The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences
Download
Publications Copernicus
Download
Citation
Articles | Volume XLI-B8
https://doi.org/10.5194/isprs-archives-XLI-B8-959-2016
https://doi.org/10.5194/isprs-archives-XLI-B8-959-2016
24 Jun 2016
 | 24 Jun 2016

COMBINED ANALYSIS OF SENTINEL-1 AND RAPIDEYE DATA FOR IMPROVED CROP TYPE CLASSIFICATION: AN EARLY SEASON APPROACH FOR RAPESEED AND CEREALS

U. Lussem, C. Hütt, and G. Waldhoff

Keywords: crop type mapping, Sentinel-1, RapidEye, agriculture, Support Vector Machine, Maximum Likelihood

Abstract. Timely availability of crop acreage estimation is crucial for maintaining economic and ecological sustainability or modelling purposes. Remote sensing data has proven to be a reliable source for crop mapping and acreage estimation on parcel-level. However, when relying on a single source of remote sensing data, e.g. multispectral sensors like RapidEye or Landsat, several obstacles can hamper the desired outcome, for example cloud cover or haze. Another limitation may be a similarity in optical reflectance patterns of crops, especially in an early season approach by the end of March, early April. Usually, a reliable crop type map for winter-crops (winter wheat/rye, winter barley and rapeseed) in Central Europe can be obtained by using optical remote sensing data from late April to early May, given a full coverage of the study area and cloudless conditions. These prerequisites can often not be met. By integrating dual-polarimetric SAR-sensors with high temporal and spatial resolution, these limitations can be overcome. SAR-sensors are not influenced by clouds or haze and provide an additional source of information due to the signal-interaction with plant-architecture. The overall goal of this study is to investigate the contribution of Sentinel-1 SAR-data to regional crop type mapping for an early season map of disaggregated winter-crops for a subset of the Rur-Catchment in North Rhine-Westphalia (Germany). For this reason, RapidEye data and Sentinel-1 data are combined and the performance of Support Vector Machine and Maximum Likelihood classifiers are compared. Our results show that a combination of Sentinel-1 and RapidEye is a promising approach for most crops, but consideration of phenology for data selection can improve results. Thus the combination of optical and radar remote sensing data indicates advances for crop-type classification, especially when optical data availability is limited.