Journal of Agricultural Meteorology
Online ISSN : 1881-0136
Print ISSN : 0021-8588
ISSN-L : 0021-8588

This article has now been updated. Please use the final version.

Impact of the 2015 El Niño event on Borneo: Detection of drought damage using solar-induced chlorophyll fluorescence
Kazutaka MURAKAMIMakoto SAITOHibiki M. NODAHaruki OSHIOYukio YOSHIDAKazuhito ICHIITsuneo MATSUNAGA
Author information
JOURNAL OPEN ACCESS FULL-TEXT HTML Advance online publication

Article ID: D-24-00012

Details
Abstract

 Remotely sensed solar-induced chlorophyll fluorescence (SIF) is applicable as an indicator of changing photosynthetic activity in terrestrial ecosystems. The vegetation of Borneo has previously been affected by drought and fire during El Niño events. Changes in satellite-based SIF data during an El Niño event in Borneo in 2015 were examined using three satellites-GOSAT, GOME-2, and OCO-2-covering the whole island and its southern and northern areas, respectively. Relationships between environmental factors and vegetation damage, precipitation, fire incidence, vegetation indices, and gross primary production (GPP), which were determined using machine-learning methods, were also examined for the period 2007–2018. SIF tended to be low in dry seasons, even in normal years, possibly because of increased drought stress and/or a higher incidence of fires with less precipitation. During the dry season of 2015, there were significant reductions in SIF in southern Borneo where fires were frequent. Other vegetation indices and GPP were also lower. Serious drought conditions with frequent fires during the El Niño event might have caused ecological degradation throughout Borneo, with a significant decrease in SIF.

1. Introduction

The island of Borneo, in the Sundaland region of SE Asia, is a biodiversity “hotspot” with a unique ecosystem rich in endemic flora and fauna (Myers et al., 2000). Species richness generally enhances the resilience of ecosystems to climate change (Chapin et al., 2000; Rockström et al., 2009). However, the peatland rainforests of Borneo have been under severe pressure over recent decades due to deforestation with rapid conversion to plantations such as oil palms (e.g., Langner et al., 2007; Gaveau et al., 2013). Changes in land use have also caused loss of biodiversity in plant and insect species owing to reductions in forest leaf litter and soil nutrients (Brühl and Eltz, 2010). In developing oil palm plantations, soil and surface water in peatlands are drained, exposing accumulated organic soil and resulting in severe fires during droughts (van der Werf et al., 2008). Precipitation in Borneo varies annually and decreases markedly during El Niño–Southern Oscillation events (Salimun et al., 2014; Tangang et al., 2017), when droughts and forest fires occur more frequently, with the impact in 2015 being the most severe of the last two decades (Sloan et al., 2017). Furthermore, the incidence of forest fires in the region is likely to increase in future with increasing drought frequency (Neelin et al., 2003). These pressures may significantly affect the health and productivity of vegetation, with degradation of all ecosystem processes and functions, including food-chain primary production, carbon and energy cycles, and climate regulation (Chapin et al., 2011), as photosynthesis is a fundamental ecosystem process. The development of systems for monitoring ecosystem integrity is now more important than ever, with increasing need to assess the impact of ecosystem degradation and to ameliorate ecosystem damage during climate change.

