How do plants alter their local climate? Soil-vegetation-atmosphere-transfer (SVAT) models, are mathematical representations of the complex and dynamic processes that occur between the soil, the living plant layer (crops, forests, scrub, grassland etc.) and the atmosphere. Modelling is an important aspect of research as these systems act in a way that is impossible to otherwise understand or predict. The latter point- making future predictions probably being the most important application of SVAT modelling, for example in making predictions of future climate change scenarios as reported in the IPCC assessments.
The value of a good models is greater than the sum of its parts; integrating knowledge and understanding from across a range of disciplines from meteorology to plant physiology. However, this level of complexity can make validating a model’s predictions somewhat challenging. A method frequently used to test such models, and to help understand the systems they represent is sensitivity analysis (SA). In this paper I conduct a SA in Simsphere (a SVAT model) to test the sensitivity of the climate to two different model representations of the control of water movement through stomata. I use this example to highlight limitations of models and the need for a cautionary approach to drawing conclusions from model predictions.
Image: A heterogeneous landscape (N.Alaska) Sourced: maps.google.co.uk
SVAT Modelling: considering the biophysical control of water in land atmosphere interactions
A need for accurate predictions of weather and climate in response to change has been an underlying driving force behind the creation of Global Circulation Models/ Global Climate Models (GCM’s). Increased understanding of land atmosphere interactions and advancement in computer technology has allowed for the development of data rich, dynamic, SVAT incorporated GCMs that are used to make climate and weather predictions today (Sellers et al., 1997). Current GCMs have been applied in research and used to inform management and policy, for example they have been employed by the international panel on climate change to make predictions for future climate change under varying land use and emission based scenarios (IPCC, 2001). They have also been used to identify drivers of past significant weather events, such as linking regional drought across the US to deforestation across the same area (Gallo et al., 1999). Whether in past, present or future contexts, the common requirement of these uses is being able to understand and represent the effect of vegetation on the climate, over a range of scales.
Comparing observational data (e.g temperature, rainfall, cloud cover, plant growth) to model outputs is an essential step in assessing model realism and generality. These often occur at a trade-off to one another depending on the scale and specificity of data used for parameterisation of inputs (relating input parameters to observational data from a specific study area) and calibration (altering a model so its’ output matches observed data). Collecting such data is often time and resource expensive, however carrying out a SA (seeing how model outputs change in response to altering the input drivers) demands relatively less effort, and is a process that can yield valuable information about the sensitivity of model processes and outputs. This information can then be used to assess the relative importance of model drivers, in order to increase our understanding of the system and also to help direct resources to understand and measure inputs that have the most significant influence on the model predictions, as in Olioso and Carlson’s (1996) parameter evaluation using Simsphere.
The role of leaf physiological processes in SVAT
The second and third generations of GCMs graced the sponge-like vegetation layer of the first generation with the introduction of a biophysical control over water transfer from plant to atmosphere. As a result, model predictions using SVAT became more accurate, as compared to observed values (Petropoulos et al., 2009). Gas exchange in plants occurs as a trade of to maximise CO2 uptake (for photosynthesis), whilst minimising water loss (where water is a limiting factor). Biophysical control was represented in SVAT models by a term for stomatal resistance, which simulated dynamic control of water uptake and transpiration by vegetation based on a number of climate and productivity related variables. Jarvis (1976) proposed stomatal resistance as a function of multiple inputs. Summarised by the equation: Rs= rmin. f(S). (f Ж 1 or fw). fVPD. fCO2 . fT1
Where f represents ‘a function of’, S represents solar flux,VPD for vapour pressure deficit, CO2 for the CO2 concentration and Tl for leaf temperature. A further option of a term for either f Ж 1 (leaf water status) or fw (substrate water availability) is a response to a divide in the scientific opinion regarding the origin of the signal controlling stomatal openness in vegetation. There is evidence to support both substrate water availability (Jones 1983) and leaf water status (Jarvis, 1976), as drivers of stomatal openness. These mechanisms have been represented in Simsphere using the Carlson Lynn or Deardorff model parameterizations, for leaf or root orientated control of RS respectively (Carlson, 2001).
