A Crop Simulation System for Integrating Remote Sensing and Climate Information to Reduce Model Uncertainty in Crop Yield Assessments Kiyoshi Honda1, Amor VM Ines2, and Akihiro Yui3 1IDEAS, Chubu University, Kusagai, Aichi, Japan 2International Research Institute for Climate and Society, The Earth Institute at Columbia University, NY 10964 3IHI Corporation, Tokyo, Japan 1. Background When a crop model is used to predict crop yields early in the growing season, two sources of uncertainties dominate – climate and model uncertainties. Climate uncertainty is largest early in the growing season but tends to decrease as weather data become available as growing season progresses. Model uncertainty (e.g., due to errors in model structure, modeling assumptions and other ancillary data), generally remains constant through the growing season. Skillful climate forecasts can reduce climate-related uncertainty in crop yield prediction especially at the earlier stages of the growing season, while model-related uncertainty can potentially be reduced by assimilating remote sensing data within the growing season (Ines et al., 2012; Hansen et al., 2006; de Wit et al., 2007). We present a crop simulation study in Tokachi, Hokkaido, Japan that aims to address the reduction of climate and model related errors in forecasting wheat yields. 2. METHODOLOGY Fig. 3. Sample results of Hokusin wheat variety calibration: a) sim. vs. obs. yields, b) sim. vs. obs. anthesis dates, c) sim. and obs. maturity dates, using single and multiobjective functions in NMCGA. Fig. 2. Location of the study area: Tokachi, Hokkaido, Japan 2.1 Crop model calibration We used a Noisy Monte Carlo Genetic Algorithm (NMCGA; Ines and Mohanty, 2008) to calibrate the genetic coefficients of a wheat variety Hokusin in DSSAT-CSM (Jones et al., 2003) using 14 years (1996-2010) of field experimental data from Tokachi Agro Research Institute (fig. 1, 2). Single Objective: Obj1 =Average[abs(simYield – obsYield)t], for all t Objective 1 is mean absolute error (MAE) in yield. Multiple Objectives: Δ1=Average[abs(simYield – obsYield)t], for all t Δ2=Average[abs(simADAT – obsADAT)t], for all t 3. RESULTS, CONCLUSIONS AND RECOMMENDATIONS 3.1 Model calibration Fig. 3 shows some results of the estimation of Hokusin wheat variety’s genetic coefficients using genetic algorithm. Aside from the single objective function for yields, we tested several multi-objective functions, e.g., weights combinations of (0.4,0.3,0.3), (0.2,0.4,0.4), (0.34,0.33,0.33) and (0.4,0.4,0.2). Obviously, the larger the weight we assign to the target variable, NMCGA will give importance to fitting that variable. On the average, the yield correlations (R) range between 0.4-0.6 and MAE between 0.59-0.75 Mg/ha. For ADAT, R range between 0.71-0.87, but with relatively large MAE, between 19-27 days. Apparently, MDAT is relatively easier to determine, R range between 0.87-0.90 with MAE between 5-11 days. 3.2 Reducing climate uncertainty Δ3=Average[abs(simMDAT – obsMDAT)t], for all t Fig. 4 shows how the crop simulation system can predict yield at different lead-times before harvest. Here, we merge monitored climate before the time of prediction with the realizations of climate, based on climatology, for the rest of the growing season in 2007-2008. We update yield prediction every month until the time of harvest. Note that the crop simulation system can capture the yield uncertainty characterized by climate. Obj2 = w1*(s.Δ1) + w2*Δ2 + w3*Δ3 Objective 2 is a composite mean absolute errors of yield, anthesis (ADAT) and maturity (MDAT) dates; s is a scaling factor. 2.2 Reducing climate uncertainty Fig. 1. Crop simulation system for yield prediction REFERENCES de Wit, A.J.W. and C.A. Van Diepen. 2007. Crop model data assimilation with the Ensemble Kalman filter for improving regional crop yield forecasts. Agric. For. Meteorol. 146: 38-56. Hansen, J.W. and A.V.M. Ines. 2005. Stochastic disaggregation of monthly rainfall data for crop simulation studies. Agricultural and Forest Meteorology. 131: 233-246. Hansen, J.W., Challinor, A., Ines, A.V.M., Wheeler, T., and V. Moron. 2006. Translating climate forecasts into agricultural terms: Advances and challenges . Clim. Res. 33: 27-41. Ines, A.V.M., Das, N.N., Hansen, J.W. and E. Njoku 2012. Assimilation of remotely sensed soil moisture and vegetation with a crop simulation model. Remote Sensing of Environment. In review. Ines, A.V.M. and B.P. Mohanty. 2008c. Parameter conditioning with a noisy Monte Carlo genetic algorithm to estimate effective soil hydraulic properties from space. Water Resources Research. 44, W08441, doi:10.1029/2007WR006125. Jones, J.W., Hoogenboom, G, Porter, C., Boote, K.J., Batchelor, W.D., Hunt, L.A., Wilkens, P.W., Singh, U., Gijsman, A.J. and J.T. Ritchie. 2003. The DSSAT Cropping System Model. Europ. J. Agronomy. 18: 235-265. We developed a framework for merging monitored climate information with climate forecasts along the growing season (fig. 1). Daily realizations of climate forecast is generated using a conditioned stochastic weather generator (Hansen and Ines, 2005). If climate forecast is not available or not being better than climatology, the system will generate weather realizations from climatology. 2.3 Reducing modeling uncertainty Based on the work of Ines et al. (2012) on EnKF-DSSAT-CSM-Maize, we are developing a wheat model implementation. The system will be able to assimilate remotely sensed soil moisture and leaf area index (LAI) to correct modeling errors in the prediction of wheat yield at different lead-times in the growing season. Fig. 4. Predicting wheat yield at different lead times. Note: we used climatology to generate 200 realizations of weather for 2007 – 2008. 3.3 Reducing modeling uncertainty Discrepancy between observed and predicted yield at harvest (fig. 4) can be minimized by the application of the EnKF-DSSAT-CSM-Wheat, which is still under development. Corresponding address: ines@iri.columbia.edu