Paddy Rice Mapping Based On Medium-and High-resolution Times Series Optical Remote Sensing Imageries

Paddy rice is one of the three major crops grown in the world,and its sown area accounts for nearly 15% of the world’s total arable land.It is crucial to use the remote sensing(RS)technique to monitor paddy rice and obtain important information on paddy rice distribution and planting area,which are of great significance for food security and precision agriculture.A variety of paddy rice mapping algorithms from the phenological and machine learning fields have been developed,which have been tested and applied on various RS images such as MODIS.Subsequently,a huge number of paddy rice products have been produced.However,due to the restrictions of long revisite time,cloud contamination and stripe noises,the existing paddy rice mapping algorithms do not perform well when using the medium resolution RS images for paddy rice mapping.To be specific,the paddy rice mapping results from medium resolution RS images often suffer from spatal incompleteness and temporal discontinuousness.As a result,for these RS images,how to develop several imporved paddy rice mapping algorithms and subsequently obtain the long-term paddy rice products is urgent to be solved.In view of this,this study uses the Sentinel-2,Landsat-7 and Landsat-5/7/8 imags as input,and gradually develops paddy rice mapping algorithms with more constraints and more complex situations,and finally obtains the paddy rice products with larger scales and longer series.Through these operations,this study boosts the existing paddy rice mapping studies at both the algorithmic and product levels.The present study and follow-up analyses will help improve researchers’ understanding of paddy rice mapping mission by providing methodology comparison,product reference and guidance for future research work.Specifically,the research contents and main conclusions of this study are as follows:(1)A paddy rice mapping in cloudy areas is carried out,which is achieved through the "local-global" spatial migration strategy,and called Spatial Domain Bridge Transfer(SDBT).In this study,the concept of local cloud-free region is proposed.The initial rice distribution map is obtained by using the phenology-based method in the local cloud-free area,and then the spatial filtering method is used to purify the phenological paddy rice mapping results,to generate the training samples for training the machine learning model.Finally,the classification model is globally transferred to the other regions for paddy rice mapping.The experimental results show that the proposed SDBT mehtod can realize the automatic mapping of paddy rice in cloudy areas,especially when the input image combination covers the transplanting stage and the later growth stage.At the same time,the SDBT algorithm can also achieve high accuracy when there are only images in the later growth stage,thus breaking through the bottleneck of overreliance on image acquisitions in the transplanting stage for phenological methods,which is expected to provide a technical basis for large-area and long time-series paddy rice mapping.(2)The research on automatic paddy rice mapping based on cloud platform and Landsat-7 image is carried out.In this study,deployed on the Google Earth Engine(GEE)cloud platform,the Landsat-7 oriented paddy rice mapping algorithms is developed,aiming at solving the problems of spatial discontinuity and incompleteness in Landsa7 derived paddy rice map.The experimental results show that the proposed L7 Rice algorithm can meet the needs of large-scale and long-term paddy rice mapping.Meanwhile,the comparison among different rice products and the accuracy evaluation both show that the proposed L7 Rice has better classification performance.These analyses indicate that the Landsat-7 image data can be used for rice mapping tasks.The poposed L7 Rice method and findings are supposed to shed light on improving the paddy rice mapping algorithms and RS image usability,which can offer a further technical improvement for paddy rice mapping in terms of large area and long time series.(3)This study conducts the long-term paddy rice mapping in Northeastern China(NEC)in the past 30 years and analyses the spatiotemporal changes of paddy rice cultivation,whichi is achieved by using the Landsat-5/7/8 images on GEE as input to generate the 30-m paddy rice maps from 1990 to 2020.This obtaind 31-year paddy rice maps with 30-m resolution improves the fineness of the existing rice mapping products in the NEC region in terms of temporal continuity and spatial resolution.On the basis of the 31-year continuous paddy rice maps,this study leverage the spatiotemporal analysis methods to analyze the shift shift of the paddy rice gravitional planting center,planting trends and spatial distribution characteristics.The results show that the center of paddy rice planting in the NEC has shifted significantly to the north,making Heilongjiang province replace Liaoning province as the rice planting center in NEC.In addition,paddy has rice expanded rapidly in the Sanjiang Plain and Songnen Plain,and tended to be saturated in the Sanjiang Plain.These studies and analyses provide crucial information for boosted understanding of paddy rice distribution characteristics and spatiotemporal changes in NEC,which can be used as the pioneer studies of large-area and long-term paddy rice mapping.