Click on “Download PDF” for the PDF version or on the title for the HTML version.
If you are not an ASABE member or if your employer has not arranged for access to the full-text, Click here for options.
Neural Approach of Sub-pixel Rice Landuse Classification for Optimized Irrigation Scheduling
Published by the American Society of Agricultural and Biological Engineers, St. Joseph, Michigan www.asabe.orgCitation: Paper number 062124, 2006 ASAE Annual Meeting . (doi: 10.13031/2013.20713) @2006
Authors: Manoj Karkee, Brian L. Steward, Lie Tang
Keywords: pixel mixture, artificial neural network, remote sensing, temporal vegetation signature, rice cropping practice
Irrigation scheduling optimization is carried out in the context of a complex system of agricultural practices and crop calendars. Remote sensing is being used for the monitoring of crop development, crop health, and cropping practices. However, this is possible only if the resolution is sufficiently high to classify patches of different types of crops and cropping practices. MODIS imagery is essential for national or regional scale studies, but has a spatial resolution of 1 km and thus results in sub-pixel mixing of different land covers. In the case of rice farms, one pixel may consist of some proportions of rice grown under different cropping system such as one, two and three crops per year as well as other land covers. Classification of the land area covered by individual pixels is of the great importance for irrigation scheduling. A method was developed for classifying sub-pixel rice land area using a neural network. Temporal patterns of NDVI, which can easily be remotely sensed, depend on and result from the complex relationship between NDVI and cropping practices associated with a pixel. These parameters consist of the proportions of different types of rice and their emergence dates. An artificial neural network (ANN) was used as a model inverter to estimate these parameters. The data for this research were produced using the SWAP crop growth model. The ANN produced up to 95.7% accuracy in crop proportion quantification with an average emergence date error of 9 days. This method had a low computational cost taking 1.22 microseconds per pixel classification in a candidate experiment conducted in a laboratory personal computer.(Download PDF) (Export to EndNotes)