Caveats in calculating crop specific pixel purity for agricultural monitoring using MODIS time series

摘要

Monitoring agriculture at regional to global scales with remote sensing requires the use of sensors that can provide information over large geographic extends with a high revisit frequency. Current sensors satisfying these criteria have, at best, a spatial resolution of the same order of magnitude as the field sizes in most agricultural landscapes. Research has demonstrated that crop specific monitoringis possible with medium spatial resolution instruments (such as with MODIS, 250 m at nadir) if a selection of purer time series is isolated. To do so, a mask of the target crop is necessary at fine spatial resolution in order to calculate the crop specific pixel purity at the coarser scale. Pixel purity represents the relative contribution of the surface of interest to the signal detected by the remote sensing instrument. A straightforward way to compute pixel purity is to calculate the area of the target crop that falls in the coarse spatial resolution grid. However, the observation footprint is generally larger than the squared projection of the pixel, especially when the observation is taken with high scan angles like MODIS does most of the time. Furthermore, the relative contribution within this footprint is not homogeneous: it depends on the spatial response of the sensor. This study analyses the error committed when crop specific pixel purity is calculated using the straightforward method instead of integrating the spatial response and taking into account gridding artefacts and other MODIS particularities such as the bow-tie effect. Differences caused by the orbit, i.e. whether MODIS is on a descending orbit for Terra or an ascending one for Aqua, are also explored. Finally, the consequence of overestimating the spatial response when calculating pixel purity is illustrated by analysing the effect on different agricultural landscapes.