The temperature regime of soil under 11 different mulching substrates was investigated during the field experiment at the soil station of Lomonosov Moscow State University. Four mineral (serpentine, marble, volcanic tuff, and foam glass) and seven organic (coconut chips, larch, and pine bark of fine and coarse fractions, uncolored and colored pine chips) mulching materials were selected for the study. The layer thickness was 5 cm at all sites. Soil temperatures were obtained at three depths (0 cm (directly under the mulch), 5, and 10 cm) from November 2021 to May 2023. The obtained data was used to train the state-of-the-art CatBoost model. Meteorological data were used as predictive parameters. Thus, after teaching the first iteration of the model, it was possible to recover the dropped temperature data. Subsequently, the next iteration of the model was trained to predict further soil temperatures based on meteorological data. For this purpose, the original database was split into three samples: training (from November 12, 2021, to February 20, 2023); validation (from February 20, 2023, to June 30, 2023; used for best iteration selection); and testing (from June 30, 2023, to September 29, 2023). The model has shown its effectiveness in predicting the temperature of urban soil, so the indicators of its accuracy on the test sample were as follows: R2 was 0.74; mean square error (MSE) was 2.96. At the same time, on the validation sample, they were as follows: R2 was 0.91; MSE was 4.31.