Strong Green‐Up of Tropical Asia During the 2015/16 El Niño

El Niño/Southern Oscillation (ENSO) is the main climate mode that drives the interannual variability in climate and consequently vegetation greenness. While widespread green‐up has been reported and examined in tropical America during El Niño, it remains unclear how vegetation in tropical Asia changes during the period. Here, we used four remote sensing‐based leaf area index (LAI) products to investigate changes in vegetation greenness during the 2015/16 El Niño in tropical Asia. We found a strong green‐up during the 2015/16 El Niño in tropical Asia, with its regional average LAI stronger than that of tropical America. The drivers for the green‐up vary across the region, with radiation being the main driver for continental tropical Asia, and temperature and soil water anomalies in the west and east parts of maritime tropical Asia, respectively. These findings provide important insights into the response of tropical Asia's vegetation to extreme climate anomalies.


Introduction
El Niño/Southern Oscillation (ENSO) is an irregularly recurring large-scale climate phenomenon that causes considerable interannual variability in tropical climate and carbon cycle (Braswell et al., 1997;Jones et al., 2001;Kondo et al., 2018;Rasmusson & Wallace, 1983;Zhang et al., 2019).ENSO fluctuates periodically between three phases: a neutral phase, a warm phase called El Niño, and a cold phase called La Niña (Santoso et al., 2017).In tropical forest regions, El Niño events are strongly associated with high temperatures and low precipitation (Malhi & Wright, 2004).This variation of temperature and precipitation consequently impacts vegetation dynamics, leading to anomalous increases in greenness (green-ups) or decreases (browning), which can typically be detected using remote sensing vegetation indices.
The effects of El Niño events or droughts alike on vegetation greenness has been widely explored for tropical America.Significant green-up of Amazon rainforest was first detected by remote sensing during the 2005 drought (Saleska et al., 2007), although subsequent assessments using stricter cloud screening led to debates regarding the magnitude and extent of this green-up (Anderson et al., 2010;Samanta et al., 2010).In recent years, remote sensing observations made during the 2015/2016 El Niño generally provided more consistent findings of green-up in the region (Janssen et al., 2021;J. Yang et al., 2018).The widely proposed mechanism for the green-up is positive anomalies in radiation due to clearer skies during El Niño years (Anderson et al., 2010;Asner & Alencar, 2010;Saleska et al., 2007).This mechanism is also used to explain the green-up of Amazon during the dry seasons at the sub-annual scale (Huete et al., 2006).
Meanwhile, our understanding of the impact of El Niño on vegetation in other parts of the tropics, particularly in tropical Asia, remains unclear.Due to the differences in climate, soil condition, and phylogeny between tropical Asia and America (Zhang et al., 2016), we expect different responses of tropical Asia's greenness to the climate anomalies during El Niño.Several global-scale studies suggest the possibility, reporting different direction or magnitude of changes in vegetation carbon uptake during El Niño between tropical regions (Luo et al., 2018;Palmer, 2018;Zhu et al., 2018).There are a few recent attempts to specifically investigate the changes in vegetation greenness in tropical Asia during El Niño (Nunes et al., 2019;Qian et al., 2019;Yue et al., 2017).However, these studies reported inconsistent findings in the timing (i.e., early El Niño in 2015 vs. late El Niño in 2016) and green-up locations (i.e., continental vs. maritime tropical Asia).For instance, Yue et al. (2017) found pronounced green-up in tropical Asia peaking in the early 2015 El Niño period, while Qian et al. (2019) reported green-up in 2016 and mainly for maritime tropical Asia.Additionally, these studies predominantly rely on a single remote sensing product as a proxy for greenness.Given the inconsistencies between remote sensing vegetation products in evaluating long-term trends and variability of vegetation greenness (Jiang et al., 2017;Z. Wang et al., 2022), further analysis using multiple data sets is needed.
The 2015/16 El Niño, one of the strongest El Niño event in the 21st century (Santoso et al., 2017), provided a rare opportunity to study the impact of ENSO on vegetation dynamics in the satellite era.Additionally, recent advancement in geostationary satellites with higher monitoring frequency compared to polar-orbiting satellites offers a great potential for studying vegetation dynamics in the tropics (Hashimoto et al., 2021).Here, we used remote sensing-based leaf area index (LAI) products (i.e., GIMMS, GLOBMAP, GLASS and MODIS) and climate data sets, complemented by novel observations from the Himawari-8 geostationary satellite to assess the changes in tropical Asia vegetation during 2015/16 El Niño.Specifically, we aim to analyze: (a) the impact of the 2015/16 El Niño on tropical Asia vegetation in comparison to other tropical regions, and (b) the primary climate drivers influencing vegetation changes in tropical Asia during the 2015/16 El Niño.We hypothesize that tropical Asia experienced green-up during the 2015/16 El Niño, comparable in magnitude to the observed effects in tropical America forests.We propose that the green-up is due to increased radiation associated with El Niño, drawing parallels with well-documented outcomes in tropical America.
A previous study reported inconsistencies within and between long-term LAI products (i.e., GLASS and GLOBMAP) pre-2000, likely due to the change in satellite platform from AVHRR to MODIS (Jiang et al., 2017).Additionally, Winkler et al. (2021) reported a change of trend in vegetation green-up post-2000.Therefore, to ensure consistency and reliability, we set the study period from 2001 to 2018 to gain more comparable estimates.
To support our analysis, we also examine vegetation greenness in parts of maritime tropical Asia (e.g., Sumatra and Borneo) using Advanced Himawari Imager (AHI) on board the geostationary satellite Himawari-8 (Bessho et al., 2016).We derived NDVI using top-of-atmosphere (TOA) reflectance from Himawari-8 AHI for the year 2016-2021 (acquired from https://data.nas.nasa.gov/geonex/geonexdata/HIMAWARI8/GEONEX-L1G/)(Wang et al., 2020).Prior to the NDVI calculation, the red band (0.5 km spatial resolution) was resampled to 1 km using pixel aggregation (averaging) to match the resolution of the NIR band.Himawari-8 AHI has a temporal resolution of 10 min, and we only considered images during daytime (02:00-08:00 UTC) in the analysis.To remove outliers of NDVI due to cloud contamination, we applied maximum-value composite (MVC) to the 10-min interval images for each month following (Hashimoto et al., 2021).However, instead of taking the maximum value of NDVI for each month, we took the 95% percentile NDVI value to avoid spikes in the timeseries.Using this method, we assume that the monthly clear sky NDVI for each pixel is the top 5% value across the 10-min intervals over that month.The 2016 NDVI anomaly was calculated from subtracting the averaged NDVI over 2017-2021 from the 2016 NDVI.

