Validation of North American land data assimilation system Phase 2 (NLDAS-2) air temperature forcing and downscaled data with New York State station observations

Abstract

Climate and meteorological factors including temperature and associated heat exposure metrics are related to environmental health risk factors and the incidence of health effects. While meteorological station data provide a good source of point measurements, temporal and spatially consistent temperature data are needed for health studies. Reanalysis data such as the North American Land Data Assimilation System’s (NLDAS) 12 km gridded product are an effort to resolve spatio-temporal environmental data issues. With funding from the NASA Applied Sciences Program, the NLDAS 12 km air temperature product has been downscaled to 1 km using MODIS Land Surface Temperature patterns. Limited validation of the native 12 km NLDAS reanalysis data has been undertaken. Our objective is to evaluate the accuracy of both the 12 km and 1 km downscaled products using the US Historical Climatology Network station data geographically dispersed across New York State for 2006, 2009, and 2010. These years were selected by a climatological analysis over the past 10 years that found 2006 to be a reference year with near-normal temperatures, 2009 a cool year and 2010 a warm year. Statistical methods including correlation, scatterplots, time series and summary statistics were used to determine the accuracy of the remotely sensed maximum and minimum temperature products. The effects of elevation and terrain slope on remotely sensed temperature product accuracy were determined with 10 m digital elevation data. National Land Cover Land Use data was used to determine the effect of mixed pixels on the accuracy of the NLDAS products. Results show the downscaled temperature product improves accuracy over the native 12 km temperature product with average correlation improvements from 0.82 to 0.87 for minimum and 0.76 to 0.85 for maximum temperatures. Higher terrain slope was related to lower correlations between stations and remotely sensed products. Stations in pixels with higher percentages of open water tended to have higher correlations. Our results inform health studies using remotely sensed temperature products to determine health risk from excessive heat by providing a more robust assessment of the accuracy of the 12 km NLDAS product and additional accuracy gained from the 1 km downscaled product.