Climate-driven changes in the predictability of seasonal precipitation

Precipitation is crucial for the hydrological cycle, biodiversity, and socioeconomic systems. Improved earth system models (ESMs) and climate observations have enhanced seasonal precipitation prediction, a vital tool for water and food security and ecological restoration. Seasonal precipitation predictability is linked to large-scale sea surface temperature (SST) variability, a major driver of atmospheric circulation. Studies have demonstrated SST anomaly patterns' influence on precipitation variability in various regions, including Australia, Africa, Asia, and Amazonia. However, climate modes and teleconnections are expected to change by the end of the 21st century, impacting precipitation patterns. This study examines how these changes affect seasonal precipitation predictability using CMIP6 models, focusing on the predictive skill of linear models using lagged SST as predictors. The study's importance lies in its contribution to understanding future changes in seasonal precipitation predictability and its implications for water resource management.

Existing research highlights the relationship between SST and precipitation predictability across various regions. Studies have demonstrated the influence of Indian Ocean SSTs on Australian, African, and Asian rainfall. Similarly, Atlantic and Pacific SST anomalies affect Amazonian precipitation. Pacific SST patterns are used to predict winter precipitation in the southern US. However, past work mainly focused on understanding climate modes and teleconnections, with limited investigation into the consequences of their change for seasonal precipitation predictability. This paper bridges this gap by analyzing future changes in this predictability in response to climate change.

The study uses a combination of observational data and CMIP6 model simulations. For the observational analysis (1964-2014), empirical orthogonal function (EOF) analysis was performed on global seasonal SST anomalies to extract principal components (PCs) as predictors. Multiple linear regression models were then developed to predict seasonal precipitation anomalies using these PCs, with the Nash-Sutcliffe Efficiency (NSE) used as a skill metric. A 5-fold cross-validation approach was employed to avoid overfitting. The best-performing models were identified, using the best combination of two PCs to predict precipitation for each grid cell. CMIP6 models were then selected based on their ability to capture the historical spatial patterns of precipitation predictability. These selected models were used to project future changes in predictability under SSP2-4.5 and SSP3-7.0 scenarios (2049-2099). The changes in precipitation predictability were quantified as the difference in NSE between future and historical periods. The study also investigated changes in SST and precipitation variability between the two periods to understand the mechanisms driving the changes in predictability.

The analysis of observational data revealed strong seasonal variations in precipitation predictability, with boreal winter being the most predictable season. High predictability was observed in regions like Amazonia during boreal winter, northwestern and southern US during boreal winter, and northern Australia during boreal autumn. The Maritime Continent also showed relatively high predictability across all seasons. The best-performing predictive model consistently used two PCs, often related to ENSO, PDO, AMO, and NPGO. The CMIP6 models showed that by the end of the century (SSP3-7.0), tropical precipitation predictability will increase except for northern South America and eastern India. Significant increases are projected in the Maritime Continent during boreal summer and central Africa during boreal winter. Central Asia will also experience increased predictability during boreal spring and winter. SSP2-4.5 showed similar spatial patterns but with smaller magnitudes. Analysis of SST and precipitation variability showed increased SST variability in several regions, potentially driving the changes in predictability. Northern South America is particularly vulnerable, with projected dry season precipitation declines, increased variability, and decreased predictability, increasing wildfire risk. Central and northern Africa, and the Maritime Continent may see enhanced resilience due to increased predictability.

The findings address the research question by demonstrating that climate change significantly alters SST-precipitation relationships, consequently impacting seasonal precipitation predictability. The increased interannual variability of SST and precipitation contributes to the observed changes. The study's significance lies in identifying regional hotspots where changes in predictability will severely challenge water resource management, especially in areas where water is scarce or where precipitation is already extremely high or low. The increased predictability in certain regions might offer opportunities for improved resource management, though these opportunities should be considered alongside the significant challenges posed by increased variability and decreased predictability elsewhere. The results emphasize the need for improved understanding of causal SST-precipitation relationships to enhance seasonal precipitation prediction and ensure sustainable water resource management in a changing climate.

This study demonstrates that climate change is projected to alter the predictability of seasonal precipitation, impacting regional water management. Increased interannual variability and changes in SST-precipitation relationships will drive these changes. Hotspots of both increased and decreased predictability were identified, highlighting the need for improved understanding of these relationships for effective climate adaptation. Future research should focus on dissecting the causal mechanisms behind these changes and improving prediction models to account for these changes.

The study relies on CMIP6 models, which have inherent uncertainties. The analysis focuses on linear relationships between SST and precipitation, potentially neglecting non-linear interactions. The study's focus on SST as a predictor might not capture all sources of precipitation predictability. The selection of the 10 best-performing CMIP6 models might introduce bias in the results.