Fingerprint of rice paddies in spatial–temporal dynamics of atmospheric methane concentration in monsoon Asia

Agriculture (e.g., rice paddies) has been considered one of the main emission sources responsible for the sudden rise of atmospheric methane concentration (XCH4) since 2007, but remains debated. Here we use satellite-based rice paddy and XCH4 data to investigate the spatial–temporal relationships between rice paddy area, rice plant growth, and XCH4 in monsoon Asia, which accounts for ~87% of the global rice area. We find strong spatial consistencies between rice paddy area and XCH4 and seasonal consistencies between rice plant growth and XCH4. Our results also show a decreasing trend in rice paddy area in monsoon Asia since 2007, which suggests that the change in rice paddy area could not be one of the major drivers for the renewed XCH4 growth, thus other sources and sinks should be further investigated. Our findings highlight the importance of satellite-based paddy rice datasets in understanding the spatial–temporal dynamics of XCH4 in monsoon Asia.


Supplementary Figure 11
Supplementary Figure 11. Consistency of spatial distributions between paddy rice croplands and atmospheric methane concentration but for Indonesia and Malaysia. The  (1 st   26 row), India with winter wheat and rice (2 nd row), Cambodia and Vietnam with multiple cropping systems (3 rd row), and Indonesia and Malaysia with much cloud (4 th row). The first and second columns are the frequency of Pearson's correlation coefficients between XCH4 and EVI for the whole year (Ry) and summer-fall season from May to November (R5-11) in the four regions, respectively. The third and fourth columns are the frequency of significance levels of Pearson's correlation between the two for the whole year (Py) and summer-fall season from May to November (P5-11) in these regions, respectively. n means the number of gridcells with valid value.    Supplementary Fig. 13), the important OM for CH4 emission from rice paddies, and then further controls CH4 emissions of rice paddies 19,24 . Some studies also indicated that the seasonal variation of CH4 emission could be predictable from rice production 25, 26 .

Supplementary Note 2. MODIS data processing
The three Moderate Resolution Imaging Spectroradiometer (MODIS) data products, including the 8-day composite surface reflectance product (MOD09A1) 27 , land surface temperature (LST) product (MYD11A2) 28 , and land cover type product (MCD12Q1) 29 were used for paddy rice mapping.
For the MOD09A1 product, we identified bad observations (clouds, cloud shadows, and snow) in two steps. First, the quality control flag layer was used to extract the clouds and cloud shadows from each observation. Second, we further recognized clouds by using the blue reflectance of ≥ 0.2 40, 41 . The snow cover observations were also identified and excluded by using NDSI and the NIR band (NDSI > 0.40 and NIR > 0.11) 38,39 . All the observations identified as clouds, cloud shadows, or snow cover were excluded. Next, we filled the gaps in the time series LST data by using the linear interpolation approach 42 . The 8-day smoothed LST product was used to determine the thermal growing season of vegetation, which was used to identify the temperature-based time window for rice transplanting. The first and last dates of stable minimum temperatures higher than 5°C in continuous three 8-day intervals were identified as the start and end of thermal growing season ( Supplementary Fig. 21). The resultant maps of the start and end dates of thermal growing season were resampled to 500 m using the nearest neighbor method to match the

Supplementary Note 3. Mapping rice paddies with time series MODIS data
The EVI and LSWI within the time window of flooding and transplanting were used to identify the observations with signals of flooding and transplanting, and a pixel was assumed to be a "potential or likely" rice paddy if one or more observations were identified in that manner (Eq. (4)- (7)).
where Ti is the  Fig. 22a). Second, we used the forest mask derived from the analysis of the Phased Array type L-band Synthetic Aperture Radar (PALSAR) data in 2010 at 50 m spatial resolution 46 . It was aggregated to 500 m to be spatially consistent with the MODIS data and a 70% threshold was used to extract forest ( Supplementary Fig. 22b). Third, we generated maps of sparsely vegetated land (e.g., saline and alkaline land, build-up) using annual maximum EVI less than 0.4 as a threshold 31 , and permanent water body was also included in this mask ( Supplementary Fig. 22c). Fourth, we used two available maps of  Fig. 22e). Fifth, we generated topographic masks to remove regions above 2600 m above sea level (asl.) 48 and with a slope greater than 4° where it was unsuitable for paddy rice planting, by using Shuttle Radar Topography Mission (SRTM) 90 m Digital Elevation Model (DEM) data ( Supplementary Fig. 22f, g). Last, the length of LST >= 5 °C less than 100 days was used as an annual temperature-based mask throughout the study area ( Supplementary Fig. 22h), which can help remove the noises occurring in cold region or montane areas. We first generated annual paddy rice maps in monsoon Asia during 2000-2015 by excluding these masks from annual potential rice paddy (flooding and transplanting) layers.

Supplementary Note 4. Accuracy assessment of MODIS-based paddy rice maps
The accuracy assessment of land cover maps is a critical component of land cover mapping.
To assess the accuracy of annual paddy rice maps in monsoon Asia, we have conducted ground truth data based validation in different study areas by using ground survey data or crowd-sourced field photo data in high latitude regions of Asia 49 . Given the huge land area in this region, it was challenging and beyond our capacity to collect national-level ground survey data for monsoon Asia. Therefore, in this paper, we focused on the comparisons with the existing paddy rice products. We

1) Accuracy assessment of paddy rice maps in China
The NLCD datasets agreed well with the MODIS-based paddy rice maps in China.
The spatial patterns of MODIS-based rice paddies were consistent with those of the NLCD-  Supplementary Fig. 23). All the slopes of linear regressions were close to 1.

2) Accuracy assessment of paddy rice maps in Korea Peninsular and Japan
The existing local paddy rice maps were used for the comparison of MODIS-based RapidEye-based paddy rice map in North Korea ( Supplementary Fig. 24a-d), even in the eastern and southern regions where small rice fields distributed sparsely 11 . The statistical analysis at provincial level also showed a significant correlation between the MODISbased and RapidEye-based paddy rice maps (R 2 =0.95, P<0.01, Supplementary Fig. 24m).
In South Korea, the resultant paddy rice map agreed with the OECD-derived paddy rice map in 2008 ( Supplementary Fig. 24e-h). The area statistics at provincial level was also significantly correlated with that of the MODIS-based paddy rice map (R 2 =0.95, P<0.01, Supplementary Fig. 24n). In Japan, the AVNIR-2-based paddy rice map showed a consistency with the MODIS-based paddy rice map ( Supplementary Fig. 24i-l). The AVNIR-2-based rice paddy area statistics at provincial level was significantly correlated with that of the MODIS-based paddy rice map (R 2 =0.80, P<0.01, Supplementary Fig. 24o).

3) Accuracy assessment of paddy rice map in South Asia
The showed significant correlations at both levels (R 2 = 0.92 for provincial comparison, R 2 = 0.81 for prefectural comparison, P<0.01, Supplementary Fig. 25g, h).

4) Accuracy assessment of paddy rice map in Southeast Asia
The nine countries in Southeast Asia with dense paddy rice areas were considered here,  Fig. 28).
The overall accuracies of the paddy rice maps for the two regions are 91% and 81% (Supplementary Table S3). The high consistency of spatial distributions between MODISbased paddy rice maps and these existing national paddy rice maps, as well as the high correlations between MODIS-based results and the statistical data in these countries, showed that our MODIS-based paddy rice maps had a reasonably high accuracy, and can examine inter-annual variations of rice paddy areas. Besides the aforementioned validations and comparisons, we will release the data together with the publication and expect a crowd-sourcing validation from the users.