Investigating Long-Range Seasonal Predictability of East African Short Rains: Influence of the Mascarene High on the Indian Ocean Walker Cell

Abstract

We investigate the predictability of East African short rains at long (up to 12 month) lead times by relating seasonal rainfall anomalies to climate anomalies associated with the predominant Walker circulation, including sea surface temperatures (SST), geopotential heights, zonal and meridional winds, and vertical velocities. The underlying teleconnections are examined using a regularized regression model that shows two periods of high model skill (0–3-month lead and 7–9-month lead) with similar spatial patterns of predictability. We observe large-scale circulation anomalies consistent with the Walker circulation at short lead times (0–3 months) and dipoles of SST and height anomalies over the Mascarene high region at longer lead times (7–9 months). These two patterns are linked in time by anticyclonic winds in the dipole region associated with a perturbed meridional circulation (4–6-month lead). Overall, these results suggest that there is potential to extend forecast lead times beyond a few months for drought impact mitigation applications.

Skilful long‐lead hybrid predictions of the East African short‐rains season

“…The CHIRPS data are very high resolution (0.05 × 0.05 ) daily data. CHIRPS has been widely used for climate study and prediction in East Africa including by Bahaga et al (2019), Peng et al (2020), andMacleod et al (2021). The CHIRPS rainfall performed very well when compared with in situ observations, likely because of its direct inclusion of rain gauge data and microwave images during calibration (Kimani et al, 2017).For this study we generated seasonal-average data for the short-rains (October-December, OND) season.…”

Investigating Predictability of the TRHR Seasonal Precipitation at Long Lead Times Using a Generalized Regression Model with Regularization

“…SST is selected as the primary predictor since it can indicate perturbations in large-scale atmospheric circulations "anchored" in ocean memory (Xie et al, 2009, Xie et al, 2016. Also, the SST field is less spatially heterogeneous compared to that of other common climate variables including geopotential height, vertical velocity of atmosphere (OMEGA) and wind velocities (Peng et al, 2020), which can help improve robustness of the regression models. Monthly SST data is collected from the Hadley Centre Sea Ice and Sea Surface Temperature (HadISST) data set with a spatial resolution of 1°by 1° (Rayner et al, 2003) over Jan 1979-Dec 2019.…”

“…There are two hyperparameters in the model: α and λ. α balances the regularization between the L-1 and L-2 norms of the regression coefficients β and is set to 0.01 for better visualization (Peng et al, 2020). λ is usually decided using a k-fold (e.g., 5-fold) cross validation (CV) (Tibshirani, 1996) and the λ value associated with minimum cross-validated mean squared errors is used (often referred to as the MinMSE λ).…”

“…While new machine learning techniques like the classification and regression tree (CART) can have better model skill in multi-class prediction of climate variables (Choubin et al, 2018;Huang et al, 2021), the elastic net logistic regression is tested to demonstrate flexibility and consistency with altered deviance functions. The multinomial regression may have more use in practical application since amplitudes of the predictand tend to be underestimated because of the regularization (Peng et al, 2020). The logistic regression is implemented by simply replacing the MSE function with the log-likelihood function for Dev (Hastie et al, 2009).…”

“…Here, we propose to use the elastic net regularization (Zou and Hastie, 2005) which linearly combines the LASSO and ridge regression regularizations. While the LASSO regularization guarantees sparsity of the model, the ridge regression regularization helps improve visualization of the regression coefficient map and therefore, interpretability of the model (Peng et al, 2020). On top of that, a pooling layer is…”