Remotely sensed rice yield prediction using multi-temporal NDVI data derived from NOAA's-AVHRR.

Area of Science:

• Agricultural Science
• Remote Sensing
• Data Science

Background:

• Remote sensing for crop yield prediction is well-established for wheat and maize but less so for other crops like rice.
• Existing methods often struggle to isolate yield variations due to environmental factors from those due to technological advancements.

Purpose of the Study:

• To develop a novel framework for provincial rice-yield prediction using remotely sensed data.
• To create a model that accounts for and removes the influence of technology, fertilizer, and management improvements on yield predictions.
• To enable the development and implementation of provincial rice-yield prediction systems.

Main Methods:

• Collected remotely sensed data over extended time frames, focusing on Normalized-Difference Vegetation Index (NDVI).
• Utilized a longer NDVI time series to build robust regression models, leveraging well-contrasted seasonal variations.
• Developed stepwise regression models for rice-yield prediction in five Chinese provinces, analyzing yield trends against time.

Main Results:

• Regression analysis revealed an annual rice yield increase of 50 to 128 kg/ha.
• NDVI showed a consistently positive influence on rice yield predictions across all models.
• The developed models demonstrated a high correlation between predicted and observed yields (1982-2004) with an overall relative error of approximately 5.82% in validation (2005-2006).

Conclusions:

• The proposed framework provides an effective and operational method for provincial-level rice-yield prediction in China.
• The methodology is adaptable to other crops with sufficient historical NDVI and yield data, provided yields show a significant increasing trend.
• This approach offers a reliable tool for agricultural monitoring and policy-making.