Journal of Geophysical Research: Atmospheres

CLARREO shortwave observing system simulation experiments of the twenty-first century: Simulator design and implementation



[1] Projected changes in the Earth system will likely be manifested in changes in reflected solar radiation. This paper introduces an operational Observational System Simulation Experiment (OSSE) to calculate the signals of future climate forcings and feedbacks in top-of-atmosphere reflectance spectra. The OSSE combines simulations from the Intergovernmental Panel on Climate Change (IPCC) Fourth Assessment Report for the NCAR Community Climate System Model (CCSM) with the MODTRAN radiative transfer code to calculate reflectance spectra for simulations of current and future climatic conditions over the 21st century. The OSSE produces narrowband reflectances and broadband fluxes, the latter of which have been extensively validated against archived CCSM results. The shortwave reflectance spectra contain atmospheric features including signals from water vapor, liquid and ice clouds, and aerosols. The spectra are also strongly influenced by the surface bidirectional reflectance properties of predicted snow and sea ice and the climatological seasonal cycles of vegetation. By comparing and contrasting simulated reflectance spectra based on emissions scenarios with increasing projected and fixed present-day greenhouse gas and aerosol concentrations, we find that prescribed forcings from increases in anthropogenic sulfate and carbonaceous aerosols are detectable and are spatially confined to lower latitudes. Also, changes in the intertropical convergence zone and poleward shifts in the subsidence zones and the storm tracks are all detectable along with large changes in snow cover and sea ice fraction. These findings suggest that the proposed NASA Climate Absolute Radiance and Refractivity Observatory (CLARREO) mission to measure shortwave reflectance spectra may help elucidate climate forcings, responses, and feedbacks.

1. Introduction

[2] The Earth's climate is dependent on absorbed solar radiation which is derived from differencing the incoming and outgoing solar radiation, and this difference is a powerful and highly variable driver of the climate system. Predicting how the Earth's broadband albedo will change over the 21st century requires an understanding of how the various mechanisms that control this albedo will change. Some of these are well understood, such as orbital mechanics and the solar cycle, while others are less understood, including snow or sea ice changes and shifts in cloud cover. The broadband albedo may change in response to both natural variability and secular trends in the climate state.

[3] Using the modern instrumental record to detect and quantify decadal-scale changes in top-of-atmosphere broadband albedo has proven problematic due to numerous issues including: (1) difficulties in indirectly inferring flux quantities from radiance measurements through angular distribution models [Loeb et al., 2007], (2) intercalibration issues among the various observations of the Earth's radiation budget collected since 1979 [Loeb et al., 2002], and (3) discrepancies in estimates of broadband solar irradiance [e.g., Fröhlich, 2006].

[4] More recently, there has been considerable debate surrounding the long-term stability of the Earth's broadband albedo with significant implications for the near-term trajectory of climate change. The issue has arisen because of apparent inconsistencies among direct and indirect observational records of the Earth's reflected solar radiation. While Pallé et al. [2004] looked at Earthshine data and found a steady decrease in the Earth's broadband reflectance from 1984 to 2000 followed by an increase between 2001 and 2003, Wielicki et al. [2005] found no trend in planetary broadband albedo from an analysis of CERES data coupled with other correlative satellite records from the Moderate Resolution Imaging Spectrometer (MODIS) [Justice et al., 1998] and Jason/TOPEX [Ducet et al., 2000]. This debate and additional problems encountered by researchers attempting to synthesize long-term climate records by harmonizing heterogeneous satellite instrument data [i.e., Hurrell and Trenberth, 1998] have led to the recognition of the critical importance of long-term measurement stability that is rigorously traceable to international standards for the assessment of changes in climate.

[5] In order to address this issue, the National Research Council's Space Studies Board [2007] has strongly recommended that NASA implement a satellite mission with the ability to make highly stable and accurate measurements of shortwave reflectance and longwave radiance spectra and of atmospheric refractivity. This mission, called the Climate Absolute Radiance and Refractivity Observatory (CLARREO), is being designed with a primary goal of achieving a set of benchmark measurements of climate from space, that are sensitive to the most critical, but least understood climate forcings, responses and feedbacks. The benchmarks provide data against which future measurement differences can be tracked and unambiguously ascribed to changes in the climate system. The benchmarks are achieved only through making measurements with calibrations that are rigorously traceable via multiple and independent metrological pathways to fundamental standards of length, mass, time, and temperature. The standards used for the CLARREO mission will be the SI second, the SI Kelvin, and the derived quantities of the SI Watt and SI reflectance, the latter of which can also be tied to the SI watt through high-accuracy spectral solar irradiance data such as the planned TSIS observations [Richard et al., 2008]. For this paper, spectral reflectance is a dimensionless quantity and is defined as follows:

equation image

where Rλ is the TOA reflectance, Iλ is the TOA radiance, and Fλ is the downwelling TOA flux with all quantities reported at a wavelength band centered at λ. A secondary goal of the mission is to provide a reference standard for intercalibration of other spaceborne missions. The CLARREO orbits are designed to avoid sampling biases related to diurnal and seasonal variability as well as to demonstrate its utility as a reference standard using the internal consistency of multiple independent measurements of the Earth system.

[6] The CLARREO shortwave instrument is currently being designed. The plans for the shortwave component of the mission include a series of one or more grating spectrometers with coverage from 320 to 2300 nm at 8 nm full-width half-maximum spectral resolution at 20 km spatial resolution. The instrument's key science requirement is to measure the absolute spectrally resolved nadir reflectance of solar radiation from Earth to space with a 2 σ accuracy of 0.3% so that detection times and trend accuracy are degraded by no more than 15% and 20%, respectively, relative to a perfect observing system [Leroy et al., 2008]. There is a wealth of modeling results that suggest that spectra of the Earth's reflectance will change in response to natural variability and climate forcing. However, there has been very little direct research to actually model anticipated changes in the Earth's shortwave reflectance spectra despite the potential for the spectra to contain a large amount of information about the Earth-atmosphere system.

