Drought assessment has been outpaced by climate change: empirical arguments for a paradigm shift

Accurate water availability monitoring is crucial for sustainable water resource management in the face of climate change and its impacts on human populations, food production, and ecosystem services. In the US, drought monitoring informs land managers, practitioners, and producers at local and regional scales, using various indicators (precipitation, evaporation, soil moisture, streamflow) to assess conditions across different timescales. National-scale drought monitoring, led by the National Drought Mitigation Center (NDMC), USDA, and NOAA, produces weekly drought maps forming the US Drought Monitor (USDM). The USDM categorizes drought severity, triggering emergency services and billions of dollars in federal disaster assistance annually. Most drought metrics rely on a statistical framework standardizing raw values into anomalies using probability statistics, often employing the Standardized Precipitation Index (SPI). A key consideration is the climatological length (reference period), as drought severity depends on this frame of reference, especially under climate change. While some studies advocate for 60+ years of observations for parameter stability, weather and climate scientists typically use 30-year climate normals, updated every decade. The paper argues that ignoring climate non-stationarity in operational drought metrics like the USDM leads to inaccurate assessments of current drought risk because "drought" conditions might simply be the "new normal" in a changing climate.

The literature review highlights the existing best practices for computing drought metrics, emphasizing the importance of the climatological length used as a reference frame. Many studies indicate that uncertainty in drought assessment declines with longer climatological records, with 60 years or more often recommended. However, this contrasts with the 30-year climate normal approach used by weather and climate scientists, which acknowledges and adapts to climate variability and change. The existing literature highlights the lack of consideration of non-stationarity in climate systems driven by anthropogenic climate change within operational drought metrics. The authors cite Milly et al. (2008), which argues that using long-term records in a non-stationary climate can lead to substantial biases in drought assessment. The authors also acknowledge that producers and decision-makers often rely on short-term memory and recent experiences for decision-making. Therefore, a mismatch between the historical reference frames used in drought monitoring and the shorter-term perspective of decision-makers adds to the existing challenges.

The study uses observed data from 1,934 GHCN sites across the contiguous US, focusing on precipitation dynamics. The analysis involves moving window operations, which consider only data from the period of interest. The GHCN data were processed to ensure quality and completeness, filtering for stations operating since 1950 with complete data until 2020. Data completeness was evaluated differently for 30-, 60-, and 90-day timescales. The core of the analysis involves fitting gamma probability distributions to aggregated annual precipitation time series data. Unbiased sample probability-weighted moments were computed, converted to quantile functions, and used to estimate gamma distribution parameters (rate and shape). The SPI was calculated using the cumulative distribution function (CDF) of the aggregated precipitation series. Drought metric bias was computed as the median difference between SPI values from long (70+ years) and short (30-year) periods. Monte Carlo simulations were conducted to evaluate SPI error with varying climatology lengths. Simulations included a stationary climate with a single parameter pair and 100 parameter pairs, and a non-stationary climate using 30-year moving window gamma parameters from 11 GHCN sites. The absolute SPI error was computed by comparing the SPI values from simulations to the true SPI values derived from the observed 30-year moving window distributions. The study employed inverse distance weighting techniques for bootstrap resampling for various period lengths.

The analysis reveals that traditional methods, which ignore climate change, lead to substantial biases in drought assessments. The study demonstrates that the assumption of an unchanging envelope of variability is flawed, as precipitation dynamics and PDFs are changing over time. Simulations show that in stationary climates, SPI error declines as climatology length increases from 2 to 30 observations, but shows little further decline at longer lengths. However, in non-stationary climates, the relationship between SPI error and climatology length is site and timescale specific. In some cases, error is minimized with fewer than 30 observations, while in others, error increases significantly with longer records. This contradicts the conventional wisdom that error always decreases with increasing climatology length. The analysis identifies significant differences in SPI values across the US when computed using long (70+ years) vs. short (30-year) climatologies, with these differences varying substantially across timescales. The study finds consistent spatial patterns of dry bias in the southwestern and southeastern US, particularly during periods of extreme drought, when longer periods of record were used. This dry bias was most pronounced with 30-day SPI calculations. Hot spots of dry bias, where longer record methodologies suggest drier conditions than 30-year climatologies, were prominent.

The findings highlight the limitations of using long-term historical records in drought assessment under non-stationary conditions. The study demonstrates that relying on long periods of record can lead to significant biases, particularly in regions experiencing rapid climate change. The results support the use of shorter (30-year) moving window approaches, which better reflect contemporary conditions and account for non-stationarity. This approach balances the statistical constraints of PDF parameterization while maintaining a relevant, contemporary reference frame. Using a 30-year moving window approach is consistent with existing practices and may better align with the short-term perspectives of decision-makers. While the relationship between SPI error and climatology length is site-specific, the results generally point to the limitations of using excessively long records for drought assessments in rapidly changing climates. This method also has implications for other drought metrics that use historical reference periods.

The study advocates for a paradigm shift in drought monitoring, emphasizing the need to account for climate non-stationarity. The findings support the adoption of a 30-year moving window approach to balance statistical requirements with the need for a contemporary reference frame. This approach represents a more accurate reflection of current drought risks and aligns better with the decision-making processes of stakeholders. While other computationally intensive approaches exist, the 30-year moving window is more easily integrated into existing infrastructure and drought monitoring practices. Further research and discussion are needed to refine drought assessment methodologies and better characterize risk in a changing climate.

The study primarily focuses on precipitation data and the SPI, acknowledging that other drought metrics and variables (temperature, evapotranspiration, streamflow) might exhibit different responses to non-stationarity. While the study addresses the importance of considering non-stationarity and updating reference periods for drought assessment, the results may be highly site specific, and require further research to understand the most effective approaches across diverse regions. Further analysis is also needed on using short-term memory in decision-making.