1. Introduction
Clouds are a major driver of Earth’s climate (Yao and Del Genio 1999; Andrews et al. 2012; Stephens 2005) in part because they alter Earth’s radiative balance by reflecting incoming solar radiation (e.g., the planetary albedo) and absorbing outgoing terrestrial radiation (e.g., the greenhouse effect). Specific properties of clouds—including phase, droplet concentration and size, and liquid water path—influence how much clouds absorb and scatter incoming solar radiation (Kokhanovsky 2004) and change how direct and diffuse radiation are partitioned at Earth’s surface. If an increase in the fraction of diffuse photosynthetically active radiation (PAR; 400–700 nm) reaching plant canopies enhances photosynthesis and canopy-level productivity (Roderick et al. 2001; Kanniah et al. 2013a), then clouds could also influence global climate through the carbon cycle. A combination of modeling and observation-based studies have demonstrated that plant canopy uptake of atmospheric carbon dioxide (CO2) can increase with diffuse light because of a shift in the distribution of light within the canopy. Typically under clear-sky conditions, upper-canopy leaves are near light saturation, while leaves in the lower canopy are shaded and light limited (Urban et al. 2012). When diffuse light increases, more light reaches lower-canopy leaves and their contribution to total canopy photosynthesis increases (Knohl and Baldocchi 2008; Urban et al. 2012). This has been described as the “diffuse fertilization effect,” where increases in diffuse light increase the total canopy photosynthesis. Changes in other environmental factors, such as temperature and vapor pressure deficit, have also been identified as mechanisms through which cloud cover and diffuse light can alter CO2 uptake by plant canopies (Min 2005; Gu et al. 1999).
At the ecosystem level, the effect of diffuse light on ecosystem carbon uptake, or productivity, as measured or derived by eddy-covariance towers [either as net ecosystem exchange (NEE) or gross primary productivity (GPP)], has been demonstrated to occur at multiple biomes, including deciduous forests, evergreen forests, maize fields, soy fields, grasslands, and tundra shrublands (Cheng et al. 2015; Lee et al. 2018; Alton 2008). However, there is wide uncertainty around how strongly diffuse light and cloud conditions affect ecosystems, with estimates varying from NEE rates that increase up to 150% under cloudy conditions relative to clear-sky conditions (Urban et al. 2007) to an estimated increase in GPP of 0.94% when the portion of diffuse light increases by 1% (Lee et al. 2018). In addition, much of the previous work linking the diffuse light effect on ecosystem productivity to clouds has relied on multiple methods for estimating cloud properties. These methods range from using categorical descriptions of clouds (Niyogi et al. 2004), a ratio of total surface radiation to modeled clear-sky radiation (Law et al. 2002), radiometers (Min 2005), and ceilometers (Oliphant et al. 2011). Although these studies provide an understanding of clouds, diffuse light, and ecosystem productivity at individual sites, the differences in these methods increase the difficulty in scaling results from individual sites to larger spatial scales. While some studies have evaluated the role of diffuse light using computational models at regional to global scales (e.g., Alton et al. 2007; Mercado et al. 2009; Keppel-Aleks and Washenfelder 2016), the response of plant canopy carbon uptake to different cloud conditions has not been constrained by observations on a global scale.
To identify whether the “diffuse fertilization effect” that has been found at individual ecosystems has a significant role at the global scale, we compare MODIS Level 3 COT with datasets of global terrestrial productivity. To account for differences and limitations in how productivity can be derived at the global scale, we use two datasets to represent productivity: solar-induced terrestrial chlorophyll fluorescence (SIF) and GPP upscaled from eddy-covariance tower data. We also examine the longer-term impacts of the diffuse fertilization effect using monthly-averaged datasets for both productivity and COT. Combining these three datasets at larger spatial and longer temporal scales extends previous work by 1) identifying the times, locations, and land-cover types across the globe that respond most strongly to the diffuse fertilization effect due to clouds and 2) estimating the global impact of diffuse light on Earth’s terrestrial carbon sink. By quantifying the global-scale effect of clouds and their optical properties on land–atmosphere CO2 fluxes, our results can inform future model development aimed at improving the confidence in model projections of the net effects of clouds on global climate.
2. Data and methods
2.1. Monthly COT
To examine the seasonal impacts of clouds on terrestrial productivity, we use monthly-averaged combined-phase COT from 2001 to 2015. We obtained COT data from MODIS/Terra Level 3 Atmosphere Monthly Global Product (MOD08_M3) Collection 6, which is available globally at 1° × 1° resolution. Statistics for the Level 3 monthly product are computed from the Level 2 daily product, which is available at 1 km × 1 km resolution (Hubanks et al. 2016). COT is retrieved only during the daytime using a plane-parallel model that assumes vertical and horizontal homogeneity of clouds (Pincus et al. 2012) when the solar zenith angle is less than or equal to 81.37° and the Terra overpass time at the equator is around 1030 local solar time in its descending (daytime) mode (Hubanks et al. 2016). COT parameters are classified by cloud phase, which includes ice clouds, liquid water clouds, undetermined clouds (i.e., for which the retrieval algorithm cannot determine cloud phase and the retrievals have lower confidence than those in other phase categories), along with combined-phase clouds (i.e., a combination of all cloud-phase categories: liquid water, ice, and undetermined). In this study, we use the combined-phase COT to capture the overall effect of clouds on the diffuse fraction of solar radiation.
The MOD08_M3 product uses the MODIS Collection 6 Clear Sky Restoral Algorithm (Meyer et al. 2014) to quantify the cloudiness of each pixel, leading to separate COT retrievals from “overcast” and “partly cloudy” conditions. Pixels are categorized as partly cloudy when they fail to meet the homogeneous-cloud-layer assumption, and retrievals of these partly cloudy pixels have lower confidence than overcast pixels (Pincus et al. 2012). Thus, we use monthly-averaged COT retrievals under “overcast” conditions in our study.
2.2. Carbon assimilation: SIF and GPP
Rates of carbon uptake by vegetation are difficult to measure directly across broad spatial scales. In this study, we use two different global datasets as proxies for terrestrial carbon uptake. The first dataset is based on SIF. SIF represents the broadband emission of photons by vegetation at longer wavelengths (650–800 nm) relative to PAR, which releases the excess energy absorbed by chlorophyll but not used for carbon fixation or emitted as heat (Guanter et al. 2012). Measurements of SIF are directly related to photosynthetic function (Joiner et al. 2013) and can be used as a proxy for GPP (Yang et al. 2015). We use satellite-derived SIF data from the Global Ozone Monitoring Instrument 2 (GOME-2_F) level 2 product (Joiner 2013), available from 2007 to 2015. The equator overpass time of GOME-2 is approximately 0930 local time, and the retrieval solar zenith angle is less than 70° (Joiner et al. 2013). For temporal and spatial consistency with the COT data, we regrid the daily GOME2_F level 2 SIF data from 0.5° × 0.5° resolution into monthly 1° × 1° resolution by filtering out data with retrieval errors above 5.0 m W m−2 sr−1 nm−1 (typical retrieval errors are between 0.1 and 1.0 mW m−2 sr−1 nm−1) and averaging all data points within each 1°× 1° grid cell.
The retrieval of SIF by GOME-2 takes advantage of solar Fraunhofer lines in a region of the solar spectrum that remains structurally unaffected by Earth’s atmosphere (Guanter et al. 2012). However, to account for potential effects of clouds on SIF retrieval, a screening for cloud contamination was applied to the dataset by retaining only SIF retrievals with effective cloud fractions feff of <30% (Joiner et al. 2012, 2013). COT can be converted to feff by multiplying the geometric cloud fraction fg by the transmittance parameter A that is dependent on both COT and surface reflectivity: feff = fgA (Joiner et al. 2012). To illustrate the concept of effective cloud fraction and its influence on SIF retrieval, we present two hypothetical cloud cover situations that result in feff = 30%. In one case, thick clouds completely block out a fraction of satellite footprints [each footprint is 40 km × 40 km (Joiner et al. 2018)] within a 0.5° × 0.5° grid cell, so the SIF value of this grid cell comes from the other footprints that are under clear-sky conditions and does not reflect the influence of COT on SIF. In contrast, when thin clouds uniformly cover an entire footprint, the SIF retrieval of this footprint does come from underneath the cloud, and thus reflects the influence of COT on SIF. The overall SIF retrievals likely represent a range of cloud conditions between these two extreme cases, and is likely biased toward clear-sky conditions. For typical vegetation and a COT of 10 (the threshold between the low-COT and high-COT regime, defined in section 2.4), A is approximately 0.5, indicating that roughly one-half of the undercloud SIF signal is still observable to satellites. For conditions in which COT ≤ 10 and fg ≤ 60%, feff should be ≤ 30% and the SIF retrieval is retained in our dataset. Therefore, while cloud impacts on SIF retrievals become more problematic for COT > 10, they are relatively small for COT ≤ 10. For this reason, we only include the low-COT regime (COT ≤ 10) analysis for the SIF data.
The second productivity dataset is a global GPP data product from the Biogeochemical Model-Data Integration Group of the Max Planck Institute for Biogeochemistry (MPI-BGC) (Jung et al. 2011). The MPI-BGC flux estimates were developed by upscaling site-level flux network (FLUXNET) observations of CO2 fluxes to the global scale using a machine-learning technique known as model tree ensembles (MTE) (Jung et al. 2011) for 1982–2011. Jung et al. (2009) tested the MTE approach by training it to reproduce the GPP simulated by the Lund–Potsdam–Jena Managed Land Biosphere Model (LPJmL). They showed that although the FLUXNET training dataset covers less than 0.1% of the globe, with sparse data in the tropics, the extrapolation in the MTE is reasonable when compared with the LPJmL-simulated GPP (Jung et al. 2009). Variations in solar radiation in the MTE GPP model are accounted for using a satellite-derived fraction of absorbed photosynthetically active radiation (fAPAR) product as one of the multiple training variables (Gobron et al. 2006; Jung et al. 2011), which may not capture the effects of diffuse light on the biosphere. We note that the role of clouds on diffuse light may not be fully accounted for in this model. As a result, the MTE GPP may not capture the diffuse fertilization effect in its simulations of GPP; however, we include it in this analysis as it provides the only GPP product based on modeled and observed physical processes.
We employ monthly mean GPP estimates (kg C m−2 s−1, converted to g C m−2 month−1) at a 0.5° × 0.5° spatial resolution. We regrid these data to a 1° × 1° resolution by calculating the unweighted average of the neighboring four 0.5° × 0.5° pixels for spatial consistency with the COT data.
2.3. Land-cover data
To identify relationships between COT and terrestrial carbon uptake by land-cover type, we use a land-cover product provided by the European Space Agency Climate Change Initiative (ESA-CCI) (ESA-CCI 2010). This land-cover dataset is selected for consistency with the land-cover categories implemented in the MPI-BGC GPP product. The land-cover data are regridded to 1° × 1° resolution from the original 300 m spatial resolution using the aggregation tool provided by ESA-CCI, which computes the percentage of each land-cover type in each 1° × 1° pixel.
2.4. Data analysis
We aim to identify the global response of terrestrial ecosystems to clouds and the diffuse fertilization effect. Previous analysis using MODIS Level 2 COT (1 km × 1 km resolution; 5-min data granules) and 30-min-averaged diffuse light measurements from eddy-covariance towers identified that diffuse light increases with COT up to a threshold of approximately 7 and then subsequently decreases with higher COT (Cheng et al. 2016). In our study, we round up this COT threshold to 10 to account for uncertainties in the thresholds across the globe. The threshold divides the globe into two regimes: 1) a low-COT or “diffuse positive” regime (COT ≤ 10), in which diffuse PAR increases with increasing COT, and 2) the high-COT or “diffuse negative” regime (COT > 10), in which diffuse PAR decreases with increasing COT. It is in the low-COT regime that the diffuse fertilization effect would be detected.
We then analyze the 1° × 1° data for COT, SIF, and GPP in three steps. First, to identify regions that are frequently under the low-COT regime, we map the temporal frequency of COT below 10 by calculating the cumulative distribution function (CDF) of COT for each pixel in each season (Figure 2). Second, to quantify the relationship between optically thin clouds and terrestrial carbon uptake, we identify regions around the globe where the low-COT regime frequently occurs during a region’s growing season by selecting pixels that meet the following two criteria: 1) a high likelihood of being in the low-COT regime, defined as the CDF of low COT (COT ≤ 10) greater than 20% (Figure 2), and 2) high photosynthetic activity during the growing season [June, July, and August (JJA) for the Northern Hemisphere, and December, January, and February (DJF) for the Southern Hemisphere], defined as the seasonal averaged SIF value greater than 0.3 mW m−2 sr−1 nm−1. Criterion 2 is implemented to exclude low plant productivity areas such as deserts. (Regions that meet these criteria are shown in Figure 3.) We subsequently sample nine subregions (black boxes in Figure 3) from these regions for the correlation analysis described in the next section. We focus on the growing season because it is when vegetation canopies have the highest photosynthetic activity. Consequently, this would be the time period when COT would have the greatest effect on carbon uptake. Additionally, most biomass burning in the tropics occurs during the dry season (Crutzen and Andreae 1990), when the effect of aerosols on diffuse light may interfere with that of clouds. As biological growing seasons tend to coincide with geographical wet seasons, the confounding effect of aerosols can be avoided in regions where biomass burning may be important for the carbon response (Rap et al. 2015). Finally, the influence of other confounding factors, like leaf phenology, on the efficiency of carbon uptake can be minimized by focusing on the growing season.