Remote sensing using Earth-observation satellites is a powerful tool for monitoring temporospatial variations in vegetative productivity over large scales. Satellite-based vegetation indices such as the normalized difference vegetation index (NDVI) and enhanced vegetation index (EVI) have been applied in detecting ecosystem degradation (e.g., Mitchard and Flintrop, 2013; Eckert et al., 2015) and estimating temporal changes in gross primary production (GPP) (e.g., Wang et al., 2005; Muraoka et al., 2013), with these indices indicating vegetation structure and photosynthetic capacity (Huete et al., 2002). Over the last decade there has been increasing focus on solar-induced chlorophyll fluorescence (SIF) as an indicator of photosynthetic activity. Chlorophyll fluorescence is red, and near-infrared radiation is emitted from chlorophyll during photosynthesis. This has been applied at laboratory scale in elucidating the status of photosynthetic systems and effects of photo-inhibition and other plant stresses (Schreiber et al., 1995; Baker, 2008). Although the fluorescence is very weak, spectral radiometers with high wavelength resolution enable its detection within solar radiation through the use of overlapping Fraunhofer line wavelengths (Meroni et al., 2009). Joiner et al. (2011) and Frankenberg et al. (2011) have shown the first global SIF image produced by TANSO-FTS (Thermal And Near-infrared Sensor for carbon Observation-Fourier Transform Spectrometer), the main sensor aboard the Greenhouse gases Observing SATellite (GOSAT; Yokota et al., 2009). Over recent years, SIF has been monitored by multiple satellite sensors including the Global Ozone Monitoring Experiment-2 (GOME-2) aboard the Meteorological Operational Satellite Program of Europe (MetOp; Joiner et al., 2013) and Orbiting Carbon Observatory-2 (OCO-2; Sun et al., 2017) satellites. These satellite-based SIF systems have aided the elucidation of phenological changes in GPP (Joiner et al., 2014) and its spatial distribution on global scales (Frankenberg et al., 2011; Liu et al., 2017). Even on local scales, Lee et al. (2013) have shown that GOSAT SIF data have potential for use in monitoring ecosystem damage by extreme events, as in drought-stress reductions in photosynthesis in Amazonia.

However, satellite sensors currently used for satellite SIF retrieval were designed for atmospheric studies rather than vegetation monitoring, and their spatial observation patterns are not suitable for vegetation monitoring on local scales (e.g., Sun et al., 2017). GOSAT has a sparse spatial data sampling system, GOME-2 has a large footprint, and OCO-2 has higher spatial resolution but covers only a limited area of the globe owing to its narrow observation swath (10.6 km at nadir; Table 1). These characteristics of satellite observation systems complicate the application of their SIF observations in assessing vegetation variability caused by environmental stresses on local scales. Persistent cloud cover is also a major obstacle in satellite optical remote sensing, especially over the tropics (Pour et al., 2013), and relatively few SIF data are available. There is also uncertainty about the suitability of current satellite SIF data in tracking vegetation stress on local scales, with more information being required concerning the dynamic response of satellite SIF observations to environmental stresses. Such observational difficulties mean that vegetation monitoring with satellite SIF data remains challenging in certain regions. Further evaluation of satellite SIF observations on local scale would be useful to test the applicability of the satellite-based SIF systems for monitoring ecosystem degradation by climate change.

Table 1. Characteristics of the three satellites used in retrieving SIF.

Some studies showed that the applicability of satellite SIF observations to detect abrupt changes in photosynthetic activities on local scale due to drought (e.g., Qian et al., 2019) but did not assess the changes due to fires. In recent years, the impacts of drought on fire risk have been increasingly severe on Borneo, with changing precipitation patterns due to climate change. An increase in fire intensity and/or fire frequency can lead to further degradation of the terrestrial ecosystem. If the satellite SIF observations could be used to assess the impact of fires on ecosystem damage, such observations could be particularly useful system for vegetation monitoring. This study aims to explore the application of GOSAT, GOME-2, and OCO-2 SIF data in monitoring vegetation changes in Borneo due to fires during the 2015 severe El Niño year. The applicability of satellite SIF data in monitoring ecosystem damage on local scales was investigated, and the consistency of SIF observations and other vegetation indices (NDVI and EVI) was evaluated.

2. Materials and methods

2.1 Study area

Borneo Island, at 8°N-5°S by 108°E-120°E, is the second-largest island in Asia with an area of ~743,000 km2. The whole island was previously vegetated with tropical evergreen forest dominated by Dipterocarpaceae species, but has been under severe pressure of deforestation and degradation due to land-use changes since the 1950s (Langner et al., 2007; Fuller et al., 2004), with lowland and mountain rainforest and peat swamp forest being converted to industrial-scale plantations (Miettinen et al., 2016). Borneo has a tropical maritime climate with monsoon rains arriving from the Pacific and Indian oceans. The mean annual precipitation for 1968-1998 at Miri Airport in northern Borneo was 2,700 mm with little seasonal variation (Sakai et al., 1999; Kumagai et al., 2004) at Palangkaraya in southern Borneo, it was 2,800 mm for 1978-2007 with a June-October dry season (Putra and Hayasaka, 2011). Significant periodic reductions in precipitation have occurred during El Niño-Southern Oscillation climatic events, with recent El Niño events being recorded during 2009-2010 and 2015-2016 (Chen et al., 2016).

