Timely and accurate knowledge about the geospatial distribution of crops at regional to national scales is crucial for estimating crop water use, crop yield and for supply-chain logistics, crop insurance and financial market forecasting. The United States (US), a major food producer, currently lacks a spatially explicit crop data product available publicly during the growing season. The National Agricultural Statistics Service (NASS) of the US Department of Agriculture (USDA) provides crop-specific data for the current year only by the spring of the year following the current growing season. The objective of this study was to estimate acreage under corn and soybeans, two of the most important agricultural commodities consumed across the globe, by the middle of the growing season for the midwest region of the US. Time series products of satellite imagery derived from the Moderate resolution Imaging Spectroradiometer (MODIS) sensor aboard the Terra and Aqua satellites hold considerable promise for large area crop mapping in an agriculturally intensive region, given their moderate spatial resolution and near-daily temporal repeat. A Neural Network based regression approach was used to estimate crop areas. MODIS-derived spectral reflectances and vegetation indices and daily gridded climate data at 1km x 1km resolution from the Daymet data set were used to capture crop specific phenology and address the inter-annual variability in crop development respectively. The algorithm was trained at the scale of ecoregions to account for spatial variability in crop phenology. The proposed method for estimating crop acreage is the first step towards developing an in-season crop yield estimation product that could enable early identification of production shortfalls over large areas for major food producing regions and could inform trade related policy decisions.