Evaluation of NLDAS-2 and Downscaled Air Temperature data in Florida

Abstract

ABSTRACT A broad spectrum of model-derived weather datasets are available in the US. Because each product integrates atmospheric conditions with different model processes, each produces different statistical biases. This study validated air temperature from NLDAS-2 and a novel statistically downscaled NLDAS-2 against observational weather station data for the state of Florida. We statistically downscaled NLDAS-2 to a 1-km grid product using MODIS land surface temperature. We investigated mean biases and Pearson correlation coefficients between daily observational data and the two model-derived datasets. We then calculated multiple Climate Extremes Indices to further scrutinize differences in capturing extreme temperatures. Finally, we quantified potential causes of systematic NLDAS-2 biases related to distance from the coast, urban heat island, land cover, and type of observational stations. Two model-derived datasets showed similar mean biases and correspondence with observational data, underestimating maximum temperature by 1°C and overestimating minimum temperature by 2°C. Extreme temperatures were well simulated in both datasets. However, we still found overestimated extreme minimum temperatures and underestimated extreme maximum temperatures. Systematic biases tended to be higher for coastal stations and grids having a high fraction of water. Our study suggests that including physical processes covering land surface and ocean interactions may improve the model accuracy.