Integrated Use of Multi-temporal SAR and Optical Satellite Imagery for Crop Mapping in Ukraine

Wednesday, 17 December 2014
Mykola S. Lavreniuk, Nataliia Kussul and Sergii Skakun, Space Research Institute NAS Ukraine and SSA Ukraine, Kyiv, Ukraine
Information on location and spatial distribution of crops is extremely important within many applications such as crop area estimation, crop yield forecasting and environmental impact analysis [1-2]. Synthetic-aperture radar (SAR) instruments on board remote sensing satellites offer unique features to imaging crops due to their all weather capabilities and ability to capture crop characteristics not available by optical instruments. This abstract aims to explore feasibility and the use of multi-temporal multi-polarization SAR images along with multi-temporal optical images to crop classification in Ukraine using a neural network ensemble.

The study area included a JECAM test site in Ukraine which is a part of the Global Agriculture Monitoring (GEOGLAM) initiative. Six optical images were acquired by Landsat-8, and twelve SAR images were acquired by Radarsat-2 (six in FQ8W mode with angle 28 deg., and FQ20W with angle 40 deg.) over the study region. Optical images were atmospherically corrected. SAR images were filtered for speckle, and converted to backscatter coefficients. Ground truth data on crop type (274 polygons) were collected during the summer of 2013. In order to perform supervised classification of multi-temporal satellite imagery, an ensemble of neural networks, in particular multi-layer perceptrons (MLPs), was used. The use of the ensemble allowed us to improve overall (OA) classification accuracy from +0.1% to +2% comparing to an individual network. Adding multi-temporal SAR images to multi-temporal optical images improved both OA and individual class accuracies, in particular for sunflower (gains up to +25.9%), soybeans (+16.2%), and maize (+6.2%). It was also found that better OA can be obtained using shallow angle (FQ20W, 40°) OA=77% over steeper angle (FQ8W, 28°) OA=71.78%.

1. F. Kogan et al., “Winter wheat yield forecasting in Ukraine based on Earth observation, meteorological data and biophysical models,” Int. J. Appl. Earth Observ. Geoinform., vol. 23, pp. 192-203, 2013.

2. F.J. Gallego et al., “Efficiency assessment of using satellite data for crop area estimation in Ukraine,” Int. J. Appl. Earth Observ. Geoinform., vol. 29, pp. 22-30, 2014.