Environmental Variables
To evaluate the environmental variables, we used the gridded land surface Climate Research Unit and Japanese Reanalysis (CRU JRA) V2.1 data set (Harris et al., 2020;Kobayashi et al., 2015) and Global Land Evaporation Amsterdam Model (GLEAM) v3.7a data set (Martens et al., 2017a;Miralles et al., 2011).CRU JRA data set (downloaded from https://catalogue.ceda.ac.uk/uuid/10d2c73e5a7d46f4ada08b0a26302ef7) has a temporal resolution of 6 hr with a spatial resolution of 0.5°.Climate variables acquired from CRU JRA are temperature at 2 m (°C), precipitation rate (mm), shortwave (SW) radiation (W/m 2 ), and vapor pressure deficit (VPD in kPa) calculated from vapor pressure and temperature.We acquired soil moisture from GLEAM daily root-zone soil moisture data in a spatial resolution of 0.25°(acquired from https://www.gleam.eu/).All data sets were resampled to monthly step, and GLEAM soil moisture was further resampled to 0.5°using pixel aggregation method (averaging).

Plant Functional Types (PFTs)
Yearly maps of plant functional types (PFTs) from 2001 to 2020 were acquired from the MODIS Land Cover Type Product (MCD12Q1; downloaded from https://lpdaac.usgs.gov/products/mcd12q1v006/)at 0.05°spatial resolution (Friedl & Sulla-Menashe, 2019).The PFTs used in this study are derived from the International Geosphere Biosphere Programme (IGBP) classification scheme (Loveland & Belward, 1997).To match the resolution of climate data sets, the PFTs map was aggregated to 0.5°by assigning the most dominant class within the 0.5°grid.Pixels with changing PFTs over the 2001-2018 period were then excluded from the analysis.The PFTs were then further classified into forest (evergreen broadleaf forest (EBF), deciduous broadleaf forest, and mixed forest) and non-forest (other PFTs.)

