Improving crop classification with landscape stratification based on MODIS-time series

Organised by Laboratory of Geo-information Science and Remote Sensing

Thu 7 July 2016 09:30 to 10:00

Venue Gaia, gebouwnummer 101
Room 1

By Bart Driessen (the Netherlands)

This study attempts to test whether stratification based on moderate resolution MODIS imagery can be used as an alternative to stratification based on detailed soil and elevation maps. To this end, the accuracies of methods for stratification and classification of WorldView 2 & 3 imagery were compared for a series of twelve images covering a case study area in Sougoumba, Mali.

Three stratification layers have been constructed: one based on the average, one on the amplitude, and one on the seasonality of a MODIS time series. Subsequently, classification has been performed using various algorithms (RF, SVM, ML, k-NN and multinomial logistic regression) on a training set of 3792 samples and validated using a training set of 1881.

Though all stratification methods proposed have a positive influence on classification results, the methods based on the amplitude and seasonality of MODIS time series yielded the highest classification accuracies. The proposed stratification techniques have a lot of potential for upscaling but some modifications will be necessary to apply them to other areas, especially when moving towards zones with multiple growing seasons per year.

This study shows that MODIS time series can be a very useful tool for developing stratification layers, and may provide a viable alternative to stratification techniques previously tested.

Keywords: crop type classification; stratification; WorldView 2/3 imagery; timeseries; MODIS EVI/NDVI.