The role of El Niño in modulating the effects of deforestation in the Maritime Continent

The deforestation rate in the Maritime Continent (MC) has been accelerating during the past several decades. Understanding the changes in local hydro-climatological cycles as deforestation takes place is essential because the MC is suffering from frequent and extreme droughts and fires, which often occur during the dry season and are more severe during El Niños. Therefore, this study explores how deforestation affects the hydrological cycle and precipitation in the MC during El Niños, focusing on the boreal autumn season and using the coupled atmosphere–land model simulations. It is found that the precipitation over the MC increases in the deforestation experiments, and the precipitation responses can be magnified during El Niño events. A strong subsidence anomaly associated with El Niño does not prevent enhanced convection associated with local deforestation. Instead, the subsidence reduces the cloud cover in the MC region during El Niño, which increases the incoming solar radiation and increases surface temperatures. Thus, a warmer environment induced by El Niño modulates the biogeophysical feedbacks associated with deforestation that also play a critical role in more substantial land surface warming. A warmer land surface induces a more unstable atmospheric environment associated with a tendency toward enhanced local convection and lateral moisture convergence. This study highlights how the different mean climate states may modulate the impact of local land-use changes on hydroclimatological cycles in the MC, and sheds light on the state of our knowledge of interactions between the local land surface and remote large-scale atmospheric circulations.


Introduction
Forests in the Maritime Continent (hereafter MC) has been experiencing significant deforestation during the past few decades (Hansen et al 2013). Many studies have demonstrated the impact of Southeast Asia and the MC deforestation on the local and remote climates (e.g. McGuffie et al 1995, Werth and Avissar 2005, Mabuchi et al 2005a, van der Molen et al 2006, Schneck and Mosbrugger 2011, Kumagai et al 2013, Lawrence and Vandecar 2015, Chen et al 2019. Changes in precipitation are one of the local responses that attract the most attention. For example, Delire et al (2001) used an atmospheric model with a simplified land scheme to simulate the precipitation changes due to Indonesian Archipelago deforestation. They found that the wind speed increased due to a reduction in surface roughness after replacing forests with grasses. The accelerated wind transported abundant moisture from the surrounding ocean to Indonesian Archipelago and increased precipitation. Using a coupled atmosphereland climate model, Chen et al (2019) found that the surface temperatures (T S ) increase through biogeophysical feedbacks in their deforested MC experiment. The biogeophysical feedbacks involve the changes of radiation, heat fluxes, and aerodynamics related to the imposed changes in the land type. After replacing forests with grasses, the latent heat fluxes decreased, but the sensible heat fluxes increased, corresponding to the increases in T S (Chen et al 2019). The higher T S resulted in an unstable atmosphere and a tendency toward enhanced local convection; hence, precipitation increased. They also found that the local convection was strong enough to influence the local Walker circulation.
The Walker circulation is a zonal overturning circulation that exhibits strong interannual variations associated with El Niño-Southern Oscillation events. During El Niño years, sea surface temperatures (SSTs) in the tropical eastern Pacific warm and the Walker circulation weakens. This results in a subsidence anomaly over the western Pacific which tends to reduce precipitation over the MC. In boreal autumn, the subsidence and anomalous surface anticyclonic circulation over the MC are further strengthened; therefore, precipitation decreased more (Hendon 2003), which can cause extreme fire and drought events over the MC (Siegert et al 2001, van der Werf et al 2004. Supari (2018) also indicated that the local anomalously cool SSTs during September-October-November (SON) help sustain a surface anticyclone in the MC regions; hence, the entire land area in the MC experienced a negative precipitation anomaly.
The climatic responses in the MC to local anthropogenic land-use changes (e.g. Kumagai et al 2013, Takahashi et al 2017, Chen et al 2019 and the natural interannual variations associated with El Niño (Hendon 2003, Supari 2018) have been investigated. However, with more intensive deforestation in the MC region in the foreseeable future, it is worth exploring the combined effects of land-use change and interannual variations on the local precipitation. At first glance, one might assume that the joint impact is that the subsidence due to El Niño inhibits the enhanced local convection due to deforestation. However, experiments using a regional model showed that the extreme precipitation after Southeast Asian deforestation was amplified during El Niño years (Tölle et al 2017). But, the physical mechanisms behind how El Niño amplified the effect of deforestation were not explored in detail.
This research will examine the El Niño's modulation on deforestation and its impacts on the hydroclimate changes in the MC by conducting the global climate model experiments. Note that El Niño-induced subsidence over the MC is usually stronger during SON (Hendon 2003), which is also the fire season (Field et al 2009). Idealized experiments will be conducted to examine the interactions between the anomalous subsidence associated with El Niño and the circulation changes associated with deforestation in the region. The details of the model description, the model experiments, and the diagnostic methods will be given in section 2. Section 3 shows the results of precipitation responses associated with deforestation under normal and El Niño years. The discussion and conclusion will be presented in sections 4 and 5.   and the El Niño forcing over 40 • S-40 • N and 30 • E-60 • W. We followed Okumura et al (2011), which chose five El Niño (1965, 1972, 1982, 1987, and 1997) and five La Niña events (1955, 1973, 1988, 1998, and 1999) based on the SST anomaly over Niño 3.4 (5 • S-5 • N, 170-120 • W) in DJF (December-January-February). The El Niño forcing is derived from the mean of the five composite El Niño events compared to the long term mean SST between 1870 and 2003 (figure S1(a) in supplementary data (available online at stacks.iop.org/ ERL/16/054056/mmedia) minus the mean of the five composite La Niña events compared to the long term mean SST (figure S1(b); Okumura et al 2011) to reduce the gap between each cycle (figure S1(c)) during the simulations.

