Machine Learning-Aided Groundwater Level and CO 2 Emission Estimations in Tropical Peatlands using Global and Regional Scenarios

Machine Learning-Aided Groundwater Level and CO 2 Emission Estimations in Tropical Peatlands using Global and Regional Scenarios | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article Machine Learning-Aided Groundwater Level and CO 2 Emission Estimations in Tropical Peatlands using Global and Regional Scenarios Waluyo Yogo Utomo, Fitri Khusyu Aini, Muhammad Askary, Syaiful Anwar, and 2 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-9002157/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Tropical peatlands store a disproportionately large fraction of global soil carbon and emit large amounts of soil CO 2 from peat decomposition and root respiration. Accurate quantification of peatland carbon emissions requires a prediction that is sensitive to the space-time dynamics of the environmental factors. This study estimated a biweekly peat CO 2 emission representing total soil respiration (Rs) representing the whole peatland ecosystem in Rupat Island, Indonesia. The Rs was estimated using nonlinear CO 2 to groundwater level/GWL response curves reviewed from Southeast Asian chamber studies. The GWL dataset was predicted using Extreme Gradient Boosting (XGBoost), integrating dense GWL measurements (Jan-2019 to Apr-2025) against dynamic and static predictors. Spatial upscaling of predicted GWL (− 32.19 ± 5.58 cm) resulting in estimated annual Rs of 4.02 to 5.67 Mt CO 2 and totaling 28.15 to 39.66 Mt CO 2 cumulative, using general and land use-based response curves. Our study reported relatively low interannual and seasonal variations (< 1 Mt CO 2 y-1), although cultivated and drained shallow peats consistently act as emission hotspots during dry years. Our Rs estimates exceeded the global and national emission factors (EFs) and other Rs-GWL prediction methods by 1.5 to 6 times, suggesting potential for implementation in CO2 modelling at the ecosystem scale. Earth and environmental sciences/Biogeochemistry Earth and environmental sciences/Climate sciences Biological sciences/Ecology Earth and environmental sciences/Ecology Earth and environmental sciences/Environmental sciences CO2 emission groundwater level machine learning peatland hydrological unit soil respiration tropical peatlands Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Figure 8 Introduction Even though only covering only 2.5–3% of the global land area, peatlands are regarded as the most carbon-rich ecosystems with estimated 450–650 Pg of carbon storage, surpassing all the world’s forests combined. Indonesia contains approximately 36 ± 6 Pg C in its peat layer, accounting for about 6–8% of global peat-carbon stocks while occupying only 10% of tropical peatland area. Collectively, these peat layers represent nearly one-third of global soil organic carbon within a small fraction of the terrestrial surface 1 – 3 . Despite their conservation importance for harbouring biodiversity and storing carbon, peatlands are also being subjected for collection of natural resources and converted for developing forestry and agricultural commodities 4 – 6 . The development of peatland swamps requires drier condition by drain out excessive water. Lowered groundwater level/GWL could trigger peat drying and thus accelerate peat decomposition. In the Southeast Asian region, the peatlands potentially emit around 30 to 100 Mg C ha⁻¹ y⁻¹ per metre drawdown and around 2.5 Gt C cumulative since 1990 4,7 . Albeit their higher recalcitrance preventing excessive organic matter oxidation 8 , 9 , some equatorial peats are reportedly having higher sensitivity to GWL dynamics, linked to regional climate changes and inter-annual rainfall variations, i.e. , El-Nino 10 , CO 2 emission factors/EFs satisfied these dynamic conditions cannot be predicted by applying static land use-based estimates, e.g. , Tier-1 IPCC EFs 11 (for example 4 , 12 ) and Indonesian forest reference emission level/FREL 13 . It requires more detailed dataset concerning national and sub-national emission estimates that not only capable to predict emissions on inter-year and land uses, but also incorporates within-year and -land uses heterogeneities. Previous studies had been explored these possibilities using more practical proxies mainly governing CO 2 emission, e.g. , GWL 14 , 15 , precipitation-GWL 16 and subsidence 17 , or more exhaustive process-based modelling e.g. , PEATCLSM 18 , and CNM-PHM 19 . Ultimately, current studies on reporting CO 2 emissions had been shifted from calculating CO 2 emitted solely from peat degradation/Rh 4,11–13 to the real amount of CO 2 exchanged on ecosystem scale 16 , 20 – 22 . In this context, upscaling the already partitioned Rh and autotrophic respiration/Ra would introduce higher uncertainty 23 – 25 than directly estimating soil total respiration/Rs as basis dataset for ecosystem respiration to estimate net CO 2 ecosystem exchange. Despite this importance, monitoring networks in tropical peatlands remain limited, particularly in Indonesia, where in-situ GWL stations are sparse and temporal resolution is often restricted to longer aggregation periods. As a result, rapid drawdown GWL caused by El Niño and canal drainage may go undetected. These process-based GWL models require extensive site-specific calibration and high-resolution inputs 18 , 19 , 26 , which constrains scalability for large-area simulation. Furthermore, numerical modeling methods such as geostatistics 27 , might not sufficiently capture the drivers of GWL changes and variability across unevenly distributed dipwells. Emerging data-driven approaches deliver promising alternatives for modeling GWL changes in tropical peatlands. In Indonesian and Malaysian peatlands, scattered research reported GWL predictions with reasonable accuracies using polynomial regression 28 , boosted tree 29 , 30 , and neural network 31 , 32 . In northern peatlands, several studies reported successful GWL prediction using random forest and boosted tree regressors 33 , 34 . While these machine learning (ML) studies made substantial progress toward peatlands GWL prediction, most remain localized and lack generalizability across broader eco-hydrological gradients. To address these challenges, this study presented Rs estimation based on data-driven GWL prediction utilized ML regressor that is trained from long-term dipwell measurements with dynamic and static predictors representing climate, vegetation, soil, and topography. The estimated Rs follows CO 2 -GWL response function reviewed from Southeast Asian chamber studies, which is vital as building block for modelling ecosystem-level CO 2 exchange. Results GWL Prediction Results, Evaluations, and Explanations General Description of GWL and All Its Predictors The descriptive statistics for groundwater level (GWL) and all predictors measured and collected within our study area (Fig. 1 ) and period (Jan-2019 to Apr-2025) are summarized and displayed in Table 1 and Fig. 2 , respectively (naming conventions are provided in Supplementary Information 1). Our study site covers a single Peat Hydrological Unit (PHU; in Indonesian terms: Kesatuan Hidrologis Gambut/KHG) on Rupat Island, Riau, Indonesia, which harbours various peat depths, multiple peat domes, and land uses (Fig. 1 ; Supplementary Information 2). A PHU represents an integrated peatland hydrological system bounded by rivers, coastal interfaces, and swamp areas, with peats interconnected with ecosystem components, e.g. , water, vegetation, and other environmental elements 35 , 36 . Most of the measured GWL data were concentrated on − 30 to − 40 cm, averaging around − 37.45 ± 11.09 cm (average ± standard deviation) and slightly skewed towards shallow water table (Fig. 2 a and 2 b; Table 1 ). Shallow GWL conditions (+ 10 to − 20 cm) were mostly located on peat at more than 6 masl and deep peats (Fig. 2 c). All dynamic and static predictors exhibited lower chance to severe multicollinearities (r < |0.8|; Fig. 2 b), which can be tolerated by the XGBoost regressor. Precipitation, storage (representing precipitation minus evapotranspiration) and their antecedent terms (lag1 and 2) were distributed across all GWL bins. Current and lagged soil moisture proxies declined following GWL deepening (Fig. 2 c). Peats located within 5 km of the coast and natural streams mostly had moderate GWL (− 30 to − 50 cm; the latter had shallower table up to − 20 cm), but never experienced deeper GWL. Table 1 Summary statistics of GWL and its predictors. Variable Units Min Q1 Median Q3 Max Mean SD Skew Kurt GWL cm -80.00 -41.00 -37.84 -32.00 -1.00 -37.45 11.09 -0.52 -2.12 Precip_raw mm 17.83 70.91 103.55 133.07 206.60 102.00 38.31 -0.09 2.16 Precip_raw_lag1 mm 17.83 75.16 101.86 131.39 208.73 102.77 36.56 0.09 2.57 Precip_raw_lag2 mm 1.16 77.69 101.13 133.03 222.44 104.13 37.07 0.15 2.52 Precip_cum_lag2 mm 174.29 270.72 309.28 345.95 482.08 308.90 56.14 0.07 2.89 S_raw mm -517.99 -328.98 -270.11 -208.93 -64.22 -270.41 78.32 -0.16 2.38 S_raw_lag1 mm -525.94 -328.98 -275.27 -215.85 -107.14 -274.37 75.97 -0.24 2.51 S_raw_lag2 mm -498.89 -319.46 -261.78 -209.29 -63.74 -267.71 75.08 -0.27 2.54 S_cum_lag1 mm -753.88 -592.22 -546.33 -501.77 -345.35 -544.78 68.41 0.10 2.87 S_cum_lag2 mm -1166.80 -885.59 -817.64 -738.53 -466.55 -812.49 103.62 0.05 2.63 SM_raw %v/v 0.51 0.57 0.58 0.60 0.66 0.58 0.02 -0.08 2.79 SM_raw_lag1 %v/v 0.51 0.57 0.58 0.60 0.65 0.58 0.02 -0.06 2.80 SM_raw_lag2 %v/v 0.51 0.57 0.58 0.60 0.65 0.58 0.02 -0.02 2.75 peat_thick m 0.01 3.34 4.99 6.26 9.90 4.91 1.93 0.07 2.62 eucl_dist_ar m 6.05 1362.10 2553.00 3845.79 8461.29 2787.59 1685.40 0.59 2.87 Filled_subsurf masl 1.15 2.03 2.27 2.81 5.68 2.47 0.58 1.26 6.00 Filled_surf masl 3.32 5.21 6.81 8.18 9.37 6.75 1.67 -0.08 1.69 Slope_subsurf rad 0.01 0.09 0.17 0.24 0.42 0.17 0.09 0.51 2.51 Slope_surf rad 0.00 0.04 0.07 0.12 0.21 0.08 0.05 0.71 2.62 coast m 875.04 4056.51 7743.58 12687.90 18077.62 8495.61 4917.45 0.27 1.75 Figure 1 . Table 1 . Figure 2 . XGBoost GWL Model Prediction Result Figure 3 presents the temporal fluctuations in predicted versus observed GWL at selected dipwell stations representing wide range of elevation and peat depth gradients. Training and validation points (indicated in green and red, respectively) exhibited consistent results across both datasets, with several minor deviations observed during periods of high variability. Dry periods (highlighted with yellow shading) denoted intervals of reduced precipitation and occurred most frequently during seasonal drawdown events. Overall, the time-series results illustrated that the trained XGBoost model generalized throughout varied site conditions, ranging from deep peat to shallower peat. Predictive varied spatially according to peat thickness and surface elevation. At deep peat sites with thickness greater than 450 cm (SRL4_092_19_SRL and PTR_19_250_PRT), predicted GWL were stable with amplitude rarely exceeding ± 30 cm. The model also successfully predicted relatively deep GWL (approximately − 55 to -60 cm) by following the original patterns in PTR_18_250_PRT and PTR_04_250_PRT. These conditions are also reflected in spatiotemporal predicted GWL maps averaged to dry, rainy, and annual courses (Fig. 4 ). Shallower water tables (− s10 to − 30 cm) were consistently located around the central domes and less-drained peat areas. Meanwhile, deeper GWL (around − 40 to − 60 cm) were detected in peripheral zones near the coast, which are actually drained for forest plantations and agriculture (Fig. 4 ; Supplementary Information 2). The pooled GWL maps display moderate seasonal differences between the rainy and dry seasons, indicating that GWL is consistently deeper during the dry season than during the rainy season. Interannual variability from 2019 to 2024 was less pronounced, regardless of the averaging factor, with significant changes occurring during 2020 and 2022 (Fig. 4 ). Figure 3 . Figure 4 . XGBoost GWL Model Evaluation The trained XGBoost model (Fig. 5 a) achieved moderate predictive accuracy in Monte Carlo cross-validation (MCCV), with median R 2 and RMSE of 0.46 and 7.25, respectively, including a median MAD and MAPE of 3.3 and 20.5%, respectively. In contrast, the model's performance on test sets were declined and showed greater variability across all error metrics relative to the training performances, with median R 2 , RMSE, MAD, and MAPE of 0.35, 8.0, 3.8, and 24.0%, respectively. The predicted GWL showed negligible bias (median ≈ 0.0) on the training folds, whereas a slight positive median bias and increased variance were observed on the test folds. Figure 5 . XGBoost GWL Model Explanation The SHaP analysis indicated that our XGBoost models relied primarily on static and antecedent climate predictors, with soil condition proxies as moderate contributors. Individual SHaP values spanned both positive and negative ranges, and the six highest mean |SHAP| values showed negative skewness. SHaP-based permutation method (nsim = 500) detected that surface elevation, coastal proximity, and dummy year = 2020 exhibited the highest mean |SHAP| values, followed by 1.5-month antecedents of precipitation (Precip_cum_lag2) and storage (S_cum_lag2), as well as season (Fig. 5 b and 5 c). Subsurface elevation, soil and vegetation properties (Filled_subsurf, Slope_surf, peat_thick, SM) with distance to natural stream and current precipitation-storage exhibited moderate to low mean |SHAP| values, while dummy years (2019, 2023–2025) had the least contributions. Further investigations based on ordered training data (n = 5857 observations; Fig. 6 a and 6 b) suggested that the XGBoost model's predictions differ according to the magnitudes of predicted GWL. SHAP values were increased oppositely toward both distribution tails. At the driest extreme (0%, approximately at GWL − 77 cm), predictions were largely determined by negative contributions from precipitation and storage, with additional static factors. In shallow GWL ( > − 25 cm; 75% and 99% percentiles), SHAP contributions are strongly positive, mainly driven by storage, year = 2022, and the cumulative influence of several minor predictors. Near the median GWL range (− 30 to − 40 cm), SHAP values were clustered around ± 5 and showed a mixed pattern. Figure 6 . CO 2 -GWL Response Curve Simulations and Upscaling to Spatiotemporal Rs Estimations Spatiotemporal Rs Dynamics and Distributions Spatial upscaling of soil CO₂ emissions from general and a combination of general and land use-based CO 2 -GWL response curves developed by 14,15 (Supplementary Information 3) over Pulau Rupat PHU are summarized in Table 2 . The upscaled Rs utilized biweekly GWL maps (averaged to − 32.19 ± 5.58 cm) corresponding to total cumulative Rs of 28.15 to 39.66 Mt CO₂ and annual cumulative Rs around 4.02 to 5.67 Mt CO₂ y ⁻¹ emitted during the study period (Jan-2019 to Apr-2025; Fig. 7 ; Table 2 ). The variability of Rs estimated by general response curve were lower than combination of general and land use-based CO 2 -GWL response curves. The Rs emitted during the dry season were slightly higher than during the rainy season. Annualized Rs tend to be lower in deeper peat areas than in shallower peat, but opposite trends were observed on cumulative annual and total. Among land-use categories, estate plantations possessed the deepest GWL and highest cumulative Rs. Estate plantation emitted the highest average annualized Rs and the highest cumulative Rs during the study period. Primary swamp forests emitted lowest Rs, which only up to 2022 (Fig. 7 , Supplementary Information 2). Inter-annual GWL were relatively stable, except for 2020, which had the deepest mean (− 34.46 ± 5.49 cm), with the Rs estimated from land-use-based response curves exhibited an increasing trend. Application of the general CO 2 -GWL response curve (Fig. 7 a; pooled to season and annual courses similar to GWL in Fig. 4 ) produces relatively smooth spatial gradients of Rs. In contrast, the use of combined land-use-based and general response curves (Fig. 7 b) results in substantially greater spatial variation, with consistently high Rs emitted from peats developed for plantations and agriculture (Supplementary Information 2) during both dry and rainy seasons. Central and peripheral zones associated with managed land uses exhibited higher annual cumulative Rs, whereas areas corresponding to undrained peatlands showed lower Rs (Fig. 7 a and 7 b). Table 2 Estimated GWL and Rs (average ± standard deviation) in PHU Pulau Rupat, averaged across all data and several factors (season, peat thickness, major land use, and year). Supplementary Information 1. Response and selected explanatory variables used in this study Factor GWL Averaged Annualized Rs Cumulative Annual Rs Cumulative Rs During Study Period General Eq LU-Based Eq General Eq LU-Based Eq General Eq LU-Based Eq cm ----------- Mg ha − 1 y − 1 --------- ---------- Mt CO 2 y − 1 --------- ---------- Mt CO 2 --------- All Data -32.19 ± 5.58 47.22 ± 0.54 66.52 ± 22.76 4.02 5.67 28.15 39.66 Season Dry -32.61 ± 5.71 47.26 ± 0.55 67.00 ± 22.98 2.01 2.85 14.09 19.97 Rainy -31.77 ± 5.58 47.18 ± 0.55 66.03 ± 22.60 2.01 2.81 14.06 19.68 Peat thickness 3 m -32.30 ± 4.27 47.18 ± 0.44 58.19 ± 19.71 2.34 2.89 16.4 20.22 Major Land Use Primary Swamp Forest -32.12 ± 1.28 47.09 ± 0.13 0.49 ± 0.00 0.08 0.00 0.59 0.01 Secondary Swamp Forest -32.47 ± 3.98 47.18 ± 0.36 57.24 ± 4.29 0.9 1.09 6.31 7.66 Plantation Forest -31.19 ± 4.98 47.10 ± 0.39 47.31 ± 1.47 0.93 0.93 6.51 6.54 Estate Plantation -34.46 ± 5.49 47.47 ± 0.66 97.99 ± 9.76 1.12 2.32 7.86 16.24 Agriculture and Other Land Uses -30.29 ± 6.65 47.10 ± 0.59 63.13 ± 7.19 0.98 1.32 6.87 9.21 Year 2019 -32.76 ± 5.66 47.30 ± 0.54 63.45 ± 23.21 4.03 5.41 4.03 5.41 2020 -34.37 ± 6.73 47.59 ± 0.88 67.10 ± 27.74 4.05 5.71 4.05 5.71 2021 -31.99 ± 5.30 47.18 ± 0.42 65.22 ± 24.89 4.01 5.54 4.01 5.54 2022 -31.75 ± 5.55 47.16 ± 0.51 68.41 ± 22.52 4.02 5.83 4.02 5.83 2023 -31.19 ± 4.66 47.07 ± 0.32 66.82 ± 19.56 4.01 5.7 4.01 5.7 2024 -30.52 ± 4.86 47.02 ± 0.32 66.21 ± 19.56 4.01 5.64 4.01 5.64 -32.73 ± 5.15 47.22 ± 0.39 68.42 ± 20.18 4.02 5.83 4.02 5.83 Figure 7 . Table 2 . Rs Response to 10-cm GWL Changes Figure 8 a illustrates how our estimated Rs would change (positive and negative values represent emitted and avoided Rs) in response to 10 cm GWL changes (on an annual basis) according to the CO 2 -GWL response curves ( 14,15 , Supplementary Information 3). The general response curve 15 predicted that an Rs change increases incrementally as GWL decreases below − 40 cm and progressively as GWL deepens, but would exceed 100% during maximum GWL drawdowns (<-80 cm). Across all response functions 14 , 