Long‐Lead Drought Forecasting Across the Continental United States Using Burg Entropy Spectral Analysis With a Multiresolution Approach

Abstract

AbstractAmid climate change, extreme droughts have occurred more frequently, accompanied by increasing socio‐economic‐environmental losses. To be prepared for upcoming droughts, proactive drought management planning based on drought forecasting is being highlighted. This study, therefore, developed a long‐term drought forecasting model using Burg entropy spectral analysis with the maximal overlap discrete wavelet transform across the continental United States (CONUS). The model forecasted monthly self‐calibrated Palmer Drought Severity Index (scPDSI) up to median lead times of 12 months with root mean square error less than 0.8 and the modified Nash‐Sutcliffe efficiency coefficient greater than 0.6 across CONUS. The lead‐time with an acceptable accuracy was affected by how the model parameterized low‐frequency structure of scPDSI that was significantly correlated to sea surface temperature climate mode. The correlation between low‐frequency signals and Niño‐3.4 tended to be weaker for the regions located at higher elevations. Thus, higher variability in spectral structures of low‐frequency signals at higher elevation‐regions led to shorter memory/persistency of time series that shortened the forecast lead time. Besides, higher variability in precipitation linked to the variability of scPDIS also shortened the forecast lead time. To appraise the applicability of the suggested forecast model to drought management planning, drought characteristics (i.e., severity, duration, and area) were obtained by the three‐dimensional drought clustering algorithm for the forecasted and observed scPDSI. Then, risk was analyzed for drought characteristics using the vine copula. Risk analysis for drought characteristics forecasted up to a 12‐month lead time showed the potential for applicability of the model to drought mitigation plans.