The sensitivity of landscape evolution models to spatial and temporal rainfall resolution

Abstract

Abstract. Climate is one of the main drivers for landscape evolution models (LEMs), yet its representation is often basic with values averaged over long time periods and frequently lumped to the same value for the whole basin. Clearly, this hides the heterogeneity of precipitation -but what impact does this averaging have on erosion and deposition, topography, and the final shape of LEM landscapes? This paper presents results from the first systematic investigation into how the spatial and temporal resolution of precipitation affects LEM simulations of sediment yields and patterns of erosion and deposition. This is carried out by assessing the sensitivity of the CAESAR-Lisflood LEM to different spatial and temporal precipitation resolutions -as well as how this interacts with different-size drainage basins over short and long timescales. A range of simulations were carried out, varying rainfall from 0.25 h × 5 km to 24 h × Lump resolution over three different-sized basins for 30-year durations. Results showed that there was a sensitivity to temporal and spatial resolution, with the finest leading to > 100 % increases in basin sediment yields. To look at how these interactions manifested over longer timescales, several simulations were carried out to model a 1000-year period. These showed a systematic bias towards greater erosion in uplands and deposition in valley floors with the finest spatial-and temporal-resolution data. Further tests showed that this effect was due solely to the data resolution, not orographic factors. Additional research indicated that these differences in sediment yield could be accounted for by adding a compensation factor to the model sediment transport law. However, this resulted in notable differences in the topographies generated, especially in third-order and higher streams. The implications of these findings are that uncalibrated past and present LEMs using lumped and time-averaged climate inputs may be under-predicting basin sediment yields as well as introducing spatial biases through under-predicting erosion in first-order streams but overpredicting erosion in second-and third-order streams and valley floor areas. Calibrated LEMs may give correct sediment yields, but patterns of erosion and deposition will be different and the calibration may not be correct for changing climates. This may have significant impacts on the modelled basin profile and shape from longtimescale simulations.

Global sensitivity analysis of parameter uncertainty in landscape evolution models

“…This can be explained by three interrelated issues: (i) LEMs typically have a large number of model parameters; (ii) long model runtimes can make multiple simulations for SA impractical; and (iii) model behaviour can be highly non-linear (e.g. Coulthard and Van De Wiel, 2007;Larsen et al, 2014;Van De Wiel and Coulthard, 2010), leading to potentially complex SA interpretations. Large numbers of model parameters and long runtimes, in particular, make Monte Carlo methods extremely time consuming -and therefore often unviable.…”

“…There are several studies on how LEMs respond to variable forcing, process changes, and model parameters, including changes in climate variability and precipitation resolution (Armitage et al, 2018;Coulthard and Skinner, 2016;Ijjasz-Vasquez et al, 1992;Tucker and Bras, 2000), channel widths (Attal et al, 2008), vegetation (Collins, 2004;Istanbulluoglu and Bras, 2005), and variations in initial conditions (Hancock, 2006;Hancock et al, 2016;Ijjasz-Vasquez et al, 1992;Willgoose et al, 2003). Campforts et al (2017) investigated how different numerical solvers affect LEM simulation.…”

“…Conversely, investigations to ascertain the model's response to potential uncertainties (e.g from model parameterisation) can be deemed as true SA (e.g. Armitage et al, 2018;Coulthard and Skinner, 2016;Hancock et al, 2016).…”

“…Ziliani et al, 2013), results from LEMs are therefore frequently assessed qualitatively, relying on visual interpretation of the simulated landforms or cross-section profiles (e.g. Coulthard and Skinner, 2016;Hancock et al, 2010Hancock et al, , 2015Hancock and Coulthard, 2012).…”

“…Catchment outlet statistics, such as sediment yield time series, allow for comparison between simulations to indicate a catchment's response to perturbations (e.g. Coulthard and Skinner, 2016;Hancock and Coulthard, 2012). However, sediment yield time series rarely provide a sufficiently complete picture of a catchment's geomorphic response.…”