Improving Agricultural Management in a Large-Scale Paddy Field by Using Remotely Sensed Data in the Ceres-Rice Model

Compared to the small-scale situation, some constraints in large-scale rice fields have caused crop growth models to fail to reach an acceptable estimate of yield. This study was conducted to investigate the possibility of enhancing the accuracy of the CERES-Rice model prediction at a large scale through the use of Landsat 5 satellite imagery (termed assimilation'). Firstly, the model was calibrated by data taken from local research. The model accuracy was then evaluated in 110 paddy fields over 26 000ha (method A). Then the model was recalibrated by paddy yield estimated from Landsat 5 image (method B). The two methods were compared based on their results. The results revealed that RMSEn in simulating grain yield in small-scale field experiments on the Hashemi cultivar for calibration and validation of the model were 9 and 8%, respectively (R-2=0.7), which indicated the model's high accuracy in yield prediction. While RMSEn in simulating grain yield in large-scale (methods A) was 22% (R-2=0.54), the use of Landsat images in the assimilation method (method B) increased its accuracy dramatically to RMSEn of 12.7% (R-2=0.72). Copyright (c) 2016 John Wiley & Sons, Ltd. Resume Par comparaison a ce qui est obtenu a petite echelle, les modeles de croissance des cultures appliques a grande echelle peinent a estimer de facon acceptable le rendement. Cette etude a ete menee afin d'evaluer la possibilite d'accroitre l'exactitude de modele CERES-Rice a grande echelle par l'assimilation d'images du satellite LANDSAT 5. Tout d'abord, le modele a ete calibre par des donnees acquises d'une recherche locale. La precision du modele a ensuite ete evaluee dans 110 rizieres sur environ 26000ha (methode A). Ensuite, le modele a ete recalibre par le rendement estime a partir d'images LANDSAT 5 (methode B). Les resultats obtenus par les deux methodes ont ete compares. Les resultats ont revele que les RMSEn des simulations en grains de la variete Hashemi dans l'experience sur le terrain a petite echelle etaient pour la calibration et la validation de 9 et 8%, respectivement (R-2=0,7), ce qui indique une excellente precision de modele dans la prevision de rendement. Le RMSEn pour simuler le rendement en grains a grande echelle (methode A) etait de 22% (R-2=0.54), montrant que l'assimilation d'image satellite (methode B) a augmente de facon spectaculaire l'exactitude a une RMSEn de 12.7% (R-2=0,72). Copyright (c) 2016 John Wiley & Sons, Ltd.