Sensitivity analysis of dynamic crop models to assist crop science: assessing the impact of multiple traits on yield in Australian wheat.

Size: px

Start display at page:

Download "Sensitivity analysis of dynamic crop models to assist crop science: assessing the impact of multiple traits on yield in Australian wheat."

Esther McKenzie
5 years ago
Views:

Pierre Casadebaig (1), Robert Faivre (2), Karine Chenu (3) 1.

Mathématiques et Informatique Appliquées de Toulouse, INRA, Toulouse

1 Sensitivity analysis of dynamic crop models to assist crop science: assessing the impact of multiple traits on yield in Australian wheat. Pierre Casadebaig (1), Robert Faivre (2), Karine Chenu (3) 1. Agroecologies, Innovation, Ruralités, INRA, Toulouse 2. Mathématiques et Informatique Appliquées de Toulouse, INRA, Toulouse 3. Queensland Alliance for Agriculture and Food Innovation, University of Queensland, Australia

2 Context Screen Search Conclusions References Overview Approach Systems approach to crop improvement [1, 2] Taking advantage of genotypes x environment x management (GEM) interactions Canopy viewed as a system of cultivar (G), pedo-climatic factors (E) and management practices (M). Problem: improving crop yield with uncertain climate agronomy: improve management practices (choose G+M E) genetics: improve genetic material (change G E+M) Tools experiments: hard to sample climatic variability models: genotypic determinism in crop models Pierre Casadebaig 3 ieme Rencontres R 27/06/ / 14

3 Context Screen Search Conclusions References Overview Approach Systems approach to crop improvement A methodology for screening crop model parameters usable as plant traits for breeding 1. Model: APSIM platform [3] dissect complex traits 2. Explore screen: sensitivity analysis factors and distribution numerical design simulation and index computation search: variance analysis parameter environment 3. Optimization parameter combination Canopy Plant Environment Climate Soil Management Morphology Complex traits f(t, P, E, θ) Phenology Plant traits Dynamic model Genotype Response Pierre Casadebaig 3 ieme Rencontres R 27/06/ / 14

4 Context Screen Search Conclusions References Method Results Defining input factors How many parameters in a typical crop model? APSIM-wheat model 500 parameters plant-related parameters 103 independant 62 values 41 functions 90 studied after grouping number group design screen search total independant studied impacting candidate Pierre Casadebaig 3 ieme Rencontres R 27/06/ / 14

5 Context Screen Search Conclusions References Method Results Setting the variation range How to keep biological meaning without going into 90 special cases? Consensus 40% variation range 3 rules: 1. scale single value 2. scale y vector 3. scale point in x or y vector leaf expansion (potential) leaf nitrogen demand crit 0.02 default max min photosynthesis (potential) photosynthesis (temperature) 1.00 level upper 0.75 default lower factor_group Pierre Casadebaig 3 ieme Rencontres R 27/06/ / 14

Context Screen Search Conclusions References Method Results Sampling method for parameter and environmental space How to sample GEM landscape for wheat in Australia?

6 Context Screen Search Conclusions References Method Results Sampling method for parameter and environmental space How to sample GEM landscape for wheat in Australia? Genotypes (Parameters) n = 9100 Morris method [4] (sensitivity) 6 levels by factor 100 random rep. of OAT designs Environments n = CO 2 levels 3 sowing dates 3 N fertilization amount 4 locations 125 years of climatic data Pierre Casadebaig 3 ieme Rencontres R 27/06/ / 14

7 Context Screen Search Conclusions References Method Results Simulation and sensitivity index computation Simulation and output variables 8 output variables distributed computing in CSIRO ( simulations) R packages ncdf4, dplyr Sensitivity indexes main (µ ): estimation of the linear effects of inputs interaction (σ): estimation of the non-linear/interaction effects Pierre Casadebaig 3 ieme Rencontres R 27/06/ / 14

