The enormous progress in weather and extended range predictions for the Indian monsoon over the last decade has not been translated to operationalized irrigation water management tools despite many agricultural advisories from operational agencies. The limited implementation is mainly due to the resolution mismatches of forecasts and decision-needs and a lack of soil moisture monitoring networks. Sustained soil moisture monitoring suffers from the high cost to farmers in installing distributed sensors. Here we develop an irrigation water management tool for the farmers at farm scale, which starts with utilizing and merging a few available soil moisture sensors and L-band satellite observations of surface soil moisture using machine learning. Such derived soil moisture field is used as the initial condition with the multi-ensemble future rainfall for the following few weeks given the weather/extended range forecasts in a farm-scale ecohydrological model. This ecohydrological model is integrated with Monte-Carlo simulations within a stochastic optimization framework to minimize water use while not allowing the soil moisture to drop below a threshold level with a certain probability. The optimization results in water arrangement decisions 2 weeks in advance and water application decisions 1-7 days in advance. We also estimate the water storage capacity needed at farm scale for effective water utilization. We find that 20-45 % and 17-35 % water savings were achievable for Kharif and Rabi seasons, respectively, without losing any yield when applied to grape farms of Nashik, Maharashtra, India. The proposed framework is co-developed with the farmers and can be used in any region for any crops, since it is generalized and easy to transfer. This is an extension of our earlier work to an end-to-end system using satellite data for soil moisture.