Mapping and monitoring cropland burning in European Russia: a multi-sensor approach

Short lived aerosols and pollutants transported from high northern latitudes have amplified the short term warming in the Arctic region. Specifically, black carbon (BC) is recognized as the second most important human emission in regards to climate forcing, behind carbon dioxide with a total climate forcing of +1.1Wm-2. Early studies have suggested that cropland burning may be a high contributor to the BC emissions which are directly deposited above the Arctic Circle. However, accurate monitoring of cropland burning from existing active fire and burned area products is limited. Most existing algorithms are focused on mapping hotter and larger wildfire events. The timing of cropland burning differs from wildfire events and their transient nature adds a further challenge to the product development. In addition, the analysis of multi-year cloud cover over Russian croplands, using the Moderate Resolution Imaging Spectroradiometer (MODIS) daily surface reflectance data showed that on average early afternoon observations from MODIS/ Aqua provided 68 clear views per growing period (defined 1st March 2003 - 30th November 2012) with a range from 30 to 101 clear views; whereas MODIS/Terra provided 75 clear views per growing period (defined 1st March 2001 - 30th November 2012) with a range from 37 to 113 clear views. Here we present a new approach to burned area mapping in croplands from satellite imagery. Our algorithm is designed to detect burned area only within croplands and does not have the requirements to perform well outside those. The algorithm focuses on tracking the natural intra-annual development curve specific for crops rather than natural vegetation and works by identifying the subtle spectral nuances between varieties of cropland field categories. Using a combination of the high visual accuracy from very high resolution (VHR, defined as spatial resolution < 5m) imagery and the temporal trend of MODIS data, we are able to differentiate between burned and plowed cropland fields in European Russia. The VHR imagery allows for more accurate identification of field condition (burned, bare, residue) through visual interpretation and by the incorporation of the 1km MODIS Active Fire (MCD14) dataset as a means of independent validation for the selection of burned training and validation samples. Confirmed by active fire and visual assessment, these fields then serve as a subset of training data to extract a larger sample set of burned fields from VHR imagery, using the Near Infrared (NIR) band (760-900 nm). NIR showed the largest statistical differences between the burned and unburned field samples using ANOVA and post-hoc statistics with an f value (625.8) far exceeding the critical F-value of 2.665 at p < 0.05. Early-stage validation of the algorithm has shown notable improvement in accuracy over the existing MODIS-based global (MCD64 and MCD45) and regional approaches. Large confusion is found over the mollisol (dark-soil) regions compared to the lighter soil areas of the north. Further algorithm improvements, which rely on in situ observations and other auxiliary sources of information, are underway. In the future, we plan to expand applications of this algorithm to cover all Russian croplands between 2001 and 2013.