Maxar-blended Forecast: Description and Evaluation of Seamlessly Integrated Multimodel and Multitimescale Ensemble Forecast System for Climate Enterprise and Applications

For a business that wants to identify asset risks and exercise mitigation strategies in the face of climate change, the foundational first step is having access to reliable climate forecasts and projections that intersect varying timescales.

Traditionally, climate models target either near-scale forecasts or long-lead forecasts, and they have different spatial resolutions and expected outcomes. These inconsistent ways of creating climate forecasts can make it difficult for businesses that don’t specialize in weather and climate to understand what forecasts can mean for their businesses.

Maxar’s ClimateDesk addresses these challenges by collecting, unifying and synthesizing these disparate forecasts from various reputed climate modeling groups into one platform and providing a proprietary, comprehensive climate forecast solution for academia, the public sector and commercial entities that covers timescales from subseasonal out to 2100.

By processing the raw climate forecasts of daily temporal resolution to correct biases and preserve long-term trends and downscaling them to a unified 0.25-degree spatial resolution, a dynamically updated, blended forecast is generated from the present day through 2100. This Maxar-blended system provides forecasts and projections as probabilistic insights for minimum, maximum and mean surface temperatures, precipitation and wind speeds to address the needs of a wide variety of commercial and public use cases.

This presentation describes the Maxar-blended climate forecast's multimodel and multiscale ensemble system design and probabilistic forecast insights available from Maxar’s ClimateDesk interface and discusses the validation results relative to observations and comparable publicly available climate forecasts. The 2015-2023 hindcasts of the Maxar-blended forecast were evaluated using a suite of categorical metrics for various events and thresholds. The results reveal that the system produces accuracy, probability of detection, critical success index, and Heidke skill scores above 80% for subseasonal and seasonal timescales and above 60% beyond the yearly timescales. The system also produces lower values for the probability of false detection and false alarm ratios for the same events, indicating its reliability.