University of Twente Student Theses


Assessment of spatial pattern of soil salinity in coastal agricultural areas using multi-sensor approach

Salgado, Sarah Joey P. (2021) Assessment of spatial pattern of soil salinity in coastal agricultural areas using multi-sensor approach.

Link to full-text:
(only accessible for UT students and staff)
Abstract:Soil salinization is the process where water-soluble salts accumulate in the soil. It is considered as one of the most expensive soil degradation problems due to its high spatial and temporal variability. In the Philippines, there is no updated map available at the time of this study about the location, extent, and severity of areas affected by soil salinity. This research utilizes remote sensing data and ground-truth data in a machine learning algorithm (Random Forests regression) to detect and retrieve soil salinity in the rice areas of the Province of Ilocos Sur, Philippines. The province was selected as the study area due to increasing levels of soil pH that lead to salt accumulation. In addition, over-pumping of groundwater for irrigation and household use, salt-making, and extension of agricultural areas to non-suitable areas are the human activities that exacerbate this phenomenon. Moreover, the province is also one of the target areas that the Department of Agriculture – Bureau of Soils and Water Management plans to update soil salinity. The main objective of this study is to develop a method to detect the spatial pattern of soil salinity effectively. Furthermore, this study is geared up to recommend strategies for the stakeholder to combat and alleviate the adverse effects of soil salinization in the coastal rice areas of the Province of Ilocos Sur, Philippines. To achieve the research objective, satellite data from Sentinel-1, Sentinel-2, and Landsat-8 were obtained. Twenty features from Sentinel-1 were generated comprising Gamma-nought and Gray Level Co-occurrence Matrix (GLCM) bands in VV and VH polarization. Seventeen bands of vegetation and salinity indices were calculated from Sentinel-2, and the land surface temperature from the thermal bands of Landsat-8 was derived. Ancillary data composed of soil properties, climate, and geographical coordinates were also used as input for the Random Forests regression. Furthermore, two different approaches were adopted: optimized multi-sensor predictors, where the predictor variables are the collection of those variables within the importance threshold from the three different sensors individually, and multi-sensor predictors in which all variables from three different sensors are used to select the predictor variables based on the importance threshold. Accuracy assessment shows that the multi-sensor predictors method was better in detecting and retrieving soil salinity. The RMSE of this method is 0.15, R2 of 0.82, and Pearson correlation coefficient of 0.91, which indicates the excellent performance of this model. Spatial variability of predicted soil salinity shows that the coastal rice areas have higher soil salinity levels than those near mountains. Multiyear variability of soil salinity was predicted in 2017, 2018, and 2020, showing an increase of soil salinity in the area ranging from 0.04 to 0.14 decisiemens per meter (dS/m) in four years. It was found out that the integration of the three sensors is more efficient and more accurate in detecting and retrieving soil salinity than using a single sensor.
Item Type:Essay (Master)
Faculty:ITC: Faculty of Geo-information Science and Earth Observation
Programme:Spatial Engineering MSc (60962)
Link to this item:
Export this item as:BibTeX
HTML Citation
Reference Manager


Repository Staff Only: item control page