You are here

Environmental Security Risk Forecasting


OUSD (R&E) MODERNIZATION PRIORITY: Artificial intelligence/machine learning, statistical forecasting


TECHNOLOGY AREA(S): Information systems, modeling and simulation technology


OBJECTIVE: Develop computer models to forecast risk to US critical infrastructure from a range of potential climate futures. During Phase I, research will be restricted to modeling past and forecasting future wildfire potential in a chosen area containing critical infrastructure.


DESCRIPTION: NGA provides GEOINT to support national policy makers and other federal and local agencies on matters of environmental security such as humanitarian and disaster relief efforts. Much of this GEOINT provides tactical intelligence, such as complementing the picture on the ground with near real time GEOINT to build a comprehensive awareness of the operating environment (e.g., [1]). Because of the increasing incidence and severity of natural disasters correlated with climate change [2], NGA requires the capability to forecast such events, particularly in areas where critical infrastructure is present, which would allow decision makers time to implement mitigation strategies beforehand.


Currently, fire intelligence analysts compile climate and drought forecasts, regional fuel conditions, and satellite and mapping imagery into short-term and seasonal forecasts for broad regions [3]. However, these existing forecasting methods are unable to make fine-grained distinctions in at-risk areas based on small-scale variations in land use, land use change, vegetation, and, moreover, proximity to critical infrastructure.


In recent years, both the volume and resolution of commercial and publicly available satellite imagery relevant to wildfire forecasting has massively increased. Together with recent improvements in machine learning, this imagery may be used to produce high-resolution GEOINT products relevant to wildfire conditions; for example, land use change, soil moisture, and normalized difference vegetation index [4,5]. Alternative phenomenologies such as SAR and lidar may be applied to monitor conditions and changes in forest health. Additionally, new applications of machine learning have produced much more robust risk assessment modeling in a variety of fields, including fire risk [6].


PHASE I: Identify two geographic areas containing US critical infrastructure, one of which that has experienced wildfire-related damage or destruction. Complete a forensic analysis of wildfire risk based on historical remote sensing data from these areas to identify predictive variables. Suggest potential mitigation strategies that would decrease risk. Using identified predictive variables, develop a computer model that forecasts wildfire risks monthly and/or seasonally and suggests mitigation strategies. The forensic analysis and methodology used to model and forecast shall be provided to NGA and (optionally) submitted to an academic journal or conference.


PHASE II: Extend Phase I results to at least two other natural disaster types relating to environmental security of critical infrastructure (e.g., flooding, permafrost melting). Extend analysis to 6+ geographic areas per natural disaster type containing US critical infrastructure on at least two different continents. Extend duration of forecasting capability (seasonal+) and compute statistically valid uncertainty and error estimates.


PHASE III DUAL USE APPLICATIONS: Accurately forecasting environmental security risks and suggesting mitigations would have immense and broad applications. Analysts across a variety of Government and commercial sectors could utilize these forecasts to improve risk understanding and suggest mitigation strategies that could potentially prevent costly repercussions of natural disasters.



  1. Lopez, T., “DOD extends 'Firefly,' related 'FireGuard' support to extinguish wildfires,” U.S. Department of Defense News, Accessed Sept 24, 2012.
  2. “Climate and land use change,” USGS FAQ, Accessed 24 September 2012.
  3. “Fire forecasting,” USDA Science and Technology, Fire Science, Accessed 24 September 2012.
  4. Khan. S., Alarabi. L., and Basalamah. S., “Deep hybrid network for land cover semantic segmentation in high-spatial resolution satellite images,” Information 2021, 12, 230.
  5. Babaeian, E., et al. “A new optical remote sensing technique for high-resolution mapping of soil moisture,” Frontiers In Big Data, 5 November 2019,
  6. Lee, J., Lin, Y., and Madaio, M. (2018). A Longitudinal Evaluation of a Deployed Fire Risk Model. Presented at the AI for Social Good Workshop at the Neural Information Processing System Conference. (NeurIPS 2018).


KEYWORDS: Environmental security, climate, fire, forecasting, remote sensing, computer vision, machine learning, deep learning

US Flag An Official Website of the United States Government