The fire season in southern Borneo includes July-November, peaking in September-October (Fanin and van der Werf, 2017). During 2015, forest fires occurred mainly in peatland areas in central and southern Borneo due to a lack of rainfall during July-October, induced by the El Niño event (Miettinen et al., 2017). Here we consider the northern and southern areas of Borneo, which are separated by the Equator.

2.2 Land temperature, precipitation, and fire frequency

Monthly mean near-surface temperatures from the Climate Research Unit gridded Time Series (CRU TS) v. 4.05 dataset (Harris et al., 2020) were used to investigate the variability of temperature in Borneo. The CRU TS is a global 0.5°×0.5° grid dataset spatially interpolated with ground-based observation data using statistical methods.

Global Precipitation Climatology Project (GPCP) v. 1.3 data (Huffman et al., 2001) were used to assess intra- and inter-annual variations in precipitation. GPCP precipitation data are a composite of data compiled from satellite and ground-based observations with daily and 1° grid resolution. Here, the data were converted to monthly total precipitation.

The Thermal Anomalies and Fire 8-Day (MOD14A2) Version 6 product (Giglio and Justice, 2015) of the Terra/Moderate Resolution Imaging Spectroradiometer (MODIS) were used to determine forest fire frequency in the targeted area. The spatial resolution of MOD14A2 is 1 km×1 km with eight-day temporal resolution. To count monthly fire events in target pixels, all data at three levels of confidence-high, nominal, and low-were used.

2.3 Satellite-based SIF

Details of the satellite observation systems are provided in Table 1. Available data periods differed among the satellites, but all included the 2015 El Niño year.

GOSAT was launched in January 2009 to monitor temporospatial variations in atmospheric CO2 and CH4 concentrations with global coverage. GOSAT SIF was retrieved from spectral radiance data (Level 1B product version V201.202) in the short-wavelength infrared region, acquired by TANSO-FTS, the main GOSAT sensor. To retrieve SIF, a spectral fitting method was applied to a 756.0-759.1 nm window of Fraunhofer lines. Radiance spectral data involve a zero-level offset, which is a spectrally constant additive signal caused by non-linearity in the analogue circuit and analogue-to-digital converter (Kuze et al., 2012); offset-corrected GOSAT SIF data were used as described by Oshio et al. (2019). GOME-2 is an instrument aboard Meteorological Operational Satellite Program of Europe-A and B (MetOp-A and MetOp-B) satellites, designed for the measurement of total atmospheric O3 content and vertical O3 profiles. We used Level 2 SIF data (version 27); i.e., daily data of MetOp-A GOME-2 (launched October 2006). Details of SIF retrieval methods of GOME-2 are provided by Joiner et al. (2013). OCO-2 was designed to observe atmospheric CO2 concentrations and was launched by NASA in July 2014. We used the OCO-2 SIF product, Level 2 Lite SIF (V8r), with details of SIF retrieval as described by Frankenberg et al. (2014).

Cloud screening was performed on SIF data to reduce contamination by cloud scattering and absorption. For GOSAT SIF, only data with ≤30% cloud pixel in the sensor field of view of TANSO-FTS were used. Cloud pixel information was obtained from the CAI L2 cloud-flag product of the GOSAT Thermal And Near infrared Sensor for carbon Observation-Cloud and Aerosol Imager (TANSO-CAI). We used data with QC flags of “good” and “good and passed cloud check” with GOME-2, and with cloud flags of “classified clear” with OCO-2. As outliers remained after cloud screening due to signal noise in satellite sensors, SIF data beyond the range of mean ± 3σ over the whole of Borneo every month were removed. Correction for incident light intensity was then applied by the method of Joiner et al. (2011):

SIFcorrected = SIFapparent / cos (SZA)               (1)

where SIFcorrected is corrected SIF, SIFapparent is apparent SIF (i.e., SIF values provided as the satellite product), and SZA is solar zenith angle. Here, mean monthly aggregated SIF values (for the whole island, including northern and southern areas) were used to detect seasonal variability.

