Forecasting effects of weather extremes: El Nino s influence maize yields in Mexico Gideon Kruseman, Kai Sonder, Victor Manuel Hernández Rodríguez, Sergio Pérez Elizalde, Juan Burgueño Ferreira International Maize and Wheat Improvement Center (CIMMYT)
Context Second drought in succession in Southern Africa African governments and international organizations looking for alternative sources of white maize in fall 2016
Context The big question is: Where to get maize? US? Mexico?
Context US is largest producer of maize with plenty of surplus, so what s the problem? This is chicken feed not the ingredient for porridge for human consumption In SSA maize for food is white. And corn from the US is not considered fit for human consumption
Context Mexico produces more than 1.5 million MT of white maize in access of domestic demand for human consumption which is available for export
Question Will this excess production in Mexico be available for southern Africa second half 2016 when needed?
El Niño, years and intensities (http://ggweather.com/enso/oni.htm)
Quick analysis of bad el Niño years Precipitation Anomalies (mm/month) for the different seasons averaged for severe EL Nino years (1957, 1965, 1972, 1982, 1997). Data used: UNAM gridded monthly Version Released May 2007: http://iridl.ldeo.columbia.edu/sources/.unam/overview.html
May
June
July
August
September
October
Can we provide better statistical foundations ENSO à weather patterns in Mexico à yields Use historic data to construct a statistically robust model Validate the model (as much as possible) Use model to forecast 2016 yields, including confidence intervals
Objective Predict maize production in Mexico based on historical El Nino index and weather data for Mexico Determine the effect of ENSO on the variables rainfall (pp), maximum temperature (tmax) and minimum temperature (tmin) by state Predict pp, tmax y tmin taking into account ENSO effect by state Predict MEI (ENSO index) Predict maíze production given predictions of pp, tmax, tmin and MEI by state Estimate total national production levels
Data ENSO MEI index www.esrl.noaa.gov/psd/enso/mei Monthly data Complex non-linear, unknown relationship with weather patterns
Data Weather patterns in Mexico Monthly data at state level
Procedure Discover patterns in ENSO simplifying the data in a statistically robust way Link ENSO pattern to weather pattern Yield analysis capturing effects of ENSO and weather patterns in robust manner
Procedure Quantifying El Nino 1. Factor analysis to create limited number of orthogonal artificial variables to be used in the analysis Alternatives for estimating weather variables and yields 2. Time series analysis 3. Structural equation modeling using seemingly unrelated regression
Procedure Quantifying El Nino 1. Factor analysis to create limited number of orthogonal artificial variables to be used in the analysis Alternatives for estimating weather variables and yields 2. Time series analysis 3. Structural equation modeling using seemingly unrelated regression
Preliminary results: ENSO Parallel analysis suggests that the number of factors = 3 and the number of components = 3
Preliminary results: ENSO Factor Analysis using method = minres Call: fa(r = ENSO_MEIdata1, nfactors = 3, rotate = "varimax", scores = "Bartlett") Standardized loadings (pattern matrix) based upon correlation matrix MR1 MR2 MR3 h2 u2 com MARAPR1 0.10-0.06 0.77 0.60 0.398 1.0 APRMAY1 0.27 0.02 0.88 0.85 0.149 1.2 MAYJUN1 0.54 0.03 0.81 0.95 0.046 1.7 JUNJUL1 0.72 0.00 0.67 0.97 0.028 2.0 JULAUG1 0.82 0.01 0.53 0.96 0.035 1.7 AUGSEP1 0.88 0.01 0.42 0.94 0.057 1.4 SEPOCT1 0.91 0.02 0.35 0.95 0.054 1.3 OCTNOV1 0.93 0.04 0.30 0.96 0.036 1.2 NOVDEC1 0.96 0.05 0.25 0.98 0.018 1.1 DECJAN2 0.98 0.05 0.13 0.98 0.019 1.0 JANFEB2 0.98 0.11 0.06 0.97 0.032 1.0 FEBMAR2 0.91 0.26 0.11 0.90 0.097 1.2 MARAPR2 0.82 0.46 0.11 0.89 0.112 1.6 APRMAY2 0.55 0.70 0.08 0.81 0.192 1.9 MAYJUN2 0.28 0.91 0.06 0.91 0.095 1.2 JUNJUL2 0.10 0.98-0.02 0.97 0.029 1.0 JULAUG2 0.01 0.99-0.05 0.97 0.026 1.0 AUGSEP2-0.04 0.97-0.05 0.94 0.063 1.0 SEPOCT2-0.03 0.94-0.01 0.89 0.112 1.0 MR1 MR2 MR3 SS loadings 8.75 5.37 3.28 Proportion Var 0.46 0.28 0.17 Cumulative Var 0.46 0.74 0.92 Proportion Explained 0.50 0.31 0.19 Cumulative Proportion 0.50 0.81 1.00
Methods 1. Auto-regressive variable model: VAR Captures simultaneous interactions between groups of variables System of simultaneous equations Current values of variables in the model do not explain current values of other variables The explained variables of each equation are the the lags of all variables in the model Possible to include season variables and exogenous parameters
