Development of a winter wheat decision support tool using a novel k nearest neighbors precipitation forecast system

Authors: Flanagan, Paul; Zhang, Xunchang; SU, JIANHONG - University Of Texas At Arlington; GUIDRY STANTEEN, SEAN - University Of Texas At Arlington

Submitted to: American Meteorological Society
Publication Type: Abstract Only
Publication Acceptance Date: 10/29/2024
Publication Date: N/A
Citation: N/A

Interpretive Summary:

Technical Abstract: While seasonal forecasts of precipitation have improved, the issue of spatial scale (having a high enough resolution to be useful at local areas) has still not been directly solved. Given this issue, decision support tools for agriculture have suffered from a lack of usable and reliable forecast data for future growing seasons. More often than not, agricultural stakeholders have to take a “conservative” approach when considering risks versus rewards when planning for the coming growing season. To approach this issue, scientists at the USDA, working with collaborators at the University of Texas at Arlington, have developed a precipitation forecast system based off the well-known K Nearest Neighbor (KNN) methodology. This novel KNN model utilizes ordered pairs of “a” spans of “b” days (normalized data is averaged across each span of “b” days to create a generalized feature vector of “a” spans of normalized data each of which represents an average across “b” days) as meteorological input into the KNN model instead of singular days (as has been used previously). Further, a “good enough method” or GEM was developed for optimizing the monthly a and b pairs such that the root mean square error over a large number of forecasts was the smallest for that particular month, or the a,b pair produced a “good enough” hindcast using historical data. This model was tested across 6 stations spanning the Oklahoma region and shown to produce superior forecasts when compared to monthly climatology. With these results, we have begun implementing a test winter wheat decision support tool using a modified version (includes a grazing component for dual purpose winter wheat) of the Decision Support System for Agrotechnology Transfer (DSSAT) winter wheat crop model, calibrated using local winter wheat yield data gathered by partners at Oklahoma State University, to produce growing season values of relevant grazing and winter wheat yield output using the identified analog historical periods produced from the KNN precipitation forecast model. The hope is to produce a coupled precipitation forecast and winter wheat decision support tool that works (given sufficient meteorological observations for the KNN model) to produce reliable seasonal forecasts of precipitation and winter wheat yields for decision support of regional winter wheat stakeholders. Mention of trade names or commercial products in this publication is solely for the purpose of providing specific information and does not imply recommendation or endorsement by the U.S. Department of Agriculture. USDA is an equal opportunity provider and employer.