Linking Seasonal Forecast To A Crop Yield Model SIMONE SIEVERT DA COSTA DSA/CPTEC/INPE – CNPq HOMERO BERGAMASCHI UFRGS First EUROBRISA Workshop March 2008 –Paraty, RJ - Brazil
Talk Outline Aim Study region Motivation Crop model calibrated for South America Methods (Bias Correction) Results Future Direction
Aim Produce maize crop yield prediction based on climate information (seasonal forecasts). Crop yield model Climate Forecast National Statistics Maize Grain Yield Source: IBGE
The Area of Study Rio Grande Do Sul State (RS) “Long River of South” 27.2°- 29.8°S/51.2°- 56.0°W
About Maize in RS… After USA and China, Brazil is the main maize producer in the entire world, and RS is the second greatest producer nationally (IBGE, 2006). Sowing Date: Sep/Oct Harvest: Feb Crop cycle ~130 days
Santa Rosa Passo Fundo Main producer region (all shaded area in the map) Bergamaschi et al Motivation: Crop and Climate Relationship Correlation btw obs.maize yield and rainfall Data Source: crop yield (IBGE) rainfall (INMET, FEPAGRO)
Standardised rainfall and yield anom. Rainfall in all cycle (~ days) 0.56 Rainfall in 0-30 days after tasselling Bergamaschi et al. 2008
Can crop model to reproduce the interannual variability of maize yield? Source: IBGE
SOIL WATER TRANSPIRATION BIOMASS LEAF CANOPY ROOT SYSTEM Water Stress Transpiration Efficiency YIELD Development Stage Yield is a time varying fraction of Biomass Outputs Yield Gap Parameter Schematic diagram of GLAM (adapted from crop and climate group webpage-Reading) Crop model: General Large Area Model Challinor et al., 2003 GLAM is a processed based crop model, which simulates soil water budget, crop plant phenology, canopy growth, root growth, aerial dry mass and grain yield
Crop model: GLAM adapted to RS GLAM were initially tested for groundnut yield across India (Challinor et al., 2003), and it was adapted to simulate maize yield for RS (Bergamaschi et.al., 2008 in preparation.).
Calibration GLAM were based on observational data (soil and crop phenology). UFRGS, Eldorado do Sul Site, Brazil. Muller et al., 2005
Input data to GLAM: SOIL WATER TRANSPIRATION BIOMASS LEAF CANOPY ROOT SYSTEM Water Stress Transpiration Efficiency YIELD Development Stage Yield is a time varying fraction of Biomass Outputs Yield Gap Parameter Daily data required: Solar Radiation Min. Temperature Max. Temperature Rainfall Schematic diagram of GLAM (adapted from crop and climate group webpage-Reading)
Maize Grain yield Estimative using GLAM and observed weather data Observed weather data from meteorological site (P, T, Rad.) FEPAGRO, INMET GLAM OBS (IBGE)
Can GLAM be used to do crop prediction with daily seasonal forecast data?
Seasonal weather data:11 ensemble member ECMWF (single grid point)
Seasonal weather data into crop model: Forecast issued Sep Daily Precipitation (a grid point) - 11 ensemble member from ECMWF model - first month of each forecasts initialized in Sep, Oct, Nov, Dec, Jan, Feb. (RS crop cycle) Rad. & Temp. Observation Daily mean climatology for wet and dry days (1998 – 2005)
ECMWF Daily Climatology ( ) for crop cycle Mean Rainfall (mm/day) sep oct nov dec jan fev month obs Ensemble mean Indiv. Member
Monthly Mean Rainfall R : R(mm d -1 ) = I(mm wd -1 ) x f(wd d -1 ) intensity frequency d=day wd=wet day (Ines and Hansen, 2006) Rainfall decomposition:
Intensity (mm/wd) Frequency (wd/d) sep oct nov dec jan fev month sep oct nov dec jan fev month Obs. Indiv. Member Emsemble Mean
Methods - Bias Correction of daily GCM: -Frequency (wd day-1) – wd = wet day F(p gcm =0) F(p obs =0) c) P 0 – used to truncate the GCM distribution=mean frequency of rainfall abv p 0 matches the obsv. Rainfall. a) b) Daily rainfall (mm) observation GCM Ines and Hansen (2006) Cumulative distribution function
F(p i ) b) p gcm p’ gcm Methods - Bias Correction of daily GCM: -intensity (mm wd-1) Ines and Hansen (2006) observation GCM
Methods - Bias Correction of daily GCM: -Multiplicative Shift
sep oct nov dec jan fev month sep oct nov dec jan fev month sep oct nov dec jan fev month Mean Rainfall (mm/day) mult. shift uncorrected Bias correction Obs. Indiv. Member Ensemble Mean
sep oct nov dec jan fev month sep oct nov dec jan fev month sep oct nov dec jan fev month Intensity (mm/wd) mult. shift uncorrected Bias correction Obs. Indiv. Member Ensemble Mean
sep oct nov dec jan fev month sep oct nov dec jan fev month sep oct nov dec jan fev month frequency (wd/d) mult. shift uncorrected Bias correction Obs. 11 GCM Mem. Mean GCM
Hindcast of Yield– Main producer region DATA INPUT FORECAST: OBS = observed weather data UNC = uncorrected seasonal forecast MS = seasonal forecast corrected using multiplicative shift IF = seasonal forecast, intensity & frequency correction method
Future Direction Statistical downscaling – to take advantage of regional of seasonal forecast skill (spatial calibration). Use of weather generator – to reproduce daily data (temporal disaggregation). Use of space-temporal downscaled daily prediction into crop model. Compared skill of different crop yield forecast approach (grid point and spatial downscaled data).
Thanks: Andrew Challinor (The University of Leeds-UK) Caio Coelho (CPTEC- Brazil) Homero Bergamaschi (UFRGS, Brazil) Tim Wheeler (The University of Reading-UK) Jim Hansen (IRI – USA) Walter Baethgen (IRI – USA)