Model experiments
For each prescribed SST experiment, we conducted control (CTR) and deforestation (DEF) simulations to determine the effect of deforestation. In CTR, the plant functional types (PFTs) pattern in the MC is prescribed following figure S1(a) in Chen et al (2019). In DEF, the broadleaf deciduous tropical trees and broadleaf evergreen tropical trees in CTR were replaced by C4 grasses, representing the primary recovering vegetation type after deforestation (Carlson et al 2012, Chen et al 2019. This substitution was made only in the MC region (10 • S-10 • N and 90 • -150 • E, as shown in figure S3). In summary, we have four experiments to compare the deforestation effects under neutral and El Niño conditions, which will be named CTR NEU , CTR NIÑO , DEF NEU , and DEF NIÑO (table 1). In CLM4.0, a grid consists of vegetated land, glacier, wetland, urban, and lake units. The vegetated land unit includes up to 16 PFTs. All PFTs shared one soil column, resulting in less realistic heat fluxes in the heterogeneous landscapes (Schultz et al 2016). Yet, in CTR and DEF, the MC's grids are rather homogeneous, with more than 80%-90% of the sub-grid is covered by forests in CTR and grasses in DEF (figures S2 and S3). Therefore, the relatively homogeneous PFTs in our experiments are less affected by the shared-soil column issue.

Reanalysis data
We also used the zonal wind and vertical velocity data in the Japanese 55 year Reanalysis (JRA55; Kobayashi et al 2015) to confirm the performance of the meridional average zonal circulation in CESM. The reanalysis data is monthly data with a horizontal resolution of 1.25 • by 1.25 • and 17 pressure levels. Neelin and Held (1987) derived a moist static energy (MSE) budget that represents the potential energy available for the tropical convection. The MSE consists of latent, sensible, and potential energy components. It can be written in the following form:

Moist static energy
where L represents the latent heat (kJ kg −1 ), q represents the specific humidity (kg kg −1 ), C p represents the specific heat of air at constant pressure (kJ (kg × K) −1 ), T represents the temperature (K), g represents the acceleration of gravity (m s −1 ), and z represents height (m). The unit of MSE is kJ kg −1 .

