This paper presents a dielectric mixture model based approach for in-situ detection of soil-nitrates in real time. The dielectric constant of a material determines the impedance across a pair of electrodes immersed in that medium. We make accurate measurements on soil impedance over multiple frequencies using an in-situ soil-sensor we have designed. The impedance values are then used to determine the effective permittivity of the soil-bulk, which is then used to determine the concentration of individual components like soil, air, water and nutrients, e.g., nitrates using the data from measurements at multiple frequencies, and solving mixing models that involves component concentration as solution variables. The method shows good accuracy in the frequency range 1-70 MHz and can determine nitrate solution with less than 12% error. By considering 3% bound water percentage and assuming snow-like dielectric nature of bound water, this error reduces to 10%. If a parametrized apparent permittivity is considered in the vicinity of constituent particles, this error further reduces to less than 9%. Accurate nitrate detection based on our on-board, real-time, in-situ soil moisture sensors can provide an accurate real-time soil nitrate sensor and has the potential to greatly enhance agricultural production and reduce the impacts to the environment.