Crop delineation using hybrid classification procedures: a case study in Scott, Saskatchewan
Currently, crop insurance companies rarely work in co-operation with remote sensing scientists as they believe that the data quality and resolution are too low to accurately delineate crop areas and predict yields. This is due to the cost of high spatial and temporal resolution data, which generally exceeds that of sending a field team to randomly inspect cropped areas. However, methods have been initiated recently, that increase the classification accuracy of medium resolution and coarse resolution data. In this study, SPOT-4 20 m resolution images for June, July and August were provided by Agriculture Financial Services Corporation (AFSC), Alberta for the area of Scott, Saskatchewan to ascertain the classification accuracy of current methodology and evaluate the possible applications of remote sensing data. Results show that hybrid classification and using normalized difference vegetation index (NDVI) are able to produce 85% classification accuracy for a three image multi-temporal stack. Using the normalized moisture difference index with the mid-infrared band for the August image resulted in 90% classification accuracy, although average per-crop-classifications were low. The best classification result was a July-August standard multi-image stack using hybrid classification (green, red, NIR-NDVI ISODATA for each image and the near-infrared band), offering higher per-crop classification accuracy than for any single image classification. The accuracy changes little with adding the June scene to the July/August multi-image stack.
The following license files are associated with this item: