A combined geostatistical approach of data fusion and stochastic simulation for probabilistic assessment of shallow water table depth risk

Sci Total Environ. 2021 Apr 15;765:142743. doi: 10.1016/j.scitotenv.2020.142743. Epub 2020 Oct 14.


In general, water table depth risks are estimated from monitoring networks that mostly provide scarce and irregular data. When jointly analysed, environmental, agricultural and geotechnical variables, treated as stochastic spatial variables, can better describe and interpret the states of a certain system subject to estimation uncertainty. Risk assessment consists essentially in calculating the frequency (probability) with which specified criteria are exceeded or fail to be met by creating multiple stochastic realizations. The aim of this paper is to propose a novel geostatistical methodology, based on the integration into one approach of multi-source data fusion and stochastic simulation, to estimate the risk of extreme (shallow) water table depth, and illustrate a demonstrative example of application of the approach to a case study in a Cerrado conservation area in Brazil. The risk of shallow water table depth was determined by using critical thresholds for water table level and a binary transformation into an indicator variable depending on whether the conditions expressed by the threshold values are met or not. Firstly, auxiliary variables were jointly, analysed to provide a delineation of the study area into homogeneous zones. Secondly, sequential indicator simulation provided a-posteriori probabilities taking into account spatial proximity. The final maps show the most probable risk category for the whole area and spatial entropy as a measure of local uncertainty. Areas nearby watershed divisors and in the north part of the region have a high risk of shallow groundwater. Informed decision-making supported by probabilistic maps and uncertainty evaluation is essential for the success of the projects of Cerrado restoration.

Keywords: Cerrado restoration; Factor block cokriging; Sequential indicator simulation; Simple kriging with varying local means; Spatial entropy.