Spatial structure of precipitation in the Brazilian Amazonia: geostatistics with block kriging

Abstract

The Geostatistical analysis is an important tool for identifying distances over which rainfall shows spatial correlation. The objectives of the present work are: to model the spatial structure of rainfall patterns in the Brazilian rainforest using three timescales and to assess kriging interpolation for predicting rainfall data at ungauged sites. We used Geostatistical modeling combined with block kriging. The rainfall data were collected from 218 rain gauges corresponding to 64 municipalities located within 9 Amazonia states. Data corresponded to monthly mean rainfall and annual rainfall computed from historical records. The last date for data collection was December 2015. Three timescales were considered: months with minimum and maximum rainfall, January-June and July-December time periods and annual rainfall. The spherical model fitted reasonably well the semivariograms corresponding to March and January-June, the Gaussian model fitted quite well the semivariograms corresponding to September and July-December periods while the exponential model fitted the annual rainfall. The cross-validation analyses based on Mean Absolute Error and goodness-of-prediction statistics showed that kriged values (kriging maps) could be better predictors of rainfall for areas without rain gauges (unsampled zones) than mean rainfall values computed from adjacent sampled areas. It was found a significant negative statistical relationship between forest loss and rainfall occurrence which confirms the influence of deforestation on the hydrology of the Amazon basin.

Keywords

• Amazonian Rainforest
• Geostatistics
• Spatial Variability
• Hydrology