Boost PC performance: How more available memory can improve productivity
Ciat crop modeling_18may11
1. Process-based crop simulation models in support of global economic modeling weather plant above ground soil and water plant below ground Ricky Robertson Environment and Production Technology Division International Food Policy Research Institute at CIAT: 18 May 2011
2. Processed based crop models try to mimic how plants respond to their environment Growth of roots/stems/leaves/fruits Water use/stress Nutrient extraction Sunshine Rain/Irrigation Soil properties This presentation draws from experience with the DSSAT family of crop models from the user perspective
3. DSSAT consists of several framework pieces working together . . . weather plant above ground soil and water plant below ground
4. . . . each of which require parameters and data real or generated variety attributes soil characteristics initial conditions planting date
5. The outputs can be thought of in two ways Dynamic: each day’s growth, water usage, etc. Response: end results as determined by inputs HWAH = 4400.30 NUCM = 264.95 NLCM = 113.41 ETCM = 773.63 For economic modeling, “response” dominates, but sometimes “dynamic” is important (water management)
6. By running repeatedly, such models can be run globally when data are available CSIRO/A2/2050 maize 990001 HC27 rainfed 90 day spin up 25% moisture content “April” All of the details must be specified for each location
7. By running repeatedly, such models can be run globally when data are available CSIRO/A2/2050 maize 990001 HC27 rainfed 90 day spin up 25% moisture content “April” All of the details must be specified for each location
8. As users, there are interesting “what ifs” that can be done Changes in yield under different climates (rainfed maize 990001; baseline/2000 to CSIRO/A1/2050/369ppm CO2) Highest yielding variety by location (irrigated rice, choosing among DSSAT generic varieties)
9. The robustness or accuracy of results depend on the pieces weather Location specific data Environment models Plant models plant above ground soil and water plant below ground As a user, I can only look for obviously strange results
10. Data quality hinges on availability, geographic coverage, and consistency weather Downscaling of climate data to local scales (to include sunshine and rainfall distribution; microclimates) Soils probably provide the greatest opportunity for improvement plant above ground soil and water plant below ground
11. Users can make some observations to help model developers weather Global scale modeling sometimes exposes strange behavior (e.g., root water extraction in rice) Calibration of varieties depends on quality and variability in experimental data (e.g., maize yield is highest around Ames, IA) plant above ground soil and water plant below ground
12. The yield projections are incorporated into the IMPACT economic model at a regional level IMPACT runs on geographical units known as Food Production Units or FPUs
13. GCM/SRES scenario climate results are down scaled to 0.5 degree/5 minute resolution 2000 June average minimum temperature 2050 CSIRO/A2 June average rainfall Monthly averages are from Thornton and Jones’s FutureClim; daily weather is from DSSAT’s SIMMETEO
14. Planting months are chosen based on current and future climate conditions (a rule-based system) 2000 Rainfed planting month 2050 CSIRO/A2 Rainfed planting month
15. Soils are represented by 27 generic soil profiles based on the harmonized FAO soil datasets Soil profiles color coded by location Soil data must be matched to DSSAT-style soil profiles
16. The remaining inputs fall under management practices Choice of crop variety Rainfed versus irrigated sources of water Planting densities, row spacing, and transplanting details Fertilizer types, amounts, and application dates
17. DSSAT generates projected yields for each location 2000 Rainfed maize yield 2050 CSIRO/A2 Rainfed maize yield
18. Parallelization of the DSSAT runs results in major time savings serial parallelized roughly 1½ weeks on 80 processors (5 crops, rainfed/irrigated, 13 climates, 15 arc-minute resolution) would take roughly 96 weeks on a single processor
19. SPAM 2000 areas are used to weight the projected yields when aggregating to FPUs Rainfed maize physical area in 2000 The Spatial Production Allocation Model data are available from http://mapspam.info/
20. Vector FPU boundaries are placed over top of the raster yield projections 2000 Rainfed maize yield with FPU boundaries in South Asia
21. Projected yields from DSSAT are aggregated up to the FPU-level for use in IMPACT By crop and rainfed/irrigated... Find total SPAM area within FPU Find total production (SPAM area × DSSAT yield) within FPU Compute area-weighted-average yield as total production / total area