Search

The Arctic is at the forefront of global climate change, warming at a rate 4 times faster than the global average. Changes are particularly apparent during winter, which lasts for around 8 months of year. CO2 emissions at this time of year make a considerable contribution to the annual Arctic carbon budget, with enhanced soil CO2 losses due to winter warming exceeding growing season carbon uptake under future climatic conditions. However, high uncertainty surrounds estimates of winter CO2 fluxes across the Arctic region, which vary by a factor of three and a half, with considerable variation between measured and simulated fluxes. Simulations of the Community Land Model (CLM5.0) were examined to address uncertainties impacting simulations of Arctic carbon fluxes; firstly of snow insulation and soil temperatures, and secondly rates of net ecosystem exchange of CO2. Improving simulated snow properties and the corresponding heat flux is important, as wintertime soil temperatures are an important control on subnivean soil respiration and hence impact Arctic winter carbon fluxes. The default model configuration was found to be inappropriate, with a change to the parameterisation of snow thermal conductivity required in order to correct for a cold soil temperature bias and better represent the insulation provided to the soil by the basal depth hoar not physically represented in CLM5.0. Simulated CO2 emissions were then examined through sensitivity testing of the parameterisation of relationships between soil temperature, moisture and respiration as well as snow thermal conductivity. The default value of the minimum soil moisture threshold for decomposition prevented soil respiration for the majority of the winter, with no CO2 emission simulated between November and mid-May, in contrast to observations showing steady CO2 emission throughout the winter. Failure to simulate CO2 emissions for over half of the year has a considerable impact on the simulation of the annual Arctic carbon budget. Changes to the model parameterisation, most crucially a decrease in the minimum moisture required for soil decomposition, allowed emissions to occur throughout the winter. Maintaining continuous wintertime eddy covariance measurements, typically used for model evaluation, in remote Arctic locations is very challenging. Consequently, as other measurement techniques would be advantageous, preliminary investigations with new low-cost CO2 sensors are presented and evaluated. Low-cost sensor measurements are used to examine spatial and temporal patterns in CO2 flux, and how these vary with changes in snow properties across the footprint of the eddy covariance tower. Measurements of CO2 flux using both new and established methods show similar magnitudes of CO2 release during March and April 2022, and asimilar relationship between CO2 and subnivean temperatures as previously shown across the Pan-Arctic. Future development of low-cost sensors should allow uncertainties to be reduced and an improvement in our understanding of the processes governing carbon fluxes from Arctic environments throughout the snow-covered season.

Forest canopies are complex 3-D structures at the interface between the atmosphere and the land surface which greatly affect radiative processes, especially across seasonally snow-covered domains. The increase in process complexity associated with these areas continues to be a source of uncertainty in land surface modelling. Model development has been hampered by a limited amount of in-situ radiation measurements, together with contrasting spatial scales of measurements and model resolution. Here, a bespoke cable car system was used to measure incoming and outgoing shortwave and longwave radiation below an evergreen forest stand, while an uncrewed aerial vehicle (UAV) system, equipped with up- and down-looking shortwave radiation sensors, was used to measure land surface albedo (LSA) above alpine, sub alpine and boreal forest stands. These in-situ measurements were combined with point-scale simulations of the Community Land Model Version 5.0 (CLM5), enabling process level assessment of algorithms used within global climate modelling frameworks. Analysis of diurnal radiation patterns, obtained via the cable car and UAV systems, revealed canopy structural shading of the snow surface as a main control on both the sub-canopy shortwave radiation budget and overall LSA. Furthermore, diurnal patterns of measured LSA revealed a strong dependency on both solar azimuth and zenith angles. Corresponding CLM5 simulations did not adequately represent the measured spatial and temporal variability in LSA and sub-canopy incoming shortwave radiation. In sparse forested environments, CLM5 performed especially poorly, as LSA was overestimated by up to 66%. The use of effective Plant Area Index (PAI) values as a simple first-order correction for this discrepancy between measured and simulated LSA substantially improved model results (64-76% RMSE reduction). That being said, such large biases suggest the need for a more robust solution, especially as the use of effective PAI values did not improve the ability of CLM5 to replicate diurnal variability in LSA and sub-canopy shortwave radiation. Hence, a time-varying transmissivity for direct shortwave radiation was integrated into CLM5, meaning that directionality of solar irradiance could be taken into account. Results with this modified version of CLM5 showed measured variability of sub-canopy incoming shortwave radiation was replicated more accurately, suggesting this approach may help to decrease uncertainty in LSA simulations across seasonally snow-covered forested environments. This has far reaching implications for simulations of the snow albedo feedback strength over the entire Northern Hemisphere Extratropics.

