Filters for Forecasting Crop Health: Analyzing and Projecting the Temporal Evolution of Landsat NDVI Data using Dynamic Linear Models and Bayesian Filters — 9a — Kamal Albousafi, Dr. Jung-Han Kimn

Accurately forecasting food availability is a critical task. One approach involves utilizing remote sensing data, such as satellite images, to observe the health of crop fields using different Vegetation Indices (VI). The Normalized Difference Vegetation Index (NDVI) provides a sound metric to track the “greenness” of crops over time. In this research, we develop statistical models that capture the dynamics of NDVI time series data to make better predictions of its future values. The median NDVI of the pixels of a farm located in Edmunds County, South Dakota is obtained using imagery from the Landsat 5 and Landsat 8 satellites, which produce images every 16 days. Furthermore, we obtained weather data, such as drought intensity and average temperature, to predict how NDVI evolves over time in our corn/soybean field of interest. Three models—dynamic linear models, the Kalman Filter, and the Extended Kalman Filter—are employed to model the NDVI time series using weather data. Our findings show that the Extended Kalman Filter forecasts NDVI, and therefore crop health, most accurately out of the tested models for our crop field of interest.

South Dakota State University
Dr. Hossein Moradi