Surface energy balance equation
The surface energy balance equation can be used to analyze the T S changes associated with changes in the radiative and surface heat fluxes (Chen and Dirmeyer 2016). The equation is written as follows: where the unit of all term is W m −2 . R net is the net surface radiation. R net can be presented in heat flux or radiation form. In the radiation form, SW net is the surface net shortwave flux, LW in is the incoming surface longwave flux, and εσT 4 S is the surface outgoing longwave flux based on the Stefan-Boltzmann law, where ε is surface emissivity, and σ is the Stephan-Boltzmann constant. For R net , SW net , and LW in , the downward direction is positive. In the heat fluxes, SH is surface sensible heat flux, LH is surface latent heat flux, and G is ground heat flux. SH can be divided into two components: ground SH and vegetation SH. The amount of vegetation SH depends on the temperature difference between the canopy and the overlying air. The ground SH is from bare soil and soil beneath the canopy. Since the canopy covered more than 80% of the MC (figures S2 and S3), the amount of ground SH is mostly from the soil beneath the canopy. LH can also be divided into three components: canopy evaporation (CE), canopy transpiration (CT), and ground evaporation (GE). Both bare soil and the ground beneath the canopy can produce GE. Same as ground SH, the GE is mostly from the latter. G is on the order of 10 −1 , which is much smaller than either the LH (10 1 -10 2 ) or SH (10 1 ). The contribution of G will be ignored in this study. For SH and LH, the upward direction is positive.

Statistical methods
Student T-test is used to test whether the difference between CTR NEU and CTR NIÑO , DEF NEU and DEF NIÑO , and (DEF-CTR) NEU and (DEF-CTR) NIÑO is significant. To test the difference between (DEF-CTR) NEU and (DEF-CTR) NIÑO , we used pooled sample variance to obtain the variance of (DEF-CTR) NEU and (DEF-CTR) NIÑO from CTR NEU and CTR NIÑO ; DEF NEU and DEF NIÑO , respectively. Also, the similarity of the variances of two samples is tested by F-test.

Results
To understand how El Niño modulates the deforestation's impact on the precipitation over the MC, we focus on SON because the El Niño-induced subsidence anomaly over the MC is the strongest during this season (Hendon 2003). This is also shown in figure 1(a) using the JRA55 reanalysis data and in figure 1(b) using the model simulations. The location and magnitude of the subsidence in the model are slightly different from those in the reanalysis. The subsidence in the model (reanalysis) is located to the west of 150 • E (165 • E). The magnitude of the model's subsidence is weaker than that in reanalysis, but the location of the subsidence center is somewhat similar.
The large-scale circulation is unfavorable for the convection over the MC in the CTR NIÑO experiment ( figure 1(b)). Thus, one might expect the strong subsidence would inhibit any anomalous convection (and therefore reduce any anomalous precipitation) associated with deforestation. To estimate the changes in precipitation associated with deforestation for different mean states, we quantify land precipitation over the MC. Table 2 shows an increase precipitation associated with deforestation in the NEU experiment of 0.59 mm d −1 ((DEF-CTR) NEU ) and an increase of 0.79 mm d −1 in the NIÑO experiment ((DEF-CTR) NIÑO ). There are two implications of these results. First, the precipitation over the MC is enhanced by deforestation in both experiments, though the mean climate states are different (i.e. climatological and El Niño SSTs). The local forcing, deforestation, is strong enough to force local convection even though the large-scale circulation is unfavorable (under El Niño). Second, the increase in precipitation due to deforestation was larger in El Niño experiment by about 0.2 mm d −1 . The amplification suggests that El Niño could modulate the effects of deforestation. This result is counter to our expectation that the subsidence induced by El Niño would inhibit the convection in general. However, the 34% increase in precipitation (0.2 mm d −1 ) is not large enough to pass a significance test due to the large interannual variability in both the NEU and NIÑO experiments (7.7 mm d −1 and 17.5 mm d −1 , respectively).
To explore why and how El Niño amplifies (rather than inhibits) the increases in precipitation associated with deforestation in the MC, we analyzed the vertical profiles of MSE. The MSE not only quantifies the potential energy that the atmosphere provides for convection but also includes information about the dominant component of convection (Chen et al 2019). Figure 2 illustrates the vertical MSE profile anomalies. We found that the MSE anomaly in NIÑO (black line in figure 2(b)) is larger (more convective potential and unstable troposphere) than that in NEU (figure S6(Db) in Chen et al 2019). The larger (smaller) convective potential in NIÑO (NEU) is consistent with the larger (smaller) increases in precipitation (table 2). Chen et al (2019) showed that the reduced latent heat fluxes after deforestation decreased lowlevel moisture. This can be seen in the Lq ′ term below 900 hPa in NIÑO (blue line in figure 2(a)). However, the vertical profile of Lq ′ above 900 hPa is different in two climate mean states with a more positive Lq ′ in NIÑO (blue line in figure 2(b)). This larger amount of moisture corresponds to greater integrated (850-950 hPa) low-level moisture convergence (9.94 W m −2 ) and integrated (100-950 hPa) omega (−68.52 Pa s −1 ) as shown in table 2. The moisture convergence and convection anomalies in (DEF-CTR) NEU are only 4.5 W m −2 and −50.84 Pa s −1 , respectively. The deforestation most likely drives local convection and moisture convergence in  figure S4. both climate states, while their magnitude is somewhat different depending on the large-scale conditions.

