Adapting observationally based metrics of biogeophysical feedbacks from land cover/land use change to climate modeling

To assess the biogeophysical impacts of land cover/land use change (LCLUC) on surface temperature, two observation-based metrics and their applicability in climate modeling were explored in this study. Both metrics were developed based on the surface energy balance, and provided insight into the contribution of different aspects of land surface change (such as albedo, surface roughness, net radiation and surface heat fluxes) to changing climate. A revision of the first metric, the intrinsic biophysical mechanism, can be used to distinguish the direct and indirect effects of LCLUC on surface temperature. The other, a decomposed temperature metric, gives a straightforward depiction of separate contributions of all components of the surface energy balance. These two metrics well capture observed and model simulated surface temperature changes in response to LCLUC. Results from paired FLUXNET sites and land surface model sensitivity experiments indicate that surface roughness effects usually dominate the direct biogeophysical feedback of LCLUC, while other effects play a secondary role. However, coupled climate model experiments show that these direct effects can be attenuated by large scale atmospheric changes (indirect feedbacks). When applied to real-time transient LCLUC experiments, the metrics also demonstrate usefulness for assessing the performance of climate models and quantifying land–atmosphere interactions in response to LCLUC.


Introduction
Many modeling and observational studies have examined feedbacks between land and atmosphere manifest in the energy and water cycles [1][2][3][4][5]. Recent studies have shown land cover/land use change (LCLUC) can alter surface climate through biogeophysical feedbacks, which include the modification of energy, moisture and momentum exchanges between the land and atmosphere [6][7][8][9]. If sufficiently large areas are involved in LCLUC, the feedback to the atmosphere can extend from regional to global scales [10]. For a better understanding of climatic variability and future climate projections, current Earth system modeling efforts now consider LCLUC as one of the categories of anthropogenic forcing [11]. However, few studies have explored the plausibility of simulated biogeophysical feedbacks within climate because of the lack of extensive observations or the development of meaningful biogeophysical metrics [12].
The development of flux tower networks that measure the land-atmosphere exchange of energy, moisture, and carbon provides a good platform to quantify and address the uncertainties in land surface models [13]. Some flux towers have been deployed in a manner that makes it possible to quantify the observed biogeophysical feedback of LCLUC. By using neighboring flux tower sites with different land cover conditions, several studies have documented the influence of LCLUC on surface climate, especially on surface temperature [14][15][16][17][18][19][20][21]. Among these studies, Lee et al [14] developed a method to examine the 'intrinsic biophysical mechanism' (hereafter IBPM) that separates the biogeophysical effects into three components associated with radiative forcing, surface roughness and the Bowen ratio. Also based on the surface energy balance, a more thorough decomposed temperature metric (DTM) was proposed by Luyssaert et al [15] to analyze the change in surface temperature due to changes in incoming radiation, surface albedo, ground heat, sensible and latent heat fluxes. These approaches were developed with observed data, but they can provide meaningful metrics to assess biogeophysical feedbacks of LCLUC in climate models.
Most current model-based studies simply identify the climatic feedback from LCLUC by calculating the difference between runs that prescribe two contrasting land cover conditions without distinguishing the different feedbacks that arise. Direct feedbacks include alterations of absorbed solar radiation due to albedo changes, and perturbations to the partitioning of net radiation between sensible, latent and ground heat fluxes [8]. Indirect feedbacks are also important [22,23]. For instance, changes in air temperature may influence circulations or the distribution of snow cover, which in turn affects surface temperature through albedo feedbacks; the effect on humidity may alter cloud distributions that influence incoming radiation at the land surface. Most current observation-based studies have focused on the direct feedback [14,16,17]. However, metrics like IBPM and DTM may help us disentangle the direct and indirect biogeophysical feedbacks of LCLUC.
In this paper, IBPM and DTM are used as the basis to investigate direct and indirect feedbacks in FLUX-NET observations [24] and the Community Earth System Model (CESM) to demonstrate their potential for the study of the climatic impacts of LCLUC. Section 2 provides a detailed description of the metrics, the models and data used, and the experimental design. Section 3 presents results from the application of the two metrics. Section 4 includes discussion and conclusions.