For those who want an overview of applications and limitations of modelling vegetation and climate but may be less inclined to read through the technical jargon, I would suggest skipping over this middle bit ‘Parameterisation ‘ and ‘Sensitivity of Stomatal Resistance and PBL’ section :)
A standard model run using parameters as described by Carlson (2001) was conducted to test the sensitivity of atmospheric fluxes and energy balance, indicated by temperature, the bowen ratio and PBL height, to transpiration by vegetation. The sensitivity to the two different model representations of stomatal resistance was tested in interaction with varying soil moisture availability (MS) (range from 0-1). MS is important to consider in this case, after having been being identified as one of the main drivers of evapotranspiration in SVAT models (Carlson and Boland, 1978, Audrey et al., 1986). Parameters important to the testing of the model in this context are listed as follows. Parameterisation used a standard crop vegetation with 50% land cover, and a leaf area index (LAI) of 7. Soil properties included soil moisture availability (standard 0.5), root moisture availability (standard 0.75) and an average substrate temp of 24.6 oC. When using the Carlson Lynn model, a wilting point value was changed from the standard 0.8 to 0.12 to match values quoted in current literature (Carlson, 2001). Sensitivity was calculated using percentage change in inputs compared to the given standard, and the responding percentage change in PBL height. This format was used to standardise units to allow direct comparisons of the sensitivities of different parameters.
Sensitivity of Stomatal Resistance and PBL
Using both stomata resistance models PBL, air temperature in the foliage, at 1.3 m and at 50m dropped as soil moisture was increased, however this effect was more predominant under the C/L model (Table1). PBLMAX represents the maximum depth, throughout a 24-hour period, of the mixing layer in the atmosphere. PBL depth is driven by the turbulent convection of sensible heat, which in turn is driven by a temperature gradient from foliage to atmosphere, as illustrated in Figure 1. Thus as the temperature in foliage increases, PBL deepens, assuming all other exchanges remain constant.
Regarding changes in stomatal resistance, Figure 2.ii illustrates that the initial model parameterisation caused a change of 75 W m-2 (at 0.5 soil moisture), compared to a smaller change of 2 W m-2 (C/L) to 10 W m-2 (DD) when soil moisture was altered under either model respectively. Thus stomatal resistance is more sensitive to model choice than soil moisture changes. Stomatal resistance model parameterisation (DD or C/L) resulted in a change of 300m-450m in PBL height (Table.1, Figure.1), with the greatest change corresponding to the highest SM value (Figure 2.i), which can be explained by the increasing deviation of stomatal resistances’ calculated by the two different models (Figure.2.ii).
Under standard soil moisture parameterisation (0.5) the Deardorff model produced a higher stomatal resistance, thus lower conductance of water, and greater Bowen ratio (Table.1). The Bowen ratio reflects latent to sensible heat ratio, therefore in this incidence a greater number means the flux of sensible heat is higher under DD model than the C/L model. Therefore using the DD model resulted in a greater temperature gradient between foliage and atmosphere, hence a larger mixing layer due to stronger turbulent convection, and a deeper PBL (Figure.1).The C/L model minimum stomatal resistance was less sensitive to increasing soil moisture than DD (Figure 2.ii). This is likely to directly relate to the difference in parameters used in the models’ equation. C/L uses water leaf status instead of soil moisture availability, creating the capacity for leaf storage of water, hence greater resilience against changes in soil moisture availability.
Sensitivity analysis of SVAT models
This SA has been conducted by making simple changes in input parameters, and is presented in the context of a wealth of SA and GSA model analysis reported in the literature for Simsphere and alternate SVAT models. Deardorff (1978) first noted the relationship between of increasing evapotranspiration from vegetation, and a decrease in sensible heat flux. However there remained conflicting evidence depending on the model parameterisation used as noted by Sun and Bosilovich (1995), who provided further evidence of soil moisture content as the most sensitive input parameter for the modelling of energy fluxes, and noted the critical role of vegetation in determining the bowen ratio of energy fluxes. Global sensitivity analysis (GSA) as described in Petropoulos et al. (2009b) has been used to identify the relative importance of multiple input parameters, and importantly, consider their interactions in the process. These finding also conclude soil moisture, vegetation cover, and topography as the most sensitive model input in Simsphere, but did not consider sensitivity to changes in model physics. This study has additionally highlighted the relative sensitivity of the choice between models of stomatal resistance, and their respective representations of water limitation of transpiration on atmosphere fluxes and PBL.
