We use cloud properties from the GCM Oriented CALIPSO Cloud Product (GOCCP v3.1.2; Chepfer et al., 2010; Cesana et al., 2012; Guzman et al., 2017) over the period 2008–2020. We do not use data collected between 2006 and 2007 because the laser tilted off nadir in November 2007, which introduced some change in the CALIPSO signal. In this product (hereafter, CALIPSO–GOCCP), lidar profiles are classified into three types: clear-sky profile when no cloud is detected, thin cloud profile when one or several cloud layers and a surface echo are detected, and opaque cloud profile when one or several cloud layers are detected but no surface echo is detected. Surface echo is not detected typically when the profile contains a cloud with visible optical depth >3–5 depending on the cloud microphysical properties. The cloud base height corresponds to the lowest cloud layer detected. From this classification, five fundamental cloud properties for CRE studies are derived.

COpaque: the opaque cloud cover, i.e., the number of opaque cloud profiles divided by the total number of profiles within a 2∘ × 2∘ latitude–longitude grid box.

An opaque cloud (Fig. 1b) is characterized by three altitudes. When the lidar is based on the ground, we measure the altitude of the lowest cloud layer (ZBase), the altitude where the lidar beam is fully attenuated (ZFA-G), and ZTOpaque-G, which is the average of the two. When the lidar is onboard a satellite, we measure the highest cloud layer (ZTop), the altitude where the lidar beam is fully attenuated (ZFA), and the average of the two (ZTOpaque).

Figure 2 illustrates the mean 2∘ × 2∘ latitude–longitude gridded values of these five variables over the period 2008–2020. At global scale, opaque clouds are more numerous (42 %; Fig. 2a) than thin clouds (25 %; Fig. 2b) in CALIPSO–GOCCP v3.1.2. Note that these numbers are different from CALIPSO–GOCCP v3.1.1 (35 % and 36 %, respectively), where the threshold used to detect surface echo, which influences the identification of opaque clouds, was lower because CALIPSO–GOCCP v3.1.1 (Guzman et al., 2017) was applied only to nighttime data since noise is lower during nighttime than daytime. CALIPSO–GOCCP v3.1.2 is applied to nighttime and daytime observations. As expected, the multiyear, annual mean opaque and thin cloud altitudes (Fig. 2c, d) reach maxima (>9 km) in the presence of deep convective clouds over the warm pool and over tropical continents and minima (<3 km) in subsidence regions such as over stratocumulus along the western coast of continents. The thin cloud emissivity (Fig. 2e) is larger along the intertropical convergence zone (ITCZ), in the continental regions, and around the Antarctic Peninsula.

In this subsection, we describe the already existing global surface LW CRE datasets derived from satellite measurements, against which we will evaluate our new satellite retrieval.

As the retrieval of the surface CRE from space observations is not direct, we will evaluate the surface LW CRE retrieved from space against that derived from surface radiation measurements collected directly at ground-based sites. For this purpose, we selected three sites located in different regions.

The first site is located in the Arctic, where constraining radiative transfer is challenging with the limited cloud, available atmospheric temperature and humidity profile observations (Kay et al., 2015) and where the surface CRE may influence the Greenland ice-cap melt (van Trich et al., 2016; Hofer et al., 2017; Shupe et al., 2013). This Summit station (Shupe et al., 2013; Gallagher et al., 2018) is located at the top of the Greenland ice cap (72.6∘ N–38.5∘ W) with an elevation of 3250 m. Summit is unique because it is the only place where we have enough observations to make a robust assessment of the surface CRE over Greenland (Lacour et al., 2018). Here, the clear-sky flux is computed using a radiative transfer algorithm with measurements of temperature and humidity profiles (e.g., REFs), while the all-sky flux is measured directly using a pair of upward- and downward-looking broadband pyrgeometers (e.g., Shupe and Intrieri, 2004; Intrieri et al., 2002).

The second site is located at continental mid-latitudes. This Site Instrumental de Recherche par Télédétection Atmosphérique (SIRTA, Haeffelin et al., 2005; Chiriaco et al., 2018) is located in France (48.7∘ N–2.2∘ E) with an elevation of 156 m. The data are part of the Baseline Surface Radiation Network (BSRN; Ohmura et al., 1998; Driemel et al., 2018). At SIRTA, the clear-sky flux is a parameterization made from the surface humidity, the integrated moisture content over the atmospheric column and the air temperature at 2 m. The details are given in Dupont and Haeffelin (2008), and this product has also been used in Rojas et al. (2021). The resulting clear-sky uncertainty is approximately ±5 W m-2.

The third site is located in the tropical belt, where clouds influence the global climate and heat transport (Loeb et al., 2016) and where extensive deep convective clouds reach the cold tropical tropopause. Here, the surface LW CRE is small, since much of the surface downward LW radiation originates from emission by the moist near-surface layers of the atmosphere (Prata, 1996). This Kwajalein station (KWA, Roesch et al., 2011), which is also part of the BSRN, is located in the northern Pacific Ocean (8.72∘ N–167.73∘ E) with an elevation of 10 m.

