Velocity by Booz Allen

From Science to Practical Climate Resilience THE VISION, BARRIERS, AND NOVEL AI APPROACHES TO OPERATIONALIZE CLIMATE SCIENCE Prachi Sukhatankar Contributions from Christopher Holder and Marie Nguyen MISSION SPOTLIGHT: CLIMATE I n 1985, Carl Sagan—one of the most prominent scientists of our times—testified in front of Congress about increased greenhouse gas emissions and

Climate resilience, as the term suggests, means being prepared in the face of climate adversities and recovering to the same state of well-being or better. Adversities include more frequent and extreme weather events, such as extended droughts, heat waves, and ocean warming, and the direct and indirect impact of those events on natural ecosystems, built infrastructure, and daily life. The impact of these events and our aspirations toward resiliency are closely related to conditions across various domains of civilization, such as society, economics, the environment, government, and technology (see Figure 1 on page 70). enhanced the understanding of our dynamic planet and the different forces at play. Several decades later, as we look to build a climate-resilient nation and globe, we are uniquely positioned to leverage climate models and other diverse datasets and to capitalize on the rise of AI, harnessing its power of pattern detection across massive amounts of structured and unstructured data to sort signal from noise, better extract insights, and inform climate adaptation and impact mitigation. The future era of climate resilience requires far more than sophisticated climate science and scientific computing. It demands a commitment to translating advanced climate science and purposefully embedding that intelligence into operational use where it is most relevant: at the hyperlocal level.

their potential global and intergenerational impact. He urged progress in climate science and research and indicated how that could be done by leveraging historical data. Climate science has come a long way since its inception in the late 1950s. The coupled Earth system models, primarily developed by federal agencies and research centers such as NOAA, NASA, the National Center for Atmospheric Research (NCAR), Intergovernmental Panel on Climate Change (IPCC) and others rely on sophisticated algorithms and advanced high performance computing to process large-scale datasets. This combination of science and technology ushered in a new era that significantly

Barriers and Opportunities for Operationalizing the Science for Climate Resilience Accessing relevant climate information is difficult for many reasons. While Earth-observing satellites capture hundreds of terabytes of data daily, the downlink stations can process only a fraction of that data, which is further transformed into analysis-ready data and stored online before being made available for broader analytical use. Simulations of climate scenarios are often stored in large nonintuitive repositories that use unique file-naming structures and file formats. For example, basic access to data may require that the systems ingest files with names like r1i1p1f1 1 , with a NetCDF, BUFR, or GRIB2 file structure, where latency is often as critical as accuracy. The files stored on repositories can be enormous—on the order of petabytes—and may cover only a few years of a single global climate variable. Downloading and working with multiple variables over a time frame of several

Moreover, new mechanisms can fuel powerful scientific collaboration and operationalization, so that small interventions done at scale yield big impacts. This opportunity to drive increased collaboration and operationalization is where novel AI approaches can help in powerful ways. The rise of ubiquitous AI—increasingly embedded in an ever-growing array of applications and devices—offers a unique chance to bridge the gap between advanced research and on-the-ground operations. Federal organizations and the climate community can harness the power of AI tools to better bring the science to localized operations. From infrastructure improvements to proactive disaster planning, bringing climate insights to decision-making can achieve the level of scale and speed required to respond to rapidly evolving climate threats. But what barriers to the operationalization of climate science lie in front of us, and how do we flip the narrative through novel approaches to integrate data, algorithms, and compute power, so that these challenges become opportunities to build real-world resilience?

1 The naming convention of r1i1p1f1 specifies the “variant-ID.” The letter “r” represents the ensemble (aka realization) number; “i” is for the initial conditions used in the model; “p” stands for the physics parameterization; and “f” represents the forcing conditions.