A semi-objective circulation pattern classification scheme for the semi-arid Northeast Brazil

The semi-arid Northeast Brazil (NEB) is just recovering from a very severe water crisis induced by a multiyear drought. With this crisis, the question of water resources management has entered the national political agenda, creating an opportunity to better prepare the country to deal with future droughts. In order to improve climate predictions, and thus preparedness in NEB, a circulation pattern (CP) classification algorithm offers various options. Therefore, the main objective of this study was to develop a computer aided CP classification based on the Simulated ANnealing and Diversified RAndomization clustering (SANDRA) algorithm. First, suitable predictor variables and cluster domain setting are evaluated using ERA-Interim reanalyses. It is found that near surface variables such as geopotential at 1,000 hPa (GP(1,000)) or mean sea level pressure (MSLP) should be combined with horizontal wind speed at the upper 700 hPa level (UWND700). A 11-cluster solution is favoured due to the trade-offs between interpretability of the cluster centroids and the explained variances of the predictors. Second, occurrence and transition probabilities of this 11-cluster solution of GP(1,000) and UWND700 are analysed, and typical CPs, which are linked to dry and wet conditions in the region are identified. The suitability of the new classification to be potentially applied for statistical downscaling or CP-conditional bias correction approach is analysed. The CP-conditional cumulative density functions (CDFs) exhibit discriminative power to separate between wet and dry conditions, indicating a good performance of the CP approach.

Automated summary

The semi-arid Northeast Brazil has recently faced a severe water crisis due to a multiyear drought, prompting the need for improved climate predictions and water resource management. A computer aided CP classification based on the SANDRA algorithm was developed, analyzing occurrence and transition probabilities of an 11-cluster solution of GP(1,000) and UWND700 to identify typical CPs linked to dry and wet conditions in the region. The new classification shows potential for statistical downscaling and CP-conditional bias correction, with CP-conditional cumulative density functions demonstrating good performance in separating wet and dry conditions.