Arctic and subarctic regions are amongst the most susceptible regions on Earth to global warming and climate change. Understanding and predicting the impact of climate change in these regions require a proper process representation of the interactions between climate, carbon cycle, and hydrology in Earth system models. This study focuses on land surface models (LSMs) that represent the lower boundary condition of general circulation models (GCMs) and regional climate models (RCMs), which simulate climate change evolution at the global and regional scales, respectively. LSMs typically utilize a standard soil configuration with a depth of no more than 4 m, whereas for cold, permafrost regions, field experiments show that attention to deep soil profiles is needed to understand and close the water and energy balances, which are tightly coupled through the phase change. To address this gap, we design and run a series of model experiments with a one-dimensional LSM, called CLASS (Canadian Land Surface Scheme), as embedded in the MESH (Modélisation Environmentale Communautaire – Surface and Hydrology) modelling system, to (1) characterize the effect of soil profile depth under different climate conditions and in the presence of parameter uncertainty; (2) assess the effect of including or excluding the geothermal flux in the LSM at the bottom of the soil column; and (3) develop a methodology for temperature profile initialization in permafrost regions, where the system has an extended memory, by the use of paleo-records and bootstrapping. Our study area is in Norman Wells, Northwest Territories of Canada, where measurements of soil temperature profiles and historical reconstructed climate data are available. Our results demonstrate a dominant role for parameter uncertainty, that is often neglected in LSMs. Considering such high sensitivity to parameter values and dependency on the climate condition, we show that a minimum depth of 20 m is essential to adequately represent the temperature dynamics. We further show that our proposed initialization procedure is effective and robust to uncertainty in paleo-climate reconstructions and that more than 300 years of reconstructed climate time series are needed for proper model initialization.

Groundwater arsenic in Uruguay is an important environmental hazard, hence, predicting its distribution is important to inform stakeholders. Furthermore, occurrences in Uruguay are known to variably show dependence on depth and geology, arguably reflecting different processes controlling groundwater arsenic concentrations. Here, we present the distribution of groundwater arsenic in Uruguay modelled by a variety of machine learning, basic expert systems, and hybrid approaches. A pure random forest approach, using 26 potential predictor variables, gave rise to a groundwater arsenic distribution model with a very high degree of accuracy (AUC = 0.92), which is consistent with known high groundwater arsenic hazard areas. These areas are mainly in southwest Uruguay, including the Paysandú, Río Negro, Soriano, Colonia, Flores, San José, Florida, Montevideo, and Canelones departments, where the Mercedes, Cuaternario Oeste, Raigón, and Cretácico main aquifers occur. A hybrid approach separating the country into sedimentary and crystalline aquifer domains resulted in slight material improvement in a high arsenic hazard distribution. However, a further hybrid approach separately modelling shallow (50 m) resulted in the identification of more high hazard areas in Flores, Durazno, and the northwest corner of Florida departments in shallow aquifers than the pure model. Both hybrid models considering depth (AUC = 0.95) and geology (AUC = 0.97) produced improved accuracy. Hybrid machine learning models with expert selection of important environmental parameters may sometimes be a better choice than pure machine learning models, particularly where there are incomplete datasets, but perhaps, counterintuitively, this is not always the case.