For almost 40 years now, satellite ocean color (OC) sensors, from CZCS (1978--1986) to SeaWiFS (1997--2010) to MODIS (1999--present) and to VIIRS (2012--present), have provided a global view of many useful products representing in-water constituents in the Earth’s oceans, especially in open ocean areas. These OC sensors, mostly onboard polar orbiting satellites, take measurements at 700 - 900 km above the Earth’s surface, and therefore can provide moderate to high spatial resolution. However, the temporal resolution of these OC sensors is low and hence unable to capture short time variations of the marine ecosystem. Geostationary OC sensors, such as GOCI (2000--present) and the planned GEO-CAPE mission, take measurements from a geosynchronous orbit and are therefore capable of making hourly measurements to study the diurnal variation of ocean properties. However, the spatial coverage of these geostationary OC sensors is limited. The Earth Polychromatic Imaging Camera (EPIC) onboard the Deep Space Climate Observatory (DSCOVR) located at the L1 point between the Earth and the Sun, make spectral measurements of the entire sunlit Earth surface about every 108 minutes while the Earth rotates. These unique measurements provide not only high temporal resolution suitable for short-term variability studies, because the same areas of interest will appear in a series of images with good viewing geometry, but also global spatial coverage. Hence, any area of interest can be studied. To explore the scientific potential of such unique datasets to serve the ocean color community, we present retrievals of spectral remote sensing reflectances from ocean and inland water areas. Traditional ocean color atmospheric correction (AC) algorithms do not work for EPIC due to the limited number of spectral bands available. Our AC and OC algorithms, however, which are based on coupled atmosphere/ocean radiative transfer model simulations and machine learning techniques, have been modified to be applicable to the EPIC spectral bands. A novel machine learning based cloud masking scheme is also proposed to address challenges associated with turbid coastal water and heavy aerosol conditions when traditional threshold-based cloud masking algorithms often fail. The new cloud mask algorithm also has adjustable sensitivity levels that work in conjunction with our AC and OC algorithms to retrieve more areas of an image while maintaining the accuracy of the retrieved parameters. We present ocean color products retrieved using these algorithms that are cross-validated with MODIS and VIIRS products.