The contribution of El Niño
In the CTR NIÑO simulation, T S was warmed by 0.16 • C (table 3) compared to the T S in CTR NEU simulation due to the reduced cloud cover (−5.2%) and increased incoming solar radiation (16.75 W m −2 ). The warmer land resulted in stronger atmospheric instability, leading to the potentials of low-level convergence and convection. However, given the relatively dry conditions, why is there more moisture convergence in the MC in DEF NIÑO experiment? The MC is located in the warm pool of the western Pacific, which is a source of abundant moisture even when MC's SSTs decrease somewhat during El Niño. Therefore, the land T S rises after deforestation in NIÑO with more local land convections associated with more low-level convergence from the surrounding ocean to the MC's land. This results in greater precipitation after deforestation in NIÑO (0.79 mm d −1 ) than that in NEU (0.59 mm d −1 ). In sum, El Niño is associated with a warmer land surface in the MC that produces greater convective instability related to a deforested MC, in which stronger convection and larger amounts of precipitation in (DEF-CTR) NIÑO than that in (DEF-CTR) NEU is apparent. The following section presents more detailed analysis of land-air interactions to understand the possible reasons for the amplified deforestation effect under El Niño, and also to explore the El Niño's modulation on the changes of biogeophysical feedbacks, including surface radiation, heat fluxes, and aerodynamics associated with deforestation.

Differences in land-air interactions in different climate states
The changes in biogeophysical feedbacks and their relations with the T S changes associated with deforestation can be quantified in equation (2) Dirmeyer 2016, Chen et al 2019), which represents the land-air interactions. By analyzing the energy balance equation (equation (2)), we can link the changes in heat and radiative fluxes to the changes of T S associated with deforestation. Table 2 showed that most of the radiative and heat fluxes changes after deforestation are larger in NIÑO than those in NEU, especially the T S and LH. Their amplification through El Niño's modulation was significant (0.12 • C for T S , and −3.36 W m −2 for LH); the details will be discussed below.

Deforestation effects for neutral conditions
After we replaced the forest with grass in NEU, LH decreased 13.2 W m −2 mainly because of a reduction in canopy evapotranspiration. Upon decomposing the LH, we find a decrease in CE (−6.22 W m −2 ) and CT (−18.56 W m −2 ) but an increase in GE (11.58 W m −2 ). CE decreases because of the smaller leaf area index (LAI) in the grasses than that in the forests, leading to lower canopy interception. The CT decreases not only due to the decreases of LAI but also due to weaker aerodynamics and transpiration efficiency after deforestation. Replacing the forests with grasses reduces the roughness and weakens the aerodynamics. The flatter plain decreases turbulence and reduces heat fluxes; therefore, when the forest is replaced with grass in our simulation, CT decreases. GE should also decrease because the ground under canopy supposed to be affected by the weakened aerodynamics. However, it increases because the grasses extract less soil moisture compared to the forests for transpiration. Thus, more water remains on the ground to be evaporated. By adding all components of LH together, the total LH decreases (by 13.2 W m −2 ) in (DEF-CTR) NEU . On the other hand, SH in DEF NEU increases by 6.39 W m −2 compared to CTR NEU to compensate for the decrease in LH. Furthermore, the two components of SH exhibit changes of opposite sign. The vegetation SH decreases (−1.12 W m −2 ) because of the reduction in aerodynamics. The ground SH did not decrease by the reduction in aerodynamics; instead, it increases (7.51 W m −2 ) to compensate for the decreases in LH and vegetation SH. Its increase also reflects the increases in T S . Clouds and albedo also play an essential role in changing the radiative fluxes. A 1% decrease in lowclouds in (DEF-CTR) NEU results in more incoming SW, but an increase in surface albedo and higher midcloud cover (1.83%) reflecting more SW. Overall, the net SW decreases by 1.48 W m −2 in DEF NEU compared to CTR NEU . On the other hand, the incoming LW increases by 2.31 W m −2 , mainly due to increases in mid-cloud cover. Chen and Dirmeyer (2016) used equation (2) to analyze the contributions from the surface radiative and heat flux components to changes in T S . We attribute the rising T S in DEF NEU to decreases in LH and increases in the downward LW. After balancing the heat flux and radiation terms, the surface gains extra energy in DEF NEU compared to CTR NEU . In other words, T S must increase to emit more LW to balance the extra energy absorbed by the surface. Therefore, we find that T S over land in the MC in DEF NEU is 1.06 • C higher than that in CTR NEU .

