Ground-based measurements of column-averaged carbon dioxide molar mixing ratios in a peatland fire-prone area of Central Kalimantan, Indonesia

Tropical peatlands in Indonesia have been disturbed over decades and are a source of carbon dioxide (CO2) into the atmosphere by peat respiration and peatland fire. With a portable solar spectrometer, we have performed measurements of column-averaged CO2 dry-air molar mixing ratios, XCO2, in Palangka Raya, Indonesia, and quantify the emission dynamics of the peatland with use of the data for weather, fire hotspot, ground water table, local airport operation visibility and weather radar images. Total emission of CO2 from surface and underground peat fires as well as from peatland ecosystem is evaluated by day-to-day variability of XCO2. We found that the peatland fire and the net ecosystem CO2 exchange contributed with the same order of magnitude to the CO2 emission during the non-El Niño Southern Oscillation year of July 2014-August 2015.

In this article, we report monitoring of column-averaged dry-air molar mixing ratios, XCO 2, in an Indonesian fire-prone area over the dry and wet seasons in Palangka Raya, Central Kalimantan, using a portable solar photometer for field measurements. The location of our observation station is shown in Fig. 1. We used a Fiber-Etalon Sun-photometer for column Carbon dioxide (FES-C), which is attached with a fiber Fabry-Perot optical device 21 . It probes larger sample volumes than in situ and smaller scales than sensors onboard satellites, and is suitable for measurement of CO 2 emission from local peatlands.
We discuss variabilities of XCO 2 with use of weather data and peatland-specific parameters including airport visibility records, hotspot counts, GWL data and weather radar images. The MODerate resolution Imaging Spectroradiometer (MODIS) hotspot data 22 are used to estimate the contribution of surface and underground fires to XCO 2 , from which CO 2 and particulate matter are produced 23 . Peatland fires characterized by low intensity burning spread slowly into peat deposits below the surface for long periods of time, contributing to emissions.

Results
Background levels of XCO 2 . A time series of XCO 2 data at Palangka Raya is shown in Fig. 2A along with environmental parameters. Increments in XCO 2 above the background levels, ΔXCO 2 , are due to peatland fires and the net ecosystem CO 2 exchange. Data missing in XCO 2 from mid-September to October 2014 was due to thick haze and dense smoke coverage that weakened the solar spectral intensity.
In estimation of the XCO 2 background levels, we use the annual trend obtained by fitting the GOSAT data on sea between the Kalimantan and Java Islands between 21 st July, 2009-30 th May, 2015 as described Supplementary S1. The GOSAT annual trend is 2.00 ± 0.10 (1σ) ppm/y. Hereafter, the uncertainty represents one standard deviation. The mean background level of FES-C is 395.8 ppm between 1 st December, 2014-6 th January, 2015 and 19 th May-2 nd July, 2015 in non-fire wet seasons that are characterized by high GWL, low temperature and high visibility. At Bukit Kototabang, Sumatra, Indonesia, a meteorological station (0.165°S, 100.295°E) has been performing surface measurement of CO 2 concentrations that irregularly varied with ±5 ppm due to peat fire and respiration, the mean value of which is 393.4 ppm for the non-fire season between July -November 2014. Combination of the FES-C non-fire wet season data with the annual trend coefficient of 2.00 ppm/y gives the best-fit background level in units of ppm: ± . + . ± . × . Background n n ( ) (394 3 1 2) (2 00 0 10) where n is day after 1 st July, 2014.  Increments in XCO 2 by peat respiration. The CO 2 emission from peat respiration or aerobic peat oxidation has a strong correlation with lowering GWL 7 . As marked by the two blue double arrows in Fig Increments in XCO 2 by peatland fires. Figure 2 shows that the fire season in the beginning-to-middle of September 2014 is characterized by high ΔXCO 2, large hotspot counts, low visibility, low GWL and low precipitation. ΔXCO 2 increased up to 10.0 ppm (mean 6.0 ppm). The increments are caused mostly by peatland fires because the surface peat that is fresh and less decomposed is destroyed by fire and the peat respiration activities are reduced 7 .