Over the three ground-based sites, the radiative flux measurements at the surface are carried out using two Kipp and Zonen CM22 pyrgeometers, which measure in the spectral range of 4.5–40 µm.

Vaillant de Guélis et al. (2017a, b) retrieved the TOA LW CRE from the five CALIPSO–GOCCP cloud properties presented in Fig. 2: the opaque cloud cover, the opaque cloud altitude, the thin cloud cover, the thin cloud altitude, and thin cloud emissivity. In adapting their approach to the surface instead of the TOA, we developed a method to retrieve the surface LW CRE from the same five CALIPSO–GOCCP cloud properties. The method we have developed is based on simple parameterization. This will allow us to, in future work, more easily decompose the temporal variations of the surface LW CRE into several components in order to identify which cloud variables have driven the variations of the surface LW CRE during the last 13 years. The following physical differences exist between the surface and the TOA.

In order to get simple notation and because we are only interested in the CRE at the surface in the LW domain in this study, the surface LW CRE will simply be denoted “CRE” in the following equations.

To infer “CRE”, the net LW radiative fluxes over all types of scenes (FAllskynet) may be compared with corresponding fluxes where the influence of clouds has been removed (FCloudy-freeskynet). Then, we define the surface LW CRE as follows: 1CRE=FAllskynet-FCloudy-freeskynet. Using downwelling (↓) and upwelling (↑) fluxes, the surface LW CRE is expressed as follows: 2CRE=FAllsky↓-FAllsky↑-FCloudy-freesky↓-FCloudy-freesky↑. Rearranging the terms on the right-hand side of this equation, we get 3CRE=FAllsky↓-FCloudy-freesky↓-FAllsky↑-FCloudy-freesky↑, which can also be expressed as 4CRE=CRE↓-CRE↑, where CRE↓ represents the surface CRE on the LW downward fluxes and CRE↑ the surface CRE on the LW upward fluxes. CRE↑ does not exceed 1 W m-2 in the annual global average (Allan, 2011) and in the radiative transfer computations. Therefore, the error in the surface properties plays a minor role.

Nevertheless, in the LW domain, clouds can warm the surface, changing the surface temperature, which is then related to the upwelling LW radiation. This is a subtle but important issue and is dependent to some degree on the surface type (i.e., land surface will warm more than ocean). If “CRE” is determined in a hypothetical way, one could assume that the surface temperature is the same. However, this does not capture the full impact of the clouds. To understand the full impact of the clouds, one would need to consider the adjustments of all other parameters, most importantly the surface temperature. In this study we assume that the surface temperature is the same under clouds and clear skies, consistent with the definition used in previous satellite-derived products (e.g., Kato et al., 2018; L'Ecuyer et al., 2019).

We use a radiative transfer code to compute the surface LW CRE due to an opaque cloud (CREOpaque) or an optically thin cloud (CREThin) in an atmospheric column fully overcast by that cloud. In these 1D atmospheric columns, molecules and clouds are evenly distributed within each layer, and each layer is considered infinite and homogeneous. For a single column fully overcast by an opaque cloud, we derived a parameterization between CREOpaque and the opaque cloud altitude ZTOpaque (see Sect. 2.1). For the single column fully overcast by a thin cloud, we derived a parameterization between CREThin, the thin cloud altitude ZTThin (see Sect. 2.1) and the thin cloud emissivity εThin, as in Vaillant de Guélis et al. (2017).

The radiative transfer simulations are performed with GAME (Dubuisson et al., 2004). This radiative transfer code computes LW fluxes at 50 different levels with a vertical resolution of 1 km in the first 25 levels. The fluxes are spectrally integrated between 5 and 200 µm, consistent with CERES measurements. We prescribe various surface temperatures and the atmospheric profiles of humidity, temperature, ozone and pressure based on ERA-Interim reanalysis (Dee et al., 2011) over oceans and lands for each month and 2∘ latitude. Humidity and temperature profiles over land for January are presented in Fig. A2 in Appendix A. Figure A3 presents the seasonal and latitudinal behavior of the first layer of the humidity and temperature profiles (from the surface to 1 km above the surface) over ocean and over land. We perform all-sky fluxes through radiative transfer computations for numerous combinations of cloud opacity and vertical distribution. We prescribe the vertical extent of each cloud, the effective size of cloud particles and the infrared optical thickness. For a column fully overcast by an opaque cloud, the cloud is represented by a 1 km-thick cloud layer with an emissivity close to 1 at ZFA-G (ZTop) above optically uniform cloud layers for different vertical extents with a vertically integrated emissivity equal to 0.8. For a column fully overcast by a thin cloud, the cloud is represented by optically uniform cloud layers with vertically integrated emissivities equal to 0.1, 0.3, 0.5 or 0.7. The cloud top altitude varies according to latitude and can reach 17 km in tropical regions and only 11 km in polar regions. For instance, the cloud top altitude at a latitude of 39∘ N takes 11 different values ranging between 2 and 13 km, and for each cloud top value, the cloud base altitude takes all possible values between 1 km above the surface and the cloud top altitude minus 1 km. Clear-sky fluxes are defined by recalculating fluxes after removing clouds with the same humidity and temperature profiles.

