A Comparison of the CHILD and Landlab Computational Landscape Evolution Models and Examples of Best Practices in Numerical Modeling of Surface Processes

Friday, 19 December 2014
Nicole M Gasparini1, Daniel E. J. Hobley2, Gregory E Tucker3, Erkan Istanbulluoglu4, Jordan Marie Adams5, Sai Siddhartha Nudurupati4 and Eric W.H. Hutton6, (1)Tulane University of Louisiana, New Orleans, LA, United States, (2)Univ of Colorado, Boulder, CO, United States, (3)Univ Colorado, Boulder, CO, United States, (4)Univ of Washington, Seattle, WA, United States, (5)Tulane University, New Orleans, LA, United States, (6)Community Surface Dynamics Modeling System, Boulder, CO, United States
Computational models are important tools that can be used to quantitatively understand the evolution of real landscapes. Commonalities exist among most landscape evolution models, although they are also idiosyncratic, in that they are coded in different languages, require different input values, and are designed to tackle a unique set of questions. These differences can make applying a landscape evolution model challenging, especially for novice programmers. In this study, we compare and contrast two landscape evolution models that are designed to tackle similar questions, but the actual model designs are quite different. The first model, CHILD, is over a decade-old and is relatively well-tested, well-developed and well-used. It is coded in C++, operates on an irregular grid and was designed more with function rather than user-experience in mind. In contrast, the second model, Landlab, is relatively new and was designed to be accessible to a wide range of scientists, including those who have not previously used or developed a numerical model. Landlab is coded in Python, a relatively easy language for the non-proficient programmer, and has the ability to model landscapes described on both regular and irregular grids. 

We present landscape simulations from both modeling platforms. Our goal is to illustrate best practices for implementing a new process module in a landscape evolution model, and therefore the simulations are applicable regardless of the modeling platform. We contrast differences and highlight similarities between the use of the two models, including setting-up the model and input file for different evolutionary scenarios, computational time, and model output. Whenever possible, we compare model output with analytical solutions and illustrate the effects, or lack thereof, of a uniform vs. non-uniform grid. Our simulations focus on implementing a single process, including detachment-limited or transport-limited fluvial bedrock incision and linear or non-linear diffusion of material on hillslopes. We also illustrate the steps necessary to couple processes together, for example, detachment-limited fluvial bedrock incision with linear diffusion on hillslopes. Trade-offs exist between the two modeling platforms, and these are primarily in speed and ease-of-use.