Urban areas have both the greatest spatial complexity of land surface features and the highest concentrations of people and economic activity exposed to urban heat islands (UHI) and extreme precipitation events. However, the forecast problem is complicated by urban areas having: i) a high fraction of the LCLU variability (buildings, greenspace, water features) at scales finer than the horizontal grid resolution of operational numerical weather prediction (NWP) models, and ii) a strong vertical aspect to the surface features and their interaction with meteorological processes.  Results from our recent NASA IDS project revealed the causal links in how anthropogenic heating and the impact of the vertical dimension of urban spaces on upward motion impact the position, spatial extent and magnitude of UHI and extreme precipitation events.  Current "operational" NWP models fail to capture these effects.  We seek here to represent these effects in the operational NWP models and deliver this important capability to NASA's Land-Earth System Digital Twins (L-ESDT) effort. We focus the study on the DOE's four Urban Integrated Field Laboratory (UIFL) cities.

Recent progress in hybrid physics - machine learning models suggests that improved representations of the LCLU can be learned from high-resolution research grade NWP models and the wealth of available remote sensing data. The overarching goal of this project is to design an optimal data-model pipeline between the stream of satellite remote sensing data and operational NWP models to accurately capture the effect of LCLU on UHI and extreme precipitation.  The research objectives are: 1) Develop a comprehensive set of urban surface forcing fields relevant to UHI and precipitation extremes over the four UIFL study regions for model training and validation; 2) Develop a machine learning solution that computes optimal "effective" LCLU parameters at the coarse operational NWP scale from high resolution remote sensing data for use in classic land surface models (e.g. NOAH-MP); 3) Augment the physics-based LSM with a machine learning booster stage that captures the net effect of interactions between high resolution LCLU information and meteorological processes at the sub-grid scale for operational NWP; and 4) Demonstrate skill improvement to operational NWP for UHI and extreme precipitation, including an attribution decomposition that demonstrates the relative contributions to the skill improvement across the different remote sensing data sources and model components.

The land surface remote sensing data are at a much finer resolution than the scales of the NWP models.  This project will use high resolution research grade WRF with a multilayer urban canopy model to generate target surface flux fields and UHI and precipitation fields for a range of conditions at each of the UIFL study regions. Once validated, these fields will form the targets to train a model to identify effective LCLU parameters for the coarse operational NWP models (NAM) and for learning a hybrid physics - machine learning supplement to the land surface model to capture the subgrid scale effects. 

The outcome will be an optimal representation of the evolving LCLU information of urban regions in the NASA's Land-Earth System Digital Twins (L-ESDT) effort. The impacts of this work will be directly felt in in two important societal areas: 1) use of land digital twin for urban design scenario analysis to explore the mitigative value of alternative zoning regulations, potential new regulations for building energy efficiency, and proposed green infrastructure retrofits; and 2) improved operational NWP forecasts will drive the model predictive control logic of smart infrastructure systems to realize a significant improvement in urban resilience.