Biophysical Variables Retrieval over Russian Winter Wheat Fields Using Medium Resolution as a Precursor of Sentinel-2
d'Andrimont, Raphaël1; Waldner, François1; Bartalev, Sergey2; Plotnikov, Dimitri2; Kleschenko, Alexander3; Virchenko, Oleg3; Roerink, Gerbert4; de Wit, Allard4; Defourny, Pierre1
1Université catholique de Louvain (UCL), BELGIUM; 2Space Research Institute of Russian Academy of Sciences (IKI), RUSSIAN FEDERATION; 3Russian Agrometeorological Institute (AMI), RUSSIAN FEDERATION; 4Alterra, NETHERLANDS
Winter wheat production in the Russian Federation (about 10% of the wheat world production) represents one of the sources of uncertainty for the international market. In particular, adverse weather conditions may induce winter kill resulting in large yield’s losses. This is the reason why the EU-FP7 MOCCCASIN project targets improving the monitoring of winter-wheat in Russia with a focus on winter-kill damage and its impacts on yield.. The objectives of this study were (1) to retrieve Green Area Index (GAI), Fraction of Absorbed Photosynthetically Active Radiation (FAPAR) and the Cover fraction (FCOVER) on a daily basis as an input for crop growth modeling in the Russian context and (2) to characterize their spatial and temporal variability at pixel, field and regional scale. As a prerequisite, the winter wheat fields must be discriminated before the end of the season in order to focus the biophysical variables and the crop growth monitoring to the crop of interest.
As the field’s sizes are very large in Russia, the ratio between the spatial resolution and the field size using medium resolution such as MODIS or MERIS is of the same order as the ratio between high resolution imagery, such as Sentinel-2 and Landsat-8, and smaller fields. This experience highlights the critical impact of both spatial and temporal resolution for operational crop monitoring, demonstrating the gaps between nominal and effective temporal resolution due to clouds and atmospheric perturbations and the needs for multi-sensor approach or very high temporal resolution.
The Tula region (about 26 000 km2) was selected as it is an important winter wheat producer sensitive to winter kill located 120 km away from Moscow . Different intensive field campaigns were carried out by the Space Research Institute of Russian Academy of Sciences (IKI) and by the Russian Agro-meteorological Institute (AMI) for calibration and validation purposes. In addition to crop type inventory along transects, IKI and AMI visited on a regular basis along two growing seasons, i.e. 2011 and 2012, 20 winter wheat fields distributed in two different areas. The field measurements included the crop type, the crop stage, the Leaf Area Index as derived from hemispherical photographs and representative soil spectra. Indeed, the so-called black soil background may influence the spectral signature in the early development stages of the growth cycle, corresponding to the period of observation for the winter kill. All the available MERIS 7-day multispectral composite at 300m were received from the MERIS-Culture processing chain to cover the whole 2011 growing season. In parallel, MODIS Collection 5 daily surface reflectance from both Terra and Aqua platforms, for which atmospherically corrected reflectance is available at a spatial resolution of 250 m in the red and near-infrared spectral bands (MYD09GQ, MOD09GQ), were compiled in two time series from September to August. Time series of daily homogeneous observations was shown to be a powerful instrument for winter wheat vegetation mapping. Using automatic locally adaptive routine jointly with preprocessed MODIS time series, IKI produced winter crop map with accuracy of 82% within Tula region. Alternative approaches using MERIS and MODIS have been also analyzed. From this winter wheat mask, only the pixels dominated by the winter wheat crop were selected for the biophysical variables retrieval. The model of the MODIS spatial response was applied to the 250m MODIS L2G grid to derive the expected purity of the time series with regards to the crop of interest.
Biophysical variables was then retrieved from multispectral reflectance time series for the selected grid cells using an approach combining radiative transfer modeling and neural networks. The algorithm is the same as the one used for the CYCLOPES product which was here adapted to retrieve the variables from the MODIS instruments on-board of Aqua and Terra platforms and calibrated for the Tula Oblast integrating soil spectra from the field. The accuracy of the biophysical variables retrieval were then assessed using the LAI data measured on the ground. Finally, the biophysical variables are used as inputs for a crop growth model (WOFOST) by Altera to estimate the yields. The impacts of the satellite-derived variables assimilation in the model is then analysed and discussed with regards to the winter kill issue.