Ningxia rice area remote sensing estimation on large scale based on multi-temporal OLI data

Abstract: Current researches mostly focus on method and accuracy selection under the condition where data sources are rich but rarely study regional application of multiphase operation land imager (OLI). In this study, Ningxia Hui Autonomous Region was selected for analysis on regional application potential of multiphase OLI data. In order to objectively obtain Ningxia rice area spatial distribution information, to lay a technical foundation for regional crop remote sensing monitoring, and according to the principle of spectral consistency, this paper divided the study area into 6 ground types of rice, sparse forest and shrub, dry land/woodland, abandoned land, wetland/water bodies, and others. The period before July 10, 2016 was taken as the early stage of rice growth. The normalized difference vegetation index (NDVI), infrared reflectance (IR) and short waved index (SWI) were established by choosing the data of OLI carried by US LandSat-8 and using the images taken in the 6 periods of March 11th, April 12 th, April 28 th, May 30th, June 15 th and July 1st. Based on the analysis of dynamic change of 3 indexes of NDVI, IR and SWI, especially on maximum NDVI, minimal IR, and minimal SWI, a decision tree was established, and the identification of rice types in the study area was conducted by using images between March 11 and July 1 of 2016. The basic processes of decision tree classification were: firstly the ground objects such as cities and towns and deserts were eliminated by using maximum NDVI from March to June; the sparse forest and shrubs were eliminated by using maximum NDVI from March to April; the dry land/woodlands were eliminated by using minimum IR from May to June; then wetland/water bodies were eliminated by using minimum IR from March to April; finally, the abandoned lands were eliminated by using minimum SWI from May to June. The remaining pixels were taken as rice. The accuracy verification was conducted by using the highly accurate GF-2 remote sensing (the resolution was 4 m) survey results of rice area background of this region. The extraction accuracy of GF-2 was as high as 99% above. The results showed that the planting area by GF-2 was 91 910 hm2 and the rice planting area was 88 030 hm2 by the OLI data. The total extraction error was only -4.22% with the Kappa coefficient of 0.83; the user's classification accuracy of rice spatial distribution was 85.11% with the mapping accuracy of 81.67%. Among the total rice area, the area in Pingluo, Helan, Yingchuan, Qingtunxia, Lingwu, Shapotou, Litong, Yongning, Zhongning, Dawukou and Huinong accounted for 27.71%, 16.76%, 13.69%, 11.87%, 9.93%, 6.72%,5.34%,3.27%, 2.24%, 1.60% and 0.87%, respectively. The rice was mostly distributed in the north of Yellow River Irrigation Area. The extraction area based on different phases was different. The rice area proportion of 129/033，129/034 and 130/034 images was 60.41%, 32.88% and 6.71%, respectively. Compared with the user’s accuracy of maximum likelihood supervised classification algorithm on the rice area extraction of 76.98% and the mapping accuracy of 61.66% in this area, the method used in this paper showed an increase of 8.13 percentage points in the user’s accuracy, and an increase of even 20.01 percentage points in the mapping accuracy. The result shows that, the method proposed here of establishment of decision classifying tree by using the satellite images of early stage OLI remote sensing time series of rice growth before July 10, and based on the analysis of changing pattern of time series of staple crops can accurately extract the staple crop planting area, and it is a potential method for regional crop area remote sensing monitoring operations.