El Niño/Southern Oscillation (ENSO) Index
The development of 2015/16 El Niño started in early 2015.Based on bi-monthly Multivariate ENSO index (MEI.v2; acquired from https://psl.noaa.gov/enso/mei/),strong El Niño (i.e., MEI > 0.5) started in May 2015 and persisted for another year before it dissipated in the second half of 2016.In our analysis, we used the entire months of 2015-2016 to derive the 2015/16 anomalies.

Data Analysis
Monthly and yearly LAI and environmental variables were calculated by averaging the values (or sum for precipitation) of each pixel at the respective timescales.Yearly LAI and environmental variables were then detrended to get yearly anomalies for each pixel.The detrending was done by fitting a linear regression to the yearly timeseries and then subtracting the predicted values from the observed yearly values.Similarly, anomalies of Geophysical Research Letters 10.1029/2023GL106955 monthly LAI and environmental variables of each pixel were calculated by subtracting the trends from the original timeseries of each month.
Yearly and monthly anomalies for tropical Asia, tropical Africa, and tropical America were obtained by averaging the LAI and climate anomalies by region.Here, tropical regions were defined as the region between 23°S and 23°N (excluding Australia).Similarly, yearly and monthly anomalies for each PFT in tropical Asia were obtained by averaging the LAI and climate anomalies according to the PFT map.Lastly, we calculated the ensemble mean of LAI anomalies from the four different remote sensing products along with their uncertainties at each timestep for each region and each PFT in tropical Asia.Note that three to four LAI products consistently exhibited the same direction of LAI anomaly in >85% of the pixels (Figure S1 in Supporting Information S1).This robust consistency between products supports our approach of using the ensemble mean of LAI anomalies in further analysis steps.All data processing steps were done in MATLAB v9.13.0 (R2022b) (The MathWorks Inc., 2022).
Descriptive statistics for monthly anomalies for the entire 2015/16 El Niño were computed for each region and PFT in tropical Asia.A one-sample t-test was also conducted to evaluate the significant difference between the mean monthly anomalies and zero.
To analyze the sensitivity of vegetation greenness to environmental variables, we used multiple linear regression to fit monthly anomalies of LAI from 2001 to 2018 with climate anomalies as the independent variables.Based on the model selection, the best models (i.e., lowest Bayesian Information Criterion (BIC)) consistently contain the combination of SW radiation, T, and a water-related variable (e.g.,VPD, soil moisture, or precipitation) (Tables S1 and S2 in Supporting Information S1).Although VPD and precipitation perform better than soil moisture, they are both strongly correlated to SW radiation and T (Figure S2 in Supporting Information S1).Thus, we chose soil moisture as one of the predictors to minimize multicollinearity between the independent variables.The formula of the regression is given below: where y is the monthly LAI anomaly, x i are the independent variables (i.e., SW radiation, T, and soil moisture), and β i is the regression coefficient, which is the measure of linear sensitivity of LAI to each environmental variable input x i .All steps of statistical analysis were done in R v4.2.2 (R Core Team, 2022).

