Fusion of MODIS and Landsat 8 images to generate high spatial-temporal resolution data for mapping autumn crop distribution

Abstract

Mapping autumn crop distribution via remote sensing is influenced by the same growth period of crops. Identifying autumn crops using low temporal resolution or low spatial resolution remote sensing data is difficult because the spectral signatures of different crops are similar. An effective approach is the use of remote sensing data with high temporal-spatial resolution in solving the problem of “foreign bodies with spectrum, ” which lowers the accuracy of autumn crop distribution mapping. In this study, the Spatial Temporal Data Fusion Approach (STDFA) is employed to generate remote sensing data with high temporal-spatial resolution, namely, Red, NIR, and NDVI. The Red and NIR data are smoothened using the TIMESAT program. Meanwhile, the phenology indices from the time-series NDVI data are calculated by the filtered Red and NIR data. The four data types, namely, Red, NIR, NDVI, and phenology indices, are used to construct fifteen kinds of 30 m resolution simulated remotely sensed data for the identification of autumn crops. The applicability of the different dimensions of the data in autumn crop identification is then verified using a support vector machine. The test data are derived from the visual interpretation of the results of unmanned aerial images. A high mapping accuracy is achieved with the autumn crop classification results from the different data sets. The crop classification results of the generated remote sensing image data and the corresponding bands of Landsat 8 and MODIS are compared. The analysis of precision and crop spatial distribution reveals that the Red+phenology data set effectively identifies autumn crops in terms of spatial position and distribution details. The data set achieves accuracies of 91.76% and 82.49% for paddy producers and users, respectively, and 85.80% and 74.97% for corn produces and users, respectively. The overall accuracy achieved for both paddy and corn reaches 86.90%. The Red, NIR, NDVI, and phenology data sets generated by the STDFA can effectively distinguish the type of autumn crops. The increase in the dimension of high spatial-temporal data and the accuracy of classification are not positively correlated, with the former showing a slight correlation with stability to some extent. Compared with the classification results of the MODIS data, the remote sensing images with high spatial-temporal resolution show higher classification precision and better crop spatial distribution. The findings of the study can therefore serve as preliminary experimental basis for utilizing remote sensing images with high spatial-temporal resolution in the identification of autumn crops.