Authors: Ning Zhang*, Ohio State University, Steven Quiring, Ohio State University
Topics: Agricultural Geography, Climatology and Meteorology
Keywords: yield prediction, soil moisture, time series model, spatial regession, NDVI
Session Type: Paper
Presentation File: No File Uploaded
Accurate crop yield prediction is essential for assessing food security and informing commodity trading. This study compared the time series model with the spatial regression model in maize yield prediction using soil moisture, climate data, and vegetation index (NDVI). The results indicate temperature has a stronger impact than precipitation on maize yield variation in the recent 50 years than before. Climate data have a higher correlation with yield in July than the accumulation over the growing season. In addition, climate data at finer temporal granularity (e.g. decade) is helpful to improve the prediction accuracy. Although soil moisture has a higher correlation with yield variation than precipitation, replacing precipitation with soil moisture does not improve prediction accuracy. The time series model is more stable than the spatial regression model, but it requires at least 15 years using climate-only data and requires 11 years using NDVI and climate data for stable performance. In contrast, the spatial regression model only requires four years for a stable prediction. Both spatial and time series models tend to underestimate yield in high yielding counties and overestimate yield in low yielding counties, and both models perform better in high yielding counties than in low yielding counties. The results from this study provide insight on data and period selection for maize yield prediction, and the methodology can be easily applied for maize yield prediction in other regions of the world.
To access contact information login