Evaluating water stress in agricultural fields is fundamental in irrigation decision-making, especially mapping the in-field water stress variability as it allows real-time detection of system failures or avoiding yield loss in cases of unplanned water stress. Water stress mapping by remote sensing imagery is commonly associated with the thermal or the short-wave-infra-red (SWIR) bands. However, integration of multi-sensors imagery such as radar imagery or sensors with only visible and near-infra-red (NIR) bands could significantly improve the in-field water stress mapping as it enables higher revisit time over the field and gap cloudy days.

In this study, a method to map in-field water stress was developed by integrating Sentinel-2 with Landsat-8, Planet SuperDove, and Sentinel-1 images. The integration starts by aligning the pixels of the different sensors into a single grid map by geo-location shift of each pixel. Afterward, the normalized-difference-water-index (NDWI) was calculated for each Sentinel-2 image, as this index was found to be highly correlated to crop water stress. Then, each pixel from Landsat-8, Planet SuperDove, or Sentinel-1 was fused, to that pixel in the NDWI time series of Sentinel-2 via a smoothing algorithm, allowing differential weights to images within the time series.

This method was assessed on a commercial alfalfa field (Israel), comparing the water stress mapping with daily field measurements of leaf-water-potential and leaf-area-index, commonly used as proxies for water stress. Results show good agreement between water stress maps generated using the method suggested here and field measurements. Further, the fused data (integration of Sentinel-2 imagery with each Landsat-8, Planet super-dove, or Sentinel-1 image) was also positively correlated to the Sentinel-2 NDWI that was left out from the process. These results outline the potential of the suggested method to generate high-frequency water stress maps, allowing more accurate decision-making in precision irrigation.