Climate change and water scarcity are reducing the available irrigation water for agriculture thus turning it into a limited resource. Today calculating and estimating crop water requirements are achieved through the ETc FAO-56 model where the effect of climate on crop water requirement is determined through the water evaporation from the soil and plant (ETref), and a calendar crop coefficient (Kc). Models that rely on spectral vegetation indices (SVI) derived from optical remote sensing became a reliable source of so-called real-time crop coefficient estimations. SVI-based Kcs are more accurate in real-time estimation of the plant water demands, but they are not adequate for water status or stress detection. The overall goal of this study is leveraging big remote satellite sensing data to determine Target Development Curves (TDCs) for the detection of unbalanced water status. The first step to achieving this goal is to enhance the sensitivity of satellite-based spectral vegetation indices (SVIs) to water stress. The hypothesis of this study was that higher sensitivity would be achieved by combining spatial and temporal measures that exist in multispectral satellite images in time series. A set of data was collected from maize fields in Israel for the year 2019. Spectral, spatial and temporal measures of vegetation indices and their combinations based on Sentinel-2 images were generated to all fields to identify measures with high sensitivity to stress that affected yield in early and peak growth periods. Average NDVI spectral time series calculated for 28 fields for the year 2019 grouped by yield levels (high, medium, and low) showed expected differences: the low yield group had lower NDVI values than the high yield group. Yet, the relative differences were relatively low (around 18%) and they were observed around 30 days after sowing (DAS) with no differences in the peak period. The spatio-spectral measure of the NDVI, i.e. its standard deviation (STD) time series has also shown expected differences where the low yield group had higher variance. The STD measure was advantageous over the NDVI as the relative differences were much higher (around 50%) and they were observed as early as 20 DAS. Yet, no differences were observed in the peak period. When a temporal measure was added, i.e. the area under the curve of the STD-NDVI, differences were further enhanced and existed also in the peak periods (around 30%). Similar analyses with similar results were conducted using other SVIs and using a data set of maize fields collected from the USA for the year 2019. Further analysis will include data from 2018 and 2020 growing seasons. The results show that a combination of spectral, spatial, and temporal measures may enhance the sensitivity of multi-spectral satellite image to water stress and may be used to further develop a library of TDCs based on them.
