Atmospheric Correction of a Seasonal Time Series of Hyperion EO-1 Images and Red Edge Inflection Point Calculation

While spaceborne satellite data has been extensively used to extract biophysical forest characteristics through reflectance features and indices, there are still many questions regarding seasonal changes in reflectance. Boreal forests have already seen changes in growth patterns from climate change, and the large scale monitoring of these forests is becoming more important. Understanding seasonal changes in reflectance in the boreal region will allow for the monitoring of seasonal biophysical changes using satellite imagery. For this monitoring to be possible the satellite imagery needs to be preprocessed and atmospherically corrected to create a time series of hemispherical-directional reflectance factors. The red edge is the abrupt change in reflectance between 680 and 740 nm seen in vegetation spectra. The red edge inflection point is the wavelength, at which the slope is steepest in the red edge. The red edge inflection point is sensitive to plant chlorophyll content and has been extensively used for estimating vegetation biophysical parameters including: leaf-area index, biomass and plant health levels. Hyperion is a narrowband imaging spectrometer aboard the Earth Observer-1 satellite. Hyperion captures data across 242 spectral bands covering a spectral range of 356 to 2577 nm resulting in a nominal spectral range of 10 nm. While the high spectral resolution of Hyperion makes it possible to calculate the REIP, there is no consensus on how this should be done, with different methods producing conflicting results. This study explains the preprocessing and atmospheric correction of a seasonal time series of five Hyperion EO-1 images (Provided courtesy of the USGS) from Hyytiälä, Southern Finland (61° 51 N, 24° 17 E). The time series ranges from 31.5.2010 to 12.8.2010, covering much of the growing season and the seasonal changes in reflectance. The first derivative, four-point linear interpolation, Lagrangian interpolation, and fifth-order polynomial fitting methods for calculating the REIP are looked at to determine their applicability for Hyperion imagery using this time series. Hyperion data requires considerable preprocessing before atmospheric correction can be done. In this study the preprocessing covered: destriping, desmiling, atmospheric correction and finally geocorrection. Atmospheric correction was done using both FLAASH and ATCOR, both of which are MODTRAN based absolute atmospheric correction algorithms. The final atmospherically corrected HDRF images were evaluated using in situ handheld spectrometer reference measurements of a grass field in the area. An average RMSE value of around 3% was achieved with both algorithms. The corrected Hyperion images were also compared against two MODIS products, which also showed good agreement. The aerosol retrieval however did not work with either algorithm, on any scene. The use of a sun photometer for aerosol level estimation was also not effective. Due to the dynamics of the red edge and expected seasonal red edge inflection point trends, the fifth-order polynomial fitting method was seen as the best method for calculating the red edge inflection point. The red edge inflection point did not correlate strongly with leaf area index overall, however there was a strong correlation with individual plots. A strong correlation was observed between Hyperion red edge inflection point and understory red edge inflection point, both overall and for individual plots.