Strong Green-Up of Tropical Asia During 2015/16 El Niño
Timeseries of yearly LAI anomaly shows that 2015/16 El Niño led to varying vegetation changes across the three tropical regions-tropical Africa, tropical America and tropical Asia (Figure 1a).We found tropical Asia and tropical America showed significant green-up during the El Niño, while tropical Africa experienced a slight browning.The green-up in tropical Asia (i.e., positive LAI anomaly of 0.04) is two times stronger than that of tropical America (i.e., positive LAI anomaly of 0.02) (Figure 1b).Additionally, we observed green-ups across tropical forests during the 2015/16 El Niño event (Figure S3 in Supporting Information S1).In contrast, nonforest regions in tropical America showed little to no change, while those in Africa experienced a slight decrease (Figure S4 in Supporting Information S1).Notably, both forested and non-forested areas in tropical Asia experienced significant green-ups during both El Niño years, as opposed to tropical America that showed greenup only in forested areas in the first year.Among the detailed forest PFTs in tropical Asia, we found EBF contributed the most to green-up in the region, followed by non-forest PFTs such as wetlands, and savannas (Figures S5 and S6 in Supporting Information S1).
Across the pan tropics, green-up was more intense and widespread in 2015 relative to 2016.For tropical Asia in 2015, green-up was apparent in Malay peninsula, the Philippines, the eastern and northern part of Borneo, western Sumatera and New Guinea, while in 2016, the green-up was mostly in eastern Sumatera and southern Borneo.Generally, individual LAI products similar spatial patterns of green-up in the region (Figure S7 in Supporting Information S1).The 2016 green-up in tropical Asia was further corroborated by observations from the geostationary satellite Himawari-8 (Figure 1e).While the location of green-up hotspots varies between data sets, both Himawari-8 NDVI and ensemble LAI anomalies show a predominant occurrence of green-up across Sumatra and Borneo (Figure S8 in Supporting Information S1).

The Drivers of Green-Up Varies Spatially Across Tropical Asia
We found all three tropical regions experienced a significant (p < 0.05) increase in VPD and T and a decrease in P and soil moisture during the 2015/16 El Niño period, while the changes in SW radiation are only significant (and positive) in tropical Asia (Figures 2b-2f).Comparing tropical America and tropical Asia, we found that positive T and VPD anomalies are larger for tropical America, but positive SW radiation anomalies are larger for tropical Asia.Considering tropical Asia experienced a similar decline in water availability (i.e., P and SM) with that of tropical America, our result indicates that the higher SW radiation is a key factor driving the green-up in tropical Asia on regional average.We further examined whether there is a spatial variation in the drivers of green-up in tropical Asia, considering the SW anomalies were not homogenous across tropical Asia in 2015 and 2016 (Figure 3d).We used multiple linear regression to fit 2001-2018 monthly LAI anomalies with monthly anomalies of environmental variables (i.e., SW radiation, T, and soil moisture) as the independent variables.The model generally provides a robust estimate, in which 65.84% of the pixels are statistically significant (Figure S9 in Supporting Information S1).Overall, we found a clear spatial variation in the climate sensitivities of LAI in tropical Asia (Figures 3a-3c).The continental tropical Asia has the largest sensitivity to radiation, while the western part of maritime tropical Asia is most sensitive to temperature, and the eastern part of maritime tropical Asia is most sensitive to soil moisture.
By multiplying the monthly anomalies of environmental variables during the 2015/16 El Niño (Figures 3d-3f) by the sensitivity of each variable, we further evaluated the magnitude of green-up explained by each climate variable across tropical Asia (Figures 3g-3i).In contrast to the regional average result (Figure 2), we found that the anomalously high SW radiation mainly drove the green-up in continental tropical Asia, while the green-up in the western part of maritime tropical Asia (i.e., Borneo and Sumatra) was driven by anomalously higher T mainly, and the green-up in the eastern part of maritime tropical Asia was driven by anomalously low soil moisture.