8 Context Screen Search Conclusions References Method Results Screening for impactful parameters About a half of the parameters are not or weakly impacting yield in control conditions null low high impact on Grain Yield (t.ha 1) index main interaction x_ws_root x_weighted_temp x_sw_avail_ratio tt_emerg_limit tfac_slope temp_fac_min sw_fac_max swdf_photo_limit swdf_pheno_limit stem_n_sen_conc sfac_slope pod_n_sen_conc pod_n_init_conc N_fact_pheno n_conc_meal n_conc_grain meal_n_sen_conc leaf_no_crit grn_water_cont fasw_emerg days_germ_limit min_tpla y_rel_root_rate sen_light_slope lai_sen_light total_n_uptake_max x_sw_demand_ratio y_height stem_dm_init root_n_init_conc root_dm_init leaf_dm_init stem_n_init_conc x_sw_ratio specific_root_length node_no_correction x_temp leaf_n_init_conc x_ave_temp root_n_sen_conc y_extinct_coef_dead initial_root_depth n_conc_pod y_leaf_size y_dm_sen_frac_root co2_rue_modifier y_leaf_no_frac y_leaves_per_node y_node_app_rate N_fact_expansion sen_rate_water y_co2_te_modifier vern_sens kno3 x_maxt_senescence potential_grain_growth_rate leaf_n_sen_conc N_fact_photo tt_flowering x_temp_root_advance root_depth_rate initial_tpla kl_modifier n_conc_root xf_modifier N_fact_grain y_ratio_root_shoot n_conc_leaf shoot_lag eo_crop_factor_default fr_lf_sen_rate max_grain_size shoot_rate node_sen_rate n_conc_stem y_frac_pod tt_start_grain_fill x_temp_grain_n_fill grains_per_gram_stem y_frac_leaf y_sla tt_floral_initiation transp_eff_cf x_temp_grainfill potential_grain_filling_rate y_extinct_coef y_rue photop_sens tt_end_of_juvenile ll_modifier Pierre Casadebaig 3 ieme Rencontres R 27/06/ / 14

9 Context Screen Search Conclusions References Method Results Screening for impactful parameters Parameters mainly impact specific output variables in control conditions flow_das maturity_das grain_no grain_size grain_protein yield lai_flow biomass N_fact_grain max_grain_size tt_flowering potential_grain_growth_rate leaf_n_sen_conc sen_rate_water vern_sens y_node_app_rate N_fact_expansion kno3 y_co2_te_modifier x_maxt_senescence tt_start_grain_fill eo_crop_factor_default n_conc_root y_ratio_root_shoot root_depth_rate x_temp_root_advance initial_tpla N_fact_photo kl_modifier shoot_lag shoot_rate y_frac_pod xf_modifier n_conc_leaf n_conc_stem tt_floral_initiation photop_sens tt_end_of_juvenile y_rue ll_modifier grains_per_gram_stem y_sla y_extinct_coef transp_eff_cf y_frac_leaf fr_lf_sen_rate node_sen_rate x_temp_grain_n_fill x_temp_grainfill potential_grain_filling_rate value Pierre Casadebaig 3 ieme Rencontres R 27/06/ / 14

10 Context Screen Search Conclusions References Results Searching for candidate parameters The impact of traits is strongly affected by environment and management Yield sensitivity index (t/ha) nitrogen high control low nitrogen 0 Yield sensitivity index (t/ha) sites 0 sen_rate_water y_node_app_rate N_fact_expansion kno3 potential_grain_growth_rate leaf_n_sen_conc x_temp_root_advance tt_flowering root_depth_rate N_fact_photo n_conc_root initial_tpla y_co2_te_modifier vern_sens N_fact_grain kl_modifier y_ratio_root_shoot xf_modifier eo_crop_factor_default y_frac_pod n_conc_leaf x_maxt_senescence shoot_lag shoot_rate tt_start_grain_fill n_conc_stem fr_lf_sen_rate max_grain_size node_sen_rate x_temp_grain_n_fill grains_per_gram_stem transp_eff_cf y_frac_leaf y_sla tt_floral_initiation x_temp_grainfill potential_grain_filling_rate y_extinct_coef photop_sens y_rue tt_end_of_juvenile ll_modifier sites Emerald Merredin Narrabri Yanco Pierre Casadebaig 3 ieme Rencontres R 27/06/ / 14