Deforestation effects for El Niño conditions
The changes in the surface radiative and heat flux terms in (DEF-CTR) NIÑO resembles those in (DEF-CTR) NEU , though the changes are larger for some fluxes (table 2). This suggests that El Niño can modulate the effects of deforestation. For example, the CE in CTR NEU were 30.78 W m −2 and in DEF NEU were 24.56 W m −2 (table 4). However, the CE in CTR NIÑO was 25.87 W m −2 and in DEF NIÑO was 21.2 W m −2 (table 4). The decreases in CE in both CTR NIÑO and DEF NIÑO compared to CE in CTR NEU and DEF NEU is due to less canopy water from reduced precipitation in El Niño state (table 3). There was a decrease of 6.22 W m −2 in CE in (DEF-CTR) NEU , whereas the decrease in (DEF-CTR) NIÑO is 4.67 W m −2 . El Niño's modulation makes the CE in (DEF-CTR) NIÑO decreased less by about 1.55 W m −2 (table 2) compared to the CE in (DEF-CTR) NEU . The modulation can also be found in other heat and radiative fluxes. The changes in CT, on the contrary, are amplified by El Niño. Table 2 shows that CT decreases substantially in (DEF-CTR) NIÑO (−23.26 W m −2 ). The main reason is that transpiration in forests is more efficient than in grasses, especially in warmer climate states (57.41 W m −2 for CTR NEU , 66.52 W m −2 for CTR NIÑO ; table 4). However, the transpiration in  (table 4). Thus, the efficiency in heating the different canopy causes higher vegetation SH changes in (DEF-CTR) NIÑO (−3.22 W m −2 ) but smaller in (DEF-CTR) NEU (−1.12 W m −2 ). On the other hand, deforestation in NIÑO results in increased ground SH (9.18 W m −2 ) from a warmer T S in NIÑO (table 3). The modulation effects in the two SH components are significant (−2.1 W m −2 for vegetation SH, and +1.67 W m −2 for ground SH). But, their competing effects result in the difference in total SH between (DEF-CTR) NIÑO (5.96 W m −2 ) and (DEF-CTR) NEU (6.39 W m −2 ) is rather small (−0.43 W m −2 , table 2), implying that the SH may not be an essential term in the modulated T S (0.12 • C).
The T S increases by 1.18 • C in (DEF-CTR) NIÑO (table 2), which is significantly larger than the corresponding increase in T S in (DEF-CTR) NEU . By analyzing the detailed heat flux and radiation changes, we attribute the larger increase in T S in (DEF-CTR) NIÑO to a larger decrease in LH (−16.56 W m −2 , table 2) and a larger increase in downward LW (3.1 W m −2 , table 2). Both these changes in LH and LW contribute to the greater increase in T S . However, the cooling of the land surface was also more substantial in (DEF-CTR) NIÑO due to a significant decrease in net SW (−5.15 W m −2 , table 2). Nevertheless, the surface gains more energy in (DEF-CTR) NIÑO than it loses as compared to the net energy in (DEF-CTR) NEU . The T S increases by 1.18 • C, which is significantly larger than the increase in T S in (DEF-CTR) NEU (table 2). This result is consistent with Chapman et al (2020), who suggested the deforestation might lead to more extreme hot in Borneo under El Niño events compared to neutral conditions. The larger increase in T S in (DEF-CTR) NIÑO also contributes to more convective instabilities as shown in MSE changes (black line in figure 2(b)). Therefore, in (DEF-CTR) NIÑO , the local convection was stronger than that in (DEF-CTR) NEU , and there was more moisture convergence (9.94 W m −2 ) and larger precipitation increases (0.79 mm d −1 ) than those in (DEF-CTR) NEU .

