Permeability values are estimated based on well logs using regression algorithms, such as gradient boosting and random forest. The training data is selected from well logs for which core-analysis-based permeability values are available. The estimated permeability values are used to plan hydrocarbon production. The well logs used to build the depth blended model may include total porosity, gamma ray, volume of calcite, density, resistivity, and neutron logs. Selecting the training data may include grouping the well logs according to regions expected to have similar characteristics, choosing a subset of the well logs corresponding to wells expected to provide stable models according to pre-determined criteria, and/or identifying training zones on the chosen well logs according to one or more rules. Validation and consistency checks may also be performed.