The surface LW CRE is retrieved from parameterizations derived from radiative transfer simulations that involve five observed CALIPSO–GOCCP cloud properties. Two surface LW CRE datasets are built from the CALIPSO–GOCCP product using this theoretical relationship over the 2008–2020 period, an orbit dataset at the CALIOP footprint resolution of instantaneous cloud property observations and a 2∘ × 2∘ gridded dataset of mean cloud properties. For the orbit dataset, each lidar profile contains either an opaque or thin cloud or no cloud, and the surface LW CRE for this last category is zero. For the gridded product, at each grid point, the opaque surface LW CRE is computed from the gridded ZTOpaque and weighted by the gridded opaque cloud cover COpaque in the same way as Vaillant-de-Guélis et al. (2017). The thin surface LW CRE is computed from the gridded ZTThin and gridded εThin and then weighted by the gridded thin cloud cover CThin. The total gridded surface LW CRE is the sum of the two. 5CRE=CREOpaque+CREThin In the retrievals, we tested both ZFA and ZTOpaque for estimating the mean altitude of opaque clouds (as discussed in Sect. 3.1).

The new product name is “LWCRE–LIDAR–Ed1” for “LW Cloud Radiative Effect derived from space Lidar observations Edition 1”, and the acronyms are LWCRE–LIDAR and CRELIDAR in this study. This new monthly gridded product is available for the 2008–2020 time period at 10.14768/70d5f4b5-e740-4d4c-b1ec-f6459f7e5563 (Arouf et al., 2022), and Table C1 summarizes the data included in the dataset.

Figure 3 shows the results of numerous simulations for the opaque cloud column (Fig. 3a) and the thin cloud column (Fig. 3b) for a specific atmospheric state over oceans in January at a latitude of 39∘ N. CREOpaque decreases approximately linearly with opaque cloud altitude at a rate of 6.0 W m-2 km-1 in this atmospheric state. This figure shows that the surface LW cloud radiative effect depends mostly on the mean altitude of the cloud and only weakly on the detailed vertical cloud distribution and the cloud bottom altitude. CREThin also decreases linearly with thin cloud altitude, and the rate of decrease depends linearly on the cloud emissivity. The linearity of these relationships is consistent with Ramanathan (1977) and Vaillant de Guélis et al. (2017, 2018). It is an empirical relation derived from radiative transfer calculations that has been verified in the observation at the TOA in Vaillant de Guélis et al. (2017a, 2018). Corti and Peter (2009) also derived an empirical relationship (power laws) from radiative transfer computation. Our linear relationship can be seen as an approximation of the Corti and Peter (2009) power law.

Based on a regression, we obtain the following linear relationships between the surface LW CRE and cloud altitude and emissivity: 6CREOpaque=COpaque×aRH,T×ZTOpaque+bRH,T, 7CREThin=CThin×εThin+0.06×aRH,T×ZTThin+bRH,T, where a(RH,T) W m-2 km-1 and b(RH,T) W m-2 are constants whose values depend on the humidity and temperature profiles as discussed hereafter. For the specific case presented in Fig. 3, a=-6.0 W m-2 km-1 and b=+88.0 W m-2.

The temperature and humidity profiles in the first layers of the atmosphere largely vary according to seasons and location as presented in Fig. A3 in Appendix A. Since these are variables that influence the surface LW CRE, their variations must be taken into account in order to retrieve the global surface LW CRE.

As an example, Fig. 4a presents the opaque surface LW CRE for a standard humidity profile and Fig. 4b presents the opaque surface LW CRE for an enhanced humidity profile (shown in Fig. A4). A 10 % change in humidity in the first few kilometers of the tropical atmosphere leads to a surface LW CRE change of 7.7 W m-2 for a cloud at 1 km and by 5 W m-2 for a cloud at 4 km. To capture some variability of humidity and temperature, we have established similar relationships as in Fig. 3 for each month and latitude (every 2∘) over land and ocean. As an example, Fig. A1 shows the simulations for cloud columns for an atmospheric state over land in January at a latitude of 39∘ N (same as Fig. 3 but over land instead of ocean). At this latitude, the amount of humidity is lower over land than ocean, and therefore the LW FCloudy-freeskynet over land is lower and the surface LW CRE would be larger than over the ocean. The surface LW CRE is greater than that over ocean and decreases at a rate (a(RH,T) W m-2 km-1) of 6.5 W m-2 km-1 instead of 6.0 W m-2 km-1 over ocean. Figure 5 presents the latitudinal and seasonal behavior of the linear regression coefficients (a(RH,T) W m-2 km-1 and b(RH,T) W m-2). The shape of these coefficients' spatiotemporal variation is influenced by the shape of the seasonal cycle of humidity and temperature in the first layers of the atmosphere (Fig. A3). For instance, the behavior of the intercept (b(RH,T) W m-2) over ocean and land (Fig. 5b and d, respectively) is driven by the shape of the humidity amount where the largest humidity amount (in tropical regions) causes the smallest intercept coefficients. The seasonal cycle of the surface LW CRE is more pronounced over land than over ocean because the seasonal cycles of humidity and temperature are more pronounced over land than over ocean due to the heat capacity of the surface (Chepfer et al., 2019).

