SENSITIVITY ANALYSIS APPLIED TO A COMPLEX CROP MODEL FOR
DURUM WHEAT IN THE MEDITERRANEAN
1
G.M. Richter1, M. Acutis2, P. Trevisiol2, K. Latiri3, R. Confalonieri4
Rothamsted Research, UK; 2University of Milan, Italy, 3INRA-T, Tunisia, 4Joint Research Centre, Italy
goetz.richter@bbsrc.ac.uk
Complex mechanistic models for crop growth have been widely used to assess the sustainability of agricultural
systems and the risk and impact of climatic change. Usually, these models are calibrated for site specific data,
and may not apply in other circumstances due to the implicit empiricism. We carried out a global sensitivity
analysis (SA) to rank the parameters of the STAMINA crop model in terms of their importance. STAMINA is a
generic but complex cropping system model, which simulates agro-meteorology, hydrology, crop development
and photosynthesis in hilly terrain [1] [2]. As model sensitivity depends on parameter variation and the
environment [3] we were interested in the relative effects of the environmental drivers (soil and climate), when
applying the SA to the rain-fed production of Triticum durum (Durum wheat, DW) in the Southern
Mediterranean. Our main objective here was to analyze the effect of the environment, namely the effect of water
availability (storage in the soil; rainfall) and differences in the a priori parameter distribution.
Materials and methods
The model sensitivity was determined for DW, (1) on a silty clay-loam in Volturino (Apulia Region, Italy), and
(2) on a drought-prone sandy soil in Nabeul (Tunisia). The analysis used single event data (Italy, 2002-03;
Tunisia, 1990/91) for separately calibrated parameter sets. We applied the Morris method [4], which is
parsimonious in terms of number of model runs required and suitable for the STAMINA model due to the large
number of parameters and complexity. The STAMINA software implements the WARMSIMLAB library
analyzing the sensitivity of crop parameters and other inputs. For the Italian site, a maximum of 57 crop
parameters were selected, characterised by mean values and standard deviation according to calibration,
assuming a normal distribution. For the Tunisian site we reduced the number of parameters to 32. A maximum of
464 simulations were executed as a result of 8 trajectories and 4 levels. μ* and σ are the mean and standard
deviation of the change in yield per change of parameter.
Results
In Volturino, 5 out of 57 parameters were not important for yield formation, in Nabeul 3 out of 32 crop
parameters were irrelevant. For both sites, the range of μ* and σ spanned over several orders of magnitude and
we display the results on a log-log scale. Parameters were grouped according to their role in the crop simulation:
establishment, development, photosynthesis, partitioning, water stress and senescence. In Volturino, parameters
which determine flowering date (Tbase, GDD), leaf area increase (Tbase, LAI0) and early allocation to the shoot
were highly sensitive. Parameters for establishment ranked very highly overall, directly or indirectly (Figure
1a). In Nabeul, among the top 15 ranks of sensitivity, four parameters were related to early growth (Figure 1b),
four to plant development, and two each to photosynthesis and senescence (start, rate of wilting). Management
(e.g. sowing date) can play a crucial role in mitigating the establishment characteristics of crops.
(a)
10000
1000
Sowing date
1000
100
σ
σ
(b)
10000
10
establish
development
assimilate
partition
senescence
water stress
1
100
10
1
1
10
100
μ*
1000
10000
1
10
100
μ*
1000
10000
Figure 1 Graphical display of Morris sensitivity measures (μ*, σ) for crop model input parameters applied to
durum wheat experiments in (a) Italy (Volturino, 2003), and (b) Tunisia (Nabeul; 1991)
The parameters on which CO2 assimilation depends were highly important in Nabeul (μ* >3000 for assimilation
rate at light saturation, ~700 for initial light use efficiency), much less in Volturino (320). Both analyses revealed
that the parameters for the temperature response curve of photosynthesis were less important. The water stress
parameters ranked similar sensitivity (9 at Volturino; 12 at Nabeul), stomatal resistance ranked 18 at Nabeul.