Third, we conduct a linear regression analysis for each selected region to investigate the correlation between COT and carbon uptake using SIF and GPP during the peak growing season. Here, the peak growing season is defined specifically for each region as the three consecutive months with the greatest SIF or GPP values (the sum of three consecutive months using the regional averages of SIF or GPP). Using monthly data from each year and each 1° × 1° pixel within a region, we conduct a linear regression analysis to quantify the relationship between low (≤10) COT and SIF, and the relationship between low and high (> 10) COT and GPP. We use data for years when both COT and the carbon uptake data are available. That is, for the COT–SIF relationship we use COT and SIF data from 2007 to 2015, and for the COT–GPP relationship we use COT and GPP data from 2001 to 2011. Within each region, we use the land-cover classification of the ESA-CCI land-cover product to analyze regions based on land-cover type. We then evaluate the COT–GPP relationship for three land-cover types in our selected regions. Specifically, shrublands, forests, and croplands because these are the dominant land-cover types contributing to terrestrial carbon uptake (Beer et al. 2010). The ESA-CCI land-cover data specifies the percentage of each land-cover type within each 1° × 1° pixel. For the forest category, we sum the percentages of evergreen broadleaf, deciduous broadleaf, evergreen needleleaf, deciduous needleleaf and mixed forests. For shrublands, we include mosaic tree, shrub/herbaceous cover, and shrubland categories. For croplands, we include all crop categories (irrigated, rain-fed, or postflooding). A pixel is categorized as shrubland, forest, or cropland if it has over 50% areal coverage in one of these land-cover categories.
3. Results
3.1. Low-COT regions
The global seasonal climatology of COT is illustrated in Figure 1. Based on prior ground-based, site-level analysis (Cheng et al. 2016), we identify the diffuse positive regime as where COT ≤ 10 and diffuse light increases with clouds, implying a higher likelihood for clouds to positively affect terrestrial carbon uptake. Globally, the low-COT regime occurs predominantly at low latitudes. This could be explained by the contrasting effects of condensation and precipitation on the liquid water content of clouds, and thus COT (Tselioudis et al. 1992). Cloud liquid water content increases with temperature, whereas precipitation reduces liquid water content and counteracts the positive effect of condensation on COT. At low latitudes where temperature is high, precipitation forms more efficiently than at colder latitudes (Tselioudis et al. 1992). When the effect of precipitation on COT dominates over condensation, liquid water content of clouds and COT decreases. The exception at low latitudes is a narrow band of high COT around the equator that is most pronounced over the Pacific and Atlantic Oceans, where COT varies seasonally from ~15 to 30. Locations of this high-COT band are consistent with the intertropical convergence zone (ITCZ), which shifts between 9°N in Northern Hemisphere summer and 2°N in Northern Hemisphere winter (Schneider et al. 2014). Areas flanking the ITCZ band (i.e., between approximately 0° and 30° in both hemispheres) have COT lower than 10 throughout the year, but only a few of these areas are located over continents.
Global distribution of MODIS Level 3 COT seasonal climatology (the mean of three monthly means of each season from 2001 to 2015) for (a) December, January and February (DJF), (b) March, April and May (MAM), (c) June, July and August (JJA), and (d) September, October and November (SON).
Citation: Earth Interactions 23, 6; 10.1175/EI-D-17-0035.1
We identify regions that are frequently under a low-COT regime based on the cumulative probability function of low COT across the globe (Figure 2). Consistent with Figure 1, the tropics and subtropics have high frequencies of cloud optical thickness within the low-COT regime. These regions, including portions of the Amazon, southern Africa, Indonesia, southern and southeastern Asia, and northern Australia, are under low COT at least 20% of the time and up to 90% during their dry seasons (DJF for the Northern Hemisphere and JJA for the Southern Hemisphere). Additionally, extremely dry areas, such as the Sahara (Africa), the Taklamakan Desert (northwestern China), and the southwestern United States, also show up to 90% frequencies of low COT during their dry seasons.
Global CDF of COT less than 10 (i.e., the low-COT regime or “diffuse positive" regime) for (a) DJF, (b) MAM, (c) JJA, and (d) SON.
Citation: Earth Interactions 23, 6; 10.1175/EI-D-17-0035.1
Regions frequently under the low-COT regime are shown in Figure 3. We analyze the relationships between COT and terrestrial carbon uptake in nine subregions. Five of them are in the Northern Hemisphere: southeastern United States (SEUS; 30°–33°N, 82°–90°W) the Sahel (SAH; 2°–10°N, 20°–30°E), southeastern Asia (SEAS; 13°–17°N, 100°–104°E), India (IA; 13°–15°N, 77°–80°E), and Spain (SPN; 37°–41°N, 2°–10°W); and four are in the Southern Hemisphere: eastern Amazon (EAM; 15°–20°S, 50°–60°W), southern Africa (SAF; 17°–20°S, 20°–27°E), Indonesia (IO; 5°–8°S, 138°–142°E), and northeastern Australia (NEAU; 13°–18°S, 130°–133°E). These subregions represent the global land area that is frequently under the low-COT regime. As a result, these nine regions are where the diffuse fertilization effect could potentially be the most significant to terrestrial carbon uptake, and we focus only on these subregions in our analysis. Outside of these regions, diffuse light is unlikely to promote productivity because cloud optical thicknesses are too high (COT > 10) and diffuse and direct light both decline with cloud optical thickness (Cheng et al. 2016).
Selected regions that satisfy both 1) frequency (CDF) of the low-COT regime greater than 20% and 2) SIF greater than 0.3 mW m−2 sr−1 nm−1 during (a) JJA for Northern Hemisphere and (b) DJF for Southern Hemisphere. SIF values are represented by colored contours and are masked to show regions where COT probability > 20%. Black boxes are regions sampled for the correlation analysis of COT and SIF. Gray land areas indicate low-COT probability < 20%.
Citation: Earth Interactions 23, 6; 10.1175/EI-D-17-0035.1
Based on the MTE GPP product, we integrate the annual mean GPP of these nine subregions and find that they account for 5.8 Pg C yr−1 of the total global annual estimate of 118.3 Pg C yr−1, or 4.9% of the global annual mean GPP. This low fraction indicates that even if diffuse light has a strong effect on GPP, it would only influence ~5% of the global GPP based on the spatial distribution of low-COT regimes and its coincidence with the landmass.
