Climate Clustering of Brazil Using Extreme Indices

Abstract

Brazil is a vast country with a diverse range of biomes and ecosystems. Due to this complexity, predicting the impact of extreme weather events in each region is challenging. Additionally, Brazil experiences a high frequency of extreme events, such as floods and droughts, which result from various climatic factors and have distinct effects on agriculture and society. To analyze the recurrence and intensity of these events, it is essential to cluster regions with similar historical patterns of extreme weather. This work calculated 46 monthly extreme indices - like spei [2], mean wind, total of precipitation, txn (less max temperature of the month), etc. - for each grid of 0.1^◦×0.1^◦ in Brazil, approximately 7700 points on your territory of 1961 to 2019. These indices are calculated using [4] dataset. To reduce the dimensionality and non-linear dependency of the data, Principal Component Analysis (PCA) is applied to the indices. After that, four strategies are used to cluster the regions: K-means on federal units with normalized original indices, k-means on federal units with PCA indices, k-means on hydrographic basins with normalized original indices and k-means on hydrographic basins with PCA indices [3]. The optimal number of clusters is found using the elbow method [1]. The results show that the number of clusters fluctuated between 5 and 8. The results of the clustering are shown in the Figure 1. Note that colors have no meaning or interpretation, they are used to differentiate the clusters. [...]