15 , the largest Rs changes (306% and 246%) were observed in estate plantation and agriculture and other land uses, respectively, if GWL deepened dramatically from 0 cm to − 80 cm. Other major land uses (secondary swamp forest and forest plantation) are also experiencing more than 100% Rs changes under similar conditions. Estate plantation exhibited intermediate changes; a 10-cm GWL drawdown between − 30 and − 60 cm would generate 20–30% Rs changes, which would further decrease with shallower GWL changes. Rs responses were not reciprocal with respect to water table rise and decline. For example, the emission increase associated with a 10-cm GWL declination from − 40 to − 50 cm exceeded the reduction avoided by a 10-cm GWL rise from − 50 to − 40 cm. Across all curves, proportional changes were greater under water-table decline than under an equivalent rise. Comparison to Other Emission Factors Annual averaged and total cumulative emitted Rs estimated using combination of general and land use-based CO 2 -GWL response curves were consistently higher (66.52 ± 22.76 Mg CO₂ ha - 1 y ⁻¹ and 39.66 Mt CO 2 ) than linear curves from previous Rs reviews 37 , 38 and static Rh-based EFs derived from IPCC Tier 1 11 and Indonesian FREL 13 (Fig. 8 b). However, the general Rs estimates from this study falls within the same range as both authors 37 , 38 and Rh from Indonesian FREL 13 which totalling around ~ 4 Mt CO₂ yr⁻¹ for Rupat Island (equivalent to ~ 45–50 Mg ha⁻¹ yr⁻¹). Emission estimated from IPCC Tier 1 were lowest compared to all studied EFs in terms of annual averaged (Fig. 8 b.1) and total cumulative emission (Fig. 8 b.2). Figure 8 . Discussion This study highlighted the spatiotemporal Rs estimation via ML-assisted GWL maps in a single tropical peatland hydrological system ( Indonesian KHG/ PHU Rupat Island). Our study provides alternative dynamic estimation of tropical peatland Rs-based EF, alongside current Rh-based global IPCC Tier 1 11 and regional (Indonesian FREL 13 ) static, land use-based EF estimations ( e.g. , 4,12 ). On the other hand, our XGBoost regressor are trained from GWL dipwell that possess comparable measurement density than any previous models 16 , 27 , 28 , 31 in terms of spatial and temporal scales. The predicted GWL (median cal R 2 0.46; val R 2 0.35; Fig. 5 a) showed various hydrological regimes (Figs. 3 and 4 ) representing water storage dynamics and trade-offs between precipitation against evapotranspiration and lateral outflow 39 . Pooled GWL (Table 2 ) suggested that at the PHU level, the averaged GWL were relatively shallow (> -40 cm), regardless of the partitioning factors. However, some drained and cultivated areas at the western and southern sides of the PHU experienced deep GWL during 2020 and 2022 (-45 to -60 cm; Fig. 4 ; Supplementary Information 2), which resulted in elevated Rs up to 120 Mg CO 2 ha - 1 y - 1 (Fig. 7 ). This condition may correspond to a dry condition underwent at the preceding and corresponding years 40 , 41 . The 6-years predicted GWL mean in primary and secondary peat swamp forests were − 32.12 ± 1.28 and − 32.47 ± 3.98 cm, respectively, which are deeper than undrained peat swamp forests reported by previous modelling study 16 . Even though the peat swamp forests in the studied PHU are undrained, their small areas and location within the peat dome resulting in the GWL are much more influenced by the surrounding by drained and cultivated peats 42 , 43 (Fig. 4 ). Differently from 16 , 22 , 37 , 38 , our study reported the averaged GWL at estate and forest plantations sites around half of their estimates (− 34.46 ± 5.49 and − 31.19 ± 4.98 cm; Table 2 ), which were within the recommended best practice threshold for GWL 44 . Deep-peat at the central PHU (northern dome) consistently emit low Rs due to their forest covers, while surrounding shallow peats serve as recurrent high-emission hotspots across seasons and years (Fig. 7 ). Dry-season and drought-year conditions increase emissions by approximately 20 to 30 Mg CO₂ ha⁻¹ yr⁻¹ and contribute substantially to annual totals. In contrast, rainy-season rewetting reduced, but did not eliminate emissions in drained areas (Fig. 7 a-b). Our simulation demonstrated that drainage-induced drying leads to disproportionately large increases in CO₂ emissions, which are not fully mitigated by short-term rewetting (Fig. 8 a). The resulting dynamic emission estimated during study period (39.66 Mt CO₂ y⁻¹) broadly correspond to higher Rs emanates from forest plantation and agriculture and other land uses (Table 2 ), exceeding Tier-1 IPCC 11 and Indonesian FREL 13 estimates by over six times (Fig. 8 b) but close to previous chamber and subsidence studies 45 , 46 . Variations in Rs across land-use categories in Table 2 may closely correspond to inter-annual land-use transitions (Supplementary Information 2) and GWL patterns (Fig. 4 ; Table 2 ). During this period, PHU Pulau Rupat underwent a marked shift toward managed land uses, particularly when primary swamp forest, bare/open land, and swamp/marsh were gradually converted to estate and forest plantations. These conditions were reflected in Fig. 3 , as deeper GWL areas were persistently concentrated throughout area dominated by both plantations during any seasons, corroborating previous reports 16 , 47 . Those areas remained moderately to deeply drained even during the rainy season, indicating that seasonal precipitation did not offset drainage-induced drawdown, taking into account for the uncertainty of our climate dataset. In comparison, areas historically classified as primary swamp forest exhibited shallower and more homogeneous GWL, consistent with their lower cumulative Rs. Cumulative Rs remained stable across years, even though there are ongoing land-use changes (Fig. 7 ; Supplementary Information 2). Simulated Rs responses to 10-cm GWL shifts (Fig. 8 a) showed that emission increases under GWL declination consistently exceeded avoided emission under equivalent rise at any response curves 14 , 15 , as also reported in previous reports 48 . However, proportional changes in Rs under moderate GWL fluctuations do not directly translate into equivalent changes in long-term peat carbon stocks. Within more typical hydrological ranges (shallower than approximately − 30 to − 50 cm), a substantial portion of Rs (commonly 40–70%) corresponds to root respiration, which represents contemporary plant carbon cycling rather than peat carbon loss. Strong seasonality and increased intensity under a changing climate 5 , 49 , 50 and widespread subsidence 51 may expose tropical peatlands to sustained inundation during rainy season and experiencing rapid GWL drawdown during dry season. This condition, in the extreme cases ( e.g. , El Nino 10 , 0 to − 80 cm; Fig. 8 a), would generate more than 100–200% emission changes generally, with the most drastic found on forest plantation (306%). At deep GWL, Rs predominantly originated from enhanced peat oxidation 52 , which in long-term conditions would severely impact old sequestered carbon 1 , 7 , 53 . Differences between the Rs generated in this study and the emission magnitudes implied by the Indonesian FREL 13 and IPCC Tier-1 11 EFs (Fig. 8 b) mainly result from differences in emission accounting frameworks and boundaries. This study, together with 37 , 38 quantifies total soil respiration (Rs) by including both autotrophic and heterotrophic components across all peatland conditions. The resulting Rs can be used as more robust basis for upscaling RE compared to partitioned Ra and Rh, considering recent shifting of EF calculation approach favoring actual ecosystem CO 2 exchange 16 , 21 , 22 . In contrast, both IPCC Tier-1 and Indonesian FREL framework takes into account only for heterotrophic respiration (Rh) associated with anthropogenic peat decomposition and explicitly excludes emissions from undisturbed peat swamp forests. Consequently, total peat CO₂ emissions represented within the IPCC Tier-1 and FREL frameworks may exceed those Rs reported here (Table 2 ; Fig. 8 b) when aggregated over extensive disturbed areas, despite lower individual EFs. Further uncertainty is introduced by the IPCC Tier-1 factors, which assume negligible emissions from peatlands classified as natural, including bushland, swamps, and logged peat swamp forests. Both IPCC Tier 1 and partially of Indonesian FREL are consider reporting conventions specifically for accounting how much peat material has been degraded. This assumption does not reflect underlying biophysical processes, particularly for Indonesian peat swamp forests, which have been currently reported as net CO 2 emitters 16 , 22 , 47 , 54 despite some higher sequestration reported from undrained Malaysian peat swamp secondary forests 55 . These reports suggested that Rs emitted from peat swamp forests cannot be neglected. From measurements covering 1997–2016, 16 reported that all undrained peat swamp forests in Indonesia and Malaysia were small net CO 2 sinks (− 0.86 ± 10.9 Mg ha - 1 y - 1 ) and acted as large net CO 2 emitters during dry years (13.1 ± 12.9 Mg ha - 1 y - 1 ). Furthermore, drained peat swamp forests currently act as net CO 2 emitters in normal, wet, and dry years. Based on 55 report on higher GPP in coastal peat swamp forests compared to inland peat swamp forests 47 , which is similar to our research site 56 (Fig. 1 ). Therefore, our reported Rs originating from primary and secondary forests (Table 2 ) might be offset by higher CO 2 sequestered by the forest vegetation. Beyond differences in accounting perspective, the IPCC Tier-1 and FREL approaches are inherently static and do not capture the sensitivity of peat CO₂ emissions to climate-driven hydrological variability 11 , 16 . Static emission factors are applied homogeneously across land-use classes and remain constant across seasons and years. This approach might obscure substantial spatial variation in drainage intensity within the same land-use category. In contrast, the biweekly GWL prediction adopted in this study captured spatial variability within land-use classes. It is also related to temporal variability associated with seasonal and interannual climate fluctuations (Fig. 3 , 4 , and 7 ; Table 2 ). Our results illustrated that climate-driven changes in GWL fluctuation modulated Rs at biweekly timescales, even within similar land-use classifications, particularly in managed peatlands where drainage infrastructure restricts water table recovery. Similar to our study, 57 reported that IPCC and Canadian national EFs were mostly dependent on CO 2 research conducted on warm periods and suggested the revised EFs around 45% lower. In tropical peatlands, this condition can be related to the measurements being mostly conducted within the drier season or dry conditions in the rainy season, which allows chambers to be placed securely on the peat surface to measure CO 2 effluxes. However, that undermines the specific conditions representing long-rainy season and La Nina period, where all peat surfaces are inundated or GWL are very shallow, thus resulting in CO2 emission that are lower than averages. Also, the EFs do not account for lowered CO 2 emission due to long-term shallower GWL as affected by large-scale restoration projects 58 – 60 . These dynamics are also not represented by land-use-only EFs, which average emissions over space and time. Our method is considered a dynamic approach that explicitly link CO₂ emissions to GWL variability. This is, therefore, provide a more temporally responsive and spatially explicit representation of peat carbon fluxes under fluctuating climatic conditions. Methods Study Site Description This study was conducted on the Rupat Island Peat Hydrological Unit (PHU; In Indonesian terms: Kesatuan Hidrologis Gambut/KHG) in Riau Province, Indonesia (Fig. 1 ). A PHU represents an integrated peatland hydrological system bounded by rivers, coastal interfaces, and swamp areas. The peats in PHU are interconnected with ecosystem components, e.g. , water, vegetation, and other environmental elements 35 , 36 . The PHU Rupat Island covers an area of approximately 118.510,16 ha and situated on Rupat Island in the Malacca Strait off the eastern coast of Sumatra, Indonesia. The peatland in this PHU is developed across the coastal (southern dome) and riverine (northern dome) backswamps at the island’s middle to southern sides, with two main domes having very deep peats (around 10 to 12 m depth; Fig. 1 ). The PHU had a variety of land uses, dominated mostly by mosaics of the remains of peat swamp forest and shrublands at the dome center, surrounded by forest and estate plantations alongside fragmented cultivated lands, settlements, and aquaculture ponds in the peripheral areas with thinner peats (Supplementary Information 2). Data Acquisition and Pre-processing This study leveraged data-driven modelling to predict groundwater level (GWL) using a combination of dynamic, static, and time-related predictors (Supplementary Information 1). The GWL data were gathered from 134 automatic loggers and manual dipwells located throughout PHU Rupat Island as part of the regulatory compliance to the Indonesian Ministry of Environment and Forestry (MoEF; now Ministry of Environment-Environmental Control Agency of Indonesia/KLH-BPLH). We selected the most observed period covering 2019 to 2025 from Simatag0,4m platform ( https://sippeg.menlhk.go.id/apps/sippegapps/tmat ) and obtained (aggregated) 8,327 biweekly records in total. All readings exceeding + 10 cm were excluded (which may indicate input errors), while the entire negative GWL was selected. All predictors were selected for their hydrological relationship with GWL spatiotemporal dynamics in tropical peatland landscapes. The constructed model conceptually represents the influence of climate, vegetation, topography, and soil on peatland hydrology. We also included several proxies for temporal autocorrelation, including lagged variables and year and season dummies. The dynamic inputs representing climate and vegetation drivers consisted of 158 biweekly precipitation (Precip) and evapotranspiration (ET) composites (2019–2025), resampled at 10-meter resolution. Biweekly Precip was aggregated from daily 0.05° CHIRPS 61 . ET estimates were derived from the MODIS MOD16A2 8-day evapotranspiration product 62 . We selected the VV band from Sentinel-1, which represents surface radar scattering, as a proxy for peat soil moisture (SM). Furthermore, we decomposed partial storage proxy (S) as an additional dynamic predictor, following S = Precip – ET, where S represents both storage and lateral outflow, and GWL is technically an upper table of S 39,53 . Since the studied PHU is separated from other higher grounds by a large river (Fig. 1 ), we assumed that the studied peatland is ombrotrophic and thus received negligible additional lateral inflow. To address temporal autocorrelation depicting antecedent water content, we add cumulative lags for Precip and ET and averaged lags for SM over the preceding biweekly and monthly windows (lag1 and lag2, respectively). Optical and multispectral imageries, as well as their spectral indices from Sentinel-2 and Landsat 8/9, were also collected but later dropped due to insufficient acquisition and high cloud cover during the biweekly window. Static predictors represent peat and topographic drivers, incorporating peat thickness (calibrated using 1,111-field samplings), distance to major natural stream, elevation, and slope. Specifically, for the LiDAR-derived elevation model, we first filled the sinks and pits, then divided the elevation into surface and subsurface rasters by subtracting the surface elevation from the peat thickness. The slope was then computed for both rasters. The inclusion of surface and subsurface elevations and their derivatives was intended to proxy acrotelmic and catotelmic flows following the scalar model mechanism formulated by 39 . All terrain features except slopes ( i.e. , flow direction and accumulation, topographic wetness index, terrain roughness and ruggedness indices) were excluded due to artifacts and less variability generated by extensive flat slopes. We appended all static rasters to align with the biweekly dynamic stacks. To control for interannual and seasonal autocorrelation, eight dummy factors (year 2019–2025 and dry-rainy seasons, respectively) were hot-encoded and included in the dataset. The final predictors (totalling 27) were derived from an initial pool of 42 candidate variables after applying a multicollinearity screening criterion of |r| < 0.8. All ET rasters were dropped due to their multicollinearity with S. All raster preprocessing, stacking, and model input generation were conducted in R and RStudio 63 , using terra 64 and tidyverse 65 . Data Acquisition and Pre-processing This study leveraged data-driven modelling to predict groundwater level (GWL) using a combination of dynamic, static, and time-related predictors (Supplementary Information 1). The GWL data were gathered from 134 automatic loggers and manual dipwells located throughout PHU Rupat Island as part of the regulatory compliance to the Indonesian Ministry of Environment and Forestry (MoEF; now Ministry of Environment-Environmental Control Agency of Indonesia/KLH-BPLH). We selected the most observed period covering 2019 to 2025 from Simatag0,4m platform ( https://sippeg.menlhk.go.id/apps/sippegapps/tmat ) and obtained (aggregated) 8,327 biweekly records in total. All readings exceeding + 10 cm were excluded (which may indicate input errors), while the entire negative GWL was selected. All predictors were selected for their hydrological relationship with GWL spatiotemporal dynamics in tropical peatland landscapes. The constructed model conceptually represents the influence of climate, vegetation, topography, and soil on peatland hydrology. We also included several proxies for temporal autocorrelation, including lagged variables and year and season dummies. The dynamic inputs representing climate and vegetation drivers consisted of 158 biweekly precipitation (Precip) and evapotranspiration (ET) composites (2019–2025), resampled at 10-meter resolution. Biweekly Precip was aggregated from daily 0.05° CHIRPS 61 . ET estimates were derived from the MODIS MOD16A2 8-day evapotranspiration product 62 . We selected the VV band from Sentinel-1, which represents surface radar scattering, as a proxy for peat soil moisture (SM). Furthermore, we decomposed partial storage proxy (S) as an additional dynamic predictor, following S = Precip – ET, where S represents both storage and lateral outflow, and GWL is technically an upper table of S 39,53 . Since the studied PHU is separated from other higher grounds by a large river (Fig. 1 ), we assumed that the studied peatland is ombrotrophic and thus received negligible additional lateral inflow. To address temporal autocorrelation depicting antecedent water content, we add cumulative lags for Precip and ET and averaged lags for SM over the preceding biweekly and monthly windows (lag1 and lag2, respectively). Optical and multispectral imageries, as well as their spectral indices from Sentinel-2 and Landsat 8/9, were also collected but later dropped due to insufficient acquisition and high cloud cover during the biweekly window. Static predictors represent peat and topographic drivers, incorporating peat thickness (calibrated using 1,111-field samplings), distance to major natural stream, elevation, and slope. Specifically, for the LiDAR-derived elevation model, we first filled the sinks and pits, then divided the elevation into surface and subsurface rasters by subtracting the surface elevation from the peat thickness. The slope was then computed for both rasters. The inclusion of surface and subsurface elevations and their derivatives was intended to proxy acrotelmic and catotelmic flows following the scalar model mechanism formulated by 39 . All terrain features except slopes ( i.e. , flow direction and accumulation, topographic wetness index, terrain roughness and ruggedness indices) were excluded due to artifacts and less variability generated by extensive flat slopes. We appended all static rasters to align with the biweekly dynamic stacks. To control for interannual and seasonal autocorrelation, eight dummy factors (year 2019–2025 and dry-rainy seasons, respectively) were hot-encoded and included in the dataset. The final predictors (totalling 27) were derived from an initial pool of 42 candidate variables after applying a multicollinearity screening criterion of |r| < 0.8. All ET rasters were dropped due to their multicollinearity with S. All raster preprocessing, stacking, and model input generation were conducted in R and RStudio 63 , using terra 64 and tidyverse 65 . GWL Model Training The GWL data and the extracted input data frame were extensively checked and cleaned by applying summary statistics, histograms, and correlations, and by iteratively removing outliers using the interquartile range (IQR) method. The cleaned dataset was then partitioned into a training and a validation matrix for further processing. Due to its widely known quick training time and accurate prediction reported previously 14 , 15 , 66 , 67 , Extreme Gradient Boosting (XGB) was selected as a regression algorithm to predict GWL. The final hyperparameters selected were: nrounds = 1460, eta = 0.01, max_depth = 9, colsample_bytree = 0.6, min_child_weight = 3, and subsample = 1.0 with out-of-bag/OOB and fold performance evaluated by root mean squared error/RMSE. Tuning was conducted gradually from broad to narrow ranges via 5-fold cross-validation. Lastly, 10-fold cross-validation was repeated 10 times to evaluate the final model. The training process was executed in R using caret 68 and XGBoost packages 66 . GWL Model Evaluation The final XGB model was evaluated using a custom Monte Carlo Cross-Validation (MCCV). This approach was applied separately to the complete training and testing datasets to examine model stability and error generalization. The MCCV procedure consisted of 25 repeated random resampling, with 80% of the data allocated for training and 20% for validation (reps = 25; train_frac = 0.8). To maintain meaningful data splits, a minimum of 100 samples was required for the training partition and 50 for the validation partition. In each of the 25 MCCV iterations, a new XGB model was refitted with a fixed, previously tuned hyperparameters. Standard evaluation indicators, i.e. , R 2 , RMSE, MAE, and BIAS, were calculated for each training and testing subset. All results were summarized using boxplots. The entire evaluation method was conducted in R using tidyverse 65 . GWL Model Explanation We interpreted the model functioning through Shapley value of game theory explanation 69 . The Shapley value was determined by employing kernel SHAP approach 70 , focusing on tree-specific Kernel SHAP 71 , 72 . SHAP values were computed for the training dataset using a Monte Carlo estimation approach, which assigned an additive importance score to every feature for each individual prediction. Global model interpretation was assessed by aggregating these values, specifically ranking feature importance based on the mean absolute SHAP value to identify the dominant drivers of GWL dynamics besides the combined visualization of their directionality and magnitude. SHAP dependence plots were analyzed to characterize non-linear relationships and interaction effects between primary drivers and their contribution to the model output. Furthermore, a stacked force plot ordered by predicted GWL was constructed to reveal cumulative feature effects and model stability across the entire prediction range from shallow to deep levels. To examine local model behavior, we generated individual force plots for representative GWL quantiles (0%, 25%, 50%, 75%, and 99%). All interpretation workflows, including the visualization of feature contributions and interaction effects, were conducted using the fastshap 73 , shapviz 74 , and SHAPforxgboost packages 75 . Upscaling GWL Prediction to Rs Estimation using CO 2 -GWL Response Function This study used the resulting biweekly GWL maps as a base to upscale Rs emissions using the CO 2 -GWL response curve applied by 14,15 in the seven Indonesian peatland units (Peatland Hydrological Unit/PHU or Kesatuan Hidrologi Gambut/KHG) in Sumatra and Kalimantan (Supplementary Information 3). They developed logistic general and land-use-based response curves after compiling 25 chamber-based emission studies conducted in Southeast Asian peatlands (organic layer > 50 cm), representing total soil respiration (Rs; autotrophic + heterotrophic respirations). The first authors applied a general response curve, whereas the latter employed a combination of general and land-use-based response curves to generate biweekly Rs maps from XGB-predicted GWL maps with the same spatiotemporal resolutions. Specifically for the response curve in the primary forest, bi-logistic functions derived from 52 were selected. Moreover, the upscaled results were compared to state-of-the-art similar compilations and Rs-GWL responses developed previously by 37,38 , as well as Rh-based EFs from Tier-1 IPCC 11 and Indonesian forest reference emission level/FREL 13 . To further visualize the response curves, this study simulated the relative change in peat Rs (reported as a percentage) in response to a 10-cm change in GWL (from the current annual GWL to the new target annual GWL), following 48 . Declarations Acknowledgements The authors gratefully thanked all office and field staff in helping during field campaign. Special acknowledgement to the IMPLI Project in supporting funds for this study. Author contributions statement Conceptualization, W.Y.U., B.B.; methodology, data collection, formal analyses, and visualization W.Y.U.; writing, original draft preparation, W.Y.U.; review and editing, W.Y.U., B.B., S.A., S.D.T., F.K.A.; supervision, S.A., S.D.T., W.Y.U., F.K.A., M.A.; project administration, W.Y.U., F.K.A., M.A.; funding acquisition, W.Y.U, F.K.A., M.A. Competing Interests The authors declare no competing interests. Funding This research was funded by Integrated Management of Peatland Landscapes in Indonesia/IMPLI Grant Number No. 2P1GN48A, by GEF-6 (Global Environment Facility-6) 2021–2026. References Page, S. E., Rieley, J. O. & Banks, C. J. Global and regional importance of the tropical peatland carbon pool. Glob Chang. Biol. 17 , 798–818 (2011). Xu, J., Morris, P. J., Liu, J. & Holden, J. P. E. A. T. M. A. P. Refining estimates of global peatland distribution based on a meta-analysis. Catena (Amst) . 160 , 134–140 (2018). Anda, M. et al. Revisiting tropical peatlands in Indonesia: Semi-detailed mapping, extent and depth distribution assessment. Geoderma 402 , (2021). Miettinen, J., Hooijer, A., Vernimmen, R., Liew, S. C. & Page, S. E. From carbon sink to carbon source: extensive peat oxidation in insular Southeast Asia since 1990. Environ. Res. Lett. 12 , 24014 (2017). Girkin, N. T. et al. Tropical peatlands in the Anthropocene: The present and the future. Anthropocene 40 , 100354 (2022). Austin, K. G. et al. Mismatch Between Global Importance of Peatlands and the Extent of Their Protection. Conserv Lett 18 , (2025). Hooijer, A. et al. Current and future CO2 emissions from drained peatlands in Southeast Asia. Biogeosciences 7 , 1505–1514 (2010). Hodgkins, S. B. et al. Tropical peatland carbon storage linked to global latitudinal trends in peat recalcitrance. Nat Commun 9 , (2018). Mander, Ü., Öpik, M. & Espenberg, M. Global peatland greenhouse gas dynamics: state of the art, processes, and perspectives. New Phytologist vol. 246 94–102 Preprint at (2025). https://doi.org/10.1111/nph.20436 Dadap, N. C. et al. Climate change-induced peatland drying in Southeast Asia. Environmental Res. Letters 17 , (2022). Hiraishi, T. et al. Supplement to the 2006 IPCC Guidelines for National Greenhouse Gas Inventories: Wetlands : Methodological Guidance on Lands with Wet and Drained Soils, and Constructed Wetlands for Wastewater Treatment . (IPCC, Intergovernmental Panel on Climate Change, 2014). (IPCC, Intergovernmental Panel on Climate Change, 2014). (2013). Sasmito, S. D. et al. Half of land use carbon emissions in Southeast Asia can be mitigated through peat swamp forest and mangrove conservation and restoration. Nature Communications 16 , (2025). Budiharto et al. National Forest Reference Level for Deforestation, Forest Degradation and Enhancement of Forest Carbon Stock. In the Context of Decision 12/CP.17 Para 12 UNFCCC . (2022). https://redd.unfccc.int/files/2nd_frl_indonesia_final_submit.pdf Aini, F. K. et al. AsiaFlux2025, Pangkalan Kerinci, Riau, Indonesia,. A Synthesis of Groundwater Level-Driven CO2 Dynamics: A Review of Different Land-Use Systems in Sumatra and Borneo, Indonesia. in AsiaFlux Conference 2025 (2025). Pulunggono, H. B. et al. AsiaFlux2025, Pangkalan Kerinci, Riau, Indonesia,. CO2 Emission Estimation from Modelled Groundwater Table in Indonesian Tropical Peatland Ecosystems. in AsiaFlux Conference 2025 (2025). Hirano, T. et al. Impact of Land Use Change and Drought on the Net Emissions of Carbon Dioxide and Methane From Tropical Peatlands in Southeast Asia. AGU Advances 6 , (2025). Tay, C. et al. Satellite radar advances carbon emissions accountability over tropical peat. Commun Earth Environ 6 , (2025). Apers, S. et al. Tropical Peatland Hydrology Simulated With a Global Land Surface Model. J Adv. Model. Earth Syst 14 , (2022). Urzainki, I. et al. A process-based model for quantifying the effects of canal blocking on water table and CO2 emissions in tropical peatlands. Biogeosciences 20 , 2099–2116 (2023). Cochrane, M. A. Spatiotemporal Variations in Hydrology Drive Greenhouse Gas Emissions in Tropical Peatlands. AGU Advances 7 , (2026). Hirano, T. et al. Large variation in carbon dioxide emissions from tropical peat swamp forests due to disturbances. Commun Earth Environ 5 , (2024). Deshmukh, C. S. et al. Net greenhouse gas balance of fibre wood plantation on peat in Indonesia. Nature 616 , 740–746 (2023). Rankin, T., Roulet, N., Humphreys, E., Peichl, M. & Jӓrveoja, J. Partitioning autotrophic and heterotrophic respiration in an ombrotrophic bog. Front Earth Sci. (Lausanne) 11 , (2023). Rankin, T. E., Roulet, N. T. & Moore, T. R. Controls on autotrophic and heterotrophic respiration in an ombrotrophic bog. Biogeosciences 19 , 3285–3303 (2022). Girkin, N. T., Turner, B. L., Ostle, N., Craigon, J. & Sjögersten, S. Root exudate analogues accelerate CO2 and CH4 production in tropical peat. Soil. Biol. Biochem. 117 , 48–55 (2018). Urzainki, I. et al. Canal blocking optimization in restoration of drained peatlands. Biogeosciences 17 , 4769–4784 (2020). Mukhaiyar, U. et al. The generalized STAR modelling with three-dimensional of spatial weight matrix in predicting the Indonesia peatland’s water level. Environ Sci. Eur 36 , (2024). Widiarso, B. et al. Predicting peatland groundwater table and soil moisture dynamics affected by drainage level. Sains Tanah . 17 , 42–49 (2020). Hikouei, I. S. et al. Using machine learning algorithms to predict groundwater levels in Indonesian tropical peatlands. Sci. Total Environ. 857 , 159701 (2023). Hikouei, I. S. et al. Machine-learning based spatiotemporal performance analysis of degraded tropical peatland groundwater level numerical model. Groundw. Sustain. Dev. 29 , 101413 (2025). Li, L. et al. Estimation of Ground Water Level (GWL) for Tropical Peatland Forest Using Machine Learning. IEEE Access. 10 , 126180–126187 (2022). Yonekura, K. et al. Prediction of groundwater level in Indonesian tropical peatland forest plantations using machine learning. Artificial Intell. Geosciences 6 , (2025). Koch, J. et al. Water-table-driven greenhouse gas emission estimates guide peatland restoration at national scale. Biogeosciences 20 , 2387–2403 (2023). Bechtold, M. et al. Large-scale regionalization of water table depth in peatlands optimized for greenhouse gas emission upscaling. Hydrol. Earth Syst. Sci. 18 , 3319–3339 (2014). Widyatmanti, W. et al. Codification to secure Indonesian peatlands: From policy to practices as revealed by remote sensing analysis. Soil. Secur. 9 , 100080 (2022). Regulation No 57/2016. Perlindungan Dan Pengelolaan Ekosistem Gambut (Indonesian Government, 2014). Prananto, J. A., Minasny, B., Comeau, L., Rudiyanto, R. & Grace, P. Drainage increases CO2 and N2O emissions from tropical peat soils. Glob Chang. Biol. 26 , 4583–4600 (2020). Novita, N. et al. Geographic Setting and Groundwater Table Control Carbon Emission from Indonesian Peatland: A Meta-Analysis. Forests 12 , 832 (2021). Cobb, A. R. & Harvey, C. F. Scalar Simulation and Parameterization of Water Table Dynamics in Tropical Peatlands. Water Resour. Res. 55 , 9351–9377 (2019). Fauziah, Prasetyo, L. B., Saribanon, N. & Hayati, N. Vulnerability of peatland fires in bengkalis regency during the ENSO El nino phase using a machine learning approach. MethodsX 14 , 103128 (2025). Utomo, B., Oktavia, M., Susilo, Y. & Putri, M. K. Identification of drought endemic areas in Musi Banyuasin regency. Kuwait J. Sci. 50 , 168–173 (2023). Astiani, D., Burhanuddin, B., Curran, L. M., Mujiman, M. & Salim, R. Effects of Drainage Ditches on Water Table Level, Soil Conditions and Tree Growth of Degraded Peatland Forests in West Kalimantan. Indonesian J. Forestry Res. 4 , 15–25 (2017). Evans, C. D. et al. Rates and spatial variability of peat subsidence in Acacia plantation and forest landscapes in Sumatra, Indonesia. Geoderma 338 , 410–421 (2019). Parish, F., Afham, A. & Lew, S. Y. (Serena). Role of the Roundtable on Sustainable Palm Oil (RSPO) in Tropical Peatland Management. in Tropical Peatland Eco-management 509–533Springer Singapore, (2021). 10.1007/978-981-33-4654-3_18 Hooijer, A. et al. Subsidence and carbon loss in drained tropical peatlands. Biogeosciences 9 , 1053–1071 (2012). Jauhiainen, J., Hooijer, A. & Page, S. E. Carbon dioxide emissions from an Acacia plantation on peatland in Sumatra, Indonesia. Biogeosciences 9 , 617–630 (2012). Hirano, T. et al. Effects of disturbances on the carbon balance of tropical peat swamp forests. Glob Chang. Biol. 18 , 3410–3422 (2012). McCalmont, J. et al. Short- and long-term carbon emissions from oil palm plantations converted from logged tropical peat swamp forest. Glob Chang. Biol. 27 , 2361–2376 (2021). Weaver, M. M., Garner, A. J., Samanta, D., Mann, M. E. & Horton, B. P. Seasonal variations of tropical cyclone genesis and landfall patterns impacting Southeast Asia in a warmer climate. Commun. Earth Environ. 6 , 866 (2025). Loisel, J. et al. Expert assessment of future vulnerability of the global peatland carbon sink. Nat. Clim. Chang. 11 , 70–77 (2021). Dadap, N. C. et al. Drainage Canals in Southeast Asian Peatlands Increase Carbon Emissions. AGU Advances 2 , (2021). Ishikura, K. et al. Carbon Dioxide and Methane Emissions from Peat Soil in an Undrained Tropical Peat Swamp Forest. Ecosystems 22 , 1852–1868 (2019). Page, S., Hooijer, A., Rieley, J., Banks, C. & Hoscilo, A. The tropical peat swamps of Southeast Asia: in Biotic Evolution and Environmental Change in Southeast Asia 406–433 (Cambridge University Press, doi: 10.1017/CBO9780511735882.018 . (2012). Deshmukh, C. S. et al. Conservation slows down emission increase from a tropical peatland in Indonesia. Nat. Geosci. 14 , 484–490 (2021). Kiew, F. et al. Carbon dioxide balance of an oil palm plantation established on tropical peat. Agric Meteorol 295 , (2020). RePPProT Regional Physical Planning Programme for Transmigration (Reppprot) . (1987). He, H. & Roulet, N. T. Improved estimates of carbon dioxide emissions from drained peatlands support a reduction in emission factor. Commun Earth Environ 4 , (2023). Lestari, I., Murdiyarso, D. & Taufik, M. Rewetting Tropical Peatlands Reduced Net Greenhouse Gas Emissions in Riau Province, Indonesia. Forests 13 , 505 (2022). Hooijer, A. et al. Benefits of tropical peatland rewetting for subsidence reduction and forest regrowth: results from a large-scale restoration trial. Sci Rep 14 , (2024). Novita, N. et al. Strong climate mitigation potential of rewetting oil palm plantations on tropical peatlands. Sci. Total Environ. 952 , 175829 (2024). Funk, C. et al. The climate hazards infrared precipitation with stations—a new environmental record for monitoring extremes. Sci Data 2 , (2015). Qiaozhen Mu, M., Zhao & Steven, W. Running. MODIS Global Terrestrial Evapotranspiration (ET) Product (NASA MOD16A2/A3) Algorithm Theoretical Basis Document Collection 5 . (2013). RCoreTeam., R. A Language and Environment for Statistical Computing. Preprint at (2026). Robert, J. et al. terra: Spatial Data Analysis, version 1.7–82. Preprint at (2026). (2026). Hadley Wickham, M. & Girlich Kirill Müller & Davis Vaughan. tidyverse: Easily Install and Load the ‘Tidyverse’. Preprint at (2025). Chen, T., Guestrin, C. & XGBoost: A Scalable Tree Boosting System. in Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining 785–794ACM, (2016). 10.1145/2939672.2939785 Hikouei, I. S. et al. Using machine learning algorithms to predict groundwater levels in Indonesian tropical peatlands. Sci. Total Environ. 857 , 159701 (2023). Max et al. caret: Classification and Regression Training. Preprint at (2025). Shapley, L. S. Cambridge University Press,. A value for n-person games. in The Shapley Value 31–40 (1988). 