[7] An Observing System Simulation Experiment (OSSE) represents one tool that has been extensively utilized to assess the scientific utility of proposed satellite instruments such as CLARREO. This tool allows mission planners to simulate instrument measurements within a controlled, well-understood, and realistic framework by pairing Earth simulations from climate models, reanalyses, or chemical transport models, with a radiative transfer and a satellite trajectory model that emulates what a proposed instrument would observe. OSSEs have been utilized by the weather forecasting community [Arnold and Dey, 1986; Atlas, 1997; Lord et al., 1997] and have been increasingly employed for a variety of NASA missions [Atlas et al., 2003; Lee et al., 2010] to evaluate both proposed and notional instruments.

[8] OSSEs are well suited for climate-monitoring mission planning for several reasons. First, the state of the climate system is reported at a variety of different time scales, is known perfectly, and can be subjected to different scenarios where the underlying forcings on and feedbacks in the system are allowed to vary. The complete and exact knowledge of the underlying (model) climate system greatly facilitates quantitative assessment of how well the hypothetical instruments meet their design requirements to detect specific properties of that system. Second, it is possible to evaluate a variety of proposed mission configurations and specifications to gauge their scientific utility and to inform the mission planning process. Third, the state of the instrument in an OSSE is perfectly known and any noise characteristics in the simulated measurements are completely prescribed. These controlled conditions help determine the optimal conditions for the mission to succeed in its objectives. Finally, all of the climate models that participated in the IPCC Fourth Assessment Report [Meehl et al., 2007] detail changes in the Earth system that will likely impact the spatial distribution of Earth's reflectance arising from variations in vegetation, snow cover, sea ice, aerosols, and clouds. Significant work has been devoted to the utilization of weather OSSEs for determining how uncertain quantities are made less uncertain with the addition of new data sets. Developing a climate OSSE requires a similar approach, though with an emphasis on climatic variables; however, climate OSSEs are much more difficult to test against real data because analysis is performed on climate conditions that have not been witnessed in the modern instrumental record.

[9] Despite these benefits, OSSEs are not perfect substitutes for observing systems precisely because they do not encapsulate all of the real-world variability that a satellite instrument may encounter. Any processes that are underrepresented in the input data (in this case, the climate model) or physics that are misrepresented in the instrument emulator will propagate into the OSSE results and will reduce the fidelity of the OSSE relative to the physical system being measured and/or the performance of the observational system. Since it is generally very difficult to address the under-representations or misrepresentations in the OSSE until actual measurements from the instrument are obtained, this simulation and analysis tool is most appropriate for establishing a likely upper bound on the expected performance of the instrument. That is, OSSE results that indicate that the proposed instrument will be able to meet its science requirements represent a necessary but not sufficient condition for mission success.

[10] With respect to the CLARREO mission, considerable research has been devoted toward determining the utility of making infrared measurements with well-known calibration by using OSSEs [Huang et al., 2010a] and detection and attribution methods [Leroy et al., 2008]. However, while sampling studies [Doelling et al., 2009] and intercalibration studies [Lukashin et al., 2009] have been undertaken for the reflected solar instrument, no studies have been undertaken to model changes in the shortwave reflectance spectrum over the 21st century. Therefore, this paper describes the development and features of an operational OSSE for shortwave reflectance measurements in order to evaluate the added value of spectral over broadband measurements. We highlight the signals that are present in the simulated data, and we conclude with recommendations for future OSSE development and analysis. This OSSE is different from those that have been utilized by the weather forecasting community in that the input data is obtained from climate models rather than retrospective meteorological analyses or reanalyses. While inputs based on current observations are appropriate for establishing how realistic the results are, climate model data ultimately are required because CLARREO's utility lies in its ability to detect the evolution of spectra under altered climate change conditions that have not been encountered in the historical periods spanned by analysis and reanalysis data sets.

[11] It is critical that the climate model OSSE provide realistic simulations of shortwave reflectance spectra that can serve as a guide for hyperspectral shortwave mission development. Therefore, the OSSE spectra must meet the following criteria: (1) The simulated reflectance spectra are fully consistent with simulations of the 21st century climate. (2) The simulated reflectance spectra for present-day conditions agree substantially with and capture the unforced (natural) variability in existing satellite-based shortwave reflectances on seasonal, annual, and interannual time scales.

[12] These criteria ensure that the signals in the simulations are accurate representations of current and projected reflectances from the Earth system. Once it has met these design objectives, the OSSE can be used to address several key questions related to the CLARREO mission: (1) What is the magnitude of reflectance spectra signals arising from aerosol and greenhouse gas forcing and does this exceed the natural variability of the climate system? (2) What is the magnitude of reflectance spectra signals arising from snow and sea ice, water vapor, and cloud feedbacks, and does this exceed the natural variability of the climate system? (3) What is the time required to detect changes in reflectance spectra relative to changes in broadband albedo for the same underlying forcings and feedbacks? (4) How easily can changes in reflectance spectra be ascribed to changes in the underlying forces and feedbacks?

[13] This paper focuses on the first criterion design objective of the OSSE and the results to date enable this paper to address the first two CLARREO mission questions. Therefore, this paper is organized as follows: In section 2, the design and features of this climate model OSSE will be described in detail with a focus on the first criterion. In section 3, the results of 21st century simulations of an anthropogenically forced and anthropogenically unforced climate will be presented within the context of the shortwave reflectance measurements. Finally, section 4 will present a discussion of the lessons learned from this climate model OSSE, the limitations of the simulations, and paths forward for future climate OSSE research with a focus on the second OSSE criterion.

2. Methodology

[14] The OSSE has been developed to the point where it can be used as a satellite instrument emulator tool to make recommendations for mission design and configuration. The inputs to the OSSE take the form of climate model integrations and can be broadly divided into descriptions of the top-of-atmosphere (TOA) and surface boundary conditions and the atmospheric state including radiatively active gaseous and condensed species. The outputs from the OSSE are generated by a pair of radiative transfer codes. Each of these components is described in detail below.