Discussion
The framework of land-air interactions is another perspective by which to measure El Niño's effects on the impacts of deforestation. These effects are manifested by the changes in surface radiative and latent and sensible heat flux. We found greater decreases in LH in (DEF-CTR) NIÑO than that in (DEF-CTR) NEU because of larger CT decreases. However, the increased SH in (DEF-CTR) NIÑO is similar to that in (DEF-CTR) NEU because of the competing changes in the two components of SH. Moreover, the mid-and high-cloud cover increases in (DEF-CTR) NIÑO compared to (DEF-CTR) NEU with less downward SW and more downward LW at the surface. El Niño's modulation on the land-air interactions result in enhanced radiation and heat flux exchanges in (DEF-CTR) NIÑO . The T S increases more in (DEF-CTR) NIÑO than that in (DEF-CTR) NEU . Higher T S in (DEF-CTR) NIÑO induces more atmospheric instability, which could increase more moisture convergence associated with more increased precipitation in (DEF-CTR) NIÑO than those in (DEF-CTR) NEU . However, this study only pointed out that different vegetation types have different responses under various atmospheric conditions. How each vegetation type react to different atmospheric states; moreover, would a fully coupled land-atmosphere-biosphere model amplify the reaction? These questions need to be further investigated by including more realistic PFT phenology for the oil palms, such as CLM-Palm (Fan et al 2015). Furthermore, our model results have a wet bias over the northern MC under the El Niño conditions (not shown). It will be deserved to explore how it may affect the results shown in this study. La Niña's modulation is also deserved to be further explored.
Deforestation has effects similar to those of an urban heat island. The increases in T S , corresponding to the urban heat island effects, contribute to an increase in convective rainfall. Thielen et al (2000) showed significant changes in heat fluxes and local precipitation due to urbanization. They proposed that the unstable environment is characterized by positive buoyancy associated with sensible heat flux variations and has the potential to trigger convection. Their results might support those of Takahashi et al (2017) who demonstrated the heating in Borneo associated with deforestation, which could enhance the amplitude of the diurnal cycle in precipitation there. Deforestation in the MC acts similar to a heat island effect with a different spatiotemporal scale. Since the MC has suffered from drought and fire events, our finding implies that the increasing precipitation tendency associated with deforestation might influence the frequency and magnitude of drought and fire events during the dry season. Such alterations might be more efficient during El Niño events.

Conclusion
This study investigates the impacts of a deforested MC on local precipitation during neutral and El Niño years. We examine these impacts using the coupled atmosphere-land model and design deforestation experiments under neutral and El Niño SST conditions. The results can be summarized in figure 3 that in the neutral years, when the land surface properties are changed to reflect deforestation, the LH fluxes decrease so that the surface retains more energy. In contrast, the SH fluxes increase to compensate for the decrease in LH fluxes. The net surface solar radiation also decreases due to the larger surface albedo and increased mid-cloud cover, resulting in larger downward longwave radiative fluxes. Thus, the surface gains more energy via the reduced LH fluxes and increased downward LW fluxes, resulting in increases in T S . The same biogeophysical feedbacks associated with deforestation were found in the NIÑO simulation; however, El Niño amplifies these feedbacks because of warmer environment. During El Niño, the tropical western Pacific atmospheric circulation is characterized by a subsidence anomaly. The large-scale descending motion decreased cloud cover and increased the downward solar radiation, leading to increased surface temperature. Such a warmer environment could amplify those biogeophysical feedbacks associated with deforestation and increase more T S . The atmosphere becomes more unstable, favoring stronger convection in (DEF-CTR) NIÑO compared to (DEF-CTR) NEU . Therefore, there is more precipitation in (DEF-CTR) NIÑO than (DEF-CTR) NEU . This study indicates how the different climate states may alter the impact of local landuse changes on hydroclimatological cycles in the MC and provides another perspective on our knowledge of interactions between the local land surface and remote SST forcings.

Data availability statement
The data that support the findings of this study are available upon reasonable request from the authors.