10.1017/CBO9780511528446.003 Lundberg, S. M. & Lee, S. I. A unified approach to interpreting model predictions. in NIPS’17: Proceedings of the 31st International Conference on Neural Information Processing Syste 4768–4777Curran Associates Inc., New York, United States, (2017). Lundberg, S. M. et al. Explainable machine-learning predictions for the prevention of hypoxaemia during surgery. Nat. Biomed. Eng. 2 , 749–760 (2018). Lundberg, S. M. et al. From local explanations to global understanding with explainable AI for trees. Nat. Mach. Intell. 2 , 56–67 (2020). Brandon Greenwell. fastshap: Fast Approximate Shapley Values. Preprint at. (2025). Michael Mayer & Adrian Stando. shapviz: SHAP Visualizations. Preprint at (2025). Liu, Y., Just, A. & Mayer, M. SHAPforxgboost: SHAP Plots for ‘XGBoost’. Additional Declarations No competing interests reported. Supplementary Files SupplementaryInformation.docx Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-9002157","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":603431320,"identity":"56d90ca9-0351-4a52-846a-5560dcf1668c","order_by":0,"name":"Waluyo Yogo Utomo","email":"data:image/png;base64,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","orcid":"","institution":"IPB University","correspondingAuthor":true,"prefix":"","firstName":"Waluyo","middleName":"Yogo","lastName":"Utomo","suffix":""},{"id":603431322,"identity":"548641f1-63c2-443e-ad63-0b0dfb4e4188","order_by":1,"name":"Fitri Khusyu Aini","email":"","orcid":"","institution":"Ministry of Environment-Environmental Control Agency of Indonesia (MoE-EPA)","correspondingAuthor":false,"prefix":"","firstName":"Fitri","middleName":"Khusyu","lastName":"Aini","suffix":""},{"id":603431324,"identity":"a93d67e1-f623-459f-9fac-4b68cb32f14e","order_by":2,"name":"Muhammad Askary","email":"","orcid":"","institution":"Ministry of Environment-Environmental Control Agency of Indonesia (MoE-EPA)","correspondingAuthor":false,"prefix":"","firstName":"Muhammad","middleName":"","lastName":"Askary","suffix":""},{"id":603431325,"identity":"4e9510c1-b016-4343-86df-dfb86634d607","order_by":3,"name":"Syaiful Anwar","email":"","orcid":"","institution":"IPB University","correspondingAuthor":false,"prefix":"","firstName":"Syaiful","middleName":"","lastName":"Anwar","suffix":""},{"id":603431326,"identity":"b84ad357-bfd1-41b9-8860-c59da2ac9a92","order_by":4,"name":"Suria Darma Tarigan","email":"","orcid":"","institution":"IPB University","correspondingAuthor":false,"prefix":"","firstName":"Suria","middleName":"Darma","lastName":"Tarigan","suffix":""},{"id":603431327,"identity":"bb869695-447c-41d9-8473-dfe8d05901c2","order_by":5,"name":"Baba Barus","email":"","orcid":"","institution":"IPB University","correspondingAuthor":false,"prefix":"","firstName":"Baba","middleName":"","lastName":"Barus","suffix":""}],"badges":[],"createdAt":"2026-03-01 14:38:11","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-9002157/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-9002157/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":104405617,"identity":"da668b58-b332-43be-b4fd-5c1d7abae5bb","added_by":"auto","created_at":"2026-03-11 12:23:27","extension":"jpg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":191123,"visible":true,"origin":"","legend":"\u003cp\u003ePHU Rupat Island location depicting its peat thickness and GWL piezometer spatial distributions. Indonesian and global peatland coverage is derived from \u003csup\u003e2,3\u003c/sup\u003e, respectively.\u003c/p\u003e","description":"","filename":"1.jpg","url":"https://assets-eu.researchsquare.com/files/rs-9002157/v1/6abb2de5eac40a1d36e6a934.jpg"},{"id":104291673,"identity":"8aa3b82e-0679-419e-8eb6-c99fd890b0de","added_by":"auto","created_at":"2026-03-10 07:06:21","extension":"jpg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":218331,"visible":true,"origin":"","legend":"\u003cp\u003eGeneral information concerning GWL and its predictors. \u003cstrong\u003e(a) \u003c/strong\u003eall GWL distributions (all, train, and test datasets), \u003cstrong\u003e(b) \u003c/strong\u003ecorrelations of all predictors (excluding all dummy variables) and \u003cstrong\u003e(c) \u003c/strong\u003eboxplots depicts the distribution of all explanatory variables within 10 cm-bins of GWL.\u003c/p\u003e","description":"","filename":"2.jpg","url":"https://assets-eu.researchsquare.com/files/rs-9002157/v1/df57ae95b2b5c5a32a2f8427.jpg"},{"id":104291671,"identity":"a2d026d6-2c94-438e-86aa-f63b1c597556","added_by":"auto","created_at":"2026-03-10 07:06:21","extension":"jpg","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":488188,"visible":true,"origin":"","legend":"\u003cp\u003eTemporal dynamics of predicted against observed GWL (classified as training and validation datasets). NA’s flagged GWLs (displayed as grey dots) are due to outlier predictors after 10-cm GWL binning (see Fig. 2-upper)\u003c/p\u003e","description":"","filename":"3.jpg","url":"https://assets-eu.researchsquare.com/files/rs-9002157/v1/2d70c2df2c42f9fa2ea0fa11.jpg"},{"id":104291672,"identity":"bdced932-e9ae-4692-b93e-598347ff6c30","added_by":"auto","created_at":"2026-03-10 07:06:21","extension":"jpg","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":144599,"visible":true,"origin":"","legend":"\u003cp\u003eSpatiotemporal dynamics of predicted GWL, pooled by rainy and dry seasons and annual courses. Uncertainty datasets are presented in Supplementary Information 4.\u003c/p\u003e","description":"","filename":"4.jpg","url":"https://assets-eu.researchsquare.com/files/rs-9002157/v1/407dd170dc0a90d316d376dc.jpg"},{"id":104405304,"identity":"47baac18-298c-4856-80c4-01e490fb36fb","added_by":"auto","created_at":"2026-03-11 12:22:32","extension":"jpg","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":195522,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003e(a) \u003c/strong\u003eAccuracy metrics derived from MCCV calculation. \u003cstrong\u003e(b.1)\u003c/strong\u003e Beeswarm plot (left) and \u003cstrong\u003e(b.2)\u003c/strong\u003e importance barplot (right) representing distribution of predictors’s SHAP values and importance of each predictor computed from mean |SHAP| value.\u003c/p\u003e","description":"","filename":"5.jpg","url":"https://assets-eu.researchsquare.com/files/rs-9002157/v1/fbf9543c2e073586a5267386.jpg"},{"id":104291678,"identity":"62186966-3da3-4861-8993-8ef49f25eba7","added_by":"auto","created_at":"2026-03-10 07:06:22","extension":"jpg","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":155373,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003e(a) \u003c/strong\u003eSHAP-based force plot of individual observation representing percentiles and mean GWL and \u003cstrong\u003e(b)\u003c/strong\u003e dynamic of predictor’s contributions over all GWL ranges as represented by stacked force plot of 10 most influential variables plus the rest. Note that x axis (representing GWL) is ordered descending.\u003c/p\u003e","description":"","filename":"6.jpg","url":"https://assets-eu.researchsquare.com/files/rs-9002157/v1/3ff2b43d1b48a3620d8360b8.jpg"},{"id":104291677,"identity":"91acba05-bec6-49e5-85ce-4eddc716dbfc","added_by":"auto","created_at":"2026-03-10 07:06:21","extension":"jpg","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":247587,"visible":true,"origin":"","legend":"\u003cp\u003eCO\u003csub\u003e2\u003c/sub\u003e emission from Rs (upscaled to annual Mg CO\u003csub\u003e2\u003c/sub\u003e ha\u003csup\u003e-1\u003c/sup\u003e y\u003csup\u003e-1\u003c/sup\u003e) emitted within dry and rainy seasons and annual total (cumulative over 24-biweekly within a year) in PHU Pulau Rupat. The Rs maps are estimated using \u003cstrong\u003e(a)\u003c/strong\u003e general CO\u003csub\u003e2\u003c/sub\u003e-GWL response curve and \u003cstrong\u003e(b)\u003c/strong\u003e combination of land use-based and general CO\u003csub\u003e2\u003c/sub\u003e-GWL response curves\u003csup\u003e14,15\u003c/sup\u003e. Uncertainty datasets are presented in Supplementary Information 5 and 6.\u003c/p\u003e","description":"","filename":"7.jpg","url":"https://assets-eu.researchsquare.com/files/rs-9002157/v1/294a2dcec45509b1f26e2735.jpg"},{"id":104405594,"identity":"e8b606b4-5f3b-4540-aff7-4d2e1dff3fb6","added_by":"auto","created_at":"2026-03-11 12:23:22","extension":"jpg","order_by":8,"title":"Figure 8","display":"","copyAsset":false,"role":"figure","size":239550,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003e(a) \u003c/strong\u003esimulation of CO\u003csub\u003e2\u003c/sub\u003e emission changes based on the 10-cm change of new annual target GWL relative to current annual GWL following \u003csup\u003e48\u003c/sup\u003e . \u003cstrong\u003e(b.1) \u003c/strong\u003ecomparison of annual averaged Rs (Mg CO\u003csub\u003e2\u003c/sub\u003e ha\u003csup\u003e-1\u003c/sup\u003e y\u003csup\u003e-1\u003c/sup\u003e) and \u003cstrong\u003e(b.2) \u003c/strong\u003eannual cumulative averaged Rs (Mt CO\u003csub\u003e2\u003c/sub\u003e ha\u003csup\u003e-1\u003c/sup\u003e y\u003csup\u003e-1\u003c/sup\u003e) emitted from pulau Rupat PHU based on dynamic EFs employing general and combination of general\u003csup\u003e15\u003c/sup\u003e and land use-based CO\u003csub\u003e2\u003c/sub\u003e-GWL response curves\u003csup\u003e14\u003c/sup\u003e, published reports \u003csup\u003e37,38\u003c/sup\u003e as well as static EFs from Tier 1 IPCC (depicted as IPCC)\u003csup\u003e11\u003c/sup\u003e and Indonesian FREL\u003csup\u003e13\u003c/sup\u003e\u003c/p\u003e","description":"","filename":"8.jpg","url":"https://assets-eu.researchsquare.com/files/rs-9002157/v1/377807421bf6639ba14761c6.jpg"},{"id":107706048,"identity":"1915a55d-f953-43cd-b877-697bda5a44d8","added_by":"auto","created_at":"2026-04-24 09:17:14","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":2576501,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-9002157/v1/b0a4cad5-6a9c-4232-9a59-3d90436d7b00.pdf"},{"id":104291679,"identity":"998915e7-eb38-44e0-991b-5e8818497cc0","added_by":"auto","created_at":"2026-03-10 07:06:22","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":3674432,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementaryInformation.docx","url":"https://assets-eu.researchsquare.com/files/rs-9002157/v1/9ad04706d14b51b72be07fa6.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"Machine Learning-Aided Groundwater Level and CO 2 Emission Estimations in Tropical Peatlands using Global and Regional Scenarios","fulltext":[{"header":"Introduction","content":"\u003cp\u003eEven though only covering only 2.5\u0026ndash;3% of the global land area, peatlands are regarded as the most carbon-rich ecosystems with estimated 450\u0026ndash;650 Pg of carbon storage, surpassing all the world\u0026rsquo;s forests combined. Indonesia contains approximately 36\u0026thinsp;\u0026plusmn;\u0026thinsp;6 Pg C in its peat layer, accounting for about 6\u0026ndash;8% of global peat-carbon stocks while occupying only 10% of tropical peatland area. Collectively, these peat layers represent nearly one-third of global soil organic carbon within a small fraction of the terrestrial surface\u003csup\u003e\u003cspan additionalcitationids=\"CR2\" citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e\u003c/sup\u003e. Despite their conservation importance for harbouring biodiversity and storing carbon, peatlands are also being subjected for collection of natural resources and converted for developing forestry and agricultural commodities\u003csup\u003e\u003cspan additionalcitationids=\"CR5\" citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e\u003c/sup\u003e. The development of peatland swamps requires drier condition by drain out excessive water. Lowered groundwater level/GWL could trigger peat drying and thus accelerate peat decomposition. In the Southeast Asian region, the peatlands potentially emit around 30 to 100 Mg C ha⁻\u0026sup1; y⁻\u0026sup1; per metre drawdown and around 2.5 Gt C cumulative since 1990\u003csup\u003e4,7\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eAlbeit their higher recalcitrance preventing excessive organic matter oxidation\u003csup\u003e\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e,\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e\u003c/sup\u003e, some equatorial peats are reportedly having higher sensitivity to GWL dynamics, linked to regional climate changes and inter-annual rainfall variations, \u003cem\u003ei.e.\u003c/em\u003e, El-Nino\u003csup\u003e\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e\u003c/sup\u003e, CO\u003csub\u003e2\u003c/sub\u003e emission factors/EFs satisfied these dynamic conditions cannot be predicted by applying static land use-based estimates, \u003cem\u003ee.g.\u003c/em\u003e, Tier-1 IPCC EFs\u003csup\u003e\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e\u003c/sup\u003e (for example \u003csup\u003e\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e,\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e\u003c/sup\u003e) and Indonesian forest reference emission level/FREL\u003csup\u003e13\u003c/sup\u003e. It requires more detailed dataset concerning national and sub-national emission estimates that not only capable to predict emissions on inter-year and land uses, but also incorporates within-year and -land uses heterogeneities. Previous studies had been explored these possibilities using more practical proxies mainly governing CO\u003csub\u003e2\u003c/sub\u003e emission, \u003cem\u003ee.g.\u003c/em\u003e, GWL\u003csup\u003e\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e,\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e\u003c/sup\u003e, precipitation-GWL\u003csup\u003e\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e\u003c/sup\u003e and subsidence\u003csup\u003e\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e\u003c/sup\u003e, or more exhaustive process-based modelling \u003cem\u003ee.g.\u003c/em\u003e, PEATCLSM\u003csup\u003e\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e\u003c/sup\u003e, and CNM-PHM\u003csup\u003e\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e\u003c/sup\u003e. Ultimately, current studies on reporting CO\u003csub\u003e2\u003c/sub\u003e emissions had been shifted from calculating CO\u003csub\u003e2\u003c/sub\u003e emitted solely from peat degradation/Rh\u003csup\u003e4,11\u0026ndash;13\u003c/sup\u003e to the real amount of CO\u003csub\u003e2\u003c/sub\u003e exchanged on ecosystem scale\u003csup\u003e\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e,\u003cspan additionalcitationids=\"CR21\" citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e\u003c/sup\u003e. In this context, upscaling the already partitioned Rh and autotrophic respiration/Ra would introduce higher uncertainty\u003csup\u003e\u003cspan additionalcitationids=\"CR24\" citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e\u003c/sup\u003e than directly estimating soil total respiration/Rs as basis dataset for ecosystem respiration to estimate net CO\u003csub\u003e2\u003c/sub\u003e ecosystem exchange.\u003c/p\u003e \u003cp\u003eDespite this importance, monitoring networks in tropical peatlands remain limited, particularly in Indonesia, where in-situ GWL stations are sparse and temporal resolution is often restricted to longer aggregation periods. As a result, rapid drawdown GWL caused by El Ni\u0026ntilde;o and canal drainage may go undetected. These process-based GWL models require extensive site-specific calibration and high-resolution inputs\u003csup\u003e\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e,\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e,\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e\u003c/sup\u003e, which constrains scalability for large-area simulation. Furthermore, numerical modeling methods such as geostatistics\u003csup\u003e\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e\u003c/sup\u003e, might not sufficiently capture the drivers of GWL changes and variability across unevenly distributed dipwells. Emerging data-driven approaches deliver promising alternatives for modeling GWL changes in tropical peatlands. In Indonesian and Malaysian peatlands, scattered research reported GWL predictions with reasonable accuracies using polynomial regression\u003csup\u003e\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e\u003c/sup\u003e, boosted tree\u003csup\u003e\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e,\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e\u003c/sup\u003e, and neural network\u003csup\u003e\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e,\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e\u003c/sup\u003e. In northern peatlands, several studies reported successful GWL prediction using random forest and boosted tree regressors\u003csup\u003e\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e,\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e\u003c/sup\u003e. While these machine learning (ML) studies made substantial progress toward peatlands GWL prediction, most remain localized and lack generalizability across broader eco-hydrological gradients.\u003c/p\u003e \u003cp\u003eTo address these challenges, this study presented Rs estimation based on data-driven GWL prediction utilized ML regressor that is trained from long-term dipwell measurements with dynamic and static predictors representing climate, vegetation, soil, and topography. The estimated Rs follows CO\u003csub\u003e2\u003c/sub\u003e-GWL response function reviewed from Southeast Asian chamber studies, which is vital as building block for modelling ecosystem-level CO\u003csub\u003e2\u003c/sub\u003e exchange.\u003c/p\u003e"},{"header":"Results","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003eGWL Prediction Results, Evaluations, and Explanations\u003c/h2\u003e \u003cdiv id=\"Sec4\" class=\"Section3\"\u003e \u003ch2\u003eGeneral Description of GWL and All Its Predictors\u003c/h2\u003e \u003cp\u003eThe descriptive statistics for groundwater level (GWL) and all predictors measured and collected within our study area (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e) and period (Jan-2019 to Apr-2025) are summarized and displayed in Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e and Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e, respectively (naming conventions are provided in Supplementary Information 1). Our study site covers a single Peat Hydrological Unit (PHU; in Indonesian terms: Kesatuan Hidrologis Gambut/KHG) on Rupat Island, Riau, Indonesia, which harbours various peat depths, multiple peat domes, and land uses (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e; Supplementary Information 2). A PHU represents an integrated peatland hydrological system bounded by rivers, coastal interfaces, and swamp areas, with peats interconnected with ecosystem components, \u003cem\u003ee.g.