2.1. Climate Model Inputs

[15] Solar reflectance spectra are produced from single ensemble members of simulated climate states generated by the Community Climate System Model (CCSM3) Version 3 [Collins et al., 2006a]. CCSM3 is a coupled climate model with components representing the atmosphere, ocean, sea ice, and land surface connected by a flux coupler. CCSM3 is designed to produce realistic simulations on decadal to millennial time scales of continental-scale dynamics, variability, and climate change.

[16] The input data are obtained from climate model simulations produced in support of the IPCC Fourth Assessment Report (AR4) [Intergovernmental Panel on Climate Change (IPCC), 2007] and are publicly available from the Earth System Grid [Bernholdt et al., 2005]. Simulations for this paper have focused on comparing and contrasting reflectance spectra associated with the A2 emissions scenario [IPCC, 2000] and the constant CO2 concentration scenario [Meehl et al., 2005, 2007]. The former scenario prescribes steadily increasing levels of well-mixed greenhouse gases including CO2, CH4, N2O, and CFCs, as well as sulfate and carbonaceous aerosol loading that peaks at 2050 and decreases to 1990 level by 2100. The result of this scenario is a tripling of CO2 by the end of the 21st century relative to pre-industrial conditions, very little change in aerosol optical depth, and considerable surface warming and snow and sea ice retreat across the multimodel ensemble assessed in the AR4. The latter scenario prescribes time-invariant concentrations of the well-mixed greenhouse gases and aerosols described above fixed to levels prescribed for 2000 CE.

[17] The state parameters from the atmospheric and land surface models input to the OSSE have been computed and archived on a horizontal spectral-Eulerian T85 (85 wave number triangular truncation) grid with an equatorial zonal resolution of approximately 1.41° (154 km). The vertical profiles of atmospheric thermodynamic properties, trace gases, and condensed species are resolved onto 26 hybrid sigma levels from the surface to a constant pressure surface of 2 hPa in the middle of the stratosphere.

[18] With the exception of some high-frequency temporal “snapshots,” the instantaneous fields from the CCSM simulations of the 21st century are averaged and archived as monthly means. The climate states used in the OSSE are extracted from these time-mean model fields. This implies that the OSSE can only capture temporal variability on monthly and longer time scales. On shorter time scales, the solar illumination and the states of the atmosphere and land surface are held fixed (i.e., persist) from one orbit of the simulated satellite to the next. The reasons for this choice of monthly mean inputs are twofold: first, a primary objective of the CLARREO mission is to record measurements that will be critical for long-term climate change evaluation, so the simulation of higher-frequency features is of secondary importance. Second, the OSSE simulations are very computationally intensive (as described below), and it is not feasible at this stage of project development to perform global simulations of CLARREO reflectance spectra over the 21st century at shorter time scales. Alternate implementations of the radiative transfer (e.g., using principal component methods to create look-up tables) may permit greater time resolution.

[19] We therefore specify the radiatively active atmospheric species with data from monthly mean concentrations of these species obtained from the CCSM3 archive of simulations for the IPCC AR4. For the condensate-free atmosphere, we input three-dimensional distributions of temperature, water vapor, carbon dioxide, methane, nitrous oxide, CFC-11, and CFC-12 as prescribed by the CCSM3 simulations. Aerosols and clouds are also described in detail by the archived simulations. For the aerosols, we input mass mixing ratio profiles of 12 aerosol types including sulfate, sea salt, dust in four grain-size categories, black and organic hydrophobic and hydrophilic carbonaceous species, and volcanic emissions. For clouds, we input vertical profiles of liquid and ice cloud water content, areal extent (fraction), and particle/droplet/crystal effective radii.

2.2. Radiative Transfer

[20] The simulated reflectance spectra corresponding to the evolving climate states are computed using a widely utilized one-dimensional general-purpose radiative transfer code called MODTRAN [Berk et al., 1999] version In the OSSE, MODTRAN calculates both the zenith-propagating spectral radiances and corresponding upwelling hemispherical spectral fluxes at wavelengths between 300 and 2500 nm. To generate this spectral radiance and flux information, the OSSE invokes MODTRAN using the DISORT2 radiative solver [Stamnes et al., 1988] with a 15 cm−1 native spectral resolution band model based on the HITRAN 2008 database [Rothman et al., 2009] and with eight discrete-ordinate streams. The radiances and fluxes are degraded with a Gaussian spectral response function with a full-width half-maximum of 8 nm corresponding to the design specifications of the CLARREO radiometer.

[21] MODTRAN allows up to 4 aerosol motifs, which we set to sulfate, sea salt, carbonaceous, and dust species obtained from combinations of the 12 aerosol fields predicted by CCSM. Aerosol optical properties including the extinction and absorption coefficients and the phase functions for each aerosol motif are specified as functions of wavelength for both radiation codes using the aerosols optics adopted in the Community Atmosphere Model CAM3 [Collins et al., 2006b] and derived from the climate model outputs using the aerosol mixing rules used in the CCSM radiation code [Collins et al., 2004]. Sulfate, sea salt, and hydrophilic carbonaceous aerosols exhibit significant hygroscopicity and therefore have optical properties that are strong functions of in situ relative humidity [Hess et al., 1998]. This hygroscopic behavior is also identical between CCSM and MODTRAN.

[22] Liquid cloud optical properties are specified as a function of wavelength and liquid droplet effective radius and are identical to those used in the CCSM radiation code [Slingo, 1989]. Ice cloud optical properties are separately specified as a function of wavelength and ice crystal effective radius and are also identical to those used in the CCSM radiation code [Ebert and Curry, 1992]. (Note that we recognize that these are “legacy” optical parameterizations and have adopted them just for consistency with the version of CCSM used for the IPCC AR4. We can readily update the cloud optics in the OSSE to current, state-of-the-art representations as needed.) The CCSM radiation code contains a robust cloud-overlap scheme based on the independent-column approximation [Collins, 2001] and this scheme is implemented in an identical manner for the radiance spectra through up to 15 calls to the MODTRAN code for each grid box with different cloud configurations produced by the cloud profile generator within the CCSM code. The subgrid geometrical cloud overlap is assumed to be maximum random for both codes.