In order to take the surface elevation in the simulation into account, we consider the surface temperature to be the temperature of the atmospheric layer located at the same altitude as the surface elevation with respect to sea level, and we discard all layers located between sea level and the altitude of surface elevation. We then performed numerous radiative transfer simulations corresponding to different clouds, as described in Sect. 3.2.

The results presented in Fig. 6 show the sensitivity of the surface LW CRE to the surface elevation over continents in January at 39∘ N. As the surface elevation increases, the atmosphere is drier, so FCloudy-freeskynet decreases and the surface LW CRE increases. The same cloud with the same cloud properties (i.e., same altitude and emissivity) will warm a surface with a high elevation more than a low elevation. For instance, an opaque cloud at an altitude of 5.5 km m.s.l. (mean sea level) will warm a surface at sea level by ∼ 58 W m-2 and a surface with an elevation of 4 km m.s.l. by ∼ 102 W m-2. These results are consistent with Wang et al. (2019), who found that the surface LW CRE increases over the Summit station in Greenland due to the dry atmosphere at high elevations. We performed radiative transfer simulations for different surface elevations at all latitudes and months (not shown) and used these to retrieve the surface LW CRE from space-based lidar observations over land. Thus, the regression coefficients over land also depend on surface elevation, with a 100 m resolution (a(RH,T,SE) W m-2 km-1, b(RH,T,SE) W m-2).

Figure 7 (first panel) show the CALIPSO–GOCCP cloud vertical mask (Guzman et al., 2017) for two different parts of an orbit, both in the tropical region. The blue areas over green areas represent the opaque clouds. The blue areas over white areas represent thin clouds. The second line represents the instantaneous surface LWCRE–LIDAR derived from CALIPSO–GOCCP instantaneous cloud properties (opaque cloud altitude, thin cloud altitude and emissivity; CRELIDAR), as described in Sect. 3.3. As expected, the surface LWCRE–LIDAR is larger for opaque clouds (Fig. 7a, ∼ 22 W m-2) than for thin clouds (Fig. 7b, ∼ 5 W m-2) for almost the same atmosphere.

Figure 8a shows the map of the surface LWCRE–LIDAR derived from the CALIPSO–GOCCP product over the 2008–2020 time period.

In annual global means, clouds radiatively warm the surface in the LW domain by 27.0 W m-2. CRELIDAR is maximal in the Southern Ocean (∼ 50–65 W m-2), where the warm opaque low clouds are numerous, as already stated by L'Ecuyer et al. (2019) and Henderson et al. (2013). There are also particularly high values in the North Atlantic (>∼ 55 W m-2) observed between Svalbard and Greenland. In the tropics, clouds typically radiatively warm the surface in the LW domain by only ∼ 15 W m-2. The moist tropical oceanic atmosphere enhances the downward clear-sky fluxes, which decreases the surface LW CRE over these oceans. The maximum tropical CRELIDAR (∼ 30 to ∼ 40 W m-2) is produced by warm opaque low oceanic stratocumulus clouds along the western coast of the continents.

Over continents, the weakest CRELIDAR (<∼ 5 W m-2) occurs over the Wadi Abadi basin in the Egyptian desert (25∘ N, 33∘ E), a cloud-free region most of the time (80 %). The largest CRELIDAR (∼ 60–65 W m-2) occurs over the Tibet Autonomous Region (29∘ N, 97∘ E), where the opaque cloud cover is high (58 %) and the mean surface elevation is high (∼ 4.42 km) over 2.5 million km2. Here, the high amount of moisture is uplifted towards southern Tibet, amplified by Rayleigh distillation as the vapor moves over the Himalayan mountains (He et al., 2015), which enhances the formation of opaque clouds.

CALIPSO space-based lidar differentiates well opaque clouds from thin clouds. Therefore, we can decompose the CRELIDAR into contributions due to opaque clouds (CREOpaque: Fig. 8b) and thin clouds (CREThin: Fig. 8c). This decomposition shows that 85 % (23.0 W m-2) of the overall annual global mean CRELIDAR (27.0 W m-2) is produced by opaque clouds. Their effect is maximal (∼ 50–55 W m-2) over the extra-tropical oceans (60∘ S and 60∘ N), where low warm opaque clouds are numerous. Thin clouds contribute only 15 % (4.0 W m-2) to the global CRELIDAR, and their effect is maximal (∼ 13 W m-2) over the dry continental polar regions of the Greenland and Antarctic ice sheets, where the thin cloud cover is large (∼ 40 %).