VAR(1) structural First order bivariate VAR model y = b b x +γ y +γ x +ε t 10 12 t 11 t 1 12 t 1 yt x = b b y +γ y +γ x +ε t 20 21 t 21 t 1 22 t 1 xt Error terms are Normal The two variables are endogenous Shock ε yt affects y directly and x indirectly Used to predict climate variables and production
SARIMA 2. seasonal ARIMA models Used to analyze seasonal time series When aggregating seasonal effects one obtains ARIMA model where m is the number of periods in the season Used to predict MEI
Linear regression Used to predict production by state Prod t =β 0 +β 1 tmax t +β 2 tmin t +β 3 pp1 t +β 4 ppmax t +β 5 ppa t + β 6 supc t +β 7 sups t-1 +β 8 MEI+ε t Prod:=Maize production MT tmax:=maximum temperaturein rainy season tmin:=minimum temperature at the end of the rainy season pp1:=rainfall at the start of rainy season ppmax:=maximum rainfall during rainy season ppa:=total annual rainfall supc:=area harvested sups:=area sowed MEI:= multivariate index of ENSO-MEI
Results Mexican states with highest production 2001-2011 Producción promedio en 2001-2011 2500 2000 Miles de toneladas 1500 Jalisco Chiapas Mexico Guerrero Veracruz Michoacan Puebla Oaxaca Campeche Guanajuato Tlaxcala Hidalgo Nayarit Tabasco Zacatecas Yucatan Durango Chihuahua Queretaro SLP Morelos QR Sinaloa Tamaulipas NL Colima Coahuila ascalientes DF Sonora BCN 1000 500 0
Effect of el Niño and rainfall
Effect of el Niño on rainfall 4 categories 1: Significant Positive 2: No significant positive 3: Not significant negative 4: Significant negative
Effect of ENSO on minimum Temperature
Effect of ENSO on minimum Temperature 4 categories 1: Significant Positive 2: No significant positive 3: Not significant negative 4: Significant negative
Effect of ENSO on maximum temperature
Effect of ENSO on maximum temperature 4 categories 1: Significant Positive 2: No significant positive 3: Not significant negative 4: Significant negative
50 100 150 200 250 300 Rainfall prediction
Maximum temperature prediction Forecast of series Jalisco 24 26 28 30 32 34 0 50 100 150 200
Minimum temperature prediction Forecast of series Jalisco 6 8 10 12 14 16 18 0 200 400 600 800
Prediction of MEI with ARIMA(1,0,0)(1,0,1) Pronóstico del MEI -2-1 0 1 2 3 0 200 400 600 800
Prediction model results: regression coefficients
Prediction maize yields in jalisco 1980-2017 Prediction of Maize yield in Jalisco (1980 2017) Production (Ton) 0e+00 1e+06 2e+06 3e+06 4e+06 Observed Predicted R squared: 0.7 Cor: 0.83 1980 1990 2000 2010 Year
Predicted maize yields in State of Mexico 1980-2017 Prediction of Maize yield in Mexico (1980 2017) Production (Ton) 0 500000 1000000 1500000 2000000 2500000 Observed Predicted R squared: 0.67 Cor: 0.82 1980 1990 2000 2010 Year
Predicted maize yields in Oaxaca 1980-2017 Prediction of Maize yield in Oaxaca (1980 2017) Production (Ton) 1e+05 2e+05 3e+05 4e+05 5e+05 6e+05 7e+05 8e+05 Observed Predicted R squared: 0.95 Cor: 0.97 1980 1990 2000 2010 Year
Prediction for 2016
Forecasted production of maize grain for 2016 Pronóstico 2016 3e+06 Production (ton) 2e+06 1e+06 0e+00 Aguascalientes BCN Campeche Chiapas Chihuahua Coahuila Colima DF Durango Guanajuato Guerrero Hidalgo Jalisco Mexico Michoacan Morelos Nayarit NL Oaxaca Puebla QR Queretaro Sinaloa SLP Sonora Tabasco Tamaulipas Tlaxcala Veracruz Yucatan Zacatecas State
Average yield of maize per hectare 6 Production (ton/ha) 4 2 0 Aguascalientes BCN Campeche Chiapas Chihuahua Coahuila Colima DF Durango Guanajuato Guerrero Hidalgo Jalisco Mexico Michoacan Morelos Nayarit NL Oaxaca Puebla QR Queretaro Sinaloa SLP Sonora Tabasco Tamaulipas Tlaxcala Veracruz Yucatan Zacatecas State
Prediction for 2016 Estimated total national production Total production: 12,151,755 MT Standard deviation: 608,773.7 MT Interval confidence at 95%: [10,958,558, 13,344,951] Production in previous years: 2011: 9,972,374.92 (niña) 2012: 12,720,476.63 (niño) 2013: 13,200,813.63 (niño) 2014: 13,469,138.43 (niño)
Conclusion Mexico can only provide limited amount to alleviate maize shortages in southern Africa
Thank you for your interest! Photo Credits (top left to bottom right): Julia Cumes/CIMMYT, Awais Yaqub/CIMMYT, CIMMYT archives, Marcelo Ortiz/CIMMYT, David Hansen/University of Minnesota, CIMMYT archives, CIMMYT archives (maize), Ranak Martin/CIMMYT, CIMMYT archives.