Konferans bildirisi Açık Erişim

State of art approaches, insights, and challenges for digital mapping of electrical conductivity as a dynamic soil property

2022    Kaya, Fuat; Ferhatoğlu, Caner; Turgut, Yavuz Şahin; Başayiğit, Levent

Soil electrical conductivity (EC) as a measure of soil salt content is a good indicator of nutrient and water availability or excessiveness in soils, which in return affect the productivity of soils. Therefore, mapping the spatial distribution of EC under intense agricultural management is important for managing soil fertility (e.g., fertilization and soil salinity remediation). However, mapping soil EC with high accuracy and spatial resolution remains to be challenge among digital soil mappers due to being a highly dynamic soil property. In this study, random forest (RF) was applied to map soil EC in an agricultural plain around the lake Manyas in the northwestern Türkiye. Fifty soil samples and a unique set of environmental predictors (aka covariates) were used to build a predictive soil EC model. The covariates were produced from Sentinel-2 optical satellite images-based vegetation and salinity indices as well as produced from Sentinel-1 with different polarizations (i.e., VV and VH), and terrain attributes representing the topography at varying scales were produced. Twelve environmental variables were selected to be relevant to predicting soil EC after using a correlation-based feature selection procedure. Resulting model performance was evaluated by root-mean-square-error (RMSE) of 10-fold cross-validation (CV). RF predicted soil EC with an RMSE of 0.07 dS m-1. Per each soil prediction in the final soil EC map, an uncertainty map was created using a sensitivity-based approach. The uncertainty map revealed the areas that were more difficult to accurately predict. Present study successfully mapped soil EC with acceptable error and can provide useful insights for managing soil fertility. In addition, an uncertainty map of soil EC can facilitate future soil sampling campaigns. For nearly a quarter of a century, while satellite-based remote sensing data has become the first choice for generating and updating soil survey information, in the near future, artificial intelligence techniques (e.g., ML) will be able to accompany soil surveyors in drawing map boundaries, especially in updating soil salinity phases.