A neural network integrated approach for rice crop monitoring

Within Asia, rice is a main source of nutrition and provides between 30 and 70% of the daily calories for half the world's population. The importance of rice production demands an effective rice crop monitoring system to provide food security for this region. Recent research has proven radar's capabilities in rice crop monitoring. Radar backscatter increases significantly during a short period of vegetation growth, but large spatial variations in rice crop growth occur due to shifting in the crop calendar. The significant increase in radar backscatter over a short period of time can be used to differentiate rice fields from other land covers. The inter-field variations can be used to derive information on local farmer practices and to enhance rice field mapping and yield prediction. The rice crop monitoring system developed in this project was based on these variations as applied to a neural network classification. The system delineated rice production areas for one wet and one dry season, and was able to extract information on rice cultivation as a function of different planting dates. A minimum mapping accuracy of 96% was achieved for both seasons. This information was then used in a neural network-based yield model to predict rice yield on a regional basis for the wet season. When the yields predicted by the neural network were compared with government statistics, the result was a prediction accuracy of 94%.