2.4 Biome types

Biome types of Borneo were provided by the MODIS/Terra + Aqua Land Cover Type dataset (MCD12Q1 V006) at 500 m spatial resolution (Friedl and Sulla-Menashe, 2019). The MCD12Q1 dataset was applied with the annual Leaf Area Index (LAI) classification. Two biome types were used: forest and savanna (Fig. 1). Canopy structure was based on the MCD12Q1 LAI/fPAR Biome scheme described by Myneni et al. (2002), with trees of >2 m height covering >60% of forest areas and 10%-60% of savanna areas.

Fig. 1. Land-cover map of Borneo. Dark green, forest; light green, is savanna; black, other classifications.

2.5 Satellite-based vegetation indices

Temporal variations in satellite SIF data were compared with temporal changes in the NDVI and EVI indices of MODIS/Terra, based on v. 6 MOD13A2 (Didan, 2015) indices of 16 d km-1 resolution. “Good quality” data on land, with “no clouds”, “no shadows”, and “no ice/snow” were selected according to quality-assurance flags of the product, and monthly average values were calculated.

2.6 Gross Primary Production

The GPP is a crucial factor in investigating and quantifying carbon fixation by vegetation at various scales (Field et al., 1995). We examined relationships between GPP and SIF to evaluate the relationship between environmental conditions during El Niño events and ecosystem function in Borneo. We used a GPP dataset with data-driven estimation using a machine-learning algorithm and a support vector regression technique (Ichii et al., 2017). The original data were monthly GPP values based on eddy covariance measurements at 54 observation sites with satellite observations of land surface conditions over Asia, with a 0.25°×0.25° grid.

3. Results

There was no apparent difference between the 2007-2018 (excluding 2015) and 2015 mean near-surface temperatures of the dry (June-October) and wet (November-May) seasons in Borneo, although the monthly mean temperature in southern Borneo was 1.0°C-1.3°C higher than that in the northern area (Table 2). In 2015, the overall mean dry-season temperature was 0.3°C-0.5°C above average, and precipitation was 34% and 51% (82.4 and 98.9 mm month-1) below average in northern and southern Borneo, respectively. Averaged over normal years, the difference in precipitation between wet and dry seasons was greater in southern Borneo (150.7 mm month-1) than northern Borneo (61.5 mm month-1), increasing to 258.0 and 127.2 mm month-1, respectively, in 2015. The El Niño event of 2015 thus led to severe drought conditions in southern Borneo, with higher temperatures.

Table 2. Mean near-surface temperature (°C) and precipitation (mm month-1) in Borneo, and its northern and southern areas in 2015 and in 2007–2018. The dry season is June–October, and the wet season covers the remaining months.

Average annual fire frequencies for 2007-2018, from MOD14A2 data, are mapped in Fig. 2. In both the yearly average plot and that for 2015 alone, fires occurred mainly in coastal and southern Borneo. In the yearly average plot, there were <500 fire pixel grid-1 yr-1 in a 0.5° grid, with counts being notably higher in 2015, especially in southern Borneo with a maximum of 2,233 counts. Fire counts were below 500 pixel grid-1 yr-1 in northern Borneo, even in 2015. Total fire counts in the whole of Borneo and the southern area alone were 43,253 and 37,000 pixel grid-1 yr-1 in 2015, compared with the 2007-2018 averages of 9,675 and 6,100 pixel grid-1 yr-1, respectively.

Fig. 2. The 0.5°×0.5° grid fire counts in Borneo during 2007–2018, excluding 2015 (a); and in 2015, the El Nino year (b).