\u003c/em\u003e, water, vegetation, and other environmental elements\u003csup\u003e\u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e,\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e\u003c/sup\u003e. Most of the measured GWL data were concentrated on \u0026minus;\u0026thinsp;30 to \u0026minus;\u0026thinsp;40 cm, averaging around \u0026minus;\u0026thinsp;37.45\u0026thinsp;\u0026plusmn;\u0026thinsp;11.09 cm (average\u0026thinsp;\u0026plusmn;\u0026thinsp;standard deviation) and slightly skewed towards shallow water table (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003ea and \u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eb; Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e). Shallow GWL conditions (+\u0026thinsp;10 to \u0026minus;\u0026thinsp;20 cm) were mostly located on peat at more than 6 masl and deep peats (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003ec). All dynamic and static predictors exhibited lower chance to severe multicollinearities (r \u0026lt; |0.8|; Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eb), which can be tolerated by the XGBoost regressor. Precipitation, storage (representing precipitation minus evapotranspiration) and their antecedent terms (lag1 and 2) were distributed across all GWL bins. Current and lagged soil moisture proxies declined following GWL deepening (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003ec). Peats located within 5 km of the coast and natural streams mostly had moderate GWL (\u0026minus;\u0026thinsp;30 to \u0026minus;\u0026thinsp;50 cm; the latter had shallower table up to \u0026minus;\u0026thinsp;20 cm), but never experienced deeper GWL.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eSummary statistics of GWL and its predictors.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"11\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c8\" colnum=\"8\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c9\" colnum=\"9\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c10\" colnum=\"10\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c11\" colnum=\"11\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eVariable\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eUnits\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eMin\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eQ1\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eMedian\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eQ3\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c7\"\u003e \u003cp\u003eMax\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c8\"\u003e \u003cp\u003eMean\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c9\"\u003e \u003cp\u003eSD\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c10\"\u003e \u003cp\u003eSkew\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c11\"\u003e \u003cp\u003eKurt\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGWL\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003ecm\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e-80.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e-41.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e-37.84\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e-32.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e-1.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e-37.45\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e11.09\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e-0.52\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e-2.12\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePrecip_raw\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003emm\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e17.83\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e70.91\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e103.55\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e133.07\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e206.60\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e102.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e38.31\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e-0.09\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e2.16\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePrecip_raw_lag1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003emm\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e17.83\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e75.16\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e101.86\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e131.39\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e208.73\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e102.77\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e36.56\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.09\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e2.57\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePrecip_raw_lag2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003emm\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e1.16\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e77.69\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e101.13\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e133.03\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e222.44\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e104.13\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e37.07\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.15\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e2.52\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePrecip_cum_lag2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003emm\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e174.29\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e270.72\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e309.28\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e345.95\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e482.08\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e308.90\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e56.14\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.07\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e2.89\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eS_raw\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003emm\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e-517.99\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e-328.98\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e-270.11\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e-208.93\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e-64.22\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e-270.41\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e78.32\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e-0.16\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e2.38\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eS_raw_lag1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003emm\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e-525.94\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e-328.98\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e-275.27\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e-215.85\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e-107.14\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e-274.37\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e75.97\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e-0.24\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e2.51\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eS_raw_lag2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003emm\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e-498.89\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e-319.46\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e-261.78\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e-209.29\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e-63.74\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e-267.71\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e75.08\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e-0.27\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e2.54\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eS_cum_lag1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003emm\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e-753.88\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e-592.22\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e-546.33\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e-501.77\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e-345.35\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e-544.78\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e68.41\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.10\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e2.87\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eS_cum_lag2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003emm\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e-1166.80\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e-885.59\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e-817.64\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e-738.53\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e-466.55\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e-812.49\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e103.62\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.05\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e2.63\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSM_raw\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e%v/v\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.51\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.57\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.58\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.60\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.66\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.58\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.02\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e-0.08\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e2.79\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSM_raw_lag1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e%v/v\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.51\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.57\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.58\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.60\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.65\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.58\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.02\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e-0.06\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e2.80\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSM_raw_lag2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e%v/v\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.51\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.57\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.58\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.60\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.65\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.58\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.02\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e-0.02\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e2.75\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003epeat_thick\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003em\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.01\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e3.34\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e4.99\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e6.26\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e9.90\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e4.91\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e1.93\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.07\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e2.62\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eeucl_dist_ar\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003em\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e6.05\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e1362.10\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e2553.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e3845.79\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e8461.29\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e2787.59\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e1685.40\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.59\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e2.87\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFilled_subsurf\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003emasl\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e1.15\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e2.03\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e2.27\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e2.81\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e5.68\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e2.47\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.58\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e1.26\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e6.00\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFilled_surf\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003emasl\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e3.32\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e5.21\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e6.81\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e8.18\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e9.37\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e6.75\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e1.67\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e-0.08\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e1.69\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSlope_subsurf\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003erad\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.01\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.09\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.17\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.24\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.42\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.17\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.09\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.51\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e2.51\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSlope_surf\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003erad\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.04\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.07\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.12\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.21\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.08\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.05\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.71\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e2.62\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ecoast\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003em\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e875.04\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e4056.51\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e7743.58\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e12687.90\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e18077.62\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e8495.61\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e4917.45\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.27\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e1.75\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eFigure\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e.\u003c/p\u003e \u003cp\u003eTable\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e.\u003c/p\u003e \u003cp\u003eFigure\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e.\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e\n\u003ch3\u003eXGBoost GWL Model Prediction Result\u003c/h3\u003e\n\u003cp\u003eFigure\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e presents the temporal fluctuations in predicted versus observed GWL at selected dipwell stations representing wide range of elevation and peat depth gradients. Training and validation points (indicated in green and red, respectively) exhibited consistent results across both datasets, with several minor deviations observed during periods of high variability. Dry periods (highlighted with yellow shading) denoted intervals of reduced precipitation and occurred most frequently during seasonal drawdown events. Overall, the time-series results illustrated that the trained XGBoost model generalized throughout varied site conditions, ranging from deep peat to shallower peat. Predictive varied spatially according to peat thickness and surface elevation. At deep peat sites with thickness greater than 450 cm (SRL4_092_19_SRL and PTR_19_250_PRT), predicted GWL were stable with amplitude rarely exceeding\u0026thinsp;\u0026plusmn;\u0026thinsp;30 cm. The model also successfully predicted relatively deep GWL (approximately\u0026thinsp;\u0026minus;\u0026thinsp;55 to -60 cm) by following the original patterns in PTR_18_250_PRT and PTR_04_250_PRT. These conditions are also reflected in spatiotemporal predicted GWL maps averaged to dry, rainy, and annual courses (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e). Shallower water tables (\u0026minus;\u0026thinsp;s10 to \u0026minus;\u0026thinsp;30 cm) were consistently located around the central domes and less-drained peat areas. Meanwhile, deeper GWL (around \u0026minus;\u0026thinsp;40 to \u0026minus;\u0026thinsp;60 cm) were detected in peripheral zones near the coast, which are actually drained for forest plantations and agriculture (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e; Supplementary Information 2). The pooled GWL maps display moderate seasonal differences between the rainy and dry seasons, indicating that GWL is consistently deeper during the dry season than during the rainy season. Interannual variability from 2019 to 2024 was less pronounced, regardless of the averaging factor, with significant changes occurring during 2020 and 2022 (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eFigure\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e.\u003c/p\u003e \u003cp\u003eFigure\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e.