Discussion
Increments in XCO 2 corrected for airmass transport. Our tethered balloon observation for vertical distribution of the CO 2 concentrations as described in Supplementary S2 suggests that CO 2 emitted from the ground into the atmosphere is homogeneously mixed at altitude up to ~700 m and is dispersed by prevailing south-easterly to southerly wind with an average wind reach distances of 160 km/day. (See Supplementary S3) Palangka Raya is located 136 km away from the southerly coastal area and surrounded by lowland peat swamp forest ecosystems 25 . Wind flow that travels over those surroundings reaches daily over Palangka Raya. Using the land-use map and the results of 24-hr backward trajectory calculations, we obtain fractional coverages, P daily , of the trajectories at 1500 m height travelling over the peatland and lowland from sea area as shown in Supplementary S3. If we assume that the whole peatland area in the southern vicinity of Palangka Raya is homogeneously contaminated by emitted CO 2 , then, the daily emission from peatland may be obtained by ΔXCO 2 /P daily where P daily is the ratio of trajectory distance over peatland /whole trajectory distance. Then, the total molar flux of emission per unit peatland area, F, may be obtained from the sum of daily ΔXCO 2 /P daily during the peat respiration periods of total 122 days in 2014-2015, daily air 2 where Sum = 715 ± 33 ppm and the molar column density of air, σ air = 3.5 × 10 5 mol/m 2 . The total uncertainty is due to the instrumental and background uncertainties in 122 days as well as P daily uncertainty. The uncertainty in P daily is estimated to be 25% as described in Supplementary S3. Total carbon emission by net ecosystem exchange. The total carbon emission is F × m C = 3,000 ± 140 gC/m 2 for two non-fire periods of Fig. 2A in July 2014-May 2015, where m C is the molecular weight of carbon (12 g/mol). The yearly net ecosystem CO 2 exchange controlled chiefly by oxidative peat decomposition, NEE, is 1500 ± 70 gC/(m 2 y) averaged for two non-fire seasons in 2014-2015. NEE is the difference between the ecosystem respiration and the gross primary production due to photosynthesis. Since this NEE excludes the contributions of emission from lowland and other sources during wind travelling, it should be an upper limit value.
When the contribution of CO 2 emission from the lowland area is included, P daily is defined as the ratio of peatland + lowland trajectory distance/whole trajectory distance: Sum = 475 ± 29 ppm and F × m C = 1,990 ± 120 gC/m 2 , then, NEE is 1,000 ± 60 gC/(m 2 y) averaged for 2014-2015. Hirano et al. 7  There is a large discrepancy between our estimated NEE and Hirano's reported values. This is partly due to difference in hydrological environments. The average temperature during our observation period in Fig. 2D is higher by ~1.5 °C than during Hirano's period 7 . According to Luo and Zhou 26 , there are some factors controlling the soil respiration, e.g., temperature, soil moisture and soil oxygen. In wetlands, oxygen is usually the limiting substrate for respiration, soil drying due to increased temperature could stimulate soil respiration. The discrepancy would also be due to the fact that our measurement is only for one particular observation area during a particular term.
Correlation between fire-origin XCO 2 and airport operation visibility records to evaluate missing data due to thick haze that weaken the solar spectral intensity. During the dry season, the area is covered by haze when peatland fires occur, reducing airport visibility. Kusumaningtyas et al. 27 attributed high values of scattering albedo or low visibility to surface and smoldering combustion in the peatland. Using the fire-origin ΔXCO 2 for 35 days of FES-C measurement, a linear relationship between ΔXCO 2 and visibility is empirically obtained with a correlation coefficient R = − 0.79 in Fig. 3: where L vis is the average visibility in units of km at UTC = 20−0 h as described in Supplementary S4. Several XCO 2 data in September−November are not available because of weak solar intensity due to thick haze conditions, which can be estimated by Eq. (3). Thus obtained XCO 2 are plotted by the yellow diamonds in Fig. 2A. Using both the measured and calculated values, we estimated the mean <ΔXCO 2 > to be 7.8 ppm during the fire season from 3 rd September, 2014 to 11 th November, 2014, total 70 days.
Haze area size of air contaminated with emitted species. We use the data of Fig. 2 in evaluating peatland dynamics caused by fire in combination with the data from satellite observation. Comparing Fig. 2B,C, one notices that visibility is low about two more weeks after surface fire ceases. Aerosol optical depth (AOD) images from the Terra-MODIS AOD data base 28 show dense aerosol plumes over Palangka Raya with a size of 30-100 km, an example of which is shown in Supplementary S5. To estimate quantitatively the sizes of hazy areas over Palangka Raya, we inspect our radar images measured with a C-band Doppler weather radar at the station during the fire period between 15 th September -11 th October, 2014. During the period, precipitation is below 10 mm/month and the AODs reported by the AERONET are over three 29 , suggesting that the radar images are due to fire-origin aerosols. A typical haze area image over Palangka Raya is shown in Fig. 4(left). The average radius of 48 images is 44 ± 14 km with height below 1.6 km. We assume that the air is homogeneously mixed and flows from the fire clusters all around the vicinity of Palangka Raya in Fig. 1 Estimation of active duration of underground fires from correlation between hotspot count and increment in XCO 2 . Burning of peat soil occurs under smoldering condition characterized by low heat and long duration that can last up to four months 3,9,31,32 . Hence, it is essential to estimate an active duration of the underground fire. For quantitative analysis, the hotspot data between September−November 2014 are chosen. Supplementary S6 shows that the hotspot count density over peatland is evenly distributed where hotspot counts 506−4186 are selected to analyze with high confidence. Since prevailing southerly wind travels with an average reach distance of 160 km/day during the fire season, daily ΔXCO 2 is corrected by the daily trajectory fraction P daily over peatland. Below we evaluate correlation coefficients, R, between ΔXCO 2 /P daily and hotspot count accumulated for active durations of underground fires as described in Supplementary S6. In this data analysis, the contribution of a hotspot to ΔXCO 2 /P daily is distance-normalized by 1/r 2 where r is the distance between the hotspot and the observation station. To simplify the analysis, we assume the following conditions: 1) surface fire with high temperature is detected by the MODIS satellite sensor and lasts for one day as a hotspot, 2) underground fire lasts longer but undetectable by the MODIS sensor as described in Method, and 3) hotspots emit the same amount of CO 2 until emission ceases.
The results of correlation analysis are shown in Fig. 4(right), in which R depend on the active underground fire duration of 1-30 days (including one day of surface fire) but does not on the observing fire field ranges of r = 30-100 km radii. The correlation coefficients are independent of r because hotspots distribute evenly over the peatland area. We found R ~ 0.5 with an active fire duration centered ~17 days (including one day of surface fire) for all r. Within the range of one-standard deviations shown in Fig. 4, the active fire duration is estimated to be in a range of 10-25 day centered around 17 days.
The fire image sensor on board the TET-1 satellite has detection capabilities in monitoring low-intensity peatland fire fronts through smoke. Fire front edges observed in the TET-1 images may be indicative that the fire is slowly spreading through the deeper peat layer for two weeks in the peatland of Central Kalimantan 33 .
Total emission of CO 2 per MODIS hotspot. During the fire season, Sum(ΔXCO 2 /P daily ) is 886 ± 33 ppm.
In this calculation, we include ΔXCO 2 values estimated from Eq. (3) for the thick haze period. The uncertainty includes uncertainties of instrumental and estimation of ΔXCO 2 in Eq. (3) as well as that of P daily . We obtain the conversion factor CF for emission per hotspot by;  where S(r) = πr 2 is the observing fire area, and N(r) is the total hotspot count during the fire season. According to the haze images, we estimate the active observation area for XCO 2 to be r = 44 ± 14 km. In the range of 30-60 km, the hotspot density, N(r)/S(r), is almost constant 0.16 ± 0.02 count/km 2 , due to homogeneous distribution of the hotspots over the peatland. Thus calculated conversion factor during the fire season of 2014 in Palangka Raya is CF = (8.5 ± 1.4) × 10 7 kgCO 2 /hotspot.

