Breon, FM; Vermote, E (2012). Correction of MODIS surface reflectance time series for BRDF effects. REMOTE SENSING OF ENVIRONMENT, 125, 1-9.
Surface reflectance time series measured from space borne instruments, such as the MODIS sensor, show an apparent high-frequency noise that limits their information content. A major contributor to this noise is the directional effect as the target reflectance varies with the observation geometry. The operational MODIS processing inverts the parameters of a BRDF model which are provided in the so-called MCD43C2 product with a frequency of (8 days)(-1). Recently, Vermote et al. (2009) suggested an alternative BRDF inversion method. A major assumption is that the BRDF model shape (i.e. the BRDF normalized by its overall amplitude) varies little throughout the year so that the two model parameters are linear functions of the NDVI. Consequently, a given target BRDF shape is described by four parameters (slope and intercept for the two NDVI-dependent parameters) rather than 2 parameters that change for each 8 days period. This method imposes additional constrain for the surface BRDF inversion. In this paper, we evaluate the performance of these two approaches for the correction of surface reflectance time series. We work at the 0.05 degrees (approximate to 5 km) resolution of the CMG grid and analyze a representative set of + 100 targets selected on the basis of the location of AERONET sites. The performance is quantified by the high-frequency noise in the corrected time series. We demonstrate that the performances of the two approaches are very similar. This result demonstrates that a simple four-parameter NDVI-scaled model performs as well as a more complex model with many more degrees of freedom. Besides, the four-parameter model, which is inverted on a given year, can be applied to the measurements of other years with a similar level of performance. Finally, a single "averaged" model can be applied to any target with a performance that is only slightly reduced compared to what is achieved with a model derived through a full inversion of the multi-temporal data. The proposed four-parameter BRDF model permits the reduction of noise in the reflectance time series by a factor of the order of three in the red and four in the near infrared. After correction, the reflectance time series are very clean, with an apparent noise that is approximate to 0.005 in the red band and 0.01 in the near infrared. The quality of the BRDF correction makes it possible to use the individual reflectance time-series at high temporal resolution, rather than indices based on their ratio, and thus retain more information about the vegetation dynamics. (c) 2012 Elsevier Inc. All rights reserved.