Investigating temporal trends and spatial patterns of extreme precipitation in Northern Virginia, USA

Abstract:

Increasing regional resilience to extreme hydro climatic events, such as flooding and droughts, depends on the scientific community's ability to provide reliable precipitation estimates. The primary objective of this work is to investigate precipitation patterns in Northern Virginia (NOVA), a fast-growing region in the larger Washington Metropolitan area. A random forest (RF) algorithm is adopted to downscale North American Land Data Assimilation (NLDAS) precipitation data to 1 km resolution for the past 40 years (1979-present). The algorithm combines a set of topographically corrected atmospheric forcing (also at 1km), terrain topography (Shuttle Radar Topography Mission, SRTM-Digital Elevation Model), and vegetation information (Moderate Resolution Imaging Spectroradiometer, MODIS Vegetation Indices) to predict 1-km precipitation fields. A recursive feature elimination method produces an optimal number of predictors to train an RF classification (rain/no rain) and an RF regression on the NLDAS precipitation estimates. The trained RF regression model predicts daily cumulative precipitation at higher spatial resolution (i.e., 1 km) using the downscaled atmospheric forcing (1 km). Temporal trend analyses in precipitation and temperature (NLDAS products) using the Mann-Kendall test show statistically significant (at 95% confidence) increasing trends in temperature, but non-significant trends in precipitation. Spatial pattern analyses of extreme precipitation reveal that the southeastern region is characterized by more extreme precipitation (with larger accumulation) than the northeastern part. The average length of dry spells (i.e., number of consecutive days with no precipitation) showed a declining linear trend. Work is undergoing to verify whether higher resolution (downscaled) products offer more information regarding the distribution of extreme precipitation events in the region. Better understating of past precipitation patterns is crucial to optimize future demands on the stormwater infrastructure in the region, especially if combined with future climate and population projections.