In my experience, the temp dataset is always left as the default value (cloud_dataflow). Instead, the necessary permissions are granted to the service account to enable the job to read data from BigQuery e.g. roles/bigquery.dataEditor role.