Development of an Optimization Scheme of a Crop Growth Model for Better Estimation of Country-scale Rice Yield

Abstract:

Rice, one of major staple crops, is intensely cultivated in Asia. Rice production varies year by year, primarily depending on weather conditions. A decrease in rice production sometimes makes serious problems on food security in Asian countries. Thus, prediction of rice production is one of ultimate goals for agricultural sciences.

Crop models enable us to quantitatively simulate expected production under given weather conditions. However, rice cultivars and cultivation management/technologies (fertilization, irrigation, etc.) are different among countries or regions. To obtain accurate simulated production, schemes and parameters adopted in models should be fine-tuned by countries/regions.

We have developed an optimization scheme for tuning parameters of a crop growth model. In this study, a paddy-rice growth model, MATCRO (Masutomi et al., 2016a,b), was used. MATCRO employs 65 parameters to model rice physiology. Information on cultivation management is also given as input datasets. First, we selected parameters and datasets which considerably affect rice yield through sensitivity analysis. We identified four parameters or datasets to be optimized in this study: two for physiological parameters and two for cultivation management. Second, using the simplex (Nelder-Mead) method, we optimized these parameters or values by fitting the simulated yields to observed yields for each of the top 20 rice-producing countries. We achieved fitting errors of < 0.5 tonne/ha using training data for most countries but failed for some countries where the observed yield data did not follow weather conditions. Finally, we applied multiple metrics to evaluation of model performance for yield prediction.
Such optimization helps to improve model performance at a country scale and potentially opens the way to predict rice yield at practical accuracy prior to a harvest season.

This study was supported by MEXT Coordination Funds for Promoting AeroSpace Utilization; Grant Number JPJ000959.