Comparisons between ground-based measurements and the satellite-derived products (CRELIDAR, CRE2BFLX) provide a direct evaluation of the satellite retrievals but are limited by the difference in the spatial resolution of the satellite-derived product (2∘ × 2∘) and the ground station observations (a few meters). For the satellite retrievals, we extract the monthly 2∘ × 2∘ grid box centered at each ground site. For the ground-based observations, we extract the hourly observation at CALIPSO satellite overpass time above each ground site (two observations per day) and average over each month. We consider all days of each month, even if CALIPSO has no sampling over the site, because there are only a few days where CALIPSO observations are not available (e.g., 18 d in 2008). Moreover, in this study, we are interested in an accurate representation of the surface LW CRE interannual variability, which might have significant impacts on climate-relevant processes, and not only in an accurate representation of the anomalies observed in, e.g., Rutan et al. (2015). That CALIPSO is missing some sampling over the ground-based site will likely not significantly affect the interannual variability (i.e., months of maxima/minima of the surface LW CRE). The locations of the three ground-based sites are reported on the maps (stars in Fig. 14).

Over the Summit station Greenland site, on average compared to ground-based observations, LWCRE–LIDAR underestimates the surface LW CRE by 8.5 W m-2, while 2BFLX underestimates it by 16.4 W m-2 (Fig. 9a). Averages over the 2008–2010 and 2011–2015 periods (Fig. 10) show that these biases calculated for a short period are similar to the longer periods. Over the 2008–2011 time period, CRELIDAR is close to CRE2BFLX, and both show consistent summer maxima and winter minima, with CRE2BFLX slightly smaller than CRELIDAR (0.8 W m-2). Over the 2011–2015 time period, CRELIDAR and the ground station data show similar annual cycles, and CRELIDAR remains smaller than the Greenland site (13.0 W m-2). In winter, the bias in CRELIDAR can go up to ∼ 15 W m-2 compared to the Greenland site and is partly due to CALIPSO–GOCCP missing thin cloud below 2 km above ground level in winter, as shown in Lacour et al. (2017). While this comparison suggests that LWCRE–LIDAR could be biased somewhat low compared to the ground station perspective over Greenland, it is also clear that this approach captures the annual variability with a correlation coefficient between the CRELIDAR and ground base site of 0.69 and a RMSE of 15.9 W m-2. The retrieval using ZFA instead of ZTOpaque seems to compare to the Greenland ground-based observations more favorably (correlation coefficient of 0.70 and RMSE of 15.0 W m-2) with a smaller bias (-11.6 W m-2 vs. -13.6 W m-2, Table 1).

Over the mid-latitude continental site (Fig. 9b) on average, LWCRE–LIDAR underestimates the surface LW CRE by 5.7 W m-2 compared to ground-based observations with a correlation coefficient of 0.73 and RMSE of 11.0 W m-2, while 2BFLX underestimates it by 9.4 W m-2 with a correlation coefficient of 0.67 and RMSE of 15.5 W m-2.

Over the tropical ocean site (Fig. 9c) on average, LWCRE–LIDAR underestimates the surface LW CRE by 2.3 W m-2 compared to ground-based observations, and 2BFLX underestimates it by 4.1 W m-2. This same behavior is found on the map of differences between CRELIDAR and CRE2BFLX (Fig. 14a) along the tropical Pacific and tropical Atlantic oceans, where 2BFLX underestimates the surface LW CRE compared to LWCRE–LIDAR.

Figure 10 presents the comparison of seasonal cycles between the satellite retrievals and the ground-based observations.

Over the Greenland site (Fig. 10a, d), LWCRE–LIDAR and 2BFLX find the same seasonal cycle of the surface LW CRE with maxima in July that correspond to the maximum opaque cloud cover, same as the ground-based seasonal cycle.

Over the mid-latitude continental site (Fig. 10b, e), the surface LW CRE seasonal cycles of LWCRE–LIDAR and 2BFLX are close to each other, and the two satellite-derived products show similar seasonal cycles to the ground station.

Over the tropical ocean site (Fig. 10c, f), the surface LW CRE seasonal cycle is relatively flat.

The time sampling is limited for LWCRE–LIDAR and 2BFLX as they observe each location only two times per day at about 01:30 and 13:30 local solar time (LST), and they do not implement diurnal variation correction in their algorithm. Nevertheless, diurnal variations of the cloud fraction profiles documented by the CATS/ISS lidar (Noel et al., 2018; Chepfer et al., 2019) indicate that the average of the cloud profiles collected at 01:30 and 13:30 LST is similar to the average of all the profiles collected along the 24 h, with 13:30 LST corresponding to the minimum in cloud profiles along the day and 01:30 LST corresponding to the maximum (Fig. 7 in Noel et al., 2018). However, this statement is valid only between 55∘ S and 55∘ N.

Figure 11 shows the diurnal surface LW CRE variation observed at SIRTA in France, together with LWCRE–LIDAR. This comparison suggests that the average of the two CALIPSO overpasses each day is similar to the ground-based observed daily mean over this site. The absence of diurnal cycle correction might not be an important source of error in the LWCRE–LIDAR product.

CloudSat, CALIPSO, and CERES/Aqua satellites are part of the A-Train constellation (Stephens et al., 2002) and closely matched in time (<5 min) and hence collocated by default, so we can compare them directly, assuming that the atmospheric changes occurring within 5 min are negligible.