[23] Broadband solar upwelling hemispherical fluxes are also computed for the same states using the CCSM radiative transfer (RT) parameterizations with the identical Sun-satellite geometry employed in the MODTRAN calculations. These broadband fluxes from CCSM RT may be readily compared with the spectral quadratures of the narrowband fluxes from MODTRAN. The rationale for running both radiative transfer models simultaneously is to ensure that the reflectance spectra calculated with MODTRAN are entirely consistent with peer-reviewed results from the CCSM simulations, thereby establishing the reliability of the reflectance spectra and the validity of any conclusions arising from analysis of the simulated spectra and insulating the results against concerns about radiometric precision. For these reasons, the OSSE produces broadband shortwave calculations from the extensively validated CCSM radiation code [Collins et al., 2006b] to provide unbroken traceability from the simulations for IPCC AR4 to the OSSE outputs.

[24] This traceability is achieved by subjecting the OSSE to two requirements that must be fulfilled if the data are to be trusted. The first traceability requirement is consistency between the OSSE and CCSM. We have demonstrated that the offline implementation of the CCSM RT in the OSSE agrees the same RT methods operating inline during CCSM integrations to within machine precision using instantaneous (rather than monthly averaged) states saved from CCSM3 (not shown). The second traceability requirement is that the spectral information produced by the OSSE is consistent with these broadband fluxes. In order to meet this requirement, we have developed internally consistent visible, near-IR, and narrowband surface radiative boundary conditions for CCSM RT and MODTRAN described below. We test this consistency by comparing the broadband fluxes calculated from CCSM RT and from quadrature in wavelength over the MODTRAN spectral hemispherical fluxes. Figures 1a1d shows the agreement between TOA fluxes calculated from MODTRAN and CCSM RT under all-sky and clear-sky conditions and confirms that the two codes are producing very similar broadband fluxes given identical inputs. The mean and standard deviation of the differences between the MODTRAN and CCSM radiation codes are less than 1% for all-sky and clear-sky conditions. The box-whisker diagrams in Figures 1b and 1d also indicate a relative level of agreement to within a couple of percent which, given differences in the number of streams between the two codes, differences in the spectroscopic data used to construct the parameterizations for gaseous species, differences in the spectral partitioning of the solar boundary condition, and differences in the handling of the surface boundary condition in terms of interpolating the spectral BRDF information and calculating diffuse and direct albedo quantities in the visible and near-IR, meets our second traceability requirement. The third traceability requirement for the OSSE, namely that the spectral fluxes and reflectances are consistent, is automatically met since these quantities are computed simultaneously from the same radiative transfer code operating on identical climate states.

Figure 1.

(a) Scatterplot of all grid points for the all-sky upwelling shortwave top-of-atmosphere flux calculated by the CCSM radiative transfer code and by MODTRAN for the 32,768 grid boxes from January 2090 for the A2 simulation. Also included are the Pearson correlation coefficient (r2), and the mean (μ) and standard deviation (σ) of the differences between the two codes. (b) Box-whisker plot of percentage difference between CCSM and MODTRAN for the all-sky fluxes shown in Figure 1a where the boxes indicate the 25th and 75th percentile values, the centerline of the box is the 50th percentile value, and the whisker lower and upper edges represent the 5th and 95th percentile values, respectively. (c) Same as Figure 1a but for clear-sky conditions. (d) Same as Figure 1b but for clear-sky conditions.

[25] The radiance calculations associated with the OSSE are extremely computationally intensive due to broad spectral coverage, high spectral resolution, and the multiple calls per grid box to implement the cloud-overlap approximation and sea ice treatment. The OSSE extensively utilizes the NASA High-End Computing (HEC) facilities including Pleiades, a 773-teraflop machine located at NASA Ames. Approximately 2500 CPU hours are required for a single time step (which in this case is one month) of simulation at T85 spatial resolution (1.4° latitude × 1.4° longitude). The code has been parallelized and load balanced, and it achieves an effective throughput rate of 1 month of simulation in 2.5 wall-clock hours on 1024 processing elements (PEs).

2.3. Radiative Top-of-Atmosphere and Surface Boundary Conditions

[26] The TOA boundary condition is presently specified by the Kurucz Solar Spectrum [Kurucz, 1995] with a fixed solar irradiance of 1360 W/m2 that is consistent with the most recent results from the SORCE instrument [Kopp and Lean, 2011].

[27] Shortwave measurements are critically dependent on solar satellite geometry including solar zenith and azimuth angles. The OSSE in its current configuration assumes a satellite Sun-synchronous orbit with a 1330 ascending-node local equator crossing time with only a nadir satellite-viewing angle. There is active discussion within the CLARREO Science Definition Team regarding different inclination angles for CLARREO's orbit. The OSSE includes the capability to simulate other orbital configurations, but they will not be exercised for the calculations discussed here.

[28] The surface boundary condition in the CCSM radiation code requires the specification of near-ultraviolet/visible (300–700 nm) and near-infrared (700–5000 nm) direct and diffuse narrowband albedos. Since this information is insufficient for calculating spectral radiances and since there are systematic errors in the model-generated albedos relative to NASA's albedo retrievals [Oleson et al., 2003], the OSSE utilizes a realistic set of surface reflectance boundary conditions that are consistent between the CCSM and MODTRAN radiation codes using Bi-directional Reflectance Distribution Functions (BRDFs). Because reflectance spectra are quite sensitive to surface boundary conditions, the development of this tool has focused on producing realistic shortwave signals associated with annual cycles in vegetation while also simulating the large signals associated with snow and sea ice annual cycles and trends consistent with climate models projections for the 21st century.

