B43A-0525
Data Assimilation of PROBA-V 100 m and 300 m.
Abstract:
One of the goals of the FP7 SIGMA projects is the extension of remote sensing time series to better monitor agricultural productivity at global scale. Extending these time series can be seen in differnt ways; on the one hand we are looking at the integration of different existing data sets with equal resolution e.g. SPOT-VGT and PROBA-V 1km resolution, or building new time series for Eta and Soil moisture. on the other hand we are also updating methods to extend existing time series with respect to their resolution and revisting frequency. The research presentend here will focus on the latter, focussing on the integration of PROBA-V 100 and 300m.The PROBA-V microsatellite is designed to offer a global coverage of land surfaces at four spectral bands at a spatial resolution of 300 m and 1 km with a daily revisit for latitudes 75 N to 56 S [1]. Due to the specific design, data can also be acquired at 100 m for a reduced swath, providing partial coverage (global coverage only every 5 days). This study proposes a data assimilation method that combines the 100 m data at the reduced swath with PROBA-V 300 m resolution data at the full swath. The resulting product is a synthetic product at 100 m spatial resolution, with a potential revisit time equal to the 300 m products (S10@300). Here, we concentrate on a ten day composite product (K10@100), to mitigate the effect of clouds. The goal of the proposed method is to produce continuous and cloud free time series of PROBA-V data at 100 m spatial resolution. The S10@300 and S10@100 ten day composits serve as input, with respective spatial resolutions of 300 m and 100 m. Whereas the S10@300 is obtained from all sensors onbaord the PROBA-V platform, the S10@100 is the product from the central viewing sensor only. Due to a combination of the reduced swath and potential cloud cover, the S10@100 is typically sparse (gaps). The data assimilation method follows the approach proposed in that is based on a Kalman filter. It is a recursive algorithm that uses a series of measurements and predicts the value of an unknown state, expresented here as a vector of NDVI values. Instead of assimilating data from Landsat and MODIS as in, we used data acquired from the same PROBA-V satellite platform at two different resolutions.
The results have been tested with a real application dealing with monitoring agriculture.