Improving Crop Models: Incorporating New Processes, New Approaches, and Better Datasets Jon I. Lizaso (jon.lizaso@upm.es) Technical University of Madrid 13th ESA Congress 25-29 August 2014, Debrecen, Hungary Overview Crop models improved in response to: o o o Incorporating new processes o Anthesis-Silking Interval (ASI) in maize Incorporating new approaches o Better crop/environment understanding New scientific questions Need for better accuracy (especially under stress conditions) Sink-limited kernel set in maize The need for quality and diversity of datasets 2 Early crop models Early models described canopy light capture and photosynthesis o Personal computers not available o o Hesketh, Baker & Duncan 1971, 1972; Baker, Hesketh & Duncan, 1972 Almost 50 years of model improvement Apple II released in 1977 IBM PC released in 1981 Later models incorporated development, growth and partitioning, and yield o De Wit, 1965; Monteith, 1965; Duncan et al., 1967 Better understanding New questions Better accuracy (stresses) Review: Boote et al., 2013. Plant, Cell & Environment 3 Improving models: new processes Crop simulation models are a deliberate simplification of a field grown crop Modelers decide what process to include: Objectives Models evolve: o o o Including new processes Including new approaches (substitute/complement previous) Re-parameterization or Re-calibration (quality datasets) Example of incorporating a new process: Anthesis-Silking Interval (ASI) in maize o o Yield is sink-limited Kernel set is source-limited (under most field conditions) 4 Maize monoecious plant Staminate flowers shedding pollen ASI Monoecious: Separate male & female flowers in the plant Grain yield depends on the synchrony between Anthesis & Silking for adequate pollination and kernel set Pistillate flowers with stigmata 5 Grain Yield (Mg ha-1) Incorporating new processes: ASI 8 6 Bolaños&Edmeades, 1993 Bolaños & Edmeades, 1992 Managed Drought (SC403) SC403 Managed Drought (SC513) Strong relationship of maize grain yield with ASI Especially under water stress Modern hybrids, with enhanced stress-tolerance, show similar trend Incorporated ASI simulation into CERESMaize (K. Tesfaye, pers comm) 4 2 0 0 5 10 15 20 Grain Yield (Mg ha-1) ASI (d) 20 Control Stress 16 12 8 4 Elite US Corn Belt hybrids (G. Edmeades, pers comm) 0 -30 0 30 60 90 120 ASI (GDU 10-30º C) 6 Incorporating new processes: ASI PAR Incorporated into CERESMaize v4.5 CO2 LAI k Flowering event changed from silking to anthesis RUE BAGDD (SPE) The model calculates the average shoot growth rate (SGR) during a thermal time window around flowering (MIN) Thermal time window delimited by two userspecified parameters Pop Dens Row Spac AAGDD (SPE) 1: Avg Shoot Growth Rate (SGR) Ear Growth Part Ear Barrenness ASI TEMP WSTR SRAD SLPF NSTR PSTR KSTR 7 Incorporating new processes: ASI Model assumes no stress when: ASI vs. SGR SGR > 5 g/plant day 18 16 Rel ASI (d) 14 12 10 8 Two new cultivar parameters: ASEN=3 o ASEN=7 o ASNS (ASI under no stress) ASEN (sensitivity to stress) ASEN=11 6 ASEN=15 4 ASEN=19 2 Under no stress: ASI=ASNS Under stress silk extrusion is delayed according to ASEN 0 0 2 4 6 8 Shoot Growth Rate (g/pl d) 10 8 Incorporating new processes: ASI Kernel Number ASNS (CUL) For negative ASI values (protogyny), it uses a function calculated from Lizaso et al. (2003, 2007) ASEN (CUL) Kernel Number 1 B&E 1993 400 0.8 CERES-ASI 300 0.6 200 0.4 100 0.2 0 Relative kernel set 500 Kernel Set The model estimates kernel number as a function of ASI, according to Bolaños & Edmeades (1993) Onset Lin Grain Fill 2: ASI SGR 0 -10 0 10 ASI 20 30 9 Incorporating new processes: ASI Barrenness THRE (ECO) 1.2 PLTPOP 1 4: Barrenness (EPP) Ears/pl 0.8 0.6 0.4 0.2 3: Kernel Number (KN) 0 0 2 4 6 8 Shoot Growth Rate (g/pl) 10 KN & Ear/plant SGR 1 Relative KN & Ears/pl Yield G2 (CUL) ASI 1 Relative KN 0.8 0.8 Ears