Intrinsic biophysical mechanism
The IBPM [14] is based on the surface energy balance: where R n is net surface radiation, S is net surface shortwave radiation, LW in is incoming longwave radiation, ε is surface emissivity, σ is the Stephan-Boltzmann constant, T s is surface temperature, H is sensible heat flux, LE is latent heat flux, and G is ground heat flux. Sensible heat flux is defined from the gradient relationship: expresses latent heat flux in terms of sensible heat flux; ρ is air density, C p is specific heat of air at constant pressure, T a is air temperature, r a is aerodynamic resistance and β is the Bowen ratio. Surface outgoing longwave radiation in (1) can be approximated using a Taylor series expansion with T a :   T  T  T T  T  4 . Using (2) and (3), T s can be solved from (1): where R n * is apparent net radiation (R n *≈R n ) [14], λ 0 is defined as temperature sensitivity and f is an energy redistribution factor: IBPM assumes nearby contrasting land types share the same atmospheric background, such as air temperature and incoming radiation. Ignoring changes in surface emissivity and ground heat flux, surface temperature change ΔT s can be derived by the first derivative of (5): where ΔS is the change of net shortwave radiation, and Δf is the change in the energy redistribution factor, attributable to changes in surface roughness (Δf 1 ) and Bowen ratio (Δf 2 ): Due to the assumption of identical meteorological forcings, IBPM does not account for atmospheric feedbacks from LCLUC (indirect effects), and is regarded only as a 'local perturbation superimposed on the changing background' [14]. On larger scales, the changes in the atmospheric background state cannot be ignored. Therefore, a revised IBPM is expressed as: where ΔT a is the change in air temperature. The ground heat flux is not ignored here because of its considerable change when vegetation is removed, especially over the high latitudes (shown later). On the right-hand side of (10), terms 1, 2 and 3 are similar to the original IBPM. Term 4 includes the change of the atmospheric background, which can be considered an indirect effect of LCLUC on climate. The impact of downward radiation changes is now implicitly included in term 1, which indicates the intrinsic surface sensitivity to the radiative fluxes, which we discuss later. The rest of its impact is reflected in air temperature change.

Decomposition of surface temperature changes
Full decomposition of radiative surface temperature change induced from LCLUC was proposed by Juang et al [16] and has been elaborated by Luyssaert et al [15]. This method (DTM) is also based on the surface energy and starts with the first derivative of equation ( To keep consistency with IBPM, surface emissivity changes are also ignored in (11). Therefore, DTM temperature change is: Equations (10) and (12) provide two metrics for estimating the change in surface temperature due to land use change from in situ measurements or climate model output.

Observational data
We use observational data from selected paired flux towers obtained from the AmeriFlux network [24] and the European Fluxes Database [25]. Paired sites contain one flux tower located in forest and one in nearby open land (grassland, cropland or open shrub). Such paired sites can represent local land cover change (deforestation in these cases), and their climatic differences can be considered representative of the impact of LCLUC. The selection of paired sites follow two criteria: (1) the data period should be at least one common year between the two sites; (2) the paired sites should have measurements of all necessary variables used to calculate the metrics. Eight pairs of flux towers meet the criteria (table 1), and their locations are shown in figure S1. The average linear distance between the paired sites is 14.5 km, which is less than that in previous paired-site studies [14,15]. The average meridional difference is 11.4 km, and the average elevation difference is 72 m.
Three-hour daytime and nighttime means are calculated from 30 min level-2 data. Many records would have to be omitted if calculating daily means because of missing data. Observations from 12:00 to 15:00 local standard time (LST) are used for the daytime mean, and 00:00 to 03:00 LST for the nighttime mean.
As reported by Wilson et al [26], a mean energy imbalance on the order of 20% is prevalent among FLUXNET sites. Energy imbalances exist at all these paired sites. Figure 1 shows a comparison between daytime net radiation and the sum of latent, sensible and ground heat fluxes at the 16 sites. To ameliorate imbalances, the residual is distributed to the sensible and latent heat fluxes in proportion to the Bowen ratio [27].

