Reliable crop forecasts can provide an early warning and decision support for farmers and other agricultural stakeholders for better risk management and improved decision-making. A number of studies have used dynamic crop models to forecast crop production at a regional scale with promising results, few studies have led to operational forecasting systems due to limitations in data availability and the inaccuracy of input data. A general large-area model (GLAM) for rice has been developed, which provides a useful tool for forecasting rice yield directly on the regional scale.

However, the forecast skill of GLAM-Rice is affected by the accuracy of the input dataset and its calibration methods, and the spatial and temporal scale used. It is possible to use the monthly average rainfall in the crop model for a rice yield forecast, accurate daily temperature and solar radiation are required. The forecast skill of the GLAM-Rice was higher at the national level than for individual grid cells, because when a forecast was conducted over a larger spatial scale, the errors from different sources could be offset by each other. While forcing the GLAM-Rice with the six-month seasonal forecasts, the forecast skill improved with successive monthly updates using observations. The forecast skill of the GLAM-Rice has a higher association with solar radiation than with temperature. It is therefore to conclude that if reliable solar radiation data could be provided, it would be feasible to use properly downscaled multi- model ensemble seasonal forecasts from GCMs for predicting crop yields earlier in the crop growing season.

1. INTRODUCTION

2. DATA AND METHODOLOGY

2.1 Research area

2.2. The Crop model

2.2.1 An overview of the crop model

2.2.2 Crop development

2.2.3 The water balance

2.2.4 Temperature stress

2.3 Weather data

2.4 Crop and soil dataset

2.5 Relationship between the input dataset and the model’s skill

2.6 hindcast methodology

2.7 satellite-derived Land Use and Leaf Area Index (LAI)

3. RESEARCH RESULTS

3.1 The effect of input data accuracy on forecast skill

3.1.1 solar radiation

3.1.2 Temperature

3.2 scale of input data

3.2.1 spatial scale

3.2.2 Temporal scale

3.3 seasonal hindcast of climate and rice yield

3.3.1 seasonal hindcast of climate

3.3.2 Effect of the calibration method on the forecast skill

3.3.3 Forecast skill of single GCM vs. MME

3.3.4 The effect of solar radiation on forecast skill

3.4 MoDIs derived LAI

4. CONCLUDING REMARKS

Appendix A

Appendix B