Monthly average SIF values for the whole of Borneo and its northern and southern areas are plotted in Fig. 3, based on observations by the three satellites. Although GOSAT SIF fluctuated widely by year, the normal-year averages reflected seasonal patterns, decreasing in September (dry season) in all regions and increasing in October-November. The wide fluctuations with high standard deviations might be caused by the small number of observation points in the targeted area, due to the sensor type and sampling pattern; the number of monthly GOSAT SIF sampling points (91.6 counts per month for the entire island) was about 1/30 that of GOME-2 and OCO-2 (3,314 and 3,327 counts per month, respectively). Seasonal variations in normal-year averages for GOME-2 were smaller than those of GOSAT, with values in the southern area also decreasing in September and increasing during October-November. Overall, OCO-2 SIF values were ~0.20 mW m-2 sr-1 nm-1 lower than those of GOSAT and GOME-2. Satellite SIF values change with conditions of observed spectral range, observational time, angle between viewing and sun directions, and land surface conditions (Oshio et al., 2019). Normal-year average values of OCO-2 SIF were relatively constant throughout the year but with a high standard deviation in October. In 2015, the El Niño year, monthly average SIF values decreased in October (late dry season) in southern Borneo, with values being lower than normal-year averages for all satellites. In southern Borneo, normal-year averages for October were 1.32, 1.18, and 1.07 mW m-2 sr-1 nm-1 for GOSAT, GOME-2, and OCO-2, whereas those in 2015 were 0.78, 0.85, and 0.70 mW m-2 sr-1 nm-1, respectively. The SIF decrease in October 2015 was statistically significant (p < 0.05), based on comparison of normal-year averages for the three satellites.

Fig. 3. Monthly average SIF observed by the three satellites for the whole, northern, and southern Borneo regions. Black lines and squares indicate average values for normal years; shading indicates standard deviations for normal years; gray lines indicate values of each normal year; colored lines and squares (blue for GOSAT, orange for GOME-2, and green for OCO-2) indicate values (with standard error) of means in 2015.

Relationships between satellite SIF, precipitation, and forest fire count are depicted in Fig. 4 with overall trends indicating that fire frequency increased with decreasing precipitation while SIF decreased with decreasing precipitation and increasing fire frequency. Relatively high SIF values in southern Borneo (>1.55 for GOSAT, 1.35 for GOME-2, and 1.20 mW m-2 sr-1 nm-1 for OCO-2) were observed mainly with precipitation of >100 mm per month and fire counts of ~300 per month, while GOSAT SIF data show more scatter. The SIF values start to decrease with fire count over 500-1000 per month for all satellite observations. Precipitation was heavier in northern Borneo and fire counts lower; monthly precipitation was rarely below 50 mm per month and the fire count rarely exceeded 1,000 per month, compared with often being below 50 mm and above 1,000 per month, respectively, in southern Borneo. These combined condition of precipitation and fire in northern Borneo leads to relatively little to no degradation in SIF. In 2015, severe drought with precipitation of <50 mm per month and frequent fires with a monthly count of about 10,000 extended for four months in southern Borneo. In those four months, SIF was very low at 0.78-1.34 mW m-2 sr-1 nm-1 for GOSAT, 0.84-1.15 mW m-2 sr-1 nm-1 for GOME-2, and 0.70-1.06 mW m-2 sr-1 nm-1 for OCO-2. In November 2015, when precipitation was 145 mm, fires occurred frequently (fire count 1,000) and SIF remained low for all satellite observations.

Fig. 4. Relationships among monthly average SIF values (mW m-2 sr-1 nm-1) observed by GOSAT, GOME-2, and OCO-2, fire counts, and precipitation (mm month-1) in the whole, northern, and southern Borneo regions. The color of each symbol indicates monthly average precipitation in the targeted area; large symbols with black edges indicate values in 2015. The blank squares and error bars plot the bin means of SIF and their standard deviations against fire count classes (0-5, 5-10, 10-50, 50-100, 100-500, 500-1000, 1000-5000, 5000-10000, 10000-).

4. Discussion

4.1 Impact of fire events on SIF variability

Previous studies have shown that SIF is an excellent indicator of the effects of drought on terrestrial vegetation with Lee et al. (2013), for example, demonstrating that GOSAT SIF clearly decreased in the Amazonian forest during seasons of low precipitation. Qian et al. (2019) examined temporal variations in Borneo using GOME-2 SIF data and found that SIF values decreased in drought conditions in El Niño years, with SIF being sensitive to water stress. There is debate concerning the choice of satellite SIF data in evaluating the response of SIF signals to climatic events. We found that SIF values and temporal patterns varied among the three satellites, even in monthly averages (Fig. 3). This was likely due to sampling patterns and/or temporospatial resolution varying among the satellites, as shown in Fig. 5. GOME-2 observations covered almost all of Borneo, allowing the capture of SIF variability over the whole area, but with spatial resolution being relatively coarse in detecting vegetation responses to disturbances on fine local scales as on oil-palm estates. In contrast, the fields of view of GOSAT and OCO-2 had fine spatial resolutions for local disturbances but with limited spatial coverage, especially for GOSAT, which had small numbers of sparse observations (82 observations in October 2015). Such differences in satellite characteristics resulted in wide variability in regional monthly mean SIF values. However, SIF variability in southern Borneo, where fires occurred frequently, was captured by all satellites, even in months with severe fires during El Niño 2015 (Figs 2 and 4). The remarkable reductions in SIF in 2015 among the three satellites may result from significant disturbance to vegetation by severe fire events, rather than drought alone, with a large reduction in GPP.