Figure 12 shows a comparison between the surface LW CRE from the three spatial satellite retrievals along four pieces of orbits located over regions with different atmospheres and different surfaces. Figure 12 (top panel) shows the vertical CALIPSO–GOCCP cloud mask (Guzman et al., 2017), while Fig. 12 (bottom panel) represents the comparison between the surface LW CREs.

Orbit A passes over the eastern Pacific Ocean and observes a deep convective tower, a mid-level opaque cloud at an altitude of 7 km, and a low opaque cloud. The differences in surface LW CRE between the three spatial restitutions do not exceed ∼ 5 W m-2. Nevertheless, within a small part of the orbit between 11.3 and 11.8∘ N (Fig. B1), we observe that the LWCRE–LIDAR retrieval is lower than the other two products, because the lidar does not detect a low cloud below ZFA, which is detected by CloudSat (shown in Fig. B1b).

Orbit B passes over the western Pacific Ocean and observes variable yet shallow clouds in the boundary layer (< 2 km). CALIPSO–GOCCP (90 m cross track, 330 m along orbit track) detects well shallow clouds in the boundary layer. CRELIDAR compares favorably to CRE2BFLX over the full scene in presence of shalow clouds but underestimate locally the surface LW CRE in the presence of the deep convective cloud for the same reason as in orbit A (Fig. B1).

Orbit C passes over ocean stratocumulus regions and observes a low opaque cloud. Between 12 and 19∘ S, the CRELIDAR (∼ 60 W m-2) is smaller than CRE2BFLX by ∼ 5 W m-2 and smaller than CRECERES by 15 W m-2.

Orbit D passes over Antarctica and observes opaque clouds at high (10 km) and mid-level (4–5 km) altitudes. In the presence of high opaque clouds (between 68 and 71∘ S or between 73 and 77∘ S), CRELIDAR is lower than CRECERES by up to ∼ 20 W m-2 and CRE2BFLX by up to ∼ 40 W m-2 but typically compares most favorably to CRECERES over the full scene.

These orbits show that by not including CloudSat, surface LWCRE–LIDAR is biased low by typically ∼ 10 W m-2 compared to 2BFLX and by ∼ 15 W m-2 compared to CERES-CCCM in regions of deep convection. In stratocumulus, surface LWCRE–LIDAR is biased low by typically ∼ 5 W m-2 compared to 2BFLX and ∼ 15 W m-2 compared to CERES-CCCM.

Figure 13a shows a comparison between the surface LW CRE derived from the CALIPSO–GOCCP product (90 m cross track, 330 m along orbit track) collocated with CERES–CCCM that uses full-resolution CALIPSO and CloudSat data and reports the results over 20 km CERES footprints. We consider only the CERES–CCCM footprints where all the CALIPSO–GOCCP profiles falling within this footprint are opaque and where there are more than 40 profiles. To retrieve the surface LWCRE–LIDAR at the CERES–CCCM footprint resolutions, we average all ZTOpaque falling within CERES–CCCM's footprint and compute the surface LWCRE–LIDAR using the relationships found in Sect. 4.

We see a strong correlation between CRECERES and CRELIDAR (R=0.84). Two significant departures from the one-to-one comparison line are observed: one for high values of the surface LW CRE and the second for low values. In the first pattern, for surface LW CRE greater than ∼ 70 W m-2, CRELIDAR is larger than CRECERES. This pattern corresponds to some low marine opaque clouds in mid-latitude regions (not shown). To reconcile the two products, CRELIDAR should be smaller by almost ∼ 5 W m-2. One way to reduce this difference would be to increase the altitudes of clouds but, due to attenuation of the signal in opaque clouds, the space-based lidar would already potentially overestimate the overall height of the clouds. Thus, the cloud height is likely not the source of this difference. Another way to reduce the surface LW CRE is by decreasing the cloud cover or the cloud opacity. However, thanks to its high spatial resolution, the space-based lidar measures the cloud cover with precision, and it should not overestimate the cloud opacity. Thus, the source of this apparent bias is more likely an underestimation of the humidity profiles used to retrieve the surface LW CRE in the presence of clouds. An increase in the humidity at these times would increase the FCloudy-freeskynet and therefore decrease the surface LW CRE. A final possibility for the difference is that each product has a unique estimate of the cloud cover due to vastly different fields of view. CALIOP footprints are only a small fraction of the CERES footprint, so part of the CERES footprint could be cloud-free even if the 40 CALIOP profiles are opaque. A study by Kato et al. (2010) demonstrated that the differences between CERES and CloudSat/CALIPSO cloud fractions decrease when averaged over area and time. Hence, this difference is likely not the primary source of bias when comparing large statistical datasets.

The second regime of differences among the products is for surface LW CRE less than ∼ 30 W m-2 (Fig. 13a), which corresponds to high opaque clouds over the warm pool region (not shown). Here, CRELIDAR is smaller than CRECERES. The underestimation of surface LWCRE–LIDAR compared to CERES–CCCM could be caused by the full attenuation of the laser beam in deep convective clouds such that CALIPSO–GOCCP overestimates the mean altitude of opaque clouds.

