ANALYSIS OF AGRO-ECOSYSTEMS EXPLOITING OPTICAL SATELLITE DATA TIME SERIES: THE CASE STUDY OF CAMARGUE REGION, FRANCE

The research activities presented in this manuscript were conducted in the frame of the international project SCENARICE, whose aim is to demonstrate the contribution of different technical and scientific competences, to assess current characteristics of analyzed cropping systems and to define sustainable future agricultural scenarios. Dynamic simulation crop models are used to evaluate the efficiency of current cropping systems and to predict their performances as consequence of climate change scenarios. In this context, a lack of information regarding the intra- and inter-annual variability of crop practices was highlighted for crops such as winter wheat, for the study area of Camargue. Moreover, a description of possible future cropping systems adaptation strategies was needed to formulate short term scenario farming system assessment. To perform this analysis it is fundamental to identify the different farm typologies representing the study area. Since it was required to take into account inter-annual variability of crop practices and farm diversities to build farm typologies, representative data of the study region in both time and space were needed. To address this issue, in this work long term time series of satellite data (2003-2013) were exploited with the specific aims to: (i) provide winter wheat sowing dates estimations variability on a long term period (11 years) to contribute in base line scenario definition and (ii) reconstruct farms land use changes through the analysis of time series of satellite data to provide helpful information for farm typologies definition. Two main research activities were carried out to address the defined objectives. Firstly a rule-based methodology was developed to automatically identify winter wheat cultivated areas in order to retrieve crop sowing occurrences in the satellite time series. Detection criteria were derived on the basis of agronomic expert knowledge and by interpretation of high confidence temporal signature. The distinction of winter wheat from other crops was based on the individuation of the crop heading and establishment periods and considering the length of the crop cycle. The detection of winter wheat cultivated areas showed that 56% of the target in the study area was correctly detected with low commissions (11%). Once winter wheat area was detected, additional rules were designed to identify sowing dates. The method was able to capture the seasonal variability of sowing dates with errors of ±8 and ±16 days in 45% and 65% of cases respectively. Extending the analysis to the 11 years period it was observed that in Camargue the most frequent sowing period was about October 31th (±4 days of uncertainty). The 2004 and 2006 seasons showed early sowings (late September) the 2003 and 2008 seasons were slightly delayed at the beginning of November. Sowing dates were not correlated to the seasonal rainfall events; this led us to formulate the hypothesis that sowing dates could be much more influenced by the harvest date of the preceding crop and soil moisture, which are related to rains but also to the date of last irrigations and to the wind. The second activity was related to define farm typologies. Temporal trajectories of winter and summer crops cultivated areas were estimated at farm scale level based on satellite data time series in the 2003-2013 periods. The validation demonstrated that the method was able to produce maps with high overall accuracy (OA 92%) and very low commission errors (3% for summer crops and 7% for winter crops). Omission errors were very low for summer crops (3%) and higher but within an acceptable level for winter crops (31%). Temporal trajectories of annual winter and summer crop land use at farm level were assumed as indicators of farm management (e.g. intensive monoculture farm or diversified crop producer). Trajectories were analysed through a hierarchical clustering procedure to identify farm management typologies. We were able to identify six typologies out of 140 farm samples, covering 75% of the arable land in the study area. A semantic interpretation of the farm types, allowed formulating hypothesis to describe farming systems. The size of the farms seemed to be an explanatory variable of the intensive or extensive farm management. The two main activities presented in this thesis highlighted the importance of time series spatial and temporal resolution for crop monitoring purposes. Currently, only heterogeneous remotely sensed data in terms of spatial and temporal resolutions are available for agricultural monitoring. Forthcoming sensors (i.e. ESA Sentinel-II A/B) will offer the chance to exploit coexisting high spatial and temporal resolutions for the first time. A preliminary application of an innovative methodology for the fusion of heterogeneous spatio-temporal resolution remotely sensed datasets was provided in the final section of the thesis with the aim to (i) produce high spatio-temporal resolution time series and (ii) verify the quality and the usefulness of the generated time series for monitoring the main European cultivated crops. The experiment positively demonstrated the contribution of data fusion techniques for the production of time series at high space-time resolution for crop monitoring purposes. The application of data fusion techniques in the main methodologies presented in this work appears to be beneficial. To conclude this thesis framework, satellite remotely sensed data properly analyzed has shown to be a reliable tool to study large-scale crop cultivations and to retrieve spatially and temporally distributed information of cropping systems. Remote sensing time series analyses lead to highlight patterns of intra- and inter-annual dynamics of agro-practices and were also useful to define farm typologies based on multi-temporal land use trajectories. Results contribute in enriching the studies and the characterization of the Camargue study area, in particular providing information such as sowing dates that are not available at present for the considered study area and represent a step forward in respect to the actual (static) available crop calendar informations. Moreover, the achieved results provide supplementary information layers for summarize and classify the diversity of the farm in the study area and to characterize farming systems.