A Global Meta-Analysis of Soil Salinity Prediction Integrating Satellite Remote Sensing, Soil Sampling, and Machine Learning
Haiyang Shi, Olaf Hellwich, Geping Luo, Chunbo Chen, Huili He, Friday Uchenna Ochege, Tim Van de Voorde, Alishir Kurban, Philippe De Maeyer
Abstract
Despite the growing interest among researchers, satellite-based prediction of soil salinity remains highly uncertain. The improvements in prediction accuracy reported in previous studies are usually limited to a single area. We performed a meta-analysis of regional satellite-based soil salinity predictions combined with <i>in situ</i> soil sampling and machine learning. Based on <inline-formula> <tex-math notation="LaTeX">$R^{2}$ </tex-math></inline-formula> and root-mean-square error (RMSE) collected, we evaluated the effects of various features on the model accuracy and established a Bayesian network to evaluate the joint causal effect of multifeatures. Most significant differences were found in soil sampling schemes and characteristics of the study area, including the mean and variability (averaged <inline-formula> <tex-math notation="LaTeX">$R^{2}$ </tex-math></inline-formula> of 0.75 for soil sample sets with lower salinity variation and 0.62 for others) of the salinity, climate type (<inline-formula> <tex-math notation="LaTeX">$R^{2}$ </tex-math></inline-formula> of 0.64 in arid areas and 0.74 in others), soil texture (<inline-formula> <tex-math notation="LaTeX">$R^{2}$ </tex-math></inline-formula> of 0.66 in sandy areas and 0.57 in others), and the interval between sampling date and satellite data acquisition date (<inline-formula> <tex-math notation="LaTeX">$R^{2}$ </tex-math></inline-formula> of 0.53 under the condition of over 15 days and 0.65 in others). Generally, using different satellite data has limited effects on model performance among which Sentinel-2 performed better (<inline-formula> <tex-math notation="LaTeX">$R^{2} $ </tex-math></inline-formula> = 0.72) than Landsat (<inline-formula> <tex-math notation="LaTeX">$R^{2} $ </tex-math></inline-formula> = 0.66). The sampling of subsamples for each sample should focus on their subpixel-scale spatial heterogeneity across satellite data rather than the number of subsamples. It is also necessary to select appropriate vegetation and salinity indices for different satellite data under different vegetation conditions. Among algorithms, random forests (<inline-formula> <tex-math notation="LaTeX">$R^{2} $ </tex-math></inline-formula> = 0.70) and support vector machines (<inline-formula> <tex-math notation="LaTeX">$R^{2} $ </tex-math></inline-formula> = 0.71) performed best.