Figure 13b represents the comparison between the surface LW CRE derived from the CALIPSO–GOCCP product (90 m cross track, 330 m along orbit track) collocated with the CloudSat 2BFLX product at a resolution of the CloudSat footprint (5 km). We also consider only the CloudSat footprints where all the CALIPSO–GOCCP profiles falling within this footprint are opaque and where there are more than 10 profiles, and we compute CRELIDAR by averaging all ZTOpaque falling within the CloudSat footprint.

Three significant departures from the one-to-one comparison line are observed: one for low values where CRELIDAR<CRE2BFLX, one for high values of the surface LW CRE where CRELIDAR>CRE2BFLX, and one for high values where CRELIDAR < CRE2BFLX. The first two patterns appear to be similar to Fig. 13a and show up for the same reasons as described above. The last pattern of differences among the products is for large values of surface LW CRE where CRE2BFLX is larger than CRELIDAR. This pattern corresponds to a subsample of marine opaque clouds (25 % of the opaque cloud collocated) in mid-latitude regions (not shown) where CloudSat is able to detect lower clouds than CALIPSO. Using ZFA instead of ZTOpaque in LWCRE–LIDAR retrieval would shift this pattern upward and reduce the sample (17 % vs. 25 %).

The differences shown in Fig. 13 are expected when comparing satellite products at footprint scales that use different remote sensing techniques. However, when looking at the gridded product distributions (Fig. B3) instead of instantaneous collocated data, the 2BFLX and LWCRE–LIDAR agree well.

To compare 2BFLX and LWCRE–LIDAR at gridded scale, we averaged 2BFLX initially at 1∘ × 1∘ resolution to 2∘ × 2∘ like the CALIPSO–GOCCP product.

Figure 14a shows global maps of differences between LWCRE–LIDAR and 2BFLX. This comparison gives an overview of the differences between the two surface LW CRE spatial products, but it may mask some differences given the fact that the two spatial products are averaged in time (monthly) and space (2∘ × 2∘ latitude–longitude gridded).

In the global annual mean, CRELIDAR is slightly higher compared to CRE2BFLX (0.7 W m-2).

Compared to 2BFLX (Fig. 14a), CRELIDAR is slightly larger than CRE2BFLX over tropical oceans. Over lands, CRELIDAR is slightly lower than CRE2BFLX. The maximum difference occurs over land along the western coasts of the North and South American continents and the Himalayan mountains, where the surface elevation is above 2.5 km. This difference might be due to the CloudSat CPR's long powerful pulse (Fig. B2), which generates a surface clutter echo that tends to partially mask signals from clouds forming below ∼ 1 km (Marchand et al., 2008). Over icy polar areas, the two products are very similar.

Zonal averages of the surface LW CRE for 2008–2010 (Fig. 14c) show that the surface LW CRE is generally low in tropical regions and increases towards the mid-latitudes as the atmospheric moisture decreases. Values do not vary much northward of about 50∘ N. To the south, a maximum occurs at about 60∘ S, with a decline towards the far south due to less cloudiness. Over the broad domain, reaching from 60∘ N to 60∘ S, the two satellite techniques show similar zonal means, with differences among the two typically not exceeding ∼ 3 W m-2.

Figure 15a shows the temporal evolution of the surface LW CRE anomaly from the two satellite-derived products. A decomposition, separating continents from oceans and Northern Hemisphere (NH) from Southern Hemisphere (SH), is presented in Fig. 17b–g.

The phasing of the annual cycle of CRELIDAR and CRE2BFLX anomalies is roughly similar over the 2008–2010 time period. The phasing of the annual cycle of the two products is actually quite consistent for both the NH and SH over both land and ocean. For the NH land (Fig. 17d), the CRE2BFLX is slightly larger than CRELIDAR. However, it is interesting that even over NH land the annual minima match pretty well.

The interannual variability is pretty interesting. For example, the NH winter maximum in LWCRE–LIDAR products appears to vary by up to about ∼ 3 W m-2 from year to year. That is the kind of variability that might have significant impacts on climate-relevant processes like melting of the cryosphere.

Based on the comparison of orbits (Fig. B1), we observe that when the space lidar does not see the cloud base, LWCRE–LIDAR underestimates the local surface LW CRE compared to 2BFLX. However, the deep convective opaque clouds cover a small part of the overall tropics compared to other clouds. Therefore, this effect does not dominate the global comparison (Fig. 14a), where surface LWCRE–LIDAR is contrarily slightly larger than the other satellite product. Figure 13b (where 2BFLX is about ∼ 70 W m-2) and Fig. 14a (Southern Ocean) consistently suggest that CALIPSO not seeing the cloud base leads to LWCRE–LIDAR underestimating the surface LW CRE more frequently in the extra-tropical oceanic storm tracks than elsewhere.

To test whether the differences between LWCRE–LIDAR and other satellite products come from the space lidar not seeing the cloud base, we used two different approaches.