Discussions
Using multiple long-term remote sensing-based LAI products, our study presents evidence of strong green-up in tropical Asia during the 2015/16 El Niño event.The magnitude of the green-up in tropical Asia is roughly two times that of Amazon during the El Niño event.We found stronger and more widespread green-up in 2015 instead of 2016.We further corroborated strong green-up in Peninsular Malaysia, Sumatra, and Borneo in 2016 using novel geostationary satellite observations.The drivers for the green-up vary spatially across tropical Asia, with anomalously high SW radiation driving the green-up in continental tropical Asia, and anomalously high temperature and low soil moisture the green-up in maritime tropical Asia.
The green-up in the tropics during dry seasons (Hashimoto et al., 2021;Huete et al., 2006) and dry years (e.g., 2005 and 2010 for Amazon, and El Niño years) have heavily examined (Luo et al., 2018;Saleska et al., 2007;J. Yang et al., 2018) in particular for tropical America, while other parts of the tropics remain less studied and understood (Nunes et al., 2019;Qian et al., 2019).Similar to previous findings by Yue et al. (2017), we found stronger and more widespread green-up in tropical Asia in 2015 instead of 2016.However, we found less greenups in continental tropical Asia in contrast to the strong green-up Qian et al. (2019) reported in the area for both years.These differences might result from different vegetation products used in the analysis, as Qian et al. (2019) used EVI, which has been documented to reach saturation at areas with high LAI such as forests (Alexandridis et al., 2020).
There are various proposed mechanisms surrounding the drivers of green-up.In this discussion, we will adopt the framework of Dynamic Global Vegetation Models (DGVMs), wherein LAI is approximated from gross primary productivity (GPP), carbon use efficiency (CUE), and carbon allocation (Cui et al., 2019).GPP represents the total carbon input from photosynthesis, while CUE indicates the proportion of carbon available for biomass production.Since carbon allocation to leaves is often assumed constant in DGVMs, changes in climatic variables primarily influence GPP and CUE.Following this framework, green-up is likely driven by enhanced leaf flush due to increasing GPP with higher radiation from clear sky during El Niño period, as widely reported in tropical America (Brando et al., 2010;Huete et al., 2006;Janssen et al., 2021).However, this notion has been challenged, as recent studies reported a decoupling of GPP and greenness changes in El Niño years, where GPP decreased but greenness increased (Koren et al., 2018;Qian et al., 2019;Yang et al., 2018).This indicates that the enhanced leaf flush or green-up, is not necessarily driven by the increase in GPP, but may result from reduction in autotrophic respiration, thereby increasing net primarily productivity.These processes do not have the same climate sensitivities as GPP, therefore, the climatic drivers for green-up can be very diverse and not limited solely to radiation.Note that while other possible mechanisms, such as sink limitation (e.g., a shift in carbon allocation to leaves; Körner, 2015;Doughty et al., 2014) and biological factors (e.g., leaf age and vertical shifts in leaf area; Smith et al., 2019;Wu et al., 2016) may contribute to green-up, we will not discuss them in detail here.
Indeed our study demonstrates that the drivers of green-up varied in tropical Asia.Based on the pixel-wise sensitivity analysis, light mainly drove green-up in continental tropical Asia (Figure 3g), with a proposed mechanism of more radiation-higher GPP-more leaves.This is supported by the fact that some parts of continental tropical Asia (i.e., Vietnam and peninsular Malaysia) experienced an increase in GPP during the 2015/ 16 El Niño (Luo et al., 2018;Zhu et al., 2018).Since the increase in GPP is stronger in 2015 than in 2016 in this region, this also explained why the green-up is stronger in 2015 than 2016 (Luo et al., 2018;Nunes et al., 2019).
For west maritime tropical Asia, the green-up was driven by high temperature.The potential mechanism could be: (a) higher T -higher GPP-more leaves or (b) higher T -higher biomass production-more leaves.For mechanism (a), as long as T is below the optimal T for photosynthesis, GPP would increase with T and contribute to LAI increase.Since the T increase in west maritime tropical Asia was mild (<0.5°C) during the 2015/2016 El Niño (Figure S10 in Supporting Information S1) and the average T in the region was below the optimal T of 31°C (Huang et al., 2019), this mechanism could be valid to explain the green-up.Meanwhile, there is also evidence showing biomass production efficiency increases with T, suggesting potential green-up driven by T (Collalti et al., 2020), as more biomass is accumulated for leaf growth.
For east maritime tropical Asia (i.e., New Guinea), the green-up was caused by a negative anomaly in soil water.Recent studies have suggested that soil moisture has a nonlinear impact on GPP (Fu, Ciais, Prentice, et al., 2022).Above a certain threshold, soil moisture and GPP are negatively correlated, as waterlogging leads to a decrease in nutrient availability and uptake (Li et al., 2019;Parent et al., 2008).Furthermore, in wetter regions, plant functions during normal climatic years may be limited by the anoxic soil condition.Consequently, moderate droughts as observed during El Niño periods may relieve the excessive moisture stress and increase photosynthesis (Chen et al., 2022) and plant development (Costa et al., 2023).While we are unable to examine the soil moisture threshold for New Guinea specifically, we note that tropical rainforest has the lowest threshold among the PFTs (∼15%, Fu, Ciais, Feldman, et al., 2022).Additionally, New Guinea has a higher average soil moisture throughout the year even during the El Niño period in comparison to other regions in tropical Asia (Figure S10 in Supporting Information S1), thus making this negative soil moisture-GPP, and further the negative soil moisture -LAI relationship possible (Figure 3f).Meanwhile, we found a positive soil moisture-LAI relationship in continental tropical Asia.This shift in the sign of soil moisture-LAI relationship is consistent with the different drought sensitivities between continental and maritime tropical Asia noted in previous studies at the seasonal and interannual scale (Guan et al., 2015;Zhang et al., 2016).
We also note the importance of the interaction between variables in vegetation change in El Niño years.For example, while the modeled impact of high SW radiation on green-up appears to be most prominent in continental tropical Asia, some parts of this region (i.e., Thailand and Cambodia) experienced browning during 2015/16.This is most likely due to the counteracting effect of dry soil moisture (Figure 3i).Drier conditions have been reported to induce leaf shedding, which is a plant strategy to limit transpiration during drought (Janssen et al., 2020).For example, a field study in a dry deciduous forest in Thailand reported earlier leaf shedding along with unusually complete deciduousness in some dipterocarp species during El Niño (Kaewthongrach et al., 2020).Additionally, regions in tropical Asia that experienced less green-up or even browning contains large non-forested area such as croplands, where crops were not able to access water through deep root systems during prolonged dryness as effectively as trees (Huete et al., 2006;Zhu et al., 2018).The dominant role of radiation and water combined in this region specifically is supported by Uribe et al. (2021) who reported that continental Southeast Asia is energy water co-synchronous, which means its photosynthetic activity is higher in more water and light at the same time.Decreased GPP in this region in 2015 (Luo et al., 2018;Zhu et al., 2018) further supports our argument.
Additionally, we also note that at intra-annual time scale, tropical Asia did not demonstrate green-up in dry seasons like Amazon (Yang et al., 2023;Zhang et al., 2016), and the difference implied a different mechanism in the green-up in tropical Asia than in tropical America.Within the tropical Asia, the difference between continental and maritime tropical Asia (Zhang et al., 2016), and the difference between the west and east part of maritime tropical Asia (Wang et al., 2023) have been noted by other studies in terms of their sensitivities to climate.

