From multi-temporal mapping to time series analysis with high spatial resolution data: evaluation of SPOT4 take5 data to simulate sentinel-2 contribution

Till now, the analyses for operational land cover mapping have been performed through multi-temporal approaches using high spatial resolution satellite data. Such data, as those acquired by Landsat and SPOT satellites, are nominally available every 16 to 26 days. However, cloud obstruction represents a major constraint, limiting the potential revisiting time by decreasing the number of scenes per season actually exploitable for land cover mapping. The coexistence of high spatial and temporal resolution is one of the key innovations offered by Sentinel-2, and a useful feature in land use and agriculture applications of optical remote sensing. Theoretically these features introduce the opportunity for land monitoring to move from multi-temporal approaches to time series analysis, therefore improving the land-cover mapping performances. In this framework, it is important to understand in advance the potential of Sentinel-2 data, even if at present there are no suitable operational Earth Observation data to perfectly simulate the mission. The SPOT4 Take5 experiment - supported by CNES - gives the opportunity to simulate Sentinel-2 data, providing the best test bench to evaluate the improvements offered by both high spatial and temporal resolution. The goal of this study is to understand the suitability of such data for time series analyses in land cover/use and agriculture applications, in order to assess seasonal dynamics of different covers/crops. SPOT4 Take5 level 2 dataset, acquired on the Provence area in France from January to June 2013, has been analysed in order to: i) evaluate the continuity of data provision (mainly considering cloud obstruction) to develop a complete time series on a pixel base; ii) reconstruct the same time series on an object base, exploiting ancillary information representing stable single land use elements. The two reconstructions give an estimation of the actual temporal resolution achievable by the system in the two scenarios. Moreover, the potential improvements due to the Sentinel-2 time series have been evaluated, respect to actual temporal resolution available (16-26 days), simulated by sub-sampling the whole Sentinel-2 dataset. Particularly, the results are discussed to understand the feasibility and potentials of the pixel based and object based approaches in the Sentinel-2 perspective, also taking into account the opportunity to integrate high temporal and spatial resolution data with other sensors (i.e. Landsat 8 OLI).