Study On Crop Cultivation Pattern Based On Modis Ndvi Time Series Data

The crop cultivation pattern in China presents a complex and diverse situation.The Jianghan Plain is an important crop producing area in Hubei Province,and planting is more common many times a year.As an effective monitoring method,remote sensing can monitor the spatial distribution and dynamic changes of crop cultivation in large-scale and real-time.However,due to the sensor's inner effects,climate and cloud,remote sensing images can not fully reflect the real terrain,so the authenticity of the monitoring results will be very limited.By smoothing and denoising the remote sensing images,the processed images can more realistically reflect the ground object.In addition,different crops'phenological characteristics are different.A single remote sensing image can not dynamically present changes in different stages of crop growth,and the reconstructed time series of remote sensing images can better present the growth of crops in different periods.Our study selected Tianmen City,Qianjiang City and Xiantao City(hereinafter referred to as TianQianMian)in Hubei Province as the research area.The NDVI(Normalized Difference Vegetation Index)was extracted from the 2005-2015 MOD13Q1 remote sensing images dataset and the NDVI time series curve was constructed.The weather and even the weather caused by clouds,sputum,etc.,effect the NDVI time series curve of the sensor acquisition image so that they can not present the characteristics of the real object in detail,so the images urged to be reconstructed and denoised.In this paper,the following three time series methods are used:HANTS filtering algorithm(Harmonic Analysis of NDVI Time Series),which denoises iteratively according to the amplitude and phase characteristics of time series of non-zero frequencies.The S-G(Savitzky-Golay)filtering algorithm can well preserve the shape and width of the original time series,which is important for the retention of important information;EMD(Empirical Mode Decomposition)filtering algorithm does not need to preset any basis.The function is an adaptive data processing method,so the result of the denoising process would be much higher noise ratio.Based on the advantages of the above,the reconstruction of the time series is used to realize the multiple cropping index extraction and crop planting area extraction in the large space scale of the TianQianMian area.The main research results are as follows:(1)Based on the HANTS,S-G and EMD filtering algorithms,the NDVI time series curve can be effectively reconstructed,eliminating noise generated by MODIS images caused by clouds and sputum.Although the processing results of the three filtering algorithms are inconsistent in accuracy,they can better present the variation characteristics of the features on the phenology.The correlation coefficient of-EMD filtering algorithm is better than HANTS'S and S-G's by comparing the spatial distribution of correlation coefficients pre and post filtering and the comparison of front and back precision.(2)The secondary difference method can obtain the familiar spatial distribution in actual research areas.The crop cultivation pattern results is verified by the ground verification points,and the average accuracy of the pattern is high.In the quadratic difference method,the average accuracy of the filtering result of the HANTS algorithm is higher than that of the S-G's and the EMD's.The decision tree-based arable land extraction method can better identify the planting patterns of twice a year,but the sensitivity to the crops of once a year is low.(3)Combining the phenological information of crops,using the decision tree to conduct a tentative analysis of the 2013-2015 ND VI time series curve.The results of pattern extraction showed that the TianQianMian area was mainly planted twice per year,and the results present similar in three filtering algorithms.Under the HANTS filtering algorithm and S-G filtering algorithm in the southwestern part of Qianjiang City,the planting area of crops in two seasons is more prominent,and there is no intensive trend in the EMD filtering algorithm.(4)Morphological similarity method Identifying and extracting features requires a standard feature curve as the basis for extraction.The Tianmen City area is mostly low-hill.Qianjiang City is famous for its rice-shrimp culture and is difficult to extract in Tianmen City and Qianjiang City.A standard feature curve that presents the real object condition.Therefore,the research area for the morphological similarity method is mainly Xiantao City.The planting area of twice crops in the Xiantao area identified by this method is relatively large,and it is sensitive to the identification of planting patterns twice.(5)Based on the second difference,the extraction effect of the 11-years time series data of the three filtering methods is more obvious;the decision tree-based method is more significant for the recognition of 2013-2015;The extraction method of morphological similarity is more significant in the Xiantao area.All three methods can effectively identify and extract crops in one season and two seasons,but the extraction method of cultivated land based on decision tree needs to improve the area extraction accuracy of crops in one season and one season.