11 Context Screen Search Conclusions References Results Searching for candidate parameters A subset of the most impactful parameters is less dependant to environmental effect variable co2 sites nitrogen sowing Residuals 0.00 ll_modifier tt_end_of_juvenile photop_sens tt_floral_initiation transp_eff_cf grains_per_gram_stem node_sen_rate max_grain_size fr_lf_sen_rate tt_start_grain_fill shoot_rate shoot_lag x_maxt_senescence y_frac_pod xf_modifier kl_modifier root_depth_rate x_temp_root_advance sen_rate_water potential_grain_filling_rate x_temp_grainfill x_temp_grain_n_fill n_conc_stem n_conc_leaf y_ratio_root_shoot N_fact_grain n_conc_root N_fact_photo tt_flowering leaf_n_sen_conc potential_grain_growth_rate kno3 N_fact_expansion y_node_app_rate y_rue y_extinct_coef y_sla y_frac_leaf eo_crop_factor_default vern_sens initial_tpla y_co2_te_modifier Pierre Casadebaig 3 ieme Rencontres R 27/06/ / 14

12 Context Screen Search Conclusions References Results Searching for candidate parameters The impact of parameters is linked to ressource availability ressource acquisition (ll_modifier) ressource use (y_rue) Standardized impact 2 1 Standardized impact 1.0 type ET1 ET2 ET3 ET Water stress index Nitrogen stress index Pierre Casadebaig 3 ieme Rencontres R 27/06/ / 14

13 Context Screen Search Conclusions References Conclusions Numerical exploration low-impact parameters are targets for code refactoring methods to integrate G E interactions in breeding approaches UseR! numerical design, decoupling, index computation: sensitivity [5] data maniplation and visualization: reshape2 [6], dplyr [7], ggplot2 [8] reproducible code (markdown, knitr [9], github repository) Pierre Casadebaig 3 ieme Rencontres R 27/06/ / 14

14 Context Screen Search Conclusions References References [1] Graeme Hammer, Mark Cooper, Francois Tardieu, Stephen Welch, Bruce Walsh, Fred van Eeuwijk, Scott Chapman, and Dean Podlich. Models for navigating biological complexity in breeding improved crop plants. Trends in Plant Science, 11(12): , December [2] Xinyou Yin, Paul C. Struik, and Martin J. Kropff. Role of crop physiology in predicting gene-to-phenotype relationships. Trends in Plant Science, 9(9): , September [3] B. A. Keating, P. S. Carberry, G. L. Hammer, M. E. Probert, M. J. Robertson, D. Holzworth, N. I. Huth, J. N. G. Hargreaves, H. Meinke, and Z. Hochman. An overview of APSIM, a model designed for farming systems simulation. European Journal of Agronomy, 18(3-4): , [4] Max D Morris. Factorial sampling plans for preliminary computational experiments. Technometrics, 33(2): , [5] Gilles Pujol, Bertrand Iooss, and Alexandre Janon. sensitivity: Sensitivity Analysis, URL R package version [6] Hadley Wickham. Reshaping data with the reshape package. Journal of Statistical Software, 21(12):1 20, URL [7] Hadley Wickham and Romain Francois. dplyr: a grammar of data manipulation, URL R package version 0.2. [8] Hadley Wickham. ggplot2: elegant graphics for data analysis. Springer New York, ISBN URL [9] Yihui Xie. knitr: A comprehensive tool for reproducible research in R. In Victoria Stodden, Friedrich Leisch, and Roger D. Peng, editors, Implementing Reproducible Computational Research. Chapman and Hall/CRC, URL ISBN Pierre Casadebaig 3 ieme Rencontres R 27/06/ / 14

Traits and technologies to design crop breeding systems for climate change

Genotype Breeding method Spectral signature Phenotype (yield) Environment Stress pattern Traits and technologies to design crop breeding systems for climate change SC Chapman1, MF Dreccer1, K Chenu2, D