Future precipitation variability during the early rainfall season in the El Yunque National Forest

Craig A Ramseyer; Paul W Miller; Thomas L Mote

doi:10.1016/j.scitotenv.2019.01.167

Future precipitation variability during the early rainfall season in the El Yunque National Forest

Sci Total Environ. 2019 Apr 15:661:326-336. doi: 10.1016/j.scitotenv.2019.01.167. Epub 2019 Jan 15.

Authors

Craig A Ramseyer¹, Paul W Miller², Thomas L Mote³

Affiliations

¹ Department of Geography and Geosciences, Salisbury University, Salisbury, MD, United States of America. Electronic address: caramseyer@salisbury.edu.
² Department of Oceanography and Coastal Sciences, Louisiana State University, Baton Rouge, LA.
³ Department of Geography, University of Georgia, Athens, Georgia.

PMID: 30677680
DOI: 10.1016/j.scitotenv.2019.01.167

Abstract

El Yunque National Forest, situated in the Luquillo Mountains of northeast Puerto Rico, is home to a wide range of climate-sensitive ecosystems and forest types. In particular, these ecosystems are highly sensitive to changes in the hydroclimate, even on short time scales. Current global climate models (GCMs) predict coarse-scale reductions in precipitation across the Caribbean prompting the need to investigate future fine-scale hydroclimate variability in the Luquillo Mountains. This research downscales coarse-resolution GCM RCP8.5 predictions from the IPCC CMIP5 project to the local scale to better assess future rainfall variability during the most critical period of the annual hydroclimate cycle, the early rainfall season (ERS). An artificial neural network (ANN) is developed using five field variables (1000-, 850-, 700-, and 500-hPa specific humidity and 1000-700-hPa bulk wind shear) and four derived precipitation forecasting parameters from the ERA-Interim reanalysis. During the historical period (1985-2016), the ANN predicts a binary dry (<5 mm) versus wet (≥5 mm) day outcome with 92% percent accuracy. When the historical inputs are replaced with bias-corrected data from four CMIP5 GCMs, the downscaled ensemble mean indicates a 7.2% increase in ERS dry-day frequency by mid-century (2041-2060), yielding an ERS dry-day percentage of 70% by mid-century. The results presented here show that the decrease in precipitation and wet-days is, at least in part, due to an increase in 1000-700 hPa bulk wind shear and a less favorable thermodynamic environment driven by increased mid-tropospheric warming and a stronger trade wind inversion. By regressing ERS total precipitation against dry-day frequency (R² = 0.95), the predicted mid-century dry-day proportion corresponds to a ~200-mm decrease in seasonal precipitation. In contrast, the ensemble predicts a dry-day frequency recovery back towards the historical climatological mean by end-century (2081-2100).

Keywords: Artificial neural networks; Climate change; Climate modeling; Hydroclimatology; Puerto Rico.