[29] Over nonfrozen ocean surfaces, we utilize a routine based on the Cox-Munk BRDF model [Cox and Munk, 1954; Kotchenova et al., 2006] that has been adapted to the MODTRAN code (V. Ross, personal communication, 2010). The ocean surface broadband albedo using this method agrees well with surface measurements made from a variety of different solar zenith angles [Jin et al., 2004]. Where sea ice is predicted, the OSSE uses the Ross-Li BRDF model [Wanner et al., 1995; Wanner et al., 1997; Lucht et al., 2000] with its component terms specified using in situ reflectance spectra of bare ice adapted from radiative transfer process models and direct measurements of sea ice [Briegleb and Light, 2007]. The areal extent and spatial distribution of sea ice is predicted using the Community Sea Ice Model in CCSM [Briegleb et al., 2004]. The radiative transfer is performed using area-weighted averages of ice-free and completely ice-covered conditions.

[30] Over land surfaces, the OSSE makes extensive use of the data from the MODerate Resolution Imaging Spectroradiometer (MODIS) mission, which produces an operational BRDF product based on the Ross-Li model [Wanner et al., 1997; Shuai et al., 2008]. The BRDF product that we use (MCD43C1 from Aqua) reports isotropic, volumetric, and geometric components at 0.05° × 0.05° spatial and 16 day temporal resolution. This product also contains information about the snow fraction for each pixel. Using the MODIS BRDF data set from 2003 to 2008, we have constructed a monthly averaged climatology of snow-free and snow-covered BRDF maps on the CCSM3 terrestrial model grid by first partitioning the fine-scale BRDFs using the high-resolution snow fractions and then averaging the partitioned data onto the coarser model grid. The land surface BRDF at any given time and grid point is a linear combination of the snow-free and snow-covered BRDFs weighted by the model grid snow fraction determined by the Community Land Model (CLM) component of CCSM3 [Dickinson et al., 2006]. This procedure allows us to simulate both the effects of secular trends and seasonal variations in snow cover, thereby allowing the OSSE to include projected changes in snow-cover distribution that have not been observed in the historical satellite record while simultaneously maintaining consistency with the measured reflectance spectral characteristics of snow-covered and snow-free surface types. Where no MODIS data are present for a given CCSM3 grid box due to polar night conditions, we fill the corresponding grid boxes using average snow-free and snow-covered BRDF data for land cover types as described by the MODIS MOD12C1 land cover type product [Friedl et al., 2002]. In order to maintain consistency between the BRDFs input to MODTRAN and the direct and diffuse narrowband albedos input to CCSM3, we convert the MODTRAN BRDF values to black-sky and white-sky albedos (also referred to as direct and diffuse albedos, respectively) in the visible and near-IR bands using the formulae from Strahler and Muller [1999]. This methodology ensures that the surface boundary conditions are consistent between the CCSM radiation code and MODTRAN. Figure 2a shows the Lambertian surface reflectance at 0.47 μm for all January simulations for snow conditions and Figure 2b shows the same reflectance for snow-free conditions, with the actual reflectance being a linear combination of the two values depending on CCSM3's snow fraction prognosis.

Figure 2.

(a) Snow-free Lambertian component of surface reflectance at 0.47 μm derived from MODIS climatology and used to construct land BRDF parameters. (b) Same as Figure 2a but for completely snow-covered conditions. (c) All-sky December–January–February top-of-atmosphere MODTRAN-calculated broadband albedo averaged over the A2 simulation for 2000–2009. (d) Same as Figure 2c but for clear sky.

2.4. Outputs From the OSSE

[31] Each data output file contains shortwave broadband (0.3–5.0 μm) all-sky and clear-sky upwelling and downwelling direct and diffuse flux profiles from the CAM radiation code and MODTRAN. From MODTRAN, the output files contain visible (0.3–0.7 μm) and near-infrared (0.7–2.5 μm) all-sky and clear-sky flux profiles and top-of-atmosphere upwelling and downwelling flux spectra at the CLARREO instrument resolution. Figures 2c and 2d shows the example top-of-atmosphere all-sky and clear-sky broadband albedo values for simulations of the unforced present-day climate. Figures 2c and 2d demonstrate the well-known effects of reflection from oceans, vegetation, sea ice, and snow on clear-sky broadband albedo and from scattering by clouds on all-sky broadband albedo.

[32] The OSSE produces all-sky and clear-sky radiance spectra both at the CLARREO instrument resolution and at MODTRAN's native band model resolution of 15 cm−1. This allows for immediate application of the OSSE results to different instrument spectral response functions including those of existing instruments. Figures 3a and 3b shows zonally averaged spectral reflectance fields under simulated present-day conditions produced by the MODTRAN code in the OSSE. Figures 3a and 3b contain notable features. First, at higher latitudes, the additional reflectance due to snow and sea ice is evident in near-ultraviolet/visible wavelengths. Second, the vertical stripes across Figures 3a and 3b are indicative of gaseous absorption features from CO2 in the vicinity of 2.0 μm, O2 A band at 0.76 μm and the O2 B band at 0.69 μm, and the near-infrared absorption bands of water vapor. The reflectances are reduced in narrow contiguous spectral regions corresponding to the overtone bands of the 2.7 μm primary water vapor band at nearly all latitudes. Note that in the tropics for the all-sky reflectance spectra, these bands are less prominent because there is very little water vapor above the tops of clouds in the convective regions which are generally around 192 hPa in the model. Still, the major signals seen in the total zonally averaged reflectance spectra in Figures 3a and 3b arise from sea ice and snow, while the impact of clouds is secondary. Third, a comparison of clear-sky and all-sky reflectance spectra indicates the zonal impact of clouds on the radiation budget. The stratus, subsidence, and storm-track cloud regimes are all detectable by the increased visible and near-IR reflectance in the all-sky relative to clear-sky conditions. It should also be noted that changes in the reflectance spectra contain other notable features beyond what is seen in Figures 3a and 3b and are explored in section 3.

Figure 3.

(a) All-sky December–January–February top-of-atmosphere spectral reflectance averaged over the A2 simulation for 2000–2009. (b) Same as Figure 3a but for clear sky.

