Extraction of planting areas of main crops based on Fourier transformed characteristics of time series leaf area index products
Wang Pengxin,Xun Lan,Li Li,Xie Yi,Wang Lei
DOI: https://doi.org/10.11975/j.issn.1002-6819.2017.21.025
2017-01-01
Abstract:Crop identification is a basis of crop monitoring using remote sense data. Crop acreage is essential information for food security, land management and trade decisions. In this paper, the north of the Yellow River in North China Plain was chosen as the study area, where the main crops are winter wheat, maize, and cotton. The 4-day composite MODIS time series leaf area index (LAI) data with a spatial resolution of 1 km were collected yearly for the identification of different crops in the study area. In order to obtain the distribution of crops, we used 92 phases of MODIS LAI images yearly from 2014 to 2016. To avoid the edge effect of the time series LAI caused by data processing, including Savitzky-Golay filter and the second-order differential methods, the last 2 phases of LAI images in last year and the first 2 phases of LAI images in next year were added to build the time series LAI of a year. And time series LAI of each year was analyzed respectively. Then Savitzky-Golay filter was used to denoise and reconstruct time series LAI curves. The results showed that the Savitzky-Golay filter can remove the influence of factors such as cloud, atmosphere, and so on, and the reconstructed time series LAI curves were smooth and consistent with the crop growth and development characteristics. The Fourier transform provides a new representation of the time series images, which allows analysis of the crop phenology using the amplitudes and phases of the most important periodic components. In this study, the first-order differential method was employed to study the crop planting patterns in the study area by acquiring the number of peaks of the LAI curves, and the fields of one crop a year and two crops a year were extracted. The Fourier transform method based on Savitzky-Golay filtered LAI was further employed to extract the key parameters, such as the amplitudes and phases of the time series LAI. The 11 parameters, including the amplitudes of 0-5 terms and the phases of 1-5 order were used to build a multiband image. As the phenology parameters of different crops had their own characteristics, the amplitudes and phases of LAI curves from different crops varied. Therefore, combined with the minimum distance method, the fields of winter wheat, maize and cotton were extracted respectively. The distinctive feature for identifying winter wheat-summer maize was its 2 wave peaks, and the 1st and the 3rd amplitudes were larger than other amplitudes, which reflected the seasonal variation characteristics of the time series LAI. The cycle of the 1st harmonic was the entire length of time series, which reflected the overall situation of crop growth yearly, and the 3rd harmonic reflected the full-year growth process of the crops. The numbers of peaks for spring maize, summer maize and cotton were 1. Due to that the LAI for maize was generally greater than cotton, and the mean values of the LAI curves for spring maize and summer maize were greater than cotton's, the 0 level amplitudes of spring maize and summer maize were larger than cotton's. The peak of LAI curves for spring maize occurred earlier than cotton's and summer maize's, so the 1st phase of spring maize LAI was larger than cotton's and summer maize's. With these features, the fields of spring maize, summer maize and cotton were extracted from the fields of one crop a year, and the fields of winter wheat - summer maize were extracted from the fields of two crops a year. In addition, the same methods above were employed to extract the fields of crops in the study area in 2014, 2015 and 2016 respectively. At last the identification precision of different crops was validated by combining Google Earth and the phenology characteristics of time series LAI curves. The validation results showed that the overall identification accuracies were greater than 80.00% during these 3 years. And the overall identification accuracy reached 87.08% with Kappa coefficient of 0.85 in 2015, and the user accuracies for individual crop were as follows: winter wheat - summer maize, 92.50%;spring maize, 80.00%; summer maize, 92.50%; and cotton, 85.00%. In conclusion, the first-order differential method can be applied to extract the planting areas of one crop a year and two crops a year accurately. By combining the Fourier transform method with the crop phenology parameters, the planting areas of different crops can be identified effectively, and the approaches in this study are feasible for accurately extracting the main crops distribution information of the study area.