Mapping paddy rice agriculture in a highly fragmented area using a geographic information system object-based post classification process

Abstract

National Taiwan University, Spatial Information Research Center, No. 1, Sec. 4,Roosevelt Road, Taipei 10617 TaiwanAbstract. Most paddy rice fieldsin Asia are comprised of small parcelsof land, and theweatherconditionsduringthegrowingseasonareusuallycloudy.Thisstudydevelopsageographicinfor-mationsystem (GIS) object-basedpostclassification(GOBPC)thatcombines low-cost remotelysensed and GIS data to precisely map paddy rice fields in the intensively cultivated but frag-mented growing areas which are characteristic of Asia. FORMOSAT-2 multispectral imageshave an 8-meter resolution and a one-day recurrence, making them ideal for mapping suchareas. Multitemporal images are examined to distinguish the different growth characteristicsbetween paddy rice and other types of ground cover. The pixel-based hybrid classification tech-nique is used with both the unsupervised and supervised approach to distinguish the paddy ricefields from their surroundings. In addition to the pixel-based approach, we also use GOBPC todealwithover-fragmentedparcelsoflandandtoreducetheincidenceofmisclassificationcausedby speckle or mixed pixels (mixels) in the images. A comparison is made with the pixel-basedtechnique.TheKappaindexofagreementobtainedwiththeGOBPCreaches0.095to0.291,andthere is a statistically significant improvement in the user and producer accuracy for all theclasses (z > 1.96) with McNemar’s test in the four study areas. The proposed GOBPC approachis shown to be useful in highly fragmented rice growing areas and may have the potential forother agricultural applications.