Model sensitivity simulations
The CESM version 1.2.2 is used in this study. CESM is a coupled Earth system model composed of separate climate system components for atmosphere, ocean, land, sea-ice and land-ice [28]. This study is focused on land-atmosphere interactions, so ocean, sea-ice and land-ice components have been deactivated. The Community Atmosphere Model version 5.3 (CAM5.3) [29] and Community Land Model version 4.5 (CLM4.5) [30] simulate the Earth's atmosphere and land respectively.
A set of offline (land-only) and coupled landcover-change sensitivity experiments have been  2). As mentioned before, the original IBPM was developed assuming atmospheric feedback is absent. Therefore, 'offline' simulations well represent direct effects of LCLUC, because the atmospheric forcings are the prescribed identically among different land-cover-change experiments. In offline experiments (Crl_off and AllGrass_off), CLM is driven by global atmospheric forcing data [31]. In coupled experiments (Ctrl_cpl and AllGrass_cpl), identical prescribed sea surface temperature (SST) and sea ice cover climatologies are used for each simulation with a fixed CO 2 concentration of 284.7 ppm. The prescribed SSTs are a pre-industrial climatology from 1870 to 1890 calculated from a merged product based on the monthly mean Hadley Centre sea ice and SST dataset version 1 (HadISST1) and version 2 of the National Oceanic and Atmospheric Administration weekly optimum interpolation SST analysis [32]. Two land cover scenarios are used: control runs (hereafter referred as Crl_off and Ctrl_cpl for the offline run and coupled run, respectively) have prescribed land cover conditions from 1850 [33], while 'All Grass' runs (AllGrass_off and AllGrass_cpl) use a modified global land cover condition where plant functional types (PFTs) which are not grass (non-grass) are replaced with grass. Non-   There is a larger discrepancy there in the atmospheric background, especially for incoming shortwave radiation. Returning to figure 3, the revised IBPM shows better agreement with observations, especially during night. For the different IBPM components, roughness change exhibits the largest impact (1.96±0.60 K during the day, −1.62±0.61 K at night). Grasslands or croplands are aerodynamically smoother than forest and transfer heat less effectively, thus experiencing higher surface temperatures during daytime and lower surface temperatures at night. The radiation term has a slight cooling effect during day (−0.08±0.07 K), attributable to albedo change cooling being nearly offset by more infrared radiation from the warmer surface. Ground heat flux shows a warming effect during nighttime (0.18±0.12 K). The Bowen ratio term is also small compared to roughness.  Overall, the metrics show their ability to estimate the surface temperature change based on observational data. However, caution must be observed in this analysis. First, the development of IBPM contains a linearization of surface outgoing longwave radiation (equation (4)), which neglects nonlinear interactions [36]. Second, several terms in equation (10) cannot be directly obtained from observations, and their estimations involve uncertainties. For instance, aerodynamic resistance (r a ) is calculated using equation (2) [36,37]. Redistribution of the energy balance residual may substantially change sensible heat flux and thus r a . Also, some observations yield negative resistance values, which are physically meaningless, when measured (T s −T a ) has the opposite sign as measured sensible heat flux. These records have been excluded in the calculations. Due to the uncertainties in estimating r a , the roughness-related T s change (term 3) can be up to ±20 K. A threshold is necessary, but in certain conditions observations cannot provide a reliable reference to determine the resistance threshold. We have chosen a threshold of ±5.12 K for term 3 based on its maximum value from the CESM sensitivity experiments. The model-based threshold is chosen for two reasons: (1) energy is conserved every time step; (2) the All-Grass_off experiment represents a similar land cover change to the paired sites.

