Expert-N-Spass), IN (InfoCrop), MO (MONICA), SI1 (SIMPLACE-1), SI2 (SIMPLACE-2), SA (SALUS), ST (STICS).
29