Space time modeling for drought classification and prediction

TL;DR

This paper introduces a nonstationary SPI framework using GAMs to better model and predict droughts under changing climate conditions, capturing spatiotemporal variability and extreme events.

Contribution

It presents a novel nonstationary SPI method with GAMs and flexible tail models, advancing drought prediction accuracy amid climate change.

Findings

Model accurately reproduces nonstationary drought patterns

Provides stable estimates of drought extremes

Enhances understanding of drought dynamics under climate variability

Abstract

The Standardized Precipitation Index (SPI) is a critical tool for monitoring drought conditions, typically relying on normalized accumulated precipitation. While longer historical records of precipitation yield more accurate parameter estimates of marginal distribution, they often reflect nonstationary influences such as anthropogenic climate change and multidecadal natural variability. Traditional approaches either overlook this nonstationarity or address it with quasi-stationary reference periods. This study introduces a novel nonstationary SPI framework that utilizes generalized additive models (GAMs) to flexibly model the spatiotemporal variability inherent in drought processes. GAMs are employed to estimate parameters of the Gamma distribution, while dual extreme tails flexible models are integrated to robustly capture the probabilistic risk of extreme drought events. Future drought and wet extremes events in terms of return levels are calculated using an extended generalized Pareto distribution, which offers flexibility in modeling the entire distribution of the data while bypassing the threshold selection step. Results demonstrate that the proposed nonstationary SPI model is both stable and capable of reproducing known nonstationary drought patterns, while also providing new insights into the evolving dynamics of drought. This approach represents a significant advancement in drought modeling under changing climatic conditions.