IBPM in CESM experiments
For offline land-cover-change experiments that satisfy the hypothesis in Lee et al [14], the original IBPM (equation (7)) is used to calculate surface temperature change in JJA ( figure 5). Generally, the global replacement of forest with grassland increases temperature. The model's averaged global temperature increase is 0.36 K. Cooling effects are only found over higher latitudes where boreal forests were removed. The original IBPM well captures the spatial variability of the model's actual surface temperature change, but there is a slight overestimation of the warming effects. The IBPM-estimated temperature change is 0.45 K at the global scale. The biogeophysical feedback is dominated in CLM by surface roughness changes (0.58 K). There is very little temperature change associated with radiative forcing (−0.06 K) or Bowen ratio (−0.07 K), demonstrating consistency with the paired site results. Recall Ctrl_off and AllGrass_off runs have the same incoming longwave and shortwave radiation. Therefore, the radiative change is attributable only to surface albedo change [14], which is +0.020 in JJA globally averaged over land, and leads to a decrease of 4.06 W m −2 in net shortwave radiation. Albedo in DJF has a larger increase (+0.061) due to the effect of exposed snow. The radiative cooling effect in high latitudes is −0.11 K in DJF (figure S2). Bowen ratio changes are determined by both sensible and latent heat flux changes. We found decreased sensible heat flux (globally −8.07 W m −2 ) attributable largely to decreased surface roughness, but only a slight change in latent heat flux (+1.39 W m −2 ) (figure S3). Evapotranspiration (ET) and its components in JJA have been examined (not shown). We find increased ET over tropics and boreal forest is due to increased ground evaporation and canopy transpiration, which overcomes the decreased canopy evaporation from lower leaf area index (LAI). ET decreases slightly after deforestation in mid-latitudes. The small change in ET explains the minor contribution of the Bowen ratio term in the original IBPM.
Impacts of LCLUC should not manifest only as local perturbations, but could alter the atmosphere at broader scales. Figure 6 shows the change in surface temperature in the coupled model simulations. When atmospheric feedback is included, deforestation decreases surface temperature in mid-and high latitudes, but increases temperature in the tropics. The global average change is −0.57 K. The original IBPM shows overall warming with the same spatial pattern as the offline run but reduced magnitude to 0.17 K, indicating local perturbations have been attenuated by large-scale atmospheric changes. The revised IBPM shows good agreement with the model's temperature difference (global average of −0.58 K). Therefore, we can separate the direct and indirect impacts of LCLUC on climate: direct impacts are represented by the original IBPM, while the atmospheric feedbacks (indirect impacts) are embodied predominantly by the added term ΔT a (equation (10)).

DTM in CESM experiments
DTM has also been applied to the coupled experiments ( figure 7). DTM-estimated temperature change shows good agreement with model surface temperature change except for a cold bias over the Arctic. Compared with IBPM, DTM provides a detailed breakdown of all components in the surface energy budget. Albedo change causes cooling, especially over Northern Hemisphere high latitudes. Due to the cooling, snow cover is prolonged into summer (now shown) acting as a positive feedback. Effects from incoming shortwave radiation do not show a uniform pattern, but there is consistent warming over many arid regions. Changes in longwave radiation have a cooling effect in most areas of the Northern Hemisphere. Overall, both LE and H decrease with land cover change (expect some regions in tropics for LE and some arid areas of middle latitudes for H), indicating a general warming effect (note that the sign for the LE and H term in equation (12) is negative). The decrease in LE can be associated with lower LAI, decreased temperature and decreased precipitation, and the decrease in H can be attributed to surface roughness decreases [21]. The increased LE in some tropical regions (such as the north edge of the Amazon) might be inconsistent with previous studies [38], which found reduced ET when tropical forest is converted to pasture. Such a discrepancy should be attributed to the potential problems with CLM hydrology in simulating ET [39,40]. The ground heat component shows cooling mainly at high latitudes, because more energy dedicated to snowmelt.