\u003c/p\u003e\n\u003ch3\u003eXGBoost GWL Model Evaluation\u003c/h3\u003e\n\u003cp\u003eThe trained XGBoost model (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003ea) achieved moderate predictive accuracy in Monte Carlo cross-validation (MCCV), with median R\u003csup\u003e2\u003c/sup\u003e and RMSE of 0.46 and 7.25, respectively, including a median MAD and MAPE of 3.3 and 20.5%, respectively. In contrast, the model's performance on test sets were declined and showed greater variability across all error metrics relative to the training performances, with median R\u003csup\u003e2\u003c/sup\u003e, RMSE, MAD, and MAPE of 0.35, 8.0, 3.8, and 24.0%, respectively. The predicted GWL showed negligible bias (median\u0026thinsp;\u0026asymp;\u0026thinsp;0.0) on the training folds, whereas a slight positive median bias and increased variance were observed on the test folds.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eFigure\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e.\u003c/p\u003e\n\u003ch3\u003eXGBoost GWL Model Explanation\u003c/h3\u003e\n\u003cp\u003eThe SHaP analysis indicated that our XGBoost models relied primarily on static and antecedent climate predictors, with soil condition proxies as moderate contributors. Individual SHaP values spanned both positive and negative ranges, and the six highest mean |SHAP| values showed negative skewness. SHaP-based permutation method (nsim\u0026thinsp;=\u0026thinsp;500) detected that surface elevation, coastal proximity, and dummy year\u0026thinsp;=\u0026thinsp;2020 exhibited the highest mean |SHAP| values, followed by 1.5-month antecedents of precipitation (Precip_cum_lag2) and storage (S_cum_lag2), as well as season (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003eb and \u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003ec). Subsurface elevation, soil and vegetation properties (Filled_subsurf, Slope_surf, peat_thick, SM) with distance to natural stream and current precipitation-storage exhibited moderate to low mean |SHAP| values, while dummy years (2019, 2023\u0026ndash;2025) had the least contributions. Further investigations based on ordered training data (n\u0026thinsp;=\u0026thinsp;5857 observations; Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003ea and \u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003eb) suggested that the XGBoost model's predictions differ according to the magnitudes of predicted GWL. SHAP values were increased oppositely toward both distribution tails. At the driest extreme (0%, approximately at GWL\u0026thinsp;\u0026minus;\u0026thinsp;77 cm), predictions were largely determined by negative contributions from precipitation and storage, with additional static factors. In shallow GWL (\u0026thinsp;\u0026gt;\u0026thinsp;\u0026minus;\u0026thinsp;25 cm; 75% and 99% percentiles), SHAP contributions are strongly positive, mainly driven by storage, year\u0026thinsp;=\u0026thinsp;2022, and the cumulative influence of several minor predictors. Near the median GWL range (\u0026minus;\u0026thinsp;30 to \u0026minus;\u0026thinsp;40 cm), SHAP values were clustered around \u0026plusmn;\u0026thinsp;5 and showed a mixed pattern.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eFigure\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e.\u003c/p\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003eCO\u003csub\u003e2\u003c/sub\u003e-GWL Response Curve Simulations and Upscaling to Spatiotemporal Rs Estimations\u003c/h2\u003e \u003cdiv id=\"Sec9\" class=\"Section3\"\u003e \u003ch2\u003eSpatiotemporal Rs Dynamics and Distributions\u003c/h2\u003e \u003cp\u003eSpatial upscaling of soil CO₂ emissions from general and a combination of general and land use-based CO\u003csub\u003e2\u003c/sub\u003e-GWL response curves developed by \u003csup\u003e14,15\u003c/sup\u003e (Supplementary Information 3) over Pulau Rupat PHU are summarized in Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e. The upscaled Rs utilized biweekly GWL maps (averaged to \u0026minus;\u0026thinsp;32.19\u0026thinsp;\u0026plusmn;\u0026thinsp;5.58 cm) corresponding to total cumulative Rs of 28.15 to 39.66 Mt CO₂ and annual cumulative Rs around 4.02 to 5.67 Mt CO₂ y ⁻\u0026sup1; emitted during the study period (Jan-2019 to Apr-2025; Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003e; Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e). The variability of Rs estimated by general response curve were lower than combination of general and land use-based CO\u003csub\u003e2\u003c/sub\u003e-GWL response curves. The Rs emitted during the dry season were slightly higher than during the rainy season. Annualized Rs tend to be lower in deeper peat areas than in shallower peat, but opposite trends were observed on cumulative annual and total. Among land-use categories, estate plantations possessed the deepest GWL and highest cumulative Rs. Estate plantation emitted the highest average annualized Rs and the highest cumulative Rs during the study period. Primary swamp forests emitted lowest Rs, which only up to 2022 (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003e, Supplementary Information 2). Inter-annual GWL were relatively stable, except for 2020, which had the deepest mean (\u0026minus;\u0026thinsp;34.46\u0026thinsp;\u0026plusmn;\u0026thinsp;5.49 cm), with the Rs estimated from land-use-based response curves exhibited an increasing trend. Application of the general CO\u003csub\u003e2\u003c/sub\u003e-GWL response curve (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003ea; pooled to season and annual courses similar to GWL in Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e) produces relatively smooth spatial gradients of Rs. In contrast, the use of combined land-use-based and general response curves (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003eb) results in substantially greater spatial variation, with consistently high Rs emitted from peats developed for plantations and agriculture (Supplementary Information 2) during both dry and rainy seasons. Central and peripheral zones associated with managed land uses exhibited higher annual cumulative Rs, whereas areas corresponding to undrained peatlands showed lower Rs (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003ea and \u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003eb).\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eEstimated GWL and Rs (average\u0026thinsp;\u0026plusmn;\u0026thinsp;standard deviation) in PHU Pulau Rupat, averaged across all data and several factors (season, peat thickness, major land use, and year). \u003cb\u003eSupplementary Information 1.\u003c/b\u003e Response and selected explanatory variables used in this study\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"8\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c8\" colnum=\"8\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eFactor\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eGWL\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"2\" nameend=\"c4\" namest=\"c3\"\u003e \u003cp\u003eAveraged Annualized Rs\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"2\" nameend=\"c6\" namest=\"c5\"\u003e \u003cp\u003eCumulative Annual Rs\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"2\" nameend=\"c8\" namest=\"c7\"\u003e \u003cp\u003eCumulative Rs During Study Period\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eGeneral Eq\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eLU-Based Eq\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eGeneral Eq\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eLU-Based Eq\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c7\"\u003e \u003cp\u003eGeneral Eq\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c8\"\u003e \u003cp\u003eLU-Based Eq\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003ecm\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c4\" namest=\"c3\"\u003e \u003cp\u003e----------- Mg ha\u003csup\u003e\u0026minus;\u0026thinsp;1\u003c/sup\u003e y\u003csup\u003e\u0026minus;\u0026thinsp;1\u003c/sup\u003e ---------\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c6\" namest=\"c5\"\u003e \u003cp\u003e---------- Mt CO\u003csub\u003e2\u003c/sub\u003e y\u003csup\u003e\u0026minus;\u0026thinsp;1\u003c/sup\u003e ---------\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c8\" namest=\"c7\"\u003e \u003cp\u003e---------- Mt CO\u003csub\u003e2\u003c/sub\u003e ---------\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cspan type=\"BoldItalicUnderline\" class=\"BoldItalicUnderline\" name=\"Emphasis\"\u003eAll Data\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e-32.19\u0026thinsp;\u0026plusmn;\u0026thinsp;5.58\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e47.22\u0026thinsp;\u0026plusmn;\u0026thinsp;0.54\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e66.52\u0026thinsp;\u0026plusmn;\u0026thinsp;22.76\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e4.02\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e5.67\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e28.15\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e39.66\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cspan type=\"BoldItalicUnderline\" class=\"BoldItalicUnderline\" name=\"Emphasis\"\u003eSeason\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDry\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e-32.61\u0026thinsp;\u0026plusmn;\u0026thinsp;5.71\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e47.26\u0026thinsp;\u0026plusmn;\u0026thinsp;0.55\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e67.00\u0026thinsp;\u0026plusmn;\u0026thinsp;22.98\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e2.01\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e2.85\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e14.09\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e19.97\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRainy\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e-31.77\u0026thinsp;\u0026plusmn;\u0026thinsp;5.58\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e47.18\u0026thinsp;\u0026plusmn;\u0026thinsp;0.55\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e66.03\u0026thinsp;\u0026plusmn;\u0026thinsp;22.60\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e2.01\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e2.81\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e14.06\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e19.68\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cspan type=\"BoldItalicUnderline\" class=\"BoldItalicUnderline\" name=\"Emphasis\"\u003ePeat thickness\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;3 m\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e-32.03\u0026thinsp;\u0026plusmn;\u0026thinsp;7.01\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e47.28\u0026thinsp;\u0026plusmn;\u0026thinsp;0.66\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e78.17\u0026thinsp;\u0026plusmn;\u0026thinsp;21.58\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e1.68\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e2.78\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e11.75\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e19.43\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u0026gt;\u0026thinsp;3 m\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e-32.30\u0026thinsp;\u0026plusmn;\u0026thinsp;4.27\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e47.18\u0026thinsp;\u0026plusmn;\u0026thinsp;0.44\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e58.19\u0026thinsp;\u0026plusmn;\u0026thinsp;19.71\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e2.34\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e2.89\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e16.4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e20.22\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cspan type=\"BoldItalicUnderline\" class=\"BoldItalicUnderline\" name=\"Emphasis\"\u003eMajor Land Use\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePrimary Swamp Forest\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e-32.12\u0026thinsp;\u0026plusmn;\u0026thinsp;1.28\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e47.09\u0026thinsp;\u0026plusmn;\u0026thinsp;0.13\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.49\u0026thinsp;\u0026plusmn;\u0026thinsp;0.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.08\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e0.59\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e0.01\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSecondary Swamp Forest\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e-32.47\u0026thinsp;\u0026plusmn;\u0026thinsp;3.98\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e47.18\u0026thinsp;\u0026plusmn;\u0026thinsp;0.36\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e57.24\u0026thinsp;\u0026plusmn;\u0026thinsp;4.29\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.9\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e1.09\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e6.31\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e7.66\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePlantation Forest\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e-31.19\u0026thinsp;\u0026plusmn;\u0026thinsp;4.98\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e47.10\u0026thinsp;\u0026plusmn;\u0026thinsp;0.39\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e47.31\u0026thinsp;\u0026plusmn;\u0026thinsp;1.47\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.93\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.93\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e6.51\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e6.54\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eEstate Plantation\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e-34.46\u0026thinsp;\u0026plusmn;\u0026thinsp;5.49\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e47.47\u0026thinsp;\u0026plusmn;\u0026thinsp;0.66\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e97.99\u0026thinsp;\u0026plusmn;\u0026thinsp;9.76\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e1.12\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e2.32\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e7.86\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e16.24\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAgriculture and Other Land Uses\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e-30.29\u0026thinsp;\u0026plusmn;\u0026thinsp;6.65\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e47.10\u0026thinsp;\u0026plusmn;\u0026thinsp;0.59\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e63.13\u0026thinsp;\u0026plusmn;\u0026thinsp;7.19\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.98\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e1.32\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e6.87\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e9.21\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cspan type=\"BoldItalicUnderline\" class=\"BoldItalicUnderline\" name=\"Emphasis\"\u003eYear\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e2019\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e-32.76\u0026thinsp;\u0026plusmn;\u0026thinsp;5.66\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e47.30\u0026thinsp;\u0026plusmn;\u0026thinsp;0.54\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e63.45\u0026thinsp;\u0026plusmn;\u0026thinsp;23.21\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e4.03\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e5.41\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e4.03\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e5.41\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e2020\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e-34.37\u0026thinsp;\u0026plusmn;\u0026thinsp;6.73\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e47.59\u0026thinsp;\u0026plusmn;\u0026thinsp;0.88\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e67.10\u0026thinsp;\u0026plusmn;\u0026thinsp;27.74\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e4.05\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e5.71\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e4.05\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e5.71\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e2021\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e-31.99\u0026thinsp;\u0026plusmn;\u0026thinsp;5.30\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e47.18\u0026thinsp;\u0026plusmn;\u0026thinsp;0.42\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e65.22\u0026thinsp;\u0026plusmn;\u0026thinsp;24.89\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e4.01\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e5.54\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e4.01\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e5.54\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e2022\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e-31.75\u0026thinsp;\u0026plusmn;\u0026thinsp;5.55\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e47.16\u0026thinsp;\u0026plusmn;\u0026thinsp;0.51\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e68.41\u0026thinsp;\u0026plusmn;\u0026thinsp;22.52\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e4.02\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e5.83\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e4.02\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e5.83\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e2023\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e-31.19\u0026thinsp;\u0026plusmn;\u0026thinsp;4.66\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e47.07\u0026thinsp;\u0026plusmn;\u0026thinsp;0.32\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e66.82\u0026thinsp;\u0026plusmn;\u0026thinsp;19.56\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e4.01\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e5.7\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e4.01\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e5.7\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e2024\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e-30.52\u0026thinsp;\u0026plusmn;\u0026thinsp;4.86\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e47.02\u0026thinsp;\u0026plusmn;\u0026thinsp;0.32\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e66.21\u0026thinsp;\u0026plusmn;\u0026thinsp;19.56\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e4.01\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e5.64\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e4.01\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e5.64\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e-32.73\u0026thinsp;\u0026plusmn;\u0026thinsp;5.15\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e47.22\u0026thinsp;\u0026plusmn;\u0026thinsp;0.39\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e68.42\u0026thinsp;\u0026plusmn;\u0026thinsp;20.18\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e4.02\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e5.83\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e4.02\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e5.83\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eFigure\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003e.