Conclusions
In this study, we reported the strong vegetation green-up in tropical Asia during the 2015/16 El Niño event using multiple remote sensing LAI products and geostationary satellite data set.The magnitude of the green-up is stronger than that of Amazon.The widespread green-up in tropical Asia was a result of complex interactions between temperature, soil moisture, and shortwave radiation, in which the high radiation drove the green-up in continental tropical Asia, while the high temperature and low soil moisture drove the green-up in maritime Geophysical Research Letters 10.1029/2023GL106955 tropical Asia.Our finding provides important insights into the response of the vegetation in tropical Asia to El Niño, especially with projected increase of ENSO variability under warming climate (Cai et al., 2021).

Figure 1 .
Figure 1.Green-ups in the tropics (a) timeseries of yearly leaf area index (LAI) anomalies by regions from 2001 to 2018; (b) boxplot of 2015/16 monthly LAI anomalies by regions; (c-d) spatial patterns of ensemble yearly LAI anomaly from GLASS, GLOBMAP, GIMMS, and MODIS in 2015 and 2016, respectively; (e) spatial patterns of 2016 NDVI anomaly from geostationary satellite Himawari-8 for Sumatra and Borneo.

Figure 3 .
Figure 3.The spatial patterns of (a-c) sensitivity of leaf area index (LAI) to climate variables (i.e., shortwave (SW) radiation, temperature (T), and soil moisture, respectively) based on the 2001-2018 data; (d-f) yearly anomalies of climate variables during 2015/16 El-Niño; (g-i) LAI anomalies explained by anomalies of climate variables in 2015/16, calculated as sensitivity × anomaly in 2015/16.Stippling indicates pixels where the t-test for each respective climate variable is statistically significant ( p-value < 0.05).