How do model deficiencies in SST impact global-mean temperature projections through the pattern effect?
Atmosphere-ocean general circulation models (AOGCMs) are unable to reproduce recently observed sea surface temperature (SST) trend patterns. In this project, we quantify the relevance of this SST pattern uncertainty to temperature projections through convolving Green’s functions with physically plausible SST pattern scenarios, which differ from the ones AOGCMs produce by themselves. SST pattern uncertainty has a significant impact on temperature projections, for example by almost doubling the uncertainty of warming in 2085 to 2.0K solely due to a redistribution of SST in a business-as-usual scenario. In the next few decades, the reversal of the current cooling trends in the East Pacific and Southern Ocean may lead to a period of abrupt warming unanticipated by AOGCMs through a destabilization of shortwave and lapse rate feedbacks. It is critical for climate change impact, adaptation, and mitigation assessments to incorporate this so far unaccounted for uncertainty until we thoroughly trust the evolution of SST patterns in AOGCMs.
The response of the annual- and global-mean net TOA radiation Green’s Function.
Statistical Post-Processing of Multi-Physics WRF Ensemble Forecasts
Numerical weather prediction requires a form of statistical post-processing due to representational error (model grid not matching terrain), model error, and observational/assimilation error. After Model Output Statistics was created in the early 1970s, reliable deterministic forecasts could be issued for locations in complex terrain. However, MOS provided only a single, non-probabilistic forecast, which could not account for the chaotic nature of the atmosphere. Ensemble weather prediction began in the 1990s, and similar to deterministic forecasts, had to be post-processed due to under-dispersion.
For my master’s thesis project, entitled “Statistical Post-Processing of Ensemble WRF Forecasts for Microclimatic Regions in the U.S. Northeast,” we utilized the Weather Research and Forecasting model (WRF) to produce 9 km and 3 km resolution forecasts from the Global Forecast System (GFS) model for microclimatic, agricultural regions in the U.S. Northeast. These forecasts were then statistically post-processed to generate probabilistic forecasts for temperature, specific humidity, incoming solar radiation, and precipitation. A comparison of forecast skill was conducted between these post- processed forecasts, the raw WRF output, the GFS forecasts, and forecasts from the National Weather Service’s National Digital Forecast Database (NDFD). Overall, significant improvement was observed in post-processed WRF forecasts over all other methods for all regions and variables. Furthermore, NDFD was found to be competitive with raw WRF for temperature, so that if observational data is unavailable for post-processing, the NDFD forecast method should be selected over running high resolution ensemble WRF. Finally, the 9 km post-processed WRF had the same forecast skill as the 3 km post-processed WRF, rendering the 3 km WRF unnecessary if observational data is available, saving computational cost.
Example temperature forecast initiated at 00:00Z April 2, 2017 for one Finger Lakes station. Left: ensemble member temperatures (thin blue lines) are plotted over a three-day period with observations (black). Center: the raw probabilistic forecast for 57Z (thick blue line) is plotted with the ensemble member temperature forecasts (thin vertical blue lines), the NGR probabilistic forecast (thick orange line) and observation (vertical black line). Right: the 75th, 95th, and 99th percentile post-processed CDF regions (orange) are plotted with the observations (black).
Mean absolute error (MAE) and continuous ranked probability scores (CRPS) for deterministic and probabilistic forecast methods for temperature (left), specific humidity (center), and solar radiation (right). These scores are aggregated across all forecasts for all stations.
Reliability diagram for a temperature threshold of 0°C aggregated for Finger Lakes stations. Post-processed WRF improves over raw model output even in the most extreme frost events during the growing season.
Land-Atmosphere Coupling Strength During Northeast Drought
Land-atmosphere coupling in the Northeastern United States (Northeast) was found to be negligible in previous studies. However, a flash drought during the summer of 2016 may have indicated otherwise. This period was one of the warmest and driest in the Northeast, especially in parts of New York State, with below average streamflow levels from decreased snowpack during the preceding winter, and depleted soil moisture values indicated by anomalously low PDSI. My research group believes that during the summer of 2016 a positive soil moisture-rainfall feedback developed, leading to greater interaction between the land and the atmosphere, which strengthened the drought. In future climatic regimes, conditions observed during the 2016 drought are predicted to become more frequent, for example soil moisture levels may decrease due to warmer temperatures and greater evapotranspiration. As soil moisture decreases, less moisture is available for convective initiation during the summer, decreasing precipitation, which further decreases soil moisture levels, in accordance with positive feedback theory. A positive feedback is indicative of a stronger land-atmosphere coupling under a warmer climate regime. Our goal is to identify a feedback in the Northeast under potential future soil moisture conditions. More generally, we are working to test the theoretical framework of land-atmosphere coupling in the Northeast as the climate warms.
Averaged sensible heat (solid green) and latent heat (solid red) fluxes for Ithaca, NY during a summer 2015 WRF run. Dashed lines represent a summer 2015 WRF run with all soil moisture values modified to 0.1 m^3/m^3. During the dry soil moisture scenario, the sensible heat flux is much larger than the latent heat flux, indicating a significant change in the Bowen ratio.