Limitations and assumptions
In modelling stomatal resistance and transpiration by vegetation in Simsphere a number of assumptions are made, for example, stomata resistance is a function of incoming solar radiation, which attenuates (decreases) at a constant rate through a canopy based on the principle of Beers Law (Carlson, 2001) as opposed to a more dynamic model of CO2 requirement based on net photosynthetic productivity. Secondly the change in stomatal resistance to soil moisture availability does not simulate stomata closure as a stress response to very low soil moisture conditions. The relative influence and uncertainly surrounding these processes cannot be considered by conducting simple SA, as they are not represented in the model.
The problem of temporal and spatial scales
Temporally, Simsphere only represents processes in 24-hour cycles so longer term patters and variations are overlooked, such as the accumulation of sensible heat in ground mass, long-term weather events, or seasonality effects. Seasonality and phenology are key drivers in the variation of plant processes over time (Bonan, 2008). The most marked example being the loss of leaves from many tree species of a temperate climate over winter, where energy fluxes are influenced by a decrease in transpiration and interception of water and light. In Simsphere temporal variation such as this have to be represented through continued calibration using seasonal data of plant behaviour for each cycle.
Simsphere is also limited in its ability to represent a heterogeneous distribution of vegetation and associated impacts on the atmospheric energy balance. The horizontal area over which processes occur is not predefined, thus dependent on the input parameter data. Calculations assume consistent mixing within horizontal layers, and there is no distinct distribution of biomass or vegetation spatially. Instead, single sets of parameters are assigned to a soil type and vegetation type, which are applied proportionally to the relative fraction of bare soil or vegetation (Carlson, 2001).
This has resulted in a number of natural processes being overlooked in model predictions, such as any variation caused by dynamic processes or heterogeneity in the distribution of soil or vegetation. Patch dynamics have been show to have an important effect on energy fluxes (Bonan, 2008) for example through altering the momentum of wind eddies via changing surface roughness over a patch of vegetation next to bare ground, or creating pockets of darker ground and greater LE fluxes from a patch of moist soil. Some of these issues relating to spatial heterogeneity and scale can be overcome by the integration of Simsphere with remote observation (RO) data, such as from satellites, which has allowed successful incorporation of spatial data into SVAT schemes to represent both vertical and horizontal processes (Gillies and Carlson, 1995, Peteropolus, 2009). The interpretation of RO temperature and vegetation indices has been used to recalibrate SVAT model outputs, SM, LE and H energy fluxes, as proposed for future application to data from US national polar-orbital operational environmental satellite system (Chauban et al., 2003).
Findings of this SA conducted in simsphere have illustrated that over high LAI canopies there is positive relationship between PBLMAX and soil moisture availability, due to the limitation of transpiration by stomatal resistance. This SA further highlights the sensitivity of the atmospheric energy balance to the model of stomatal resistance used, and its respective term for water limitation of transpiration. Although multiple SA studies consider soil moisture availability and vegetation cover to be a key influential parameter in vegetation-atmosphere interactions, understanding and representation of model processes also plays a significant role in the sensitivity of model predictions of water and energy fluxes. In this case the model representation was the biophysical control of water loss by vegetation. Carrying out SA only indicates the relative importance of those parameters included within the model, it does not however consider the impact of those processes or feedbacks that are misrepresented, or missing from the model all together. These unrepresented or unknown mechanisms have to be given consideration by looking at model predictions in the context of our understanding of a system’s dynamics, by comparing observational data to model data outputs, and identifying the nature of the error in our predictions.
Andre, J.C.; Goutorbe, J.P.; Perrier, A. HAPEX — Mobilhy: a hydrologic atmospheric experiment for the study of water budget and evaporation flux at the climatic scale. Bull. Am. Meteorol. Soc. 1986, 67, 138-144.