\u003c/p\u003e \u003cp\u003eTable\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e.\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e\n\u003ch3\u003eRs Response to 10-cm GWL Changes\u003c/h3\u003e\n\u003cp\u003eFigure\u0026nbsp;\u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e8\u003c/span\u003ea illustrates how our estimated Rs would change (positive and negative values represent emitted and avoided Rs) in response to 10 cm GWL changes (on an annual basis) according to the CO\u003csub\u003e2\u003c/sub\u003e-GWL response curves (\u003csup\u003e14,15\u003c/sup\u003e, Supplementary Information 3). The general response curve\u003csup\u003e\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e\u003c/sup\u003e predicted that an Rs change increases incrementally as GWL decreases below \u0026minus;\u0026thinsp;40 cm and progressively as GWL deepens, but would exceed 100% during maximum GWL drawdowns (\u0026lt;-80 cm). Across all response functions\u003csup\u003e\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e,\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e\u003c/sup\u003e, the largest Rs changes (306% and 246%) were observed in estate plantation and agriculture and other land uses, respectively, if GWL deepened dramatically from 0 cm to \u0026minus;\u0026thinsp;80 cm. Other major land uses (secondary swamp forest and forest plantation) are also experiencing more than 100% Rs changes under similar conditions. Estate plantation exhibited intermediate changes; a 10-cm GWL drawdown between \u0026minus;\u0026thinsp;30 and \u0026minus;\u0026thinsp;60 cm would generate 20\u0026ndash;30% Rs changes, which would further decrease with shallower GWL changes. Rs responses were not reciprocal with respect to water table rise and decline. For example, the emission increase associated with a 10-cm GWL declination from \u0026minus;\u0026thinsp;40 to \u0026minus;\u0026thinsp;50 cm exceeded the reduction avoided by a 10-cm GWL rise from \u0026minus;\u0026thinsp;50 to \u0026minus;\u0026thinsp;40 cm. Across all curves, proportional changes were greater under water-table decline than under an equivalent rise.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003eComparison to Other Emission Factors\u003c/h2\u003e \u003cp\u003eAnnual averaged and total cumulative emitted Rs estimated using combination of general and land use-based CO\u003csub\u003e2\u003c/sub\u003e-GWL response curves were consistently higher (66.52\u0026thinsp;\u0026plusmn;\u0026thinsp;22.76 Mg CO₂ ha\u003csup\u003e-\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e\u003c/sup\u003e y ⁻\u0026sup1; and 39.66 Mt CO\u003csub\u003e2\u003c/sub\u003e) than linear curves from previous Rs reviews \u003csup\u003e\u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e,\u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e\u003c/sup\u003e and static Rh-based EFs derived from IPCC Tier 1\u003csup\u003e11\u003c/sup\u003e and Indonesian FREL\u003csup\u003e\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e\u003c/sup\u003e (Fig.\u0026nbsp;\u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e8\u003c/span\u003eb). However, the general Rs estimates from this study falls within the same range as both authors\u003csup\u003e\u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e,\u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e\u003c/sup\u003e and Rh from Indonesian FREL\u003csup\u003e\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e\u003c/sup\u003e which totalling around ~\u0026thinsp;4 Mt CO₂ yr⁻\u0026sup1; for Rupat Island (equivalent to ~\u0026thinsp;45\u0026ndash;50 Mg ha⁻\u0026sup1; yr⁻\u0026sup1;). Emission estimated from IPCC Tier 1 were lowest compared to all studied EFs in terms of annual averaged (Fig.\u0026nbsp;\u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e8\u003c/span\u003eb.1) and total cumulative emission (Fig.\u0026nbsp;\u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e8\u003c/span\u003eb.2).\u003c/p\u003e \u003cp\u003eFigure\u0026nbsp;\u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e8\u003c/span\u003e.\u003c/p\u003e \u003c/div\u003e"},{"header":"Discussion","content":"\u003cp\u003eThis study highlighted the spatiotemporal Rs estimation via ML-assisted GWL maps in a single tropical peatland hydrological system (\u003cem\u003eIndonesian KHG/\u003c/em\u003ePHU Rupat Island). Our study provides alternative dynamic estimation of tropical peatland Rs-based EF, alongside current Rh-based global IPCC Tier 1\u003csup\u003e11\u003c/sup\u003e and regional (Indonesian FREL\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e13\u003c/span\u003e\u003c/sup\u003e) static, land use-based EF estimations (\u003cem\u003ee.g.\u003c/em\u003e, \u003csup\u003e4,12\u003c/sup\u003e). On the other hand, our XGBoost regressor are trained from GWL dipwell that possess comparable measurement density than any previous models\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e16\u003c/span\u003e,\u003cspan class=\"CitationRef\"\u003e27\u003c/span\u003e,\u003cspan class=\"CitationRef\"\u003e28\u003c/span\u003e,\u003cspan class=\"CitationRef\"\u003e31\u003c/span\u003e\u003c/sup\u003e in terms of spatial and temporal scales.\u003c/p\u003e \u003cp\u003eThe predicted GWL (median cal R\u003csup\u003e2\u003c/sup\u003e 0.46; val R\u003csup\u003e2\u003c/sup\u003e 0.35; Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e5\u003c/span\u003ea) showed various hydrological regimes (Figs.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003e and \u003cspan class=\"InternalRef\"\u003e4\u003c/span\u003e) representing water storage dynamics and trade-offs between precipitation against evapotranspiration and lateral outflow\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e39\u003c/span\u003e\u003c/sup\u003e. Pooled GWL (Table\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003e) suggested that at the PHU level, the averaged GWL were relatively shallow (\u0026gt; -40 cm), regardless of the partitioning factors. However, some drained and cultivated areas at the western and southern sides of the PHU experienced deep GWL during 2020 and 2022 (-45 to -60 cm; Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e4\u003c/span\u003e; Supplementary Information 2), which resulted in elevated Rs up to 120 Mg CO\u003csub\u003e2\u003c/sub\u003e ha\u003csup\u003e-\u003cspan class=\"CitationRef\"\u003e1\u003c/span\u003e\u003c/sup\u003e y\u003csup\u003e-\u003cspan class=\"CitationRef\"\u003e1\u003c/span\u003e\u003c/sup\u003e (Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e7\u003c/span\u003e). This condition may correspond to a dry condition underwent at the preceding and corresponding years\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e40\u003c/span\u003e,\u003cspan class=\"CitationRef\"\u003e41\u003c/span\u003e\u003c/sup\u003e. The 6-years predicted GWL mean in primary and secondary peat swamp forests were − 32.12 ± 1.28 and − 32.47 ± 3.98 cm, respectively, which are deeper than undrained peat swamp forests reported by previous modelling study\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e16\u003c/span\u003e\u003c/sup\u003e. Even though the peat swamp forests in the studied PHU are undrained, their small areas and location within the peat dome resulting in the GWL are much more influenced by the surrounding by drained and cultivated peats\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e42\u003c/span\u003e,\u003cspan class=\"CitationRef\"\u003e43\u003c/span\u003e\u003c/sup\u003e (Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e4\u003c/span\u003e). Differently from \u003csup\u003e\u003cspan class=\"CitationRef\"\u003e16\u003c/span\u003e,\u003cspan class=\"CitationRef\"\u003e22\u003c/span\u003e,\u003cspan class=\"CitationRef\"\u003e37\u003c/span\u003e,\u003cspan class=\"CitationRef\"\u003e38\u003c/span\u003e\u003c/sup\u003e, our study reported the averaged GWL at estate and forest plantations sites around half of their estimates (− 34.46 ± 5.49 and − 31.19 ± 4.98 cm; Table\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003e), which were within the recommended best practice threshold for GWL\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e44\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eDeep-peat at the central PHU (northern dome) consistently emit low Rs due to their forest covers, while surrounding shallow peats serve as recurrent high-emission hotspots across seasons and years (Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e7\u003c/span\u003e). Dry-season and drought-year conditions increase emissions by approximately 20 to 30 Mg CO₂ ha⁻¹ yr⁻¹ and contribute substantially to annual totals. In contrast, rainy-season rewetting reduced, but did not eliminate emissions in drained areas (Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e7\u003c/span\u003ea-b). Our simulation demonstrated that drainage-induced drying leads to disproportionately large increases in CO₂ emissions, which are not fully mitigated by short-term rewetting (Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e8\u003c/span\u003ea). The resulting dynamic emission estimated during study period (39.66 Mt CO₂ y⁻¹) broadly correspond to higher Rs emanates from forest plantation and agriculture and other land uses (Table\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003e), exceeding Tier-1 IPCC\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e11\u003c/span\u003e\u003c/sup\u003e and Indonesian FREL\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e13\u003c/span\u003e\u003c/sup\u003e estimates by over six times (Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e8\u003c/span\u003eb) but close to previous chamber and subsidence studies\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e45\u003c/span\u003e,\u003cspan class=\"CitationRef\"\u003e46\u003c/span\u003e\u003c/sup\u003e. Variations in Rs across land-use categories in Table\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003e may closely correspond to inter-annual land-use transitions (Supplementary Information 2) and GWL patterns (Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e4\u003c/span\u003e; Table\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003e). During this period, PHU Pulau Rupat underwent a marked shift toward managed land uses, particularly when primary swamp forest, bare/open land, and swamp/marsh were gradually converted to estate and forest plantations. These conditions were reflected in Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003e, as deeper GWL areas were persistently concentrated throughout area dominated by both plantations during any seasons, corroborating previous reports\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e16\u003c/span\u003e,\u003cspan class=\"CitationRef\"\u003e47\u003c/span\u003e\u003c/sup\u003e. Those areas remained moderately to deeply drained even during the rainy season, indicating that seasonal precipitation did not offset drainage-induced drawdown, taking into account for the uncertainty of our climate dataset. In comparison, areas historically classified as primary swamp forest exhibited shallower and more homogeneous GWL, consistent with their lower cumulative Rs. Cumulative Rs remained stable across years, even though there are ongoing land-use changes (Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e7\u003c/span\u003e; Supplementary Information 2).\u003c/p\u003e \u003cp\u003eSimulated Rs responses to 10-cm GWL shifts (Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e8\u003c/span\u003ea) showed that emission increases under GWL declination consistently exceeded avoided emission under equivalent rise at any response curves\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e14\u003c/span\u003e,\u003cspan class=\"CitationRef\"\u003e15\u003c/span\u003e\u003c/sup\u003e, as also reported in previous reports\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e48\u003c/span\u003e\u003c/sup\u003e. However, proportional changes in Rs under moderate GWL fluctuations do not directly translate into equivalent changes in long-term peat carbon stocks. Within more typical hydrological ranges (shallower than approximately − 30 to − 50 cm), a substantial portion of Rs (commonly 40–70%) corresponds to root respiration, which represents contemporary plant carbon cycling rather than peat carbon loss. Strong seasonality and increased intensity under a changing climate\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e5\u003c/span\u003e,\u003cspan class=\"CitationRef\"\u003e49\u003c/span\u003e,\u003cspan class=\"CitationRef\"\u003e50\u003c/span\u003e\u003c/sup\u003e and widespread subsidence\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e51\u003c/span\u003e\u003c/sup\u003e may expose tropical peatlands to sustained inundation during rainy season and experiencing rapid GWL drawdown during dry season. This condition, in the extreme cases (\u003cem\u003ee.g.\u003c/em\u003e, El Nino\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e10\u003c/span\u003e\u003c/sup\u003e, 0 to − 80 cm; Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e8\u003c/span\u003ea), would generate more than 100–200% emission changes generally, with the most drastic found on forest plantation (306%). At deep GWL, Rs predominantly originated from enhanced peat oxidation\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e52\u003c/span\u003e\u003c/sup\u003e, which in long-term conditions would severely impact old sequestered carbon\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e1\u003c/span\u003e,\u003cspan class=\"CitationRef\"\u003e7\u003c/span\u003e,\u003cspan class=\"CitationRef\"\u003e53\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eDifferences between the Rs generated in this study and the emission magnitudes implied by the Indonesian FREL\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e13\u003c/span\u003e\u003c/sup\u003e and IPCC Tier-1\u003csup\u003e11\u003c/sup\u003e EFs (Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e8\u003c/span\u003eb) mainly result from differences in emission accounting frameworks and boundaries. This study, together with \u003csup\u003e\u003cspan class=\"CitationRef\"\u003e37\u003c/span\u003e,\u003cspan class=\"CitationRef\"\u003e38\u003c/span\u003e\u003c/sup\u003e quantifies total soil respiration (Rs) by including both autotrophic and heterotrophic components across all peatland conditions. The resulting Rs can be used as more robust basis for upscaling RE compared to partitioned Ra and Rh, considering recent shifting of EF calculation approach favoring actual ecosystem CO\u003csub\u003e2\u003c/sub\u003e exchange\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e16\u003c/span\u003e,\u003cspan class=\"CitationRef\"\u003e21\u003c/span\u003e,\u003cspan class=\"CitationRef\"\u003e22\u003c/span\u003e\u003c/sup\u003e. In contrast, both IPCC Tier-1 and Indonesian FREL framework takes into account only for heterotrophic respiration (Rh) associated with anthropogenic peat decomposition and explicitly excludes emissions from undisturbed peat swamp forests. Consequently, total peat CO₂ emissions represented within the IPCC Tier-1 and FREL frameworks may exceed those Rs reported here (Table\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003e; Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e8\u003c/span\u003eb) when aggregated over extensive disturbed areas, despite lower individual EFs. Further uncertainty is introduced by the IPCC Tier-1 factors, which assume negligible emissions from peatlands classified as natural, including bushland, swamps, and logged peat swamp forests. Both IPCC Tier 1 and partially of Indonesian FREL are consider reporting conventions specifically for accounting how much peat material has been degraded. This assumption does not reflect underlying biophysical processes, particularly for Indonesian peat swamp forests, which have been currently reported as net CO\u003csub\u003e2\u003c/sub\u003e emitters\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e16\u003c/span\u003e,\u003cspan class=\"CitationRef\"\u003e22\u003c/span\u003e,\u003cspan class=\"CitationRef\"\u003e47\u003c/span\u003e,\u003cspan class=\"CitationRef\"\u003e54\u003c/span\u003e\u003c/sup\u003e despite some higher sequestration reported from undrained Malaysian peat swamp secondary forests\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e55\u003c/span\u003e\u003c/sup\u003e. These reports suggested that Rs emitted from peat swamp forests cannot be neglected. From measurements covering 1997–2016, \u003csup\u003e16\u003c/sup\u003e reported that all undrained peat swamp forests in Indonesia and Malaysia were small net CO\u003csub\u003e2\u003c/sub\u003e sinks (− 0.86 ± 10.9 Mg ha\u003csup\u003e-\u003cspan class=\"CitationRef\"\u003e1\u003c/span\u003e\u003c/sup\u003e y\u003csup\u003e-\u003cspan class=\"CitationRef\"\u003e1\u003c/span\u003e\u003c/sup\u003e) and acted as large net CO\u003csub\u003e2\u003c/sub\u003e emitters during dry years (13.1 ± 12.9 Mg ha\u003csup\u003e-\u003cspan class=\"CitationRef\"\u003e1\u003c/span\u003e\u003c/sup\u003e y\u003csup\u003e-\u003cspan class=\"CitationRef\"\u003e1\u003c/span\u003e\u003c/sup\u003e). Furthermore, drained peat swamp forests currently act as net CO\u003csub\u003e2\u003c/sub\u003e emitters in normal, wet, and dry years. Based on \u003csup\u003e55\u003c/sup\u003e report on higher GPP in coastal peat swamp forests compared to inland peat swamp forests\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e47\u003c/span\u003e\u003c/sup\u003e, which is similar to our research site\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e56\u003c/span\u003e\u003c/sup\u003e (Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003e). Therefore, our reported Rs originating from primary and secondary forests (Table\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003e) might be offset by higher CO\u003csub\u003e2\u003c/sub\u003e sequestered by the forest vegetation.