A GENERALIZED APPROACH TO WHEAT YIELD FORECASTING USING EARTH OBSERVATIONS: DATA CONSIDERATIONS, APPLICATION, AND RELEVANCE.

Abstract

In recent years there has been a dramatic increase in the demand for timely, comprehensive global agricultural intelligence. The issue of food security has rapidly risen to the top of government agendas around the world as the recent lack of food access led to unprecedented food prices, hunger, poverty, and civil conflict. Timely information on global crop production is indispensable for combating the growing stress on the world's crop production, for stabilizing food prices, developing effective agricultural policies, and for coordinating responses to regional food shortages.

Earth Observations (EO) data offer a practical means for generating such information as they provide global, timely, cost-effective, and synoptic information on crop condition and distribution. Their utility for crop production forecasting has long been recognized and demonstrated across a wide range of scales and geographic regions. Nevertheless it is widely acknowledged that EO data could be better utilized within the operational monitoring systems and thus there is a critical need for research focused on developing practical robust methods for agricultural monitoring. Within this context this dissertation focused on advancing EO-based methods for crop yield forecasting and on demonstrating the potential relevance for adopting EO-based crop forecasts for providing timely reliable agricultural intelligence. This thesis made contributions to this field by developing and testing a robust EO-based method for wheat production forecasting at state to national scales using available and easily accessible data. The model was developed in Kansas (KS) using coarse resolution normalized difference vegetation index (NDVI) time series data in conjunction with out-of-season wheat masks and was directly applied in Ukraine to assess its transferability. The model estimated yields within 7% in KS and 10% in Ukraine of final estimates 6 weeks prior to harvest. The relevance of adopting such methods to provide timely reliable information to crop commodity markets is demonstrated through a 2010 case study.