Characterization of cereals in a semi-arid context based on remote sensing indicators from high spatial resolution images from the Sentinel 1 and Sentinel 2 satellite in central Tunisia

Global food security is based on a limited number of species mainly cereals, maize and rice.                                    
In semi-arid region, the availability of cereals on the international market at competitive prices in relation to local production has led to a change in domestic demand in these countries and has affected the capacity of populations to cover their basic food needs. An operational early grain yield prediction system has been needed to assist policy makers in making initial assessments and planning for annual grain imports. In this context, the main objective of this study is to develop a method for the early estimation of grain and grain straw yields based on high spatial resolution optical satellite data and radar data. Thus, we used two lines of research: the first is based on analysing the relationship between vegetation index and the VH/VV ratio with cereals yields measured in situ. The second axis is based on the estimation of the cereal yields based on a combined index. This last is a combination of the radar index VH/VV and an optical index.

For the first axe, a 22 Sentinel-2 and 55 Sentinel-1 images acquired between 01/09/2017 and 31/08/2018 are used. From the optical data, three spectral indices (NDVI, EVI and EVI2) are calculated and from the Radar data, we calculated the VH/VV polarization ratio. At the same time, we realized experimental measurements made on 54 test plots of dry or irrigated cereals carried out in study area during the 2017-2018 agriculture year. The first approach based on a statistical analysis between the NDVI, EVI and EVI2 vegetation indices and the yields measured showed that NDVI is the best optical index allowing an estimate of grain yield from mid-March with a correlation coefficient R² = 69.22% for the average weight of the grains and R² = 72.38% for the average weight of the straw. Validation of estimates obtained by remote sensing shows that this approach is robust, with an error of 1.79qx/ha and 1.21 qx/ha, respectively, for seed and straw yields. The evolution of yields as a function of the VH/VV ratio was then studied for different dates. The analysis allows that an early estimate can be made the 10th of March based on this ratio with a correlation coefficient R² = 53.79% for the average weight of the seeds and R² = 56% for the average weight of the straw.

For the second axe, a combined index was developed based on the combination of the radar index VH/VV and the optical index. The results show that the most suitable combination is the one between the Radar Index and the NDVI where correlations R² = 63.64% for the average seed weight and R² = 64.03% for the average straw weight. The validation of the estimates obtained by this combined index is made with an error equal to 1.97 qx/ha and 1.31 qx/ha, respectively for the seed and straw yields.