\u003c/p\u003e \u003cp\u003eBeyond differences in accounting perspective, the IPCC Tier-1 and FREL approaches are inherently static and do not capture the sensitivity of peat CO₂ emissions to climate-driven hydrological variability\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e11\u003c/span\u003e,\u003cspan class=\"CitationRef\"\u003e16\u003c/span\u003e\u003c/sup\u003e. Static emission factors are applied homogeneously across land-use classes and remain constant across seasons and years. This approach might obscure substantial spatial variation in drainage intensity within the same land-use category. In contrast, the biweekly GWL prediction adopted in this study captured spatial variability within land-use classes. It is also related to temporal variability associated with seasonal and interannual climate fluctuations (Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003e, \u003cspan class=\"InternalRef\"\u003e4\u003c/span\u003e, and \u003cspan class=\"InternalRef\"\u003e7\u003c/span\u003e; Table\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003e). Our results illustrated that climate-driven changes in GWL fluctuation modulated Rs at biweekly timescales, even within similar land-use classifications, particularly in managed peatlands where drainage infrastructure restricts water table recovery. Similar to our study, \u003csup\u003e57\u003c/sup\u003e reported that IPCC and Canadian national EFs were mostly dependent on CO\u003csub\u003e2\u003c/sub\u003e research conducted on warm periods and suggested the revised EFs around 45% lower. In tropical peatlands, this condition can be related to the measurements being mostly conducted within the drier season or dry conditions in the rainy season, which allows chambers to be placed securely on the peat surface to measure CO\u003csub\u003e2\u003c/sub\u003e effluxes. However, that undermines the specific conditions representing long-rainy season and La Nina period, where all peat surfaces are inundated or GWL are very shallow, thus resulting in CO2 emission that are lower than averages. Also, the EFs do not account for lowered CO\u003csub\u003e2\u003c/sub\u003e emission due to long-term shallower GWL as affected by large-scale restoration projects\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e58\u003c/span\u003e–\u003cspan class=\"CitationRef\"\u003e60\u003c/span\u003e\u003c/sup\u003e. These dynamics are also not represented by land-use-only EFs, which average emissions over space and time. Our method is considered a dynamic approach that explicitly link CO₂ emissions to GWL variability. This is, therefore, provide a more temporally responsive and spatially explicit representation of peat carbon fluxes under fluctuating climatic conditions.\u003c/p\u003e "},{"header":"Methods","content":"\u003ch2\u003eStudy Site Description\u003c/h2\u003e\u003cp\u003eThis study was conducted on the Rupat Island Peat Hydrological Unit (PHU; In Indonesian terms: Kesatuan Hidrologis Gambut/KHG) in Riau Province, Indonesia (Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003e). A PHU represents an integrated peatland hydrological system bounded by rivers, coastal interfaces, and swamp areas. The peats in PHU are interconnected with ecosystem components, \u003cem\u003ee.g.\u003c/em\u003e, water, vegetation, and other environmental elements\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e35\u003c/span\u003e,\u003cspan class=\"CitationRef\"\u003e36\u003c/span\u003e\u003c/sup\u003e. The PHU Rupat Island covers an area of approximately 118.510,16 ha and situated on Rupat Island in the Malacca Strait off the eastern coast of Sumatra, Indonesia. The peatland in this PHU is developed across the coastal (southern dome) and riverine (northern dome) backswamps at the island’s middle to southern sides, with two main domes having very deep peats (around 10 to 12 m depth; Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003e). The PHU had a variety of land uses, dominated mostly by mosaics of the remains of peat swamp forest and shrublands at the dome center, surrounded by forest and estate plantations alongside fragmented cultivated lands, settlements, and aquaculture ponds in the peripheral areas with thinner peats (Supplementary Information 2).\u003c/p\u003e\u003cdiv id=\"Sec15\" class=\"Section2\"\u003e \u003ch2\u003eData Acquisition and Pre-processing\u003c/h2\u003e \u003cp\u003eThis study leveraged data-driven modelling to predict groundwater level (GWL) using a combination of dynamic, static, and time-related predictors (Supplementary Information 1). The GWL data were gathered from 134 automatic loggers and manual dipwells located throughout PHU Rupat Island as part of the regulatory compliance to the Indonesian Ministry of Environment and Forestry (MoEF; now Ministry of Environment-Environmental Control Agency of Indonesia/KLH-BPLH). We selected the most observed period covering 2019 to 2025 from Simatag0,4m platform (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://sippeg.menlhk.go.id/apps/sippegapps/tmat\u003c/span\u003e\u003cspan class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e) and obtained (aggregated) 8,327 biweekly records in total. All readings exceeding + 10 cm were excluded (which may indicate input errors), while the entire negative GWL was selected. All predictors were selected for their hydrological relationship with GWL spatiotemporal dynamics in tropical peatland landscapes. The constructed model conceptually represents the influence of climate, vegetation, topography, and soil on peatland hydrology. We also included several proxies for temporal autocorrelation, including lagged variables and year and season dummies. The dynamic inputs representing climate and vegetation drivers consisted of 158 biweekly precipitation (Precip) and evapotranspiration (ET) composites (2019–2025), resampled at 10-meter resolution. Biweekly Precip was aggregated from daily 0.05° CHIRPS\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e61\u003c/span\u003e\u003c/sup\u003e. ET estimates were derived from the MODIS MOD16A2 8-day evapotranspiration product\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e62\u003c/span\u003e\u003c/sup\u003e. We selected the VV band from Sentinel-1, which represents surface radar scattering, as a proxy for peat soil moisture (SM). Furthermore, we decomposed partial storage proxy (S) as an additional dynamic predictor, following S = Precip – ET, where S represents both storage and lateral outflow, and GWL is technically an upper table of S\u003csup\u003e39,53\u003c/sup\u003e. Since the studied PHU is separated from other higher grounds by a large river (Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003e), we assumed that the studied peatland is ombrotrophic and thus received negligible additional lateral inflow. To address temporal autocorrelation depicting antecedent water content, we add cumulative lags for Precip and ET and averaged lags for SM over the preceding biweekly and monthly windows (lag1 and lag2, respectively). Optical and multispectral imageries, as well as their spectral indices from Sentinel-2 and Landsat 8/9, were also collected but later dropped due to insufficient acquisition and high cloud cover during the biweekly window. Static predictors represent peat and topographic drivers, incorporating peat thickness (calibrated using 1,111-field samplings), distance to major natural stream, elevation, and slope. Specifically, for the LiDAR-derived elevation model, we first filled the sinks and pits, then divided the elevation into surface and subsurface rasters by subtracting the surface elevation from the peat thickness. The slope was then computed for both rasters. The inclusion of surface and subsurface elevations and their derivatives was intended to proxy acrotelmic and catotelmic flows following the scalar model mechanism formulated by \u003csup\u003e39\u003c/sup\u003e. All terrain features except slopes (\u003cem\u003ei.e.\u003c/em\u003e, flow direction and accumulation, topographic wetness index, terrain roughness and ruggedness indices) were excluded due to artifacts and less variability generated by extensive flat slopes. We appended all static rasters to align with the biweekly dynamic stacks. To control for interannual and seasonal autocorrelation, eight dummy factors (year 2019–2025 and dry-rainy seasons, respectively) were hot-encoded and included in the dataset. The final predictors (totalling 27) were derived from an initial pool of 42 candidate variables after applying a multicollinearity screening criterion of |r| \u0026lt; 0.8. All ET rasters were dropped due to their multicollinearity with S. All raster preprocessing, stacking, and model input generation were conducted in R and RStudio\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e63\u003c/span\u003e\u003c/sup\u003e, using terra\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e64\u003c/span\u003e\u003c/sup\u003e and tidyverse\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e65\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e \u003c/div\u003e\u003ch2\u003eData Acquisition and Pre-processing\u003c/h2\u003e\u003cp\u003eThis study leveraged data-driven modelling to predict groundwater level (GWL) using a combination of dynamic, static, and time-related predictors (Supplementary Information 1). The GWL data were gathered from 134 automatic loggers and manual dipwells located throughout PHU Rupat Island as part of the regulatory compliance to the Indonesian Ministry of Environment and Forestry (MoEF; now Ministry of Environment-Environmental Control Agency of Indonesia/KLH-BPLH). We selected the most observed period covering 2019 to 2025 from Simatag0,4m platform (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://sippeg.menlhk.go.id/apps/sippegapps/tmat\u003c/span\u003e\u003cspan class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e) and obtained (aggregated) 8,327 biweekly records in total. All readings exceeding + 10 cm were excluded (which may indicate input errors), while the entire negative GWL was selected. All predictors were selected for their hydrological relationship with GWL spatiotemporal dynamics in tropical peatland landscapes. The constructed model conceptually represents the influence of climate, vegetation, topography, and soil on peatland hydrology. We also included several proxies for temporal autocorrelation, including lagged variables and year and season dummies. The dynamic inputs representing climate and vegetation drivers consisted of 158 biweekly precipitation (Precip) and evapotranspiration (ET) composites (2019–2025), resampled at 10-meter resolution. Biweekly Precip was aggregated from daily 0.05° CHIRPS\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e61\u003c/span\u003e\u003c/sup\u003e. ET estimates were derived from the MODIS MOD16A2 8-day evapotranspiration product\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e62\u003c/span\u003e\u003c/sup\u003e. We selected the VV band from Sentinel-1, which represents surface radar scattering, as a proxy for peat soil moisture (SM). Furthermore, we decomposed partial storage proxy (S) as an additional dynamic predictor, following S = Precip – ET, where S represents both storage and lateral outflow, and GWL is technically an upper table of S\u003csup\u003e39,53\u003c/sup\u003e. Since the studied PHU is separated from other higher grounds by a large river (Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003e), we assumed that the studied peatland is ombrotrophic and thus received negligible additional lateral inflow. To address temporal autocorrelation depicting antecedent water content, we add cumulative lags for Precip and ET and averaged lags for SM over the preceding biweekly and monthly windows (lag1 and lag2, respectively). Optical and multispectral imageries, as well as their spectral indices from Sentinel-2 and Landsat 8/9, were also collected but later dropped due to insufficient acquisition and high cloud cover during the biweekly window. Static predictors represent peat and topographic drivers, incorporating peat thickness (calibrated using 1,111-field samplings), distance to major natural stream, elevation, and slope. Specifically, for the LiDAR-derived elevation model, we first filled the sinks and pits, then divided the elevation into surface and subsurface rasters by subtracting the surface elevation from the peat thickness. The slope was then computed for both rasters. The inclusion of surface and subsurface elevations and their derivatives was intended to proxy acrotelmic and catotelmic flows following the scalar model mechanism formulated by \u003csup\u003e39\u003c/sup\u003e. All terrain features except slopes (\u003cem\u003ei.e.\u003c/em\u003e, flow direction and accumulation, topographic wetness index, terrain roughness and ruggedness indices) were excluded due to artifacts and less variability generated by extensive flat slopes. We appended all static rasters to align with the biweekly dynamic stacks. To control for interannual and seasonal autocorrelation, eight dummy factors (year 2019–2025 and dry-rainy seasons, respectively) were hot-encoded and included in the dataset. The final predictors (totalling 27) were derived from an initial pool of 42 candidate variables after applying a multicollinearity screening criterion of |r| \u0026lt; 0.8. All ET rasters were dropped due to their multicollinearity with S. All raster preprocessing, stacking, and model input generation were conducted in R and RStudio\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e63\u003c/span\u003e\u003c/sup\u003e, using terra\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e64\u003c/span\u003e\u003c/sup\u003e and tidyverse\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e65\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e\u003ch2\u003eGWL Model Training\u003c/h2\u003e\u003cp\u003eThe GWL data and the extracted input data frame were extensively checked and cleaned by applying summary statistics, histograms, and correlations, and by iteratively removing outliers using the interquartile range (IQR) method. The cleaned dataset was then partitioned into a training and a validation matrix for further processing. Due to its widely known quick training time and accurate prediction reported previously\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e14\u003c/span\u003e,\u003cspan class=\"CitationRef\"\u003e15\u003c/span\u003e,\u003cspan class=\"CitationRef\"\u003e66\u003c/span\u003e,\u003cspan class=\"CitationRef\"\u003e67\u003c/span\u003e\u003c/sup\u003e, Extreme Gradient Boosting (XGB) was selected as a regression algorithm to predict GWL. The final hyperparameters selected were: nrounds = 1460, eta = 0.01, max_depth = 9, colsample_bytree = 0.6, min_child_weight = 3, and subsample = 1.0 with out-of-bag/OOB and fold performance evaluated by root mean squared error/RMSE. Tuning was conducted gradually from broad to narrow ranges via 5-fold cross-validation. Lastly, 10-fold cross-validation was repeated 10 times to evaluate the final model. The training process was executed in R using caret\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e68\u003c/span\u003e\u003c/sup\u003e and XGBoost packages\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e66\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e\u003ch2\u003eGWL Model Evaluation\u003c/h2\u003e\u003cp\u003eThe final XGB model was evaluated using a custom Monte Carlo Cross-Validation (MCCV). This approach was applied separately to the complete training and testing datasets to examine model stability and error generalization. The MCCV procedure consisted of 25 repeated random resampling, with 80% of the data allocated for training and 20% for validation (reps = 25; train_frac = 0.8). To maintain meaningful data splits, a minimum of 100 samples was required for the training partition and 50 for the validation partition. In each of the 25 MCCV iterations, a new XGB model was refitted with a fixed, previously tuned hyperparameters. Standard evaluation indicators, \u003cem\u003ei.e.\u003c/em\u003e, R\u003csup\u003e2\u003c/sup\u003e, RMSE, MAE, and BIAS, were calculated for each training and testing subset. All results were summarized using boxplots. The entire evaluation method was conducted in R using tidyverse\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e65\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e\u003ch2\u003eGWL Model Explanation\u003c/h2\u003e\u003cp\u003eWe interpreted the model functioning through Shapley value of game theory explanation\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e69\u003c/span\u003e\u003c/sup\u003e. The Shapley value was determined by employing kernel SHAP approach\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e70\u003c/span\u003e\u003c/sup\u003e, focusing on tree-specific Kernel SHAP\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e71\u003c/span\u003e,\u003cspan class=\"CitationRef\"\u003e72\u003c/span\u003e\u003c/sup\u003e. SHAP values were computed for the training dataset using a Monte Carlo estimation approach, which assigned an additive importance score to every feature for each individual prediction. Global model interpretation was assessed by aggregating these values, specifically ranking feature importance based on the mean absolute SHAP value to identify the dominant drivers of GWL dynamics besides the combined visualization of their directionality and magnitude. SHAP dependence plots were analyzed to characterize non-linear relationships and interaction effects between primary drivers and their contribution to the model output. Furthermore, a stacked force plot ordered by predicted GWL was constructed to reveal cumulative feature effects and model stability across the entire prediction range from shallow to deep levels. To examine local model behavior, we generated individual force plots for representative GWL quantiles (0%, 25%, 50%, 75%, and 99%). All interpretation workflows, including the visualization of feature contributions and interaction effects, were conducted using the fastshap\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e73\u003c/span\u003e\u003c/sup\u003e, shapviz\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e74\u003c/span\u003e\u003c/sup\u003e, and SHAPforxgboost packages\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e75\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e\u003ch2\u003eUpscaling GWL Prediction to Rs Estimation using CO\u003csub\u003e2\u003c/sub\u003e-GWL Response Function\u003c/h2\u003e\u003cp\u003eThis study used the resulting biweekly GWL maps as a base to upscale Rs emissions using the CO\u003csub\u003e2\u003c/sub\u003e-GWL response curve applied by \u003csup\u003e14,15\u003c/sup\u003e