Soil salinization is a significant threat to agricultural production, making accurate salinity prediction essential. This study addresses key challenges in the Yellow River Delta (YRD) soil salinity inversion, including (1) determining which Landsat 8 OLI level performs better, (2) identifying the most suitable month for salinity inversion, and (3) improving model performance and identifying important variables in modeling. Thus Landsat 8 OLI images (Level-1 and Level-2) for 12 months were collected, then images having less than 10% cloud cover were selected and processed to extract spectral values. A total of 86 sampled points were processed to measure soil salinity. Using Pearson correlation and expert insights, January 15 and August 26 were identified as suitable dates for inversion. Then, seven original bands, 29 spectral indicators, and 39 derived variables which created through six mathematical transformations, were used to construct the following three models: partial least squares regression (PLSR), random forest (RF), and backpropagation neural network (BPNN). The results showed the following: (1) The Level-1 data, after FLAASH atmospheric correction, outperforms Level-2 data. (2) January is optimal for salinity inversion. (3) Among the three models, RF outperformed the others, achieving test set R² = 0.55, RMSE = 3.4, suggesting that the combination of spectral indicators and mathematically transformed variables can effectively enhance model accuracy for predicting soil salinity in the YRD. Furthermore, SWIR1, SWIR2, CLEX, second-order difference of SWIR1, and first-order difference of SWIR2 along with NIR played a key role in modeling.