A statistical assessment of drought variability and climate prediction for Kansas

Abstract

The high-quality climate data and high-resolution soil property data in Kansas and adjacent states were used to develop drought datasets for the monthly Palmer Drought Severity Index (PDSI), Standardized Precipitation Index (SPI), and the Standardized Precipitation-Evapotranspiration Index (SPEI) over 1900 to 2014. The statistical analysis of these multiple drought indices were conducted to assess drought occurrence, duration, severity, intensity, and return period. Results indicated that the PDSI exhibited a higher frequency for every category of drought in central and western Kansas than the SPEI by up to 10%. Severe and extreme drought frequency was the highest in southwest Kansas around the Arkansas River lowlands and lowest in the southeast. The mean total drought frequency for eastern, central, and western Kansas was 36%, 39%, and 44%, respectively. The regional mean correlations between the SPI and SPEI were greater than or equal to 0.95 for all regions, but due to statistically significant increases in potential evaporation in western Kansas, the PDSI and SPEI are recommended over the SPI for meteorological and hydrological drought analysis. Drought variability of the last 115 years was analyzed through the Empirical Orthogonal Functions (EOFs) techniques and their Varimax rotations from 1900 to 2014 in Kansas. Large-scale synoptic patterns primarily dominated the Kansas spatial drought structures, especially during long-duration events. The EOFs indicated that the first principal components of drought explained approximately 70% of the drought variability across the state and demonstrated a statistically significant wetting trend over the last 115 years, oscillating at a period of about 14 years for all drought indices. The 99° W meridian line acted as the dominant transitional line demarcating the areas of Kansas’ climate and vegetation relationship as spatial drought presented. The Multivariate El Nino Index (MEI) signal , which modulates global and regional climate variabilities, provided a low-frequency indicator to couple with Kansas drought’s leading modes by varying leads of 3 to 7 months depending on the use of drought index and time steps selected. Large-scale predictors of surface temperature and precipitation are evaluated from the monthly forecasts in Climate Forecast System version 2.0 (CFSv2) from North Dakota down through central Texas (32.6 - 47.7°N and 92.8 - 104.1°W). By using singular value decomposition (SVD), the CFSv2 monthly forecasts of precipitation and 2-m temperature were statistically downscaled using ensemble mean predictions of reforecasts from 1982-2010. Precipitation skill was considerably less than temperature, and the highest skill occurred during the wintertime for 1-month lead time. Only the central and northern plains had statistically significant correlations between observed and modeled precipitation for 1-month lead time. Beyond a 1-month lead time, prediction skill was regionally and seasonally dependent. For the 3-month lead time, only the central plains demonstrated statistically significant mean anomaly correlation. After three-month lead times, the ensemble means of forecasts have shown limited reliable predictions which could make the forecast skill too low to be useful in practice for precipitation. However, temperature forecasts at lead times greater than five months showed some skill in predicting wintertime temperatures.