Bonan.G.B (2008). Review: Forests and climate change; forcings, feedbacks and the climate benefits of forests. Science, 320, 1444-1449
Carlson (2001). Simsphere Technical Manual and Workbook. Dutton Institute. E-Education Institute. [Accessed 02/10/2012]
Carlson.T.N, Boland.F.E. (1987) Analysis of urban-rural canopy using a surface heat flux/temperature model. J.Appl.Met.17, 998-1013.
Chauhan, N.S.; Miller, S.; Ardanuy, P. (2003) Spaceborne soil moisture estimation at high resolution: a microwave-optical/IR synergistic approach. Int. J. Remote Sens. 22, 4599-4646.
Deardorff. (1987) Efficient prediction of ground surface temperature and moisture with inclusion of layer of vegetation. J. Geophy. Res. 20, 1889-1903
Gallo, K.P., Owen, T.W., Easterling, D.R. and Jamason, P.F. 1999: Temperature trends of the US historical climatology network based on satellite-designated land use/land cover. J.Climate 12, 1344-1348.
Gillies, R.R.; Carlson, T.N. Thermal remote sensing of surface soil water content with partial vegetation cover for incorporation into climate models. J. Appl. Meteorol. 1995, 34, 745-756.
Intergovernmental Panel on Climate Change (IPCC) 2001: Climate Change 2001: Synthesis Report: Contribution of working groups I, II and III to the Third Assessment Report. Cambridge Press.
Jacquemin, B. and Noilhan, J.: 1990, ‘Sensitivity Study and Validation of a Land Surface Parameterization using the HAPEX-MOBILHY Data Set’, Boundary-Layer Meteorol. 52,93-134.
Jarvis.P.G (1976) The interpretation of the variations in leaf potential and stomatal resistance found in canopies in the field. Phil. Trans. R.Scotland. 273, 593-610
Jarvis, P.G.; McNaughton, K.G. (1986) Stomatal control of transpiration: scaling up from leaf to region. Adv. Ecol. Res., 15, 1-49.
Jones.H.G (1983) Plants and Microclimate. Cambridge University Press. 323,
Olioso, A.; Carlson, T.N.; Brisson, N. (1996) Simulation of diurnal transpiration and photosynthesis of a water stressed soybean crop. Agric. Forest Meteorol, 81, 41-59.
Petropoulos.G, Carlson.T.N and Wooster.M.J, (2009) An Overview of the Use of the SimSphere Soil Vegetation Atmosphere Transfer (SVAT) Model for the Study of Land- Atmosphere Interactions. Sensors. 9, 4286-4308
Petropoulos.G, Wooster.M.J, Carlson.T.N, Kennedy.M.C and Scholze.M (2009b). A global Bayesian sensitivity of the 1d Simsphere SVAT model using Gaussian model emulation. Ecological Modelling. 220, 2427-2440
Polard, D.; Thompson, S.L. (1995) Use of a land-surface-transfer scheme (LSX) in a global climate model: the response to doubling stomatal resistance. Glob. Plan. Chan.10, 130-161.
Ross, S.; Oke, T.R. (1988) Tests of three urban energy balance models. Bound. Layer Meteorol, 44, 73-96.
Sun.W.Y and Bosilovich. (1996) Planetary boundary layer and surface layer sensitivity to land surface parameters. Boun.-Lay. Meter. 77, 252-278
Sellers.P.J, Dickinson.R.E, Ramdall.D.A, Betts.A.K, Hall.F.J, Berry.J.A, Collatz.G.J, Denning.A.S, Mooney.H.A, Nobre.C.A, Sato.N, Field.C.B and Henderson-Sellers.A. (1997) Modeling the Exchanges of Energy, Water, and Carbon Between Continents and the Atmosphere. Science. 275,502-509
Wilson, M, Henderson-Sellers, A, Dickinson, R, & Kennedy, P (1987), ‘Sensitivity of the Biosphere–Atmosphere Transfer Scheme (BATS) to the Inclusion of Variable Soil Characteristics’, Journal Of Climate & Applied Meteorology, 26, 3, 341-362