Investigating the capability of multi-temporal Landsat images for crop identification in high farmland fragmentation regions

摘要

Crop identification is a critical component for grain production prediction. Identifying crop type using remote sensing techniques has been investigated for many decades. A number of different supervised methods have been developed to discriminate different crops. However, most of these methods were applied to areas with relatively large cultivated fields. In China, the cultivation policy leads to extreme complexity in the agricultural landscape, especially in summer and autumn seasons. The objective of this study was to investigate the capability of multi-temporal Landsat images for crop identification in a region with high farmland fragmentation. The study area is located in Taigu, Shanxi province, where the crop planting structure is very complicated. A total of 7 Landsat Enhanced Thematic Mapper Plus (ETM+) images were acquired from 14 October 2003 to 26 June 2004 for classification. Two most favorable classifiers, support vector machine (SVM) and maximum likelihood classifier (MLC) were selected for classification with training samples using different combinations of multi-temporal Landsat images. The overall classification accuracy and Kappa statistics estimated from the confusion matrix using validation samples were selected for evaluating all classification results. Accuracy assessment results indicated that multi-temporal ETM+ data achieved satisfactory classification accuracy (best overall accuracy 89.61%) in the study area. SVM classifier performed better than MLC when three or less Landsat images were used. The addition of the temporal dimension further increased the overall classification accuracy for both SVM and MLC, but the accuracy increased slightly for SVM classifier. The time of data acquisitions are of great importance for crop classification. Results in this paper indicated that multitemporal Landsat ETM+ data are capable for crop discrimination in regions with high farmland fragmentation. In the future, the use of Chin- Environment Satellite HJ-1A/B data for this application should be investigated in the future for the higher temporal resolution and greater spatial coverage.