per plant 0.6 0.6 0.4 0.4 0.2 0.2 0 0 0 5 10 15 ASI (d) 20 25 Onset Lin Grain Fill Model calculates barrenness as a function of SGR Since kernels are set on ears, barren ears are checked with ASI 30 10 Incorporating new processes: ASI P5 (CUL) G3 (CUL) EPP 5: Yield KN ASI Onset Lin Grain Fill Finally, yield is calculated with: o kernel number (KN) o ears/plant (EPP) o onset of linear grain fill 11 Incorporating new processes: ASI THRE (ECO) P5 (CUL) G3 (CUL) 4: Barrenness (EPP) 7: Ear Growth 5: Yield BAGDD (SPE) AAGDD (SPE) Part Ear G2 (CUL) 1: Avg Shoot Growth Rate (SGR) 3: Kernel Number (KN) 6: Onset Lin Grain Fill 2: ASI ASNS (CUL) ASEN (CUL) DSGFT (ECO) 12 Incorporating new processes: ASI Simulated / Measured Kernel Number Yield 1.4 1.2 CERES ASI 1.0 0.8 0.6 0.4 0.2 0.0 Rain n Rain N Irr n Irr N Rain n Rain N Irr n Irr N Some preliminary results indicate the new model is working reasonably well Additional testing is required under various conditions and stresses 13 Improving models: new approaches Maize grain yield is sink-limited. The potential size of the sink, kernel set, is determined around flowering However, maize kernel set is usually source-limited Maize models simulate kernel numbers: o o o Edmeades and Daynard, 1979 Light captured Photosynthetic rate Growth rate Example of incorporating a new approach: Sink-limited kernel set in maize 14 Simulating kernel set in maize If pollen becomes limited, as in hybrid seed production, or there is poor synchrony between anthesis and silking, kernel set may be sink-limited Example of incorporating a new approach that complements current procedure: sink-limited kernel set o o o Pollen dynamics Silk dynamics Relationship linking pollen & silks 15 J. Lizaso, 2005 Dynamics of pollen shed: measuring pollen rates 8 Self-adhesive traps are located daily at silks level. Fluorescence microscopy produces images that are processed with image-analysis software. This result in pollen counts as pollen grains cm-2 d-1 (Fonseca et al., 2002) 53 212 16 J. Lizaso, 2007 Dynamics of pollen shed: measuring pollen rates 500 Gauss functions adequately describe daily pollen rates for hybrids and inbreds 300 200 100 0 202 204 206 208 210 212 214 D a y o f th e ye a r 600 P3489 -1 200 d ) 198 500 -2 196 P o lle n ra te (g ra in s c m P o lle n ra te (g ra in s c m -2 -1 d ) P3394 400 400 300 200 100 0 198 200 202 204 206 208 210 212 214 D a y o f th e ye a r 17 J. Lizaso, 2007 Dynamics of pollen shed: simulating pollen rates To simulate ear-level pollen rates (grain cm-2 d-1) 2 pieces of information are required: Progression of population reaching anthesis (%) Daily pollen production from individual tassels (grain plant-1 d-1). These values can be calculated from: Total pollen produced per tassel (million grains/tassel) Duration of pollen shed per tassel (d) Total pollen produced per tassel can be field measured or estimated from tassel morphology (Fonseca et al., 2003) 18 J. Lizaso, 2007 Dynamics of pollen shed: simulating pollen rates S im u laPteod pno neis ra te A thlle llen sehs d A n th e s is 100 60 Poenlle rc en tsc m (% ))n s ) P ra P o lle illio gn ra intte sopf(g lapra nla tinns(m 40 -2 P e rc e n t o f p la n ts (% ) 1 3 00.80 80 20 0 195 200 205 210 215 220 225 -1 D a y o f th e y e a r P o lle n s h e d 0 .6 P o lle n g ra in s p la n t -1 (m illio n s ) 0 .8 0 .4 0 .2 0 .0 0 1 2 3 4 5 D a ys a fte r a n th e s is 6 7 8 2.64 millions Dekalb 611 8 days 8 pl m-2 250 80 0 .6 Dekalb Dekalb 611 611 -2 8 pl 8 pl m-2 m 200 60 105.40 40 100 0 .2 20 50 0 0 .00 11099 05 1 2 0109 5 2 205 2 30 0 2 140 215 205 5 62 1202 0 7 222155 8 DDaDays aofte ff thr eea nye yth eaaer rs is ayy o 19 Dynamics of silk appearance: measuring silk extrusion Silks are cut and ears are covered with glassine bags to prevent pollination Each day 2 cm pieces are cut from the silk bouquet and are kept in alcohol at 4º C Silks are counted and monomolecular functions are fit 20 J. Lizaso, 2007 Dynamics of silk appearance: simulating silk extrusion Silk simulation requires field measurements of: Progression of population reaching silking Pattern of silk extrusion from individual ears: Number of silks per ear Duration of silk extrusion Measurements of number of silks are facilitated by measuring the perimeter of the bouquet (Schneider, 2005) 21 J. Lizaso, 2007 Dynamics of silk appearance: simulating silk extrusion S im u la ilk pn e a ra n c e ilk inasgpio S te ilkdS esx tru 107 80 610006 S ilk in g 100 Asgrow Asgow740 740 Asgrow 740 -2 8 8plplmm -2 8 pl m -2 80 60 40 20 0 185 190 195 105 60 200 205 210 215 D a y o f th e y e a r 410004 40 103 S ilk e x tru s io n 800 -1 210002 20 N u m b e r o f s ilk s e x p o s e d e a r -1 Nu rm oefnestrg ilk pts oss(% ehda) e adr em wPb lyeerc e ilk o fesd pe lasxn -1 -1 8 811000 0 101 1 0000 1108855 600 400 200 0 0 0 1 9 01 9 2 195 1 9 52 040 220005 6 2 1200 5 2 1 58 2 1 0 2 2 0 1201252 5 2 4 6 8 D a ys a fte r s ilk in g D aDD ys rthseeilk aayyaofte ye aag rr off th yein 22 10 Linking pollen & silks: kernel set relationship SSim ilk a pnp era a te ra n c e imuulalateteddsp o lle S im u la te d s ilk a p p e a ra n c e S im u la te d p o lle n ra te 310008 -1 108 107 d 250 150 100 107 250 Asgrow 740 8 pl m-2 Dekalb 611 8 pl m-2 -2 100 -1 rtio n o fe(g kdera rn e lsssce ta (% P )d) -1 olyo lle te in NPero wp enmra e rg s ilk hm 200 50 0 190 195 200 205 D a y o f th e ye a r 210 215 N e w ly e m e rg e d s ilk s h a -1 -2 P o lle n ra te (g ra in s c m ) 300 6 8100 106 105 104 103 102 101 100 200 105 185 190 195 200 205 210 215 D a y o f th e ye a r 60 115004 3 4100 100 B a s s e tti a n d W e s tg a te , 1 9 9 4 102 20 50 101 y = 0 .9 6 x 0 < x < 100 y = 96 100 < x 2 r = 0 .9 7 10000 1 019805 11 90 0 01 9 5 1 9 5 2 0 0 220000 320005 2 0 52 140 0 0 D Dlle a yanyosfohfth e aarinr s cm R a te o f p o eth deye (gye ra 2 1251 0 5 0 02 2 0 -2 2 12652050 -1 d ) 23 220 225 Simulated kernels plant-1 Evaluating a complementary approach Lizaso et al., 2007 800 r2C = 0.17; MSDC = 15084 r2M = 0.73; MSDM = 6664 600 400 200 CERES Modified 0 0 200 400 600 800 Measured kernels plant-1 Simulated kernels plant-1 Results from seed production fields show the processes are quite predictable and our procedures capture them 1600 Yet useful for seed industry r2C = 0.77; MSDC = 23127 r2M = 0.87; MSDM = 8350 1200 800 400 Too many inputs from male and female inbreds CERES Modified When both, source- and sinklimited conditions were simulated the new model showed excellent accuracy 0 0 400 800 1200 1600 Measured kernels plant-1 24 J. Lizaso, 2005 Towards the future: the quest for quality & diverse datasets A number of current efforts to improve crop models: o o o AgMIP Program: Pilot studies on wheat, maize, rice, and ongoing work on sugarcane, potato, sorghum-millet, peanut, soybean MACSUR Project: Focusing on European agriculture, more interested in crop rotations, pastures, and livestock Model packages: DSSAT, APSIM, CropSyst, STICS, EPIC, and others Beyond the number of processes included, and the approach chosen, a permanent concern for model improvement/testing is the quality and diversity of datasets especially in areas and processes poorly represented 25 J. Lizaso, 2005 Towards the future: the quest for quality & diverse datasets Bassu et al., 2014 Relative variation of simulated yield 0.4 Morogoro, TZ Lusignan, FR Rio Verde, BR Ames, US 0.3 0.2 Field data collection must continue especially in areas and processes poorly represented AgMIP maize team showed that an ensemble of 19 models was superior simulating maize yield than any single model So, how many models are enough? 0.1 0.0 0 4 8 12 16 20 Number of models averaged As the ensemble size increased, relative variation dropped differently for each site 26 J. Lizaso, 2005 27