3. Results

[33] The first results from the OSSE span the 21st century for two emissions scenarios. The first simulation is essentially unforced with constant concentrations for all radiatively active forcing agents set to values for the year 2000 CE. It is based upon the CCSM integration labeled b30.036a, and it is treated as the control in the analysis that follows. The second simulation corresponds to the IPCC A2 emissions scenario with steadily increasing well-mixed greenhouse gas concentrations ending with a CO2 abundance of 856 ppm at 2100. It is based upon the CCSM integration labeled b30.042a and is treated as the experiment in this work. The results of the b30.036a and b30.042a integrations can both be downloaded from the Earth System Grid [Bernholdt et al., 2005].

[34] The calculated reflectance spectra contain the radiative effects of clouds, aerosols, land surface changes (in particular from snow and ice) and greenhouse gases, and characteristic features, or “signatures,” of these effects can be seen in Figures 4a4d. The signatures are computed as residuals by differencing the reflectance spectra for a U.S. Standard model atmosphere [Anderson et al., 1986] with and without the radiative agents of interest [Wetherald and Manabe, 1988]. The signatures are dictated by intrinsic optical and microphysical properties of these radiative agents (e.g., clouds) combined with effects of gaseous extinction by the surrounding atmosphere.

Figure 4.

(a) SW spectral reflectance difference signature of aerosol minus aerosol-free ocean reflectance from four different aerosol motifs at unit optical depth at 0.55 μm for a U.S. Standard model atmosphere [Anderson et al., 1986]. (b) SW spectral reflectance difference signature of overcast cloud-covered minus cloud-free ocean reflectance from a unit optical depth at 0.55 μm low (liquid) cloud layer between 700 and 750 mbar and a high (ice) cloud layer between 250 and 290 mbar. (c) One thousand times the SW spectral difference signature of a doubling of CO2, a doubling of CH4, a doubling of N2O, and a 20% increase in H2O.

[35] The aerosol signatures shown in Figure 4a are spectrally broadband and qualitatively similar for the four aerosol motifs, although the widths of the signatures are greater for sea salt and dust corresponding to the larger effective radii and lower Angstrom exponents for these predominantly natural species. The signatures of low- and high-altitude clouds (Figure 4b) are distinguished by periodic features associated with the primary and overtone absorption bands of water vapor in the near infrared present in the low-cloud signatures and largely absent in the high-cloud signatures. Since the low clouds are below most of the tropospheric vapor, the solar radiation reflected from the tops of these clouds passes through and hence is extinguished by the vapor twice before reaching the satellite sensor. Also, the difference in phase leads to different optical properties between ice and water clouds which can be seen at 1600 nm and 2200 nm. As expected, the reflectance signatures of large changes in greenhouse gases (Figure 4c) are spectrally distinct due to the narrow vibration-rotation bands and associated absorption lines of these species. In principle all of these signatures should be the easiest to detect and attribute in the simulated and observed spectra, but formal statistical analysis is necessary to interpret the signatures.

[36] While the simulations for the experiment and the control cover the 21st century, several prominent features of the system forcing and response can most readily be seen in changes to the broadband albedo, broadband fluxes, and spectral reflectances between decade-long samples from the two simulations. In Figures 5a and 5f, differences are shown between the middle and beginning of the 21st century in seasonal broadband albedo, TOA broadband flux, and spectral reflectance are shown for all-sky and clear-sky conditions. Figures 5c and 5d illustrate the high level of agreement between the changes in reflected solar flux calculated with MODTRAN and the CCSM radiative transfer algorithm under both clear-sky and all-sky conditions. The globally averaged differences between the two lines are indicated in Figures 5c and 5d and are largely due to the fact that CCSM is a two-stream code while MODTRAN is an eight-stream, multiple-scattering code. According to the prescription for the A2 SRES emissions scenario, there is a peak in aerosol forcing due to anthropogenic sulfate and carbonaceous emissions in the middle of the 21st century, and the simulated reflectance spectra indicate this forcing. Because there may be varying degrees of cloud contamination, Figure 5e is representative of trends in the shortwave reflectance measurements that are most directly measured by an instrument like CLARREO. However, depending on the instrument's footprint, clear-sky measurements may also obtainable. As shown in Figure 5f, the resulting increases in simulated visible reflectances from 2050 to 2059 relative to those from 2000 to 2009 in clear-sky conditions are quite evident in regions of the Northern Hemisphere with the largest projected anthropogenic emissions of aerosols and aerosol precursors. Spectral signatures of the declining area of the Arctic and Antarctic sea ice packs are also evident near both poles. At midcentury, the signals in the all-sky spectra exhibit both masking of the surface radiative feedbacks by overlying cloud decks and emergence of cloud feedback signals associated with the ITCZ and low-altitude cloud systems in the southern subtropics. It is important to note that these cloud radiative feedbacks differ widely across the multimodel ensemble assembled for the AR4 [IPCC, 2007], and hence the reflectance signatures of cloud feedbacks in this first realization of the OSSE are model-specific features.

Figure 5.

(a) Difference between decadal average of December–January–February all-sky TOA albedo for 2050–2059 and 2000–2009 from the A2 simulation (GHG and aerosol forcing). (b) Same as Figure 5a but for clear-sky albedo. (c) Same as Figure 5a but displaying zonally averaged all-sky TOA reflected shortwave fluxes (note that positive numbers indicate upward flux) from CCSM and MODTRAN. (d) Same as Figure 5c but for clear-sky fluxes. (e) Same as Figure 5a but for zonally averaged all-sky TOA spectral reflectances. (f) Same as Figure 5e but displaying clear-sky reflectances.

