Exploiting MODIS Observation Geometry to Identify Crop Specific Time Series for Regional Agriculture Monitoring
Duveiller, Gregory; Lopez-Lozano, Raul
European Commission Joint Research Centre, ITALY

For many remote sensing applications, the requirements in terms of revisit frequency, geographical coverage and spectral information leads to a spatial resolution that is often coarser than desired. Crop monitoring at regional to continental scales suffers from this compromise when applied to many regions around the world. As a consequence, it generally relies on a signal that has a spatial resolution close to the kilometre that is composed of reflected radiation coming from different adjacent land uses, thus diluting the information content regarding the target crop. MODIS provides a higher spatial resolution (~250m) that is closer to the size of individual fields in many agro-ecological landscapes. Studies have shown how such data can provide a subset of time series that fall adequately within the target crop fields and that characterize regional crop specific behaviour. The present research explores to what extent the adequacy of MODIS time series can be inferred from properties of the time series themselves. More specifically, we propose to better exploit some specificities of the MODIS geometry to derive a measure of signal temporal consistency, which could indicate the level of signal purity with respect to the observed land surface object.

MODIS is a whiskbroom sensor with a large swath. According to the daily change in orbit, MODIS observations for any given place from one day to the next are made with considerably different viewing angles. Consequently, the footprint of the observation varies considerably, sampling the vicinity around the centre of the grid cell in which the time series is ultimately recorded in. If these consecutive observations that have sampled the vicinity provide similar values of NDVI for instance, the resulting temporal signal should be relatively coherent, thereby indicating that the signal comes from a spatially homogeneous surface, such as a single large field covered by the same crop. If the resulting temporal signal is very noisy, it is probable that the consecutive daily observations have sampled different land uses, thus contaminating the signal and rendering it less valuable for crop specific monitoring due to its mixed nature. To evaluate to what extent this notion of temporal coherence can be used to identify crop specific time series, this study analyses how much the spatial component contributes to the temporal noise and how much noise remains from other effects such as the BRDF. This is done by applying a spatial response model for MODIS on existing crop data layers on different contrasting agro-ecological sites.