Constructing regional climate networks in the Amazonia during recent drought events

Abstract

Climate networks are powerful approaches to disclose tele-connections in climate systems and to predict severe climate events. Here we construct regional climate networks from precipitation data in the Amazonian region and focus on network properties under the recent drought events in 2005 and 2010. Both the networks of the entire Amazon region and the extreme networks resulted from locations severely affected by drought events suggest that network characteristics show slight difference between the two drought events. Based on network degrees of extreme drought events and that without drought conditions, we identify regions of interest that are correlated to longer expected drought period length. Moreover, we show that the spatial correlation length to the regions of interest decayed much faster in 2010 than in 2005, which is because of the dual roles played by both the Pacific and Atlantic oceans. The results suggest that hub nodes in the regional climate network of Amazonia have fewer long-range connections when more severe drought conditions appeared in 2010 than that in 2005.

Fig 1. Normalized drought strength over Amazon region.

(A) 2005, (B) 2010, and (C) probability distribution functions of normalized drought length.

Fig 2. Network measures and gray areas highlight the austral summer periods.

(A) link density ρ, (B) transitivity $\mathcal{T}$, (C) clustering coefficient $\mathcal{C}$, (D) average path length $\mathcal{L}$, (E) assortativity $\mathcal{R}$ and (F) modularity $\mathcal{Q}$. Gray time windows highlight the summer seasons.

Fig 3. Distribution of 50% large values of drought strength across the Amazon region in the year (A) 2005, and (B) 2010.

Fig 4. Network measures of extreme drought networks.

Gray areas highlight the austral summer periods (• spatial points of 2005, and ° spatial points of 2010). (A) link density ρ, (B) transitivity $\mathcal{T}$, (C) clustering coefficient $\mathcal{C}$, (D) average path length $\mathcal{L}$, (E) assortativity $\mathcal{R}$ and (F) modularity $\mathcal{Q}$. Gray time windows highlight the summer seasons.

Fig 5. Maps of averaged NDS values.

Locations of (A) 50% upper quantile and (B) 50% lower quantile of NDS values.

Fig 6. Map of the averaged network degrees (normalized to unit interval) for locations that are characterized by 50% upper quantile of NDS values (A), and (B) 50% lower quantile of NDS values.

The ROIs are denoted by ×.

Fig 7. Map of correlation coefficients to the ROI at (58°W, 8.5°S) as indicated by ×, where large network degrees are observed in the extreme drought networks.

(A) austral summer 2005, (B) winter 2005, (C) summer 2010, and (D) winter 2010.

Fig 8. Map of correlation coefficients to the ROI at (65°W, 5.5°S) as indicated by ×, where large network degrees are observed in the network of normal drought conditions.

(A) austral summer 2005, (B) winter 2005, (C) summer 2010, and (D) winter 2010.

Fig 9. Correlation distances to the ROI at (58°W, 8.5°S) in the extreme drought network (A, B), and respectively, to the ROI at (65°W, 5.5°S) in the network of normal drought conditions (C, D).

(A, C) summer, and (B, D) winter. Error-bars correspond to the standard deviations.

Fig 10. Delays of the ocean effects on the precipitations in the Amazon.

(A) 2005, no significant delay effects have been identified from the ENSO region, and (B) 2010.