|
Abstract:
|
Sample surveys have been the foundation of official statistics produced by the USDA’s National Agricultural Statistics Service (NASS) and other Federal statistical agencies for more than half a century. It has become evident that estimates and predictions can be improved by combining survey data with increasingly available information from diverse sources, such as administrative, weather, and remotely sensed data. In this presentation, two data science approaches that have been used at NASS will be explored. The first is to use models to combine survey and non-survey data at a pre-specified level of geography. The other approach is to combine data at a finer spatial scale and to aggregate values to produce estimates or predictions at various levels of geography. Current research efforts are focused on combining these data using machine learning methods to produce pre-season predictions and early-season estimates of acres planted to major crops at the state and national levels. The success and challenges associated with each approach will be discussed. The need for additional research will be highlighted.
|
