Habitat conversion is contributing to widespread loss of biodiversity and other critical ecosystem services, yet in many parts of the world the scale and pattern of habitat loss goes unmonitored. Decision makers at multiple scales (local to national to regional) are hungry for information on land-cover change, requiring the information to be as recent as possible in order to prioritise interventions and act upon new land-cover change patterns in a timely manner. The high temporal resolution (16 days) of MODIS NDVI data (product MOD13Q1) lends itself to being used to monitor land cover across large extents, but a combination of massive volumes of data and large amounts of noise in the time-series make the endeavor a challenge. In this paper we describe a methodology for detecting landcover changes due to human activities across Latin America, which is capable of providing near real-time monitoring of habitat loss. The methodology is based on the premise that natural vegetation follows a predictable pattern of changes in greenness from one date to the next brought about by site-specific characteristics and climatic conditions in the preceding days. We use a Bayesian-probability based neural network to learn how the greenness of a given pixel responds to a unit of rainfall (derived from the TRMM daily rainfall product 3b42), then apply the model to identify anomalies in the time series which can be attributed to human activities (i.e. non-natural fluctuations in greenness). The Terra-I project presented in this paper is the implementation of this methodology across all of Latin America and demonstrates a potentially powerful means of monitoring habitat loss at a temporal and spatial resolution relevant for decision makers. We show that a 2 month turnaround is possible with this methodology, and provide the first approximations of deforestation for several forested regions in Latin America for the year 2008.