Fig. 5. Satellite observation patterns of the three satellites over Borneo in October 2015. Red circles, light-blue rectangles, and orange dots indicate fields of view of GOSAT, GOME-2, and OCO-2, respectively.

The reduction in SIF during drought conditions coincides with reductions in photosynthesis rate on leaf (Flexas et al., 2002) and canopy (Liu et al., 2017) scales. Therefore, we speculate that the remarkable SIF reduction observed in southern Borneo during El Niño 2015 may indicate degradation of ecosystem functions, such as a decline in photosynthetic activity and/or vegetation density. The dynamics of ecosystem function in southern Borneo in response to El Niño 2015 were considered in terms of comparison of retrieved satellite-based SIF data with a GPP dataset derived from a data-driven estimation independent of satellite SIF observations (Fig. 6a). Seasonal variations of GPP estimates averaged over southern Borneo in 2015 were roughly consistent with those of SIF (correlation coefficient, r = 0.49, 0.63, and 0.56 for GOSAT, GOME-2, and OCO-2, respectively). The 2015 mean September-November GPP over southern Borneo was 6.86 g C m-2 d-1, or 0.36 g C m-2 d-1 (4.9%) below the normal-year average. This reduction in GPP in the late dry season of 2015 was significantly different (p < 0.05) from that of the normal-year average. This qualitatively distinguishes the impacts of drought stress and forest-fire damage on vegetation, whereas the auxiliary GPP dataset reflects the degradation of ecosystem functions during the severe dry season of 2015.

Fig. 6. a Monthly average modeled GPP, b NDVI, and c EVI in southern Borneo. Black lines and squares indicate average values for normal years (2007–2018, excluding 2015); shading indicates standard deviation for normal years; gray lines indicate values of each normal year; color lines and squares indicate values for 2015.

The large reduction of SIF during El Niño 2015 was considered in terms of the impact of fire events on SIF variability. The 0.5° grids over the whole of Borneo in 2015 were classified into three groups: “no fire” (n = 96); “low” for grids with fire fractions of 1-99 pixels per grid (n = 112); and “high” with fire fractions of ≥100 pixels per grid (n = 31). Box-and-whisker plots describing SIF variability for the period during the severe dry season of 2015 (July-November) are shown in Fig. 7 for the three satellites. Unrealistic negative values appeared in SIF observations owing to the scatter in satellite retrievals, but all data were used in the following analyses without arbitrary screening to remove the negative values to preserve the original probability distributions of the SIF observations. There were clear reductions in SIF variability with increasing fire frequency. For example, median values of SIF variability for “no fire”, “low”, and “high” were 1.41, 1.35, and 0.95 mW m-2 sr-1 nm-1, respectively, for GOSAT; 1.22, 1.17, and 0.89 mW m-2 sr-1 nm-1, respectively, for GOME-2; and 1.12, 1.08, and 0.83 mW m-2 sr-1 nm-1, respectively, for OCO-2. These values suggest that fire is a key environmental variable limiting the variability of SIF, with potential to suppress vegetation function in Borneo.

Fig. 7. Distributions of SIF for July–November 2015 for three satellite observations. “No fire”, “low”, and “high” fire classifications are shown.