Summary and Conclusion
We have demonstrated the approach of using solar absorption spectra recorded with a small Fabry-Perot solar photometer at a site in a source region for an estimation of the CO 2 source strength of the surrounding fires. This methodology is applied to observation of the variability of the column averaged dry-air molar mixing ratios of CO 2 in Palangka Raya, Central Kalimantan, Indonesia in July 2014−August 2015, an non-ENSO year. Our methodology is not a general one but a specific one to peatland field measurements. Assuming the conditions described in the text, the CO 2 data are analyzed in combination with the peatland-specific parameters such as hotspot counts, water levels, airport visibility records and radar images as well as weather data. We found that the increments are due to the underground smouldering fires in the fire season and the aerobic peat oxidation in the beginning of the dry season. Both sources contribute with the same order of magnitude to the enhancement in the atmospheric CO 2 concentration.

Methods
Location of the observation station and its surroundings. Monitoring of XCO 2 was performed at the Tjilik Riwut Palangka Raya BMKG Station, Palangka Raya, Central Kalimantan, Indonesia (2.224°S, 113.946°E, 10 m a.s.l.) from June 2014 to August 2015 with a portable Fiber-Etalon Sun-photometer for CO 2 , instrumental details of which are described in Supplementary S7. Visibility records and weather data were provided by the BMKG station. The station for measurement of precipitation and ground water level is located at 2.31°S and 114.06°E as shown in Fig. 1 24 . The C-band Doppler weather radar (Gematronik/Selex SI, Meteor 600 C, single / linear horizontal polarization) was located at the Palangka Raya BMKG station.
Satellite product 1: hotspot counts. We use daily active fire detection data derived from the MODIS observation as a peatland fire index, extracting the data collection 6 from the Fire Information for Resources Management System 22 . The MODIS active fire product includes the hotspot counts with high-confidence detection in a range larger than 60%. In MODIS active fire algorithm, temperature thresholds of detection potential fire pixel is 300-314 K 34 . Although the temperature of underground fire reaches above them, the spatial resolution and thick haze hamper the ability of the MODIS sensor to detect some of active smouldering fires.
Satellite product 2: carbon dioxide mixing ratios of GOSAT. The GOSAT satellite data set is a product of measurements on the Java Sea between the Kalimantan and Java Islands as shown in Fig. 1 Data Availability. Datasets during the current study are available in http://www.tonokura-lab.k.u-tokyo. ac.jp/pub/SR.html.