Boreal forests cover about a fifth of seasonally snow-covered land over the Northern Hemisphere. Skill in modelling snow cover has been shown to be lower for forested than for open areas, which was attributed to more complex processes in forests. One of these processes is the enhancement of longwave radiation beneath forest canopies, which has been found to impact the surface energy balance and rates of snowmelt. Single-layer vegetation schemes, as used in the Community Land Model version 4.5 (CLM4.5), have been found to overestimate diurnal cycles in vegetation temperature and sub-canopy longwave radiation. However, simulation of longwave enhancement and its impact on snow cover has not yet been assessed in a global climate model. Forest stand-scale forcing was used for the simulation of sub-canopy longwave radiation by CLM4.5 and to drive SNOWPACK, a snow model featuring a two-layer canopy module, as a benchmark model for CLM4.5. Simulated sub-canopy longwave radiation and longwave enhancement were subsequently assessed using measurements from forest stands located within seasonally snow-covered regions, which vary in vegetation density and cover the range of boreal plant functional types in CLM4.5. CLM4.5 was found to overestimate the diurnal range of sub-canopy longwave radiation and longwave enhancement, and simulation errors increased with decreasing cloudiness and increasing vegetation density. Implementation of a parameterization of heat storage by biomass reduced simulation errors, but only marginally affected the amplitude of diurnal ranges. In contrast, SNOWPACK simulated a small diurnal range of sub-canopy longwave radiation across the range of vegetation density. A simple correction was derived from stand-scale simulations and implemented in global simulations of CLM4.5 in order to scale diurnal cycles of sub-canopy longwave radiation and assess the consequent impact on snow cover. Overestimated diurnal cycles of sub-canopy longwave radiation were found to result in underestimated averages of longwave enhancement over the snow cover season, as nighttime underestimation outweighed daytime overestimation. This underestimated energy input to snow resulted in an underestimation of snow temperatures and a general delay of meltout across snow-covered forests. However, the impact of overestimated diurnal cycles on daily average longwave enhancement was found to change throughout the snowmelt season, due to increasing insolation and day length, which results in spatial differences in its impact on meltout. These findings indicate that multiple vegetation layers are indispensable for accurate simulation of longwave enhancement and its impact on snow cover, thereby contributing to the growing evidence of limitations in modelling vegetation as a single layer.

The overlap of seasonal snow and forests contributes significantly to global snowmelt, largely driven by net radiation, the combination of shortwave and longwave radiation. In sub-canopy longwave radiation models, unavailability of forest canopy temperature measurements often results in locally measured air temperatures being used to represent the temperature of the emitting canopy. The validity of this assumption was tested using extensive field measurements in three sub-canopy mid-latitude alpine environments, and resulted in improvements to model simulations of sub-canopy incoming longwave radiation. Three different radiometer configurations (stationary distributed, stationary linear, and moving linear) were simultaneously compared in a spruce forest in the eastern Swiss Alps, capturing the annual range of sun angles and sky conditions. The two linear configurations showed the greatest similarity in shortwave transmissivity, and the measurements of longwave enhancement were largely similar between all three configurations. Simulation of incoming longwave radiation at a point commonly partitions the up-looking hemispherical view between radiation coming from the sky (the sky-view fraction, Vf) and the forest canopy (1 - Vf). In this two-part model, using air temperature as a proxy for tree temperature at three forested sites with Vf < 0.3 resulted in model underestimations, representing canopy temperatures elevated above air temperature. Importantly, these errors were largest during sunny, clear-sky conditions, particularly along sun-lit canopy discontinuities. However, in denser canopies, measured tree trunk temperatures were cooler than local air temperature. Model estimations were improved by applying a bulk offset to air temperature (+3�C), reducing error from 11 Wm-2 to 4 Wm-2. Within sun-lit discontinuous forests, measured canopy temperatures varied between 5-25�C above air temperature and errors of the two-part model were as high as 40 Wm-2. Point-scale simulations of longwave radiation were improved by explicitly accounting for tree trunk temperatures within the canopy-view fraction, creating a three-part model (sky, trunk and canopy). Modelled estimates of incoming longwave radiation beyond the point-scale were improved through extensive canopy temperature measurements using infrared thermal imagery around forest gaps. A parametrisation to estimate canopy temperature using sub-canopy shortwave radiation and air temperature was developed for modelling sub-canopy incoming longwave radiation using the two-part model along canopy discontinuities. These findings provide a framework for incorporating sub-canopy longwave radiation within larger scale snowmelt models. Particularly, it is important to represent both canopy discontinuities, as well as canopy temperatures at sub-daily time steps. Exclusion of these factors could lead to inaccurate estimations of snowmelt initiation and runoff rates.

Snow accumulation has potential climatological, hydrological and ecological impacts at a global scale. Satellite passive microwave radiometers have the potential to provide snow accumulation data with a historical record of over 30 years, however, current data products contain unknown uncertainty and error. Snowpack stratigraphy is the spatial variation in snowpack properties caused by the layered nature of the snowpack. Snowpack stratigraphy influences the accuracy and increases uncertainty in simulations of microwave emission from snow which in turn increases uncertainty in satellite derived estimates of snow water equivalent using microwave radiometers. Two methods were developed to help better quantify snowpack stratigraphy. An improved technique for characterising snowpack stratigraphy within a snow trench was developed. Secondly a new method was developed to quantify the density of ice layers that form in snowpacks with known error and uncertainty. Snowpack stratigraphy was characterised using the improved technique across the Trail Valley Creek watershed in the Canadian Northwest Territories. Two 50 m trenches and eleven 5 m trenches were dug across the range of landcover types found in the watershed. This dataset allowed layer boundary roughness to be characterised and the properties of snow layers to be mapped with an unprecedented level of accuracy. Ice lens density was measured 60 times at three locations in the Arctic and midlatitudes at locations with coincident ground based radiometer measurements. The impact that accurate parameterisation of density has on modelled estimates of brightness temperature was quantified. Simulations of microwave brightness temperatures were conducted using snow emission models at all locations. The output of these simulations, and comparison to ground based observations where available, allowed for the characterisation of variability in brightness temperature simulations caused by stratigraphic heterogeneity. The findings presented in this thesis will inform research aiming to better characterise the satellite error budget. Improvements in this area helps improve global snow mass and snow accumulation estimates.