Fire intensity and fire frequency vary with biome type. Forest densely covered with thick-barked trees with little understory is less flammable, whereas savanna and grassland of relatively low biomass density and thick understory are more flammable (e.g., Siegert et al., 2001). SIF observations acquired in October 2015 over grids of predominantly forest and savanna areas over the whole of Borneo were studied to evaluate the ability of satellite SIF observations in distinguishing dynamics of SIF response to various fire conditions (Fig. 8). To evaluate the fire conditions, burned area fractions at the 0.5° grids were compared across the entire target grids. GOME-2 observations alone were used because of their entire Borneo coverage (Fig. 5). Dominant biomes within the fields of view of individual GOME-2 observations were classified according to MCD12Q1 land cover data. The higher the fire fractions, the lower the number of SIF observations. More SIF data were recorded during fires in savanna than in forests. The SIF values of both forests and savanna decreased with increasing fire fraction, with values in savanna being more sensitive to severe fires: 0.58, 0.42, and 0.08 mW m-2 sr-1 nm-1 for fire fractions of the 0.5° grids of 0-0.1, 0.1-0.2, and >0.2 in savanna; and 0.99, 0.53, and 0.45 mW m-2 sr-1 nm-1 in forests, respectively (Fig. 8). This is consistent with the general understanding that savanna is exposed to more flammable conditions than forests and is more susceptible to fires, although severe fires (fire fraction >0.2) rarely occur in savanna, even during the El Niño in 2015. It was difficult to isolate the impact of fire disturbances from drought stress in SIF observations, but the SIF observations under drought and fire conditions in October 2015 led us to speculate that satellite SIF observations may provide reliable data for tracking vegetation stress and damage on local scales.

Fig. 8. Relationship of GOME-2 SIF and fire fraction in October 2015 for forest (blue) and savanna (orange). Solid lines and shadows indicate mean and standard deviation of SIF values for each fire fraction. Histograms in an embedded small figure indicate number of fire counts over forest (navy) and savanna (green).

4.2 Comparison with other vegetation indices

Different types of satellite indicator are available for quantifying and understanding vegetation dynamics. The NDVI is calculated from near-infrared and visible light reflection by vegetation and is intended simply to detect living vegetation and to distinguish it from other materials such as dead vegetation, but it is less sensitive in areas of dense vegetation. The EVI is a similar indicator but uses additional wavelengths and incorporates additional information on atmospheric conditions and canopy background noise, leading to greater sensitivity to dense vegetation and less interference from atmospheric aerosols (Xiao et al., 2003). In Borneo, there is a substantial reduction in NDVI due to the high frequency of forest fires (Idris et al., 2005). Here, both NDVI and EVI exhibited seasonal variations similar to those of satellite SIF observations for southern Borneo in 2015, with a sudden reduction in September and October and recovery in November relative to the yearly average (Fig. 6b and c). This implies that significant degradation of vegetation occurred in southern Borneo during the severe drought of 2015. However, the effects of increased fires during drought conditions on vegetation dynamics and satellite observations warrants further study, particularly concerning differences in the characteristics of SIF and vegetation indices.

5. Conclusions

SIF has been used as an indicator of photosynthetic activity. We retrieved satellite-based SIF data from three satellites-GOSAT, GOME-2, and OCO-2-and examined variability in SIF data in response to the El Niño event of 2015, which is the largest event to date in the 21st century. El Niño 2015 caused large reductions in precipitation and increased forest-fire frequency in the late dry season in southern Borneo. Significant decreases in SIF were observed over the Borneo domain in all satellite data during the late dry season relative to those in normal years. When SIF data displayed prompt reductions in the late dry season of 2015, other vegetative indicators of GPP, NDVI, and EVI also began to decrease substantially, before recovering at the beginning of the rainy season. SIF fluctuations varied with fire fractions and biome type. These results suggest that ecosystem functions in southern Borneo were suppressed by drought stress and/or reduction in vegetation density caused by fire events associated with the El Niño event. The results of this study demonstrate that retrieved satellite-based SIF data with different characteristics can be used as indicators of ecosystem function, together with vegetation indices, even on local and regional scales. Further work should distinguish between the impacts of drought stress and forest-fire damage on vegetation to elucidate more clearly the impact of El Niño 2015 on the vegetation function in Borneo.

Acknowledgements

GOSAT SIF data were processed by the Research Computation Facility for GOSAT-2 (RCF2).

References
 
© Author (s)

This article is licensed under a Creative Commons [Attribution 4.0 International] license.
https://creativecommons.org/licenses/by/4.0/
feedback
Top