Application of metrics to LME
Here we apply the metrics to long-term climate simulations designed for realistic transient landcover-change experiments. Figure 8 shows JJA surface temperature change from pre-industrial to present conditions when only LCLUC is considered. There is warming over agricultural regions in India, Europe and Brazil, but cooling over North America and northern Eurasia. The global averaged change is −0.05 K. The revised IBPM captures the cooling effect of LCLUC (−0.08 K on global average). The small global change in the original IBPM indicates the direct effect has a weak contribution to global temperature in this model. This agrees with previous studies implying LCLUC has a negligible global signature, even though there are significant local impacts [8,41,42]. A diminished IBPM effect is found in LME compared with our thorough global deforestation (from no-grass to grass) experiment; LME has much less change in land cover from 1850 to 2005, during which crops and pasture PFTs expand, and their influence is likewise limited.
Applying the IBPM to the all-forcing LME experiments, the direct effect of LCLUC (figure 9(c)) shows a similar pattern to that in land-change-only experiments ( figure 8(c)), even though the atmospheric background is very different. Radiative-and roughness-related temperature changes are consistent between the experiments, but the Bowen-ratio-related changes account for certain discrepancies. The differences in IBPM are attributable to changes in vegetation phenology as the terrestrial carbon and nitrogen cycles are active in the model. Figure 10 shows the change of leaf area index (LAI) from 1850s to 2000s in land-change-only and all-forcing experiments. LAI decreases in the land-change-only experiment, especially over Eurasia. When all forcings are included, LAI increases over most regions, except parts of East China, Eastern Europe, and Central America. The decrease of LAI in the land-change-only experiment is the result of agricultural expansion and wood harvest, while the increase in the all-forcing experiment implies that other drivers, such as CO 2 fertilization, nitrogen deposition, and longer growing seasons in high latitudes, may exert a larger influence on vegetation growth than anthropogenic LCLUC [35].
DTM shows good agreement with the modeled temperature change for the land-change-only    figure 11(d)). The changes in incoming shortwave and longwave radiation cools global temperature by −0.05 K and −0.03 K, respectively. At low latitudes, these components of incoming radiation often show opposite effects. For instance, greater incoming shortwave radiation is found over central Eurasia and Brazil, where there is decreased incoming longwave radiation; both associated with reduced cloud cover. Latent heat flux decreases due to lower LAI and leads to warming (about 0.04 K globally). Meanwhile, sensible heat flux decreases and warms global surface temperature by another 0.02 K. LCLUC does not have an obvious influence on ground heat flux in this experiment, showing only a small cooling at high latitudes.

Conclusions
This paper has explored two observation-based biogeophysical metrics of LCLUC impacts on temperature and their applicability in climate modeling. IBPM and DTM well capture surface temperature changes, and provide insight into the contribution of different surface components (such as albedo, surface roughness, net radiation, and surface heat fluxes) to surface climate.
IBPM [14] is useful to identify the direct impacts of LCLUC on surface temperature. Data from paired FLUXNET sites corroborate offline model sensitivity experiments that show surface roughness effects dominate the biogeophysical feedback of LCLUC at local climate background, while albedo and the Bowen ratio effects play secondary roles. Comparing results from offline and coupled simulations, or land-change-only and all-forcing experiments, there are consistent direct effects, indicating the robustness of IBPM. IBPM requires estimation of unmeasured parameters like aerodynamic resistance and sensitive parameters like Bowen ratio.
Our revision to IBPM shows the ability to represent changes when atmospheric feedbacks (indirect effects) are also considered. Both direct and indirect effects of LCLUC can then be identified. Coupled sensitivity experiments and long-term transient LCLUC experiments suggest that indirect effects can eclipse direct effects on regional to global scales.
DTM [15] shows better agreement with simulated temperature changes in coupled climate models. It provides a straightforward depiction of contributions from all components of the surface energy balance. All the components for this metric can be directly obtained from surface flux observations (where complete) or model output without derived parameters.
Development of metrics like these will help the climate modeling community validate climate model performance in simulating the response to LCLUC. Past model comparison studies [23] struggled to reconcile conflicting temperature and flux responses to land use change among models [41]. Verifiable LCLUC metrics can be used to diagnose the climate sensitivity in Earth system models, especially for the upcoming Land-Use Model Intercomparison Project (LUMIP, https://cmip.ucar.edu/lumip). The two metrics used in this study are focused on the surface energy balance; metrics related to the surface water balance should also be considered to quantify impacts of LCLUC directly on soil moisture and runoff, and indirectly on precipitation.