[37] These and other emerging signals of climate feedbacks are amplified by the end of the IPCC simulations. In Figures 6a6f, differences in broadband albedo, zonally averaged broadband fluxes, and spectral reflectances are shown between the end and the beginning of the 21st century. Recall that in this particular emissions scenario, the concentration of CO2 will triple relative to pre-industrial abundances by the end of the Century, but aerosol burdens will be comparable to present-day values due to the imposition of air pollution controls. The resulting changes in cloud patterns and snow cover and sea ice are substantial and can be seen in the broadband albedo plots in Figures 6a and 6b. These features are also present in the reflectance spectra. One of the striking features seen in Figure 6e is the zonally averaged all-sky spectral reflectance signatures associated with changes in the location of the Intertropical Convergence Zone (ITCZ) and the poleward migration of the stratus clouds and the storm tracks. The zonally averaged reflectance spectra show the migration of the storm tracks in the southern Hemisphere but less apparent in the northern Hemisphere due to the cancellation of increase in clouds and decreased snow cover. The zonally averaged spectral reflectance features of snow and sea ice are especially evident in the clear-sky spectra but are less evident in the all-sky spectra due to CCSM's prediction of increased cloud cover over those regions where snow and sea ice recede. Also, the OSSE simulations indicate that while the clouds and snow and sea ice largely are approximately equal and opposite near the Arctic for zonally averaged all-sky broadband albedo, it is evident from the reflectance spectra that the spectral features are distinct and of opposite sign between the visible and the near-infrared. In particular, over Alaska and western Russia, snow is predicted to decrease dramatically while cloud cover is predicted to increase in a compensatory manner but there is a dipole that is evident in the spectral reflectance.

Figure 6.

Same as Figure 5, but plots describe differences between 2090 and 2099 and 2000–2009.

[38] Especially when compared to Table 1, which summarizes the mean differences between the end and the beginning of the 21st century for the control and the experiment, it is clear that both the broadband albedo and reflectance spectra are very different between the two emissions scenarios. This table indicates that the scale of the changes in broadband albedo from the experiment are roughly an order of magnitude larger than the scale of the simulation's natural variability as determined from the variability in the control run. Similarly, changes in reflectance of the experiment at 800 and 950 nm, which indicate the response in a window region and in a wing of a water vapor absorption feature, respectively, are also an order of magnitude larger as compared to the control. Also of note in Table 1 is the feature that changes in clear-sky broadband albedo and reflectance are generally larger signals than all-sky changes. Also, it should be mentioned that the percentage change in the experiment both at 800 and 950 nm are considerably larger than the percentage change in broadband albedo associated with the experiment, thereby suggesting that there may be information contained in reflectance that will hasten detection of changes in climate state variables.

Table 1. Mean Zonally Averaged Change in Broadband Albedo (Δα) for a Constant-Concentration Simulation (CC) and the A2 Scenario Simulation for December, January, and Februarya
 CC ClearA2 ClearA2/CC ClearCC AllA2 AllA2/CC All
  • a

    Also displayed are the mean changes in zonally averaged reflectance ΔR800 and ΔR950 corresponding to the CLARREO channels at 800 and 950 nm, respectively. Results are displayed both for the control simulation (constant-concentration) and the experiment simulation (A2 scenario).

equation image DJF 2090s to 2000s−0.0014−0.01269.000.0004−0.0066−16.5
equation image DJF 2090s to 2000s−0.0017−0.01699.94−0.0013−0.013210.16
equation image DJF 2090s to 2000s−0.0023−0.01486.43−0.0022−0.01235.59

4. Discussion

[39] This work describes the simulator for the first global climate shortwave observing system simulation experiment for a hyperspectral instrument covering a spectral region from the near-UV to the near-IR. OSSE calculations have been completed over the 21st century for a forced and an unforced emissions scenario as described in section 3.

[40] The simulations show that the large increase in aerosol loading by the middle of the 21st century as prescribed by the forced emission scenario is readily detectable in both clear- and all-sky reflectance spectra. The spectral signals from differencing data in the 2090s with data from the 2000s show shifts in the ITCZ and contrast strongly with those associated with shifts in storm tracks and the stratus clouds which is evidenced by the banded structure of the seasonally averaged spectral reflectance differences where lower clouds shifts occur. Moreover, these changes are qualitatively very different from the signals observed in the control simulation and occur in those parts of the spectrum where water vapor exhibits strong absorption features. Furthermore, we find that there may be added utility from CLARREO full spectral measurements over broadband or discrete band (for example, MODIS) measurements, and in subsequent papers, we will quantify the added utility of CLARREO in terms of shorter times to detection of climate change trends, etc.

[41] These findings suggest that shortwave reflectance spectra such as those from CLARREO may be able to detect changes in clouds, aerosols, sea ice, and snow cover at the middle and end of the 21st century. This is not meant to suggest that CLARREO will have to be operational for many decades before scientific utility can be derived from its measurements. Rather, shortwave reflectance spectra contain signals that are relevant to climate forcings and feedbacks. Formal methods to detect trends in the data that are statistically significant, and formal methods to attribute trends in the data to forcings and feedbacks in the climate system are both required on the time series to make determinations about the time scale at which these spectra would be useful.

[42] As mentioned in section 1, analysis based on OSSE results is predicated on using climate model results as an authentic surrogate for the real climate system. The OSSE described in this paper has several limitations that warrant discussion. First and foremost, the underlying data which form the inputs to the OSSE are based on the CCSM climate model which represents only one possible realization of the climate trajectory for the 21st century. Also, the radiative transfer emulator, while extensively validated, has its own set of limitations including the inability to capture projected changes in nonfrozen land surface vegetation or composition due to the observationally based methods we have employed for the surface radiative boundary conditions.

[43] Because of these simplifications, results derived from analysis of OSSE data probably provide an upper bound on CLARREO's performance in terms of the mission's ability to achieve its stated science requirements. Therefore, OSSE results represent a necessary but not sufficient condition for determining the possible utility of CLARREO to detect and attribute climate change since the OSSE most likely overestimates climate change signals relative to unforced internal climate variability. To compare OSSE results with the climate system, it is necessary to use existing shortwave spectral reflectance measurements and compare them with OSSE simulations of present-day conditions. For example, the continuous visible spectra from the SCIAMACHY instrument [Bovensmann et al., 1999], though originally designed to detect chemical constituents, can characterize how well spectral variability in present-day climate model simulations compares to real-world spectral variability [Roberts et al., 2009].

