Publications

Manzione, RL; Castrignano, A (2019). A geostatistical approach for multi-source data fusion to predict water table depth. SCIENCE OF THE TOTAL ENVIRONMENT, 696, UNSP 133763.

Abstract
Accurate water table depth mapping is important for water management and activity planning. The joint use of exhausted geospatial raster data with sparse field measurements could improve predictions. The aim of this work was to fuse different support data, collected with remote sensors, with point soil field observations to improve water table depth prediction. A method for multi-source data fusion is described in detail, based on multivariate geostatistics and exemplified with a case study in a conservation area of 5700 ha in the state of Sao Paulo, Brazil. TanDEM-X digital surface model with 90 m resolution and SAFER (Simple Algorithm for Evapotranspiration Retrieving) data calculated from Sentinel-2 images with 20 m resolution, were jointly used with water table depth and soil physical variables measured at 56 locations to predict water table depth in two hydrological years (2015-16 and 2016-17). Data were transformed to normal distributions using the Gaussian anamorphosis approach. A Linear Model of Coregionalization (LMC), calculated for all direct and cross-variograms of the eleven variables of study, was regularized at block support for multi-collocated block cokriging predictions. Support change correction was made to reduce punctual variance to block variances. Univariate and multivariate geostatistical interpolation methods were compared through cross validation. The uncertainty associated to the water table depths estimated by multivariate approach was lower than those by the univariate approach. Moreover, multivariate predictions incorporated the influences induced by local relief, vegetation and soil properties. Confidence interval maps, presented as uncertainty measure, reveal areas with higher and lower precision of groundwater level prediction that could be effectively used as support in land use management. (C) 2019 Elsevier B.V. All rights reserved.

DOI:
10.1016/j.scitotenv.2019.133763

ISSN:
0048-9697