First, we used ZFA instead of ZTOpaque in the LWCRE–LIDAR retrieval. By definition, ZFA is always lower in altitude than ZTOpaque. Therefore, this change should reduce the difference between the surface LWCRE–LIDAR and other surface LW CREs if the differences were due to CALIPSO missing the cloud base. Figure 14b shows that the difference between surface LWCRE–LIDAR and the other satellite product increases instead of decreases when using ZFA instead of ZTOpaque. This suggests that the differences in surface LW CRE are likely not often due to CALIPSO misrepresenting the cloud base and that, in the majority of the cases, the cloud base might not be far from ZFA. Nevertheless, contrary to the satellite retrieval intercomparison, using ZFA instead of ZTOpaque leads to slightly better agreement between LWCRE–LIDAR and ground-based retrievals (e.g., Figs. 9, 10, Table 1). Ground-based measurements derive directly the surface LW CRE. While there are certainly challenges in comparing ground-based and satellite estimates, we should consider the ground-based estimates to be of pretty high quality.

Second, we used the cloud-base height (called the CBASE dataset) described in Mülmenstädt et al. (2018) instead of ZFA to compute ZTOpaque. In the CBASE dataset, the cloud-base-height value is given at a horizontal resolution of 40 km along the CALIPSO orbit track in the portion of the orbit where clouds are opaque. Along each CALIPSO orbit, we collocated the cloud-base-height dataset with the CALIPSO–GOCCP dataset and replaced ZFA with the cloud-base-height value given in the CBASE dataset. Then we computed ZTOpaque and the surface LW CRE. CBASE values are distributed at all latitudes and are available in 33.2 % of all the CALIPSO opaque profiles because CBASE can be retrieved only when thin clouds are detected within the 40 km orbit piece that also contains opaque cloud profiles. Comparing Fig. 16a and b indicates that the subsample of opaque CALIPSO–GOCCP profiles where CBASE is documented contains both large values of surface LW CRE associated with mid- and low-level clouds located at mid-latitudes (upper right data in panel b) and small values of surface LW CRE (lower left), but it does not include the data where 2BFLX is much larger than LWCRE–LIDAR which correspond to mid-latitude oceanic opaque clouds. When replacing ZFA (Fig. 16b) with CBASE (Fig. 16c) in the LWCRE–LIDAR algorithm, the surface LWCRE–LIDAR rises slightly almost everywhere because CBASE is lower in altitude than ZFA, and surface LWCRE–LIDAR values lower than ∼18 W m-2 are no longer present. The latter correspond to both deep convective clouds and shallow boundary layer clouds. The correlation between 2BFLX and LWCRE–LIDAR is similar whether we use ZFA (0.79) or CBASE (0.78) in the LWCRE–LIDAR algorithm.

This sensitivity study suggests that using a more advanced cloud base height (here CBASE) derived from lidar measurements than ZFA in the LWCRE–LIDAR algorithm will increase the surface LW CRE value retrieved in some opaque cloud profiles slightly, but it does not fundamentally change the results.

Thus, what these results mean collectively is that (1) the inability of CALIPSO to observe the cloud base likely does have some effect (with respect to ground-based measurements). (2) This effect actually makes the comparison to other satellite products worse, which means that there are other issues (possibly also with the other satellite product), leading to further differences.

Looking for other issues that could explain the differences between satellite products, we examined humidity and temperature profiles.

Contrary to the surface LWCRE–LIDAR retrieval method, CERES-CCCM and 2BFLX retrievals of surface flux account for sub-daily variations in temperature/humidity and capture regional variations (e.g., eastern vs. western tropical Pacific), climate events (e.g., ENSO), and extreme changes over polar regions. 2BFLX uses 3-hourly atmospheric state variable data on a half-degree Cartesian latitude and longitude grid from AN-ECMWF.

As shown in Fig. 14, monthly mean gridded surface LWCRE–LIDAR is consistent with 2BFLX, even though 2BFLX uses sub-daily spatiotemporal resolutions of temperature/humidity profiles, while LWCRE–LIDAR uses monthly mean temperature/humidity profiles. Nevertheless, instantaneous surface LWCRE–LIDAR retrievals are likely more biased than the monthly mean gridded surface LWCRE–LIDAR due to the use of monthly mean temperature/humidity profiles, because monthly means miss extreme humidity and temperature profiles. To estimate the error in the instantaneous surface LWCRE–LIDAR values, we compared the instantaneous surface LWCRE–LIDAR values obtained using 6-hourly temperature/humidity profiles from ERA-I to one obtained using monthly means and 2BFLX, for 1 d at footprint scales.

Figure 17 shows that using sub-daily profiles in LWCRE–LIDAR retrieval makes the comparison to other satellite products worse at footprint scale. More analysis (not shown) indicates that, for thin clouds, surface LWCRE–LIDAR retrieved using sub-daily temperature/humidity profiles agrees better with 2BFLX (at 5 km resolution) than surface LWCRE–LIDAR retrieved using monthly mean profiles. In contrast, the agreement between LWCRE–LIDAR and other products is lower when using sub-daily temperature/humidity profiles in all other cases: opaque clouds and also thin clouds when compared to CERES-CCCM (at 20 km). Overall, this suggests that the differences between the three daily products are likely due to other causes than LWCRE–LIDAR using monthly mean temperature/humidity profiles.