The Study Of Paddy Rice Areas Extraction Based On Vegetation Phenology Characteristics

"Hunger breeds discontent". This highlights the importance of the food for human survival. As a large developing country with a population of nearly 1.4 billion, food security is the precondition of maintaining social stability. In order to ensure food security in China, the state and government put forward to resolutely defend the red line of 1.8 billion mu of arable land. Rice is one of the main food crops in China, which is of great significance to ensure food security. Therefore, grasping the information such as rice planting area, cropping system and production management timely and accurately is particularly important. With the growing of population and the rapid progress of urbanization, the appearance of occupying infield is inevitable. It is necessary to build reliable and long-term rice planting area detection system. The original method on acquiring rice planting area in agriculture wastes time and energy and has many drawbacks. The application of remote sensing brings some breakthroughs to the method on acquiring rice planting area. The large-scale rice planting area estimation is mostly based on NOAA/AVHRR data, whose spatial resolution has a great resistance to monitor. Compared with NOAA/AVHRR, EOS/MODIS has large improvement in spatial resolution, and improves the estimation accuracy of rice planting area on a large scale. The present paper takes MODIS as a data source, on the basis of the decision tree classification method, combines the change of the EVI value in the process of rice growth and rice phenophase to recognize rice, and to achieve large-scale rice planting area extraction by remote sensing. The specific research results are as follows:(1) The research uses the Savitzky-Golay filter, Asymmetric Gaussian function fitting, Double logistic fitting of TIMESAT software to reconstruct the time series EVI, to reduce the abnormal value in the data, and to improve the data utilization. By using diagram analysis, quantitative statistical analyze, it analyzes the results processed by three filtering methods, chooses AG filtering as the refactoring functions of EVI curve.(2) The research uses the time series EVI processed by AG filtering to extract the key phenology of rice (transplanting stage, heading stage, maturity stage). Taking the corresponding date of the beginning EVImin in the process of the rice growing period as the transplanting stage, taking the corresponding date of maximum value of EVI curves as the heading stage, and the maturity stage is confirmed by relative threshold method, whose corresponding date (when EVI=EVImin+ΔEVI*0.6) is regarded as maturity stage. The corresponding dates of the extracted three key growth periods are compared with the dates of the meteorological stations, which illustrates most of the sample value fall within the boundary of error of plus or minus 16 days.(3) By taking the EVI value of the process of rice growth, the key phenology and some statistics(length of growing season, EVI curve amplitude) as the class condition of decision-making tree, for different values of the installation of the condition, the study classifies the surface features types to five groups:paddy rice, water body, forest land, settlement place and others throughout the study area, and uses the Ground Truth ROIs method to obtain confusion matrix. The Overall Accuracy is 94.37%, and the kappa coefficient is 0.9127, which indicates the classification result precision is high. Take the city as the unit to statistically analyze the extracted rice planting area, and compare with the rice planting area recorded in statistical yearbook. Among the 36 cities, the precision of 27 cities is above 70%, and 8 cities is below 60%. On the whole, it is feasible to do extraction for the large-scale rice planting area using 16 days’ composed MODIS-EVI data.