We note that the seasonal average of standard deviation of COT can be large across the globe (results not shown). For example, the standard deviation of COT in SAH during JJA and EAM during DJF is approximately 14, but the values in SPN during JJA are lower (6–8). This suggests a strong variation in COT values retrieved in the low-COT regime, and thus the selection of low-COT regions may also be subject to uncertainty.
3.2. Correlation between COT and terrestrial carbon uptake
To understand the impact of COT on terrestrial carbon uptake, we evaluate the correlations between 1° × 1° COT and 1° × 1° carbon update data from the two proxies described in the Methods: SIF and GPP. Both COT and carbon uptake data used in this study are monthly values at a relatively coarse spatial resolution to extend our understanding of the diffuse fertilization effect at broad spatial and temporal scales. Because the peak growing season (defined in section 2.4) varies with location and hemisphere, we analyze the relationships between COT and carbon uptake for different time periods; for example, in Figures 4–6, July, August, and September (JAS); January, February, and March (JFM); May, June, and July (MJJ); and August, September, and October (ASO) are analyzed. Because of the potential cloud contamination of the SIF retrievals, we do not show the correlation of COT and SIF for COT > 10. The three different land-cover types (shrubs, forests, and crops) show varying relationships between COT and carbon uptake (Figures 4–6). However, linear correlations between COT and SIF are not statistically significant (p > 0.05; see Table 1), except for in two crop sites (SPN and SAF) under the low-COT regime. The slopes and the coefficients of determination R2 of these statistically significant cases are marked in boldface type in Figures 4–6. The significance of correlations between COT and GPP (p values not included in Table 1) is similar to those between COT and SIF. These results suggest that, overall, the relationships between COT and terrestrial carbon uptake are weak at the monthly temporal scale and global spatial scale. Subsequently, the diffuse fertilization effect from clouds does not appear to be a dominant driver of monthly, global carbon fluxes.
Scatterplots of (a),(b) COT vs SIF and (c),(d) COT vs GPP for shrubs in (left) SAH and (right) NEAU. The low-COT-regime regression, slope, and coefficient of determination R2 are shown in red, and those for the high-COT regime are shown in blue.
Citation: Earth Interactions 23, 6; 10.1175/EI-D-17-0035.1
Scatterplots of (top) COT vs SIF and (bottom) COT vs GPP for forests in (a),(e) SEUS, (b),(f) SAH, (c),(g) EAM and (d),(h) IO. The low-COT regime regression, slope, and R2 are shown in red, and those for the high-COT regime are shown in blue.
Citation: Earth Interactions 23, 6; 10.1175/EI-D-17-0035.1
Scatterplots of (top) COT vs SIF and (bottom) COT vs GPP for croplands in (a),(g) SAH, (b),(h) SEAS, (c),(i) IA, (d),(j) SPN, (e),(k) EAM and (f),(l) SAF. The low-COT regime regression, slope, and R2 are shown in red, and those for the high-COT regime are shown in blue. The slopes and R2 values of statistically significant correlations between COT and SIF are marked as boldface.
Citation: Earth Interactions 23, 6; 10.1175/EI-D-17-0035.1
Statistics of linear correlations between COT and SIF for different land-cover types, including p values of the low-COT regime. Land-cover type fraction (%) is defined as the ratio of the number of pixels of each land-cover type to the total number of pixels in each sampled region. The p values are only available for the dominant land-cover type of each region for which regression lines are available.
In shrublands (SAH and NEAU, Figure 4), there is approximately no change in carbon uptake (both GPP and SIF) as COT increases from 0 to 10 (effectively a zero slope), and R2 values are approximately zero (Figure 4). In SAH (Figures 4a,c) and NEAU (Figures 4b,d) under low COT, both the SIF and the GPP data show a slight positive slope, although both are very weakly correlated with COT. Under high COT, GPP in both SAH and NEAU show a positive relationship with COT but correlations are again weak and not statistically significant. This suggests that monthly carbon uptake in shrublands has a very weak response to the low-COT regime or the diffuse fertilization effect. In addition, only two shrub sites of limited area were found to be frequently under low COT, suggesting that most shrublands across the globe are unlikely affected by the diffuse fertilization effect. In the high-COT regime, we would expect the absorption of light by optically thick clouds to outweigh any enhancement in diffuse PAR and thereby reduce carbon uptake; yet the response at both shrub sites is a slight increase in GPP with COT. Overall, these results suggest that in shrublands across the globe, annual carbon uptake is unlikely to be affected by clouds in either the diffuse positive or diffuse negative regimes.
The four forest regions (Figure 5) show generally consistent relationships between COT and carbon uptake in the low-COT regime. Specifically, under the low-COT regime, SIF increases slightly with COT in two of the four regions (SAH and EAM), and GPP increases with COT in all four regions. In these four forest regions, correlations between COT and carbon uptake are very weak, with R2 values generally less than 0.01 (i.e., only less than 1% of variation in SIF or GPP is explained by COT). These results may indicate that either the diffuse fertilization effect is small in tropical and subtropical forests, or that the effect varies among forests of different types (e.g., evergreen broadleaf and deciduous broadleaf forests) at different locations, making the diffuse fertilization effect difficult to detect at monthly temporal and 1° × 1° spatial resolutions. In the high-COT regime, SEUS (Figure 5e) and IO (Figure 5h) show a slight decrease in GPP with COT, while SAH (Figure 5f) and EAM (Figure 5g) show the opposite. This observed relationship differs from previous findings, which suggests that the effect of high COT is to reduce carbon uptake of forests (e.g., Cheng et al. 2016). However, low R2 values suggest that this relationship is weak and high COT does not likely affect forest productivity at monthly and regional scales.
In croplands (Figure 6), SIF and GPP increase with COT under the low-COT regime, except for IA (Figures 6c,i) and SAF (Figures 6f,l), where SIF decreases with COT but GPP increases with COT, and SPN (Figures 6d,j) where GPP decreases with COT but SIF increases with COT. However, all correlations have low R2 values. Of all regions evaluated globally (Figures 4–6), the only two statistically significant correlations were in crop regions, yet they show opposite relationships: SIF increases with COT in SPN (Figure 6d) but decreases with COT in SAF (Figure 6f) under the low-COT regime. In the high-COT regime, the COT–GPP relationship yields the expected negative response at four of the six sites.