[44] With the simulations in hand, there is considerable work to be done in order to analyze the simulated reflectance spectra. First, it is necessary to perform trend detection studies [Leroy et al., 2008] using CLARREO spectra and instrument models for existing satellite instruments. Subsequently, it will be necessary to attribute the simulated changes in CLARREO reflectance spectra to forcing terms such as aerosols and greenhouse gases and manifestations of feedbacks in snow coverage, sea ice extent, and clouds using fingerprinting methods [e.g., Leroy and Anderson, 2010]. It should be noted that many of the simulated changes in the reflectance spectra including those arising from aerosols, snow and sea ice, and clouds do not have sharp spectral features but vary slowly with wavelength as seen in Figures 4a4c, and this will tend to complicate efforts to attribute changes in the reflectance spectra to specific forcings and feedbacks. Formal analysis using detection and attribution methods [International Ad Hoc Detection and Attribution Group, 2005] will be required to determine how separable and attributable these signals are. Also, it will be critical to explore how the different processes that contribute to climate sensitivity in CCSM [Kiehl et al., 2006] manifest themselves in the simulated reflectance spectra with a particular focus on the shortwave cloud response exhibited by CCSM3 run at different horizontal resolution. Furthermore, given the sensitivity of reflectance spectra to snow and sea ice, it is critical that the OSSE be utilized to explore CCSM3′s cryospheric feedbacks on snow [Lawrence et al., 2008] and sea ice [Holland et al., 2006], especially in the context of ensemble runs which may capture the abrupt sea ice losses that have been observed recently [Holland et al., 2011].

[45] Additionally, there are some mission parameters for CLARREO that need to be addressed with the OSSE. First, the inclination angle of the orbit for CLARREO is likely to be different from other polar-orbiting sounders and our OSSE does not currently reflect this. Work by Anderson et al. [2004] and Kirk-Davidoff et al. [2005] have found that the infrared instrument will be more likely to achieve its science objectives with an orbital inclination angle at or near true polar. This orbit exhibits a precessing local equator crossing times that will significantly impact shortwave spectra. Since reflectance spectra vary nonlinearly with incoming solar zenith angle, the OSSE needs to be reconfigured to produce simulated spectra for different orbits which can be accomplished by sampling the probability distribution function of solar zenith angles for each grid box implied by the orbit under consideration. Second, the plan for the CLARREO mission prescribes multiple formation-flying platforms, and the OSSE needs to be reconfigured accordingly and this can be achieved by calculating the PDF of solar zenith observation angles for multiple formation- or even non-formation-flying platforms.

[46] Most importantly, it is necessary to implement panspectral (simultaneous shortwave reflectance and longwave radiance) calculations to explore the synergy between the shortwave and infrared instruments, which may enhance the science value of the CLARREO mission. This approach has been successfully employed for the joint analysis of infrared and radio occultation measurements [Huang et al., 2010b] but it remains to be demonstrated that it can be employed for the shortwave and longwave spectral combination in addition to the radio occultation. As can be seen in Figure 7, it may be quite scientifically fruitful to address this question because while the IR can readily detect temperature and water vapor profiles and cirrus clouds, measurements between 320 and 2500 nm can elucidate the presence of stratus clouds. Moreover, since low-altitude clouds have been identified as dominant contributors to the intractability of the disagreement in climate sensitivity between climate models [Cess et al., 1996; Stephens, 2005; Bony et al., 2006; Dufresne and Bony, 2008], such panspectral analysis is crucial toward interpreting future measurements as they relate to important, but not well-understood climate feedbacks.

Figure 7.

(a) Simulated CLARREO shortwave spectrum over an ocean surface with a midlatitude summer model atmosphere [Anderson et al., 1986] with a solar zenith angle of 30°, clear-sky conditions, a liquid stratus cloud between 925 and 979 mbar, a cloud water path of 0.2 kg/m2, cloud fraction of 100%, cloud effective radius of 14 μm, the Cox-Munk BRDF model [Cox and Munk, 1954] with a wind speed of 5 m/s for shortwave calculations, sea surface temperature of 293.8 K, and a spectrally uniform infrared emissivity of 0.05. (b) Same as Figure 7a but for CLARREO infrared spectra.

[47] The shortwave OSSE that we have developed for this paper has been oriented around the archived results of CCSM, and this model's simulations represent only one possible realization of Earth's future climate. In furtherance of this research, it will be necessary to configure the shortwave OSSE so that it can perform simulations based on different ensemble member runs of CCSM and based on the Coupled Model Intercomparison Project Phase 5 (CMIP5) archived climate model data [Meehl et al., 2007]. The CMIP5 contain disparate cloud feedback strengths [i.e., Stephens, 2005] and it will be scientifically interesting, though computationally intensive, to explore how those differences are realized in simulated CLARREO reflectance spectra.


[48] Funding for this research was supported by the following NASA grants: NNX08AT80G, NAS2–03144, and NNX10AK27G. This work was also supported by Contractor Supporting Research (CSR) funding from Berkeley Laboratory, provided by the Director, Office of Science, of the U.S. Department of Energy under contract DE-AC02-05CH11231. Additionally, NASA High-End Computing grants SMD-08-0999, SMD-09-1397, and SMD-10-1799 provided computational resources produce the simulations. The following individuals also provided considerable assistance with this research: Don Anderson of NASA Headquarters; Tsengdar Lee of the NASA Science Mission Directorate; Gail Anderson of AFRL; Vincent Ross of Aerex Corp.; Lex Berk of Spectral Sciences, Inc.; David Young, Bruce Wielicki, Zhonghai Jin, and Rosemary Baize of NASA Langley Research Center; Crystal Schaaf and Zhousen Wang of Boston University; and the entire NASA High-End Computing technical support team of NASA User Services. We also wish to thank the three reviewers for their insightful comments and suggestions.