Identifying crop specific signals for global agricultural monitoring based on the stability of daily multi-angular MODIS reflectance time series
Abstract
Global agricultural monitoring requires satellite Earth Observation systems that maximize the observation revisit frequency over the largest possible geographical coverage. Such compromise has thus far resulted in using a spatial resolution that is often coarser than desired. As a consequence, for many agricultural landscapes across the world, crop status can only be inferred from a mixed signal of the landscape (with a pixel size typically close to 1 km), composed of reflectance from neighbouring fields with potentially different crops, variable phenological behaviours and distinct management practices. MODIS has been providing, since 2000, a higher spatial resolution (~250m) that is closer to the size of individual fields in many agro-ecological landscapes. However, the challenge for operational crop specific monitoring remains to identify in time where a given crop has been sown during the current growing season. An innovative use of MODIS daily data is proposed for crop identification based on the stability of the multi-angular signal. MODIS is a whiskbroom sensor with a large swath. For any given place, consecutive MODIS observations are made with considerably different viewing angles according to the daily change in orbit. Consequently, the footprint of the observation varies considerably, thereby sampling the vicinity around the centre of the grid cell in which the time series is ultimately recorded in. If the consecutive observations that have sampled the vicinity provide similar NDVI values (for which BRDF effects are reduced), the resulting temporal signal is relatively stable. This stability indicated that the signal comes from a spatially homogeneous surface, such as a single large field covered by the same crop with similar agro-management practices. If the resulting temporal signal is noisy, it is probable that the consecutive daily observations have sampled different land uses, thus contaminating the signal. Such time series can therefore be discarded as they are much more difficult to interpret for crop specific monitoring. The approach is demonstrated over different agro-ecological landscapes in Europe and America at regional level. Stable crop temporal signals are first identified automatically and then undergo an unsupervised classification. Clusters exhibiting the expected temporal behaviour of the dominant crops can then be labelled based on knowledge of the landscape. Such crop specific signals can then be related to official crop yield estimates at regional scale for operational yield forecasting during the remaining time life of MODIS. But more importantly, it could serve as a basis to develop a crop specific global archive of crop specific signals since 2000, which could be used as a reference for future satellite Earth observation systems.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2013
- Bibcode:
- 2013AGUFM.B33L..07D
- Keywords:
-
- 0402 BIOGEOSCIENCES Agricultural systems;
- 0480 BIOGEOSCIENCES Remote sensing