Averaging the parameter groups the overall ranking (Table 1) according to mean sensitivity (μ*) show that
parameters describing early establishment (initial LAI, temperature factor for GLAI) had the greatest impact on
yield of DW grown on a sandy soil in Tunisia, Overall, parameters related to crop establishment played a key
role in the Mediterranean environment but in Volturino these parameters were less important than at Nabeul,
possibly due to difference in soil. The sensitivity to parameters of dry matter partition is more important at
Volturino than at Nabeul, however, some of these parameters were related to early leaf development.
Table 1: Ranking of parameter groups according to mean Morris sensitivity (μ*) in relevance for
yield formation of Durum wheat (DW) in Tunisia and Italy
Group
Establishment
Development
Photosynthesis
Water stress
Senescence
Partitioning
Italy, Silty Clay Loam
Average μ* Average σ
677
1035
408
292
113
95
755
488
336
264
873
425
Tunisia, Loamy Sand
Average μ* Average σ
1501
979
1063
1121
591
380
380
415
362
326
138
175
Discussion
The SA for our crop growth model emphasises the importance of early establishment for DW, which confirms
other findings in the Mediterranean for wheat [5] and other crops [6]. It further highlights the importance of
correctly simulating photosynthesis and crop development, however. The individual and the overall ranking of
the parameter sensitivity can be different depending on the environment. At both sites, the time span of grain
filling is critically determined by flowering date and the speed of maturation. The latter is determined by the
onset of senescence and the senescence rate, in which the results of the SA agree very well for both
environments, confirming that the duration of the late growth stages is important in Mediterranean conditions [7]
either by management or variety selection. It is surprising that in both environments (soil in Volturino holds
more water) the sensitivity to changes in the water stress parameter was similar.
Further work will be presented that uses the same parameter sets in different environments and alters the soil and
weather inputs within the range of variation typical for the region. More analysis is also needed on the artefacts
possible in the SA related to (a) the variation of the parameters that are too small in relation to the mean value;
or (b) parameters that are missed out because of a priori neglecting of a process.. Finally, parameter selection
should avoid the model becoming over-parameterised and in the SA some parameters could be omitted as they
are complementary, intimately linked, like phenology and partitioning, or sequential (growth stages).
References
[1] Acutis M, Rana G, Trevisiol P, Bechini L, Laudato M, Ferrara RM, Richter GM: Integrating a spatial
micrometeorological model into the risk assessment for arable crops in hilly terrain. In: Modelling water and
nutrient dynamics in soil-crop systems. Eds: KC Kersebaum, et al.; Springer Verlag. In print. 2007
[2] Richter, GM et al.: Stability and Mitigation of Arable Systems in Hilly Landscapes. QLK5-CT-2002-01313,
Final Report to the European Commission, Brussels, 2006
[3] Francos A, Elorzab FJ et al.: Sensitivity analysis of distributed environmental simulation models:
understanding the model behaviour in hydrological studies at the catchment scale. Reliability Engineering
System Safety 79, 205–218 (2003)
[4] Saltelli A, Tarantola S, Campolongo F, Ratto M, Sensitivity Analysis in practice. A Guide …. Wiley, 2004.
[5] Botwright TL, Condon AG, Rebetzke GJ, Richards RA: Field evaluation of early vigour for genetic
improvement of grain yield in wheat. Austral J Agric Res 53(10), 1137-1145 (2002)
[6] Aguera F., Villalobos FJ, Orgaz F: Evaluation of sunflower (Helianthus annuus, L.) genotypes differing in
early vigour using a simulation model. Eur J Agronomy 7(1-3), 109-118 (1997)
[7] Hafsi M, Pfeiffer WH, Monneveux P: Flag leaf senescence, carbon content and carbon isotope
discrimination in durum wheat grown under semi-arid conditions. Cereal Res. Comm., 31(1-2), 161-168
(2003)