In this paper we consider continuous-time local model networks (LMN) to model dynamical processes with strong nonlinearities. The local model approach allows for simple black-box identification procedures using experimental data. Using the LoLiMoT algorithm the number of models can be significantly reduced and may yield insights into the nonlinearities driving the process. We propose a variation of the LoLiMoT algorithm that partitions the operating range in a more efficient manner and proves particular suited for heterogenous nonlinearities.