Using Spatio-Temporal Autoregressive Approach to Improve a Process-Driven Corn Nitrogen Uptake Model.

Corn yield response to applied nitrogen (N) varies across and within fields, and from year to year due to soil, landscape, weather and crop management factors that affect N supply - demand relationships. Statistically-based regional fertilizer N rate algorithms are limited to the region where they are developed and usually offer little scientific insight as to the biophysical components governing the variation in actual N need. A process-based N model that accounts for weather, soil, topography and management factors is needed to improve in-season N management decision support tools. However, modeling and prediction of subfield-scale, in-season corn N uptake and yield are often unsatisfactory due to the spatial and temporal heterogeneity of the dynamic interactions among management and environmental factors.

Spatial clustering, where nearby areas exhibit similar patterns, is often observed in the difference between simulated and observed corn yield. The analysis of spatial patterns where overestimations or underestimations of corn N uptake and grain yield are observed could provide insights to why the original N model fails in some situations. Spatio-temporal autoregressive approach will be used to predict the spatial patterns of the original N model prediction errors using the results of multiple site-years of N response trials. The modified N model will then be used to predict corn N uptake and yield on another dataset of multiple site-years N response trials. The forecast errors of the original and modified N model will be compared and discussed in this presentation.