4. Discussion
In this study, we analyze the relationship between cloud optical thickness and carbon uptake in shrublands, forests and croplands in nine regions across the globe using monthly global data derived from satellite products (MOD08_M3 COT, GOME-2_F SIF) and gridded estimates of productivity (MTE GPP). We identify nine regions that are frequently under a low-COT regime (COT ≤10) during their respective peak growing seasons, and are most likely to experience a positive effect from diffuse light on carbon uptake. Outside of these nine regions, the COT values are consistently larger than those previously identified as driving a diffuse fertilization. Our analysis of COT therefore suggests that at large spatial and temporal scales, there are relatively few locations globally where there is likely to be a significant diffuse fertilization effect from clouds. In some of the nine regions, both SIF and GPP show consistent relationships with COT, for example, the low-COT regimes for forests and croplands in SAH, and for croplands in SEAS and EAM. However, in many regions, the carbon uptake response to COT varies depending on the carbon uptake proxy (SIF or GPP).
Detecting the diffuse fertilization effect at regional scales is difficult, in part because there are several sources of uncertainty in analysis. For example, satellite observations of COT and SIF are subject to retrieval errors, and there are also uncertainties in the gridded MTE GPP product. Both of the carbon uptake products may be limited by their ability to accurately account for diffuse light, as the SIF product may be biased toward clear-sky and the MTE GPP product only uses potential radiation as a training variable. Our analysis method is subject to uncertainties because of the spatial resolution we used to examine the diffuse fertilization effect at larger spatial scales, and the lack of inclusion of other variables such as temperature and soil moisture. We discuss these sources of uncertainty in more depth below, although we argue that these sources of uncertainty do not alter our conclusions.
As discussed in section 2.2, one source of uncertainty may be the potential cloud contamination in SIF retrievals. To avoid cloud contamination, the GOME-2_F SIF product eliminates pixels with cloud fraction > 0.3 (Joiner et al. 2013). Therefore, the SIF data may not represent high-COT conditions well, such that we do not include them into the correlation analysis shown in Figures 4–6. In addition, other uncertainties in satellite measurements may also affect our results. Both the MOD08_M3 COT and GOME 2_F SIF measurements exhibit a wide range of standard deviation over the years analyzed (Figure 7), suggesting large spatial and interannual variability in carbon uptake within a given region. This representation of uncertainty is region dependent; for example, COT has the greatest range of standard deviation in SEUS, and the smallest in NEAU, while SIF is more variable in EAM than other regions. Despite the role of other possible sources of variation, if the diffuse fertilization effect is strong and drives the carbon response, SIF or GPP should respond to COT in the low-COT regime in at least some of the nine regions, if not all. The observations indicate, however, that there are only weak relationships between COT and carbon uptake across all regions, suggesting that on monthly time scales, the diffuse fertilization effect will not significantly affect regional carbon budgets.
Monthly time series of COT (blue lines) and SIF (orange dashed lines) in nine regions: (a) SEUS, (b) SEAS, (c) SPN, (d) SAF, (e) NEAU, (f) SAH, (g) IA, (h) EAM, and (i) IO. The shading represents ±1 std dev from the mean COT (cyan shading) or SIF (orange shading). All values are averaged over the region and then taken as the average for each month over all years (COT from 2001 to 2015; SIF from 2007 to 2015 of the original 0.5° × 0.5° resolution).
Citation: Earth Interactions 23, 6; 10.1175/EI-D-17-0035.1
Also noted in section 2.2, uncertainties in the MTE GPP product may contribute to inconsistencies of the relationship between COT and GPP in both low- and high-COT regimes. The MTE upscaling may be affected by the heterogeneity of vegetation within each grid cell, uncertainties of FLUXNET measurements and the footprint mismatch between FLUXNET data and other driver data retrieved from satellites (Jung et al. 2009). We also note that in the nine regions selected for our analysis based on the COT threshold, only SPN and NEAU contain FLUXNET towers. The MTE product is therefore based mostly on extrapolation from the machine-learning algorithm outside our regions of interest. It is possible that the MTE GPP data do not represent the true response of GPP to diffuse light in the tropical and subtropical regions we have analyzed. Despite this limitation, direct observations from Kanniah et al. (2013a) are consistent with our results. In this study, flux tower observations within NEAU showed that enhanced canopy light use efficiency (LUE) under cloudy skies is insufficient to increase GPP due to the dramatic decline in total radiation. Another source of uncertainty in the MTE GPP product is the extent to which the algorithm includes the diffuse fertilization effect due to the limited solar radiation input data used to drive the model (Jung et al. 2011). We expect, however, that if there is a significant diffuse enhancement in terrestrial carbon uptake, the SIF product should respond with greater fidelity since these observations are not predicated on any assumptions about solar radiation input. Despite the very different methodological approaches to estimating terrestrial carbon uptake represented by SIF and MTE GPP, both data products provide a similar, noncorrelative relationship between COT and carbon uptake. This suggests that the diffuse fertilization effect from clouds has a negligible long-term impact on global productivity.
While finer temporal or spatial scales may reveal different responses, our analysis is fundamentally limited by the currently available datasets. The MTE GPP is limited to 0.5° × 0.5° spatial resolution and monthly temporal resolution. While individual SIF footprints are available, it is necessary to average multiple retrievals (either in space or time) as these data are of relatively low precision. These time and space constraints, in conjunction with the fact that the MODIS Level 3 COT data are a monthly 1° × 1° product, form the rationale for the space–time resolution selected for our analysis. We further note that, while 1 km × 1 km MODIS Level 2 COT data are available, filtering and averaging these data to the 0.5° × 0.5° grid at the global scale is computationally expensive given the geometry to convert from 1 km to degrees. In addition, higher-resolution data have previously been used to study the effects of COT on GPP in temperate forests and croplands. For example, Cheng et al. (2016) used 5-min data granules from MODIS Level 2 data for COT and half-hourly or hourly AmeriFlux measurements for GPP. They examined the relationship between diffuse light and COT at two resolutions (i.e., 1 × 1 km2 and the average for a 3 × 3 km2 area) and found that the larger spatial resolution may be better at explaining the effect of diffuse light on COT because it captures more spatial heterogeneity that could affect ground-based diffuse light measurements. This comparison suggests that using a higher temporal or spatial resolution in our own study may not influence the results of the relationship between COT and carbon uptake. We anticipate that in the future, machine-learning approaches may enable the use of several different data products together to infer patterns at higher spatial and temporal resolution than is available from any single satellite (e.g., Luo et al. 2018).
