Crop condition assessment using high temporal resolution satellite images

Abstract

Early and accurate reporting of crop condition is desired for its importance and sensitivity to commodity exchange markets and other related sectors. Traditional approach to create the weekly report on crop condition relies on the survey of selected farmers. The results are subjective and inconsistent throughout the crop growing season. Selected remote sensing approaches have been implemented and evaluated against the same dataset. The most challenge for operatically using remotely sensed approach to assess crop condition is the difficult in constructing a high temporal resolution time series of consistent vegetation indices due to cloud contamination or atmospheric effect. In this study, an operational approach was developed to estimate the crop condition using a series of smoothed Normalized Vegetation Indices. Five categories of smoothing algorithms were implemented and compared. They are high order polynomial fitting, “4253H, Twice”, cubic B-Spline, Savitzky-Golay filtering, and double sigmoid kernel fitting. Surveyed data were used to evaluate the results of 48 experiments. The results showed that Savitzky-Golay filtering has a good performance on crop condition assessment. Smoothing improved accuracy of crop condition assessment.