FORMA (FORest Monitoring for Action) is a near real-time tree cover loss alert system for humid tropical forests (as defined by Hansen et al. (2008), based on WWF’s terrestrial ecoregions) spanning portions of 89 countries. It uses a cloud computing algorithm to analyze frequently updated satellite imagery along with complementary information on factors that affect tree cover loss, such as fires and precipitation. The system generates “alerts” for the world’s humid tropical forests that identify 500 × 500 meter areas where new, large-scale loss is likely to have occurred.FORMA is designed for quick identification of new areas of tree cover loss. The system analyzes data gathered daily by the MODIS sensor, which operates on NASA’s Terra and Aqua satellites. The FORMA alerts system then detects pronounced changes in vegetation cover over time, as measured by the Normalized Difference Vegetation Index (NDVI), a measure of vegetation greenness. These pronounced changes in vegetation cover are likely to indicate forest being cleared, burned, or defoliated.FORMA alerts only appear in areas where the probability of tree cover loss is greater than or equal to 50%. Tree cover loss alerts appear when and where new, large-scale loss is likely to have occurred after 2005. That is, alerts appear for a particular period when there has been significant loss in the area during or before that period. Thus the alerts should not be interpreted as an analysis of total tree cover loss area but rather as an indication of an area that has a high probability of having experienced tree cover loss or disturbance over time. The system employs advanced statistical techniques to achieve the best fit to scientifically validate information on loss, measured as a probability.While the data is available at 16-day intervals, the alerts in this layer are displayed as monthly data due to the monthly availability of the complimentary precipitation data. Upcoming upgrades to FORMA include improving the resolution to 250 × 250 meters, and expanding coverage to tropical dry forest and eventually to other biomes across the global scale.Data applicationsThe alerts have been designed for quick identification of tree cover loss as it happens. This allows for rapid response and prioritization of scarce financial and human resources dedicated to forest conservation or sustainable forest management. Armed with this information, stakeholders can use preemptive methods such as on-the-ground visits or aerial inspection with high-resolution satellite imagery (less than 5-meter pixel resolution) to investigate suspected tree cover loss areas.In addition, the alerts may be of value to a variety of researchers who study both temporal and spatial patterns related to tree cover loss areas.The alerts can be compared against other relevant data layers, such as protected areas and concessions boundaries, to evaluate the effectiveness of forest management practices across time and spatial extent.Accuracy and validationInaccuracies are an inherent part of remote sensing analysis. FORMA alerts appear in areas with a greater than 50% probability of tree cover loss. However, persistent cloud cover is a continuous issue in the tropics, and extreme flooding can also produce unreliable remotely sensed data that will result in tree cover loss “false positives” (alerts where no actual tree cover loss has occurred). Furthermore, the alerting system cannot detect all forest cover loss, whether due to the small size of the loss area, persistent cloud cover, or other explanations still being identified through GFW validation efforts.The major instances of false positives may occur as the following:- A random, "speckled" distribution of alerts across an ecoregion, or complete filling of a small ecoregion. Caused by limited or sparse training data, particularly in small ecoregions, which makes it difficult to tune the model there. As a result, alerts cannot be reliably detected. In a normal ecoregion, alerts are usually clustered.-A rapid explosion of alerts over 1-3 months covering a relatively large area. Caused by a significant, persistent drop in detected vegetation levels due to persistent cloud cover along coastlines, in mountains, or elsewhere.-Alerts in water. Caused by shifting water bodies. These alerts should be considered not necessarily as false positives but rather as ambiguous alerts requiring additional data for corroboration.The GFW team is working aggressively to address potential inaccuracies in the data through rigorous validation methods. Specifically, the GFW team is comparing the growing data set of historical alerts to other validated data sets, which are being used for similar applications.This issue brief demonstrates the spatial correlation of the alerts with the PRODES and DETER data sets, produced by the Brazilian Space Agency for the Amazon. The conclusions from the working paper help to illustrate the potential pitfalls of the algorithm, along with its strengths. Through future refinement and proposed crowdsourcing efforts, the GFW team expects the data quality of the FORMA Alerts will continue to improve.

Dataset Attributes

  • objectid
  • cartodb_id
    1 to 3253564
  • lat
    -33.05208333 to 30.77291667
  • lon
    -97.05629864 to 169.44268969
  • iso
  • gadm2
    -9999 to 217694
  • date
    1134950400000 to 1442188800000
  • shape

Related Datasets