Other environmental variables, such as temperature, precipitation, and vapor pressure deficit (VPD), are known drivers of carbon uptake (Beer et al. 2010; Nemani et al. 2002; Kanniah et al. 2012, 2013b). Wu et al. (2016) used eddy-covariance-derived carbon data with models in Amazon forests to show that solar radiation, diffuse light fraction, and VPD are responsible for 75% variations in carbon uptake on an hourly scale. However, on monthly scales, they account for only 3% of carbon uptake, while biotic factors such as leaf phenology have a stronger influence. In our analysis, we do not account for the effects of temperature and precipitation on carbon uptake, which may provide one reason for the lack of correlation with COT or the inconsistent responses between SIF and GPP to COT (Figure 5). Therefore, clouds and diffuse light may have stronger effects on terrestrial carbon uptake only on short time scales, while on monthly or longer time scales, carbon uptake is controlled by other environmental or biotic factors that affect photosynthesis. Moreover, temperature-driven interannual variations in carbon uptake of tropical ecosystems may also affect the relationship between COT and SIF or GPP (Wang et al. 2013; Clark et al. 2003). Based on the fact that we identified the potential hot spots of the diffuse fertilization in the tropics and subtropics, where radiation contributes less than 10% of spatial variations in carbon uptake of forests, shrublands, and croplands (Beer et al. 2010), the amount of incoming solar radiation may not be a strong limiting factor to photosynthesis for these land-cover types. Prior studies have observed that favorable environmental conditions (e.g., low VPD and temperature) are necessary for LUE to increase under high diffuse fraction conditions (Kanniah et al. 2013a). Without these additional favorable environmental conditions, the positive effect of diffuse light on terrestrial carbon uptake may not be detectable.
The weak relationship found here between COT and metrics of carbon uptake (i.e., SIF and GPP) contrasts findings from point-based observational studies (Gu et al. 1999; Min 2005; Min and Wang 2008; Bai et al. 2012) that suggest a significant increase in ecosystem productivity under thin cloud conditions in temperate forests and tropical shrubs. Our analysis uses different temporal (monthly) and spatial (global) scales from previous studies, which typically focus on single ground-based sites. At individual sites, local environmental factors may play more important roles in canopy carbon uptake. For example, findings of Min (2005) and Bai et al. (2012) are based on data of short time periods (growing season of only one year), which might not be consistent across multiple years due to climate interannual variability in climate. Studies of Min and Wang (2008) and Gu et al. (1999), on the other hand, are based on multiyear data available at high temporal resolution (hourly or half hourly) from FLUXNET towers, whereas our study is based on monthly satellite-derived COT and SIF data, as well as global GPP data upscaled from site measurements. These differences indicate that the importance of clouds and thus, diffuse light on terrestrial carbon uptake, may vary across temporal and spatial scales.
Our results indicate that the relationship between COT and ecosystem productivity at individual sites does not remain constant at longer time periods and broader spatial scales. This has important implications for how we can estimate the global diffuse fertilization effects from clouds. Spatial resolution may also influence other factors that affect the relationship between COT and ecosystem productivity at the spatial and temporal scales implemented in this study. For example, a 1° × 1° area will have wide ranges of water and nutrient availability in soils, which are important environmental drivers of productivity that we did not explore in the study. Clouds also exhibit a strong variability across these spatial scales, with the added temporal variability due to their intermittent nature. This spatial and temporal variability may explain some of the differences between the point-based studies and the results presented here. However, the use of the monthly satellite-derived product determines the prevailing cloud characteristics over much of the globe. If the diffuse fertilization effect was persistent, then we would expect to see a response in the carbon data at these spatial and temporal scales.
5. Conclusions
This study provides the first attempt to quantify the diffuse fertilization effect from clouds over multiple years and broad spatial scales from an observational perspective. Using satellite-derived global datasets of cloud optical thickness with solar-induced chlorophyll fluorescence and gross primary production data, we investigate the influence of clouds on terrestrial carbon uptake through changes in diffuse light and total light. While prior studies have looked at these effects at individual sites, our study broadens the spatial scope to understand the importance of the diffuse radiation effect on the global terrestrial carbon sink. We find that regions frequently under cloud conditions where diffuse light increases with cloud cover (i.e., the diffuse positive COT regime) are primarily in the tropics and subtropics. However, in these regions, COT is weakly correlated with terrestrial carbon uptake (represented by SIF and GPP) on a monthly time scale. Areas outside of these regions fall under the diffuse negative COT regime, where diffuse light decreases with COT and clouds are more likely to have a negative effect on productivity due to a reduction in light availability. Therefore, in contrast to previous findings for individual sites, we conclude that the global effect of thin clouds on terrestrial carbon uptake is weak.
Of the nine regions frequently in the low-COT regime, the total annual mean GPP accounts for only 4.9% of the global annual mean GPP. Because the diffuse light is less likely to have a positive effect on productivity in areas outside of these nine regions, the diffuse fertilization effect in the nine regions represents an estimate of the strength of the effect on a global scale. Therefore, even if local correlations between COT and carbon uptake are strong, they will likely not scale to large changes in the global carbon budget because only a small fraction of vegetation canopies are affected by the diffuse fertilization effect. Other environmental or biotic factors could be major drivers of photosynthesis and have more important impacts on terrestrial carbon uptake on a monthly time scale across the globe, such as precipitation, soil moisture, and VPD.
Acknowledgments
This work was supported in part by National Science Foundation Grant AGS 1242203 awarded to A. L. Steiner and NASA Grant NNX15AH13G awarded to G. Keppel-Aleks. We obtained the MODIS COT data from https://modis-atmos.gsfc.nasa.gov/MOD08_M3, the GOME-2_F SIF data from http://acdb-ext.gsfc.nasa.gov/People/Joiner/my_gifs/GOME_F/GOME-F.htm, the MTE GPP data from the Biogeochemical Model-Data Integration Group of the Max Planck Institute for Biogeochemistry, and the ESA-CCI global land-cover data from https://www.esa-landcover-cci.org/?q=node/158 (ESACCI-LC-L4-LCCS-Map-300m-P5Y-2010-v1.6.1).
References
Alton, P. B., 2008: Reduced carbon sequestration in terrestrial ecosystems under overcast skies compared to clear skies. Agric. For. Meteor., 148, 1641–1653, https://doi.org/10.1016/j.agrformet.2008.05.014.
Alton, P. B., P. R. North, and S. O. Los, 2007: The impact of diffuse sunlight on canopy light-use efficiency, gross photosynthetic product and net ecosystem exchange in three forest biomes. Global Change Biol., 13, 776–787, https://doi.org/10.1111/j.1365-2486.2007.01316.x.
Andrews, T., J. M. Gregory, P. M. Forster, and M. J. Webb, 2012: Cloud adjustment and its role in CO2 radiative forcing and climate sensitivity: A review. Surv. Geophys., 33, 619–635, https://doi.org/10.1007/s10712-011-9152-0.
Bai, Y., J. Wang, B. Zhang, Z. Zhang, and J. Liang, 2012: Comparing the impact of cloudiness on carbon dioxide exchange in a grassland and a maize cropland in northwestern China. Ecol. Res., 27, 615–623, https://doi.org/10.1007/s11284-012-0930-z.
Beer, C., and Coauthors, 2010: Terrestrial gross carbon dioxide uptake: Global distribution and covariation with climate. Science, 329, 834–838, https://doi.org/10.1126/science.1184984.
Cheng, S. J., G. Bohrer, A. L. Steiner, D. Y. Hollinger, A. Suyker, R. P. Phillips, and K. J. Nadelhoffer, 2015: Variations in the influence of diffuse light on gross primary productivity in temperate ecosystems. Agric. For. Meteor., 201, 98–110, https://doi.org/10.1016/j.agrformet.2014.11.002.
Cheng, S. J., A. L. Steiner, D. Y. Hollinger, G. Bohrer, and K. J. Nadelhoffer, 2016: Using satellite-derived optical thickness to assess the influence of clouds on terrestrial carbon uptake. J. Geophys. Res. Biogeosci., 121, 1747–1761, https://doi.org/10.1002/2016JG003365.
Clark, D. A., S. C. Piper, C. D. Keeling, and D. B. Clark, 2003: Tropical rain forest tree growth and atmospheric carbon dynamics linked to interannual temperature variation during 1984-2000. Proc. Natl. Acad. Sci. USA, 100, 5852–5857, https://doi.org/10.1073/pnas.0935903100.
Crutzen, P. J., and M. Andreae, 1990: Biomass burning in the tropics: Impact on atmospheric chemistry and biogeochemical cycles estimates of worldwide biomass burning. Science, 250, 1669–1678, https://doi.org/10.1126/science.250.4988.1669.
ESA-CCI, 2010: ESACCI-LC-L4-LCCS-Map-300m-P5Y-2010-v1.6.1. European Space Agency Climate Change Initiative (ESA-CCI)-Land Cover, accessed May 2016, https://www.esa-landcover-cci.org/?q=node/158.
Gobron, N., and Coauthors, 2006: Evaluation of fraction of absorbed photosynthetically active radiation products for different canopy radiation transfer regimes: Methodology and results using Joint Research Center products derived from SeaWiFS against ground-based estimations. J. Geophys. Res., 111, D13110, https://doi.org/10.1029/2005JD006511.
Gu, L., J. D. Fuentes, H. H. Shugart, R. M. Staebler, and T. a. Black, 1999: Responses of net ecosystem exchanges of carbon dioxide to changes in cloudiness: Results from two North American deciduous forests. J. Geophys. Res., 104, 31 421–31 434, https://doi.org/10.1029/1999JD901068.
Guanter, L., C. Frankenberg, A. Dudhia, P. E. Lewis, J. Gómez-Dans, A. Kuze, H. Suto, and R. G. Grainger, 2012: Retrieval and global assessment of terrestrial chlorophyll fluorescence from GOSAT space measurements. Remote Sens. Environ., 121, 236–251, https://doi.org/10.1016/j.rse.2012.02.006.
Hubanks, P., S. Platnick, M. King, and B. Ridgway, 2016: MODIS Atmosphere L3 Gridded Product Algorithm Theoretical Basis Document (ATBD) & users guide. NASA Algorithm Theoretical Basis Document ATBD-MOD-30, 124 pp., https://atmosphere-imager.gsfc.nasa.gov/sites/default/files/ModAtmo/L3_ATBD_C6_C61_2019_02_20.pdf.
Joiner, J., 2013: GOME-2 740 nm terrestrial chlorophyll fluorescence data, version 26. NASA Goddard Space Flight Center, accessed 29 July 2016, https://acd-ext.gsfc.nasa.gov/People/Joiner/my_gifs/GOME_F/GOME-F.htm.
Joiner, J., Y. Yoshida, A. P. Vasilkov, E. M. Middleton, P. K. E. Campbell, Y. Yoshida, A. Kuze, and L. A. Corp, 2012: Filling-in of near-infrared solar lines by terrestrial fluorescence and other geophysical effects: Simulations and space-based observations from SCIAMACHY and GOSAT. Atmos. Meas. Tech., 5, 809–829, https://doi.org/10.5194/amt-5-809-2012.
Joiner, J., and Coauthors, 2013: Global monitoring of terrestrial chlorophyll fluorescence from moderate spectral resolution near-infrared satellite measurements: Methodology, simulations, and application to GOME-2. Atmos. Meas. Tech., 6, 2803–2823, https://doi.org/10.5194/amt-6-2803-2013.
Joiner, J., Y. Yoshida, Y. Zhang, G. Duveiller, M. Jung, A. Lyapustin, Y. Wang, and C. J. Tucker, 2018: Estimation of terrestrial global gross primary production (GPP) with satellite data-driven models and eddy covariance flux data. Remote Sens., 10, 1346, https://doi.org/10.3390/rs10091346.
Jung, M., M. Reichstein, and A. Bondeau, 2009: Towards global empirical upscaling of FLUXNET eddy covariance observations: Validation of a model tree ensemble approach using a biosphere model. Biogeosciences, 6, 2001–2013, https://doi.org/10.5194/bg-6-2001-2009.
Jung, M., and Coauthors, 2011: Global patterns of land-atmosphere fluxes of carbon dioxide, latent heat, and sensible heat derived from eddy covariance, satellite, and meteorological observations. J. Geophys. Res., 116, G00J07, https://doi.org/10.1029/2010JG001566.
Kanniah, K. D., J. Beringer, P. North, and L. Hutley, 2012: Control of atmospheric particles on diffuse radiation and terrestrial plant productivity: A review. Prog. Phys. Geogr., 36, 209–237, https://doi.org/10.1177/0309133311434244.
Kanniah, K. D., J. Beringer, and L. Hutley, 2013a: Exploring the link between clouds, radiation, and canopy productivity of tropical savannas. Agric. For. Meteor., 182–183, 304–313, https://doi.org/10.1016/j.agrformet.2013.06.010.
Kanniah, K. D., J. Beringer, and L. B. Hutley, 2013b: Response of savanna gross primary productivity to interannual variability in rainfall: Results of a remote sensing based light use efficiency model. Prog. Phys. Geogr., 37, 642–663, https://doi.org/10.1177/0309133313490006.
Keppel-Aleks, G., and R. A. Washenfelder, 2016: The effect of atmospheric sulfate reductions on diffuse radiation and photosynthesis in the United States during 1995–2013. Geophys. Res. Lett., 43, 9984–9993, https://doi.org/10.1002/2016GL070052.
Knohl, A., and D. D. Baldocchi, 2008: Effects of diffuse radiation on canopy gas exchange processes in a forest ecosystem. J. Geophys. Res., 113, G02023, https://doi.org/10.1029/2007JG000663.
Kokhanovsky, A., 2004: Optical properties of terrestrial clouds. Earth-Sci. Rev., 64, 189–241, https://doi.org/10.1016/S0012-8252(03)00042-4.
Law, B. E., and Coauthors, 2002: Environmental controls over carbon dioxide and water vapor exchange of terrestrial vegetation. Agric. For. Meteor., 113, 97–120, https://doi.org/10.1016/S0168-1923(02)00104-1.
Lee, M. S., D. Y. Hollinger, T. F. Keenan, A. P. Ouimette, S. V. Ollinger, and A. D. Richardson, 2018: Model-based analysis of the impact of diffuse radiation on CO2 exchange in a temperate deciduous forest. Agric. For. Meteor., 249, 377–389, https://doi.org/10.1016/j.agrformet.2017.11.016.
Luo, Y., K. Guan, and J. Peng, 2018: STAIR: A generic and fully-automated method to fuse multiple sources of optical satellite data to generate a high-resolution, daily and cloud-/gap-free surface reflectance product. Remote Sens. Environ., 214, 87–99, https://doi.org/10.1016/j.rse.2018.04.042.
Mayer, B., A. Kylling, S. Madronich, and G. Seckmeyer, 1998: Enhanced absorption of UV radiation due to multiple scattering in clouds: Experimental evidence and theoretical explanation. J. Geophys. Res., 103, 31 241–31 254, https://doi.org/10.1029/98JD02676.
Mercado, L. M., N. Bellouin, S. Sitch, O. Boucher, C. Huntingford, M. Wild, and P. M. Cox, 2009: Impact of changes in diffuse radiation on the global land carbon sink. Nature, 458, 1014–1017, https://doi.org/10.1038/nature07949.
Meyer, K., S. Platnick, G. Wind, J. Riedi, N. Goddard, and S. Flight, 2014: MODIS Collection 6 Clear Sky Restoral (CSR): Filtering cloud mask “not clear” pixels. MODIS Science Team Meeting, Columbia, MD, NASA, https://modis.gsfc.nasa.gov/sci_team/meetings/201404/posters/meyer.pdf.
Min, Q., 2005: Impacts of aerosols and clouds on forest-atmosphere carbon exchange. J. Geophys. Res., 110, D06203, https://doi.org/10.1029/2004JD004858.
Min, Q., and S. Wang, 2008: Clouds modulate terrestrial carbon uptake in a midlatitude hardwood forest. Geophys. Res. Lett., 35, L02406, https://doi.org/10.1029/2007GL032398.
Nemani, R., M. White, P. Thornton, K. Nishida, S. Reddy, J. Jenkins, and S. Running, 2002: Recent trends in hydrologic balance have enhanced the terrestrial carbon sink in the United States. Geophys. Res. Lett., 29, 1468, https://doi.org/10.1029/2002GL014867.
Niyogi, D., and Coauthors, 2004: Direct observations of the effects of aerosol loading on net ecosystem CO2 exchanges over different landscapes. Geophys. Res. Lett., 31, L20506, https://doi.org/10.1029/2004GL020915.
Oliphant, A. J., D. Dragoni, B. Deng, C. S. B. Grimmond, H. P. Schmid, and S. L. Scott, 2011: The role of sky conditions on gross primary production in a mixed deciduous forest. Agric. For. Meteor., 151, 781–791, https://doi.org/10.1016/j.agrformet.2011.01.005.
Pincus, R., S. Platnick, S. A. Ackerman, R. S. Hemler, and R. J. Patrick Hofmann, 2012: Reconciling simulated and observed views of clouds: MODIS, ISCCP, and the limits of instrument simulators. J. Climate, 25, 4699–4720, https://doi.org/10.1175/JCLI-D-11-00267.1.
Rap, A., and Coauthors, 2015: Fires increase Amazon forest productivity through increases in diffuse radiation. Geophys. Res. Lett., 42, 4654–4662, https://doi.org/10.1002/2015GL063719.
Roderick, M. L., G. D. Farquhar, S. L. Berry, and I. R. Noble, 2001: On the direct effect of clouds and atmospheric particles on the productivity and structure of vegetation. Oecologia, 129, 21–30, https://doi.org/10.1007/s004420100760.
Schneider, T., T. Bischoff, and G. H. Haug, 2014: Migrations and dynamics of the intertropical convergence zone. Nature, 513, 45–53, https://doi.org/10.1038/nature13636.
Stephens, G. L., 2005: Cloud feedbacks in the climate system: A critical review. J. Climate, 18, 237–273, https://doi.org/10.1175/JCLI-3243.1.
Tselioudis, G., W. B. Rossow, and D. Rind, 1992: Global patterns of cloud optical thickness variation with temperature. J. Climate, 5, 1484–1495, https://doi.org/10.1175/1520-0442(1992)005<1484:GPOCOT>2.0.CO;2.
Urban, O., and Coauthors, 2007: Ecophysiological controls over the net ecosystem exchange of mountain spruce stand. Comparison of the response in direct vs. diffuse solar radiation. Global Change Biol., 13, 157–168, https://doi.org/10.1111/j.1365-2486.2006.01265.x.
Urban, O., and Coauthors, 2012: Impact of clear and cloudy sky conditions on the vertical distribution of photosynthetic CO2 uptake within a spruce canopy. Funct. Ecol., 26, 46–55, https://doi.org/10.1111/j.1365-2435.2011.01934.x.
Wang, W., and Coauthors, 2013: Variations in atmospheric CO2 growth rates coupled with tropical temperature. Proc. Natl. Acad. Sci. USA, 110, 13 061–13 066, https://doi.org/10.1073/pnas.1219683110.
Wu, J., and Coauthors, 2016: Partitioning controls on Amazon forest photosynthesis between environmental and biotic factors at hourly to interannual timescales. Global Change Biol., 23, 1240–1257, https://doi.org/10.1111/gcb.13509.
Yang, X., J. Tang, J. F. Mustard, J. Lee, and M. Rossini, 2015: Solar-induced chlorophyll fluorescence correlates with canopy photosynthesis on diurnal and seasonal scales in a temperate deciduous forest. Geophys. Res. Lett., 42, 2977–2987, https://doi.org/10.1002/2015GL063201.
Yao, M. S., and A. D. Del Genio, 1999: Effects of cloud parameterization on the simulation of climate changes in the GISS GCM. J. Climate, 12, 761–779, https://doi.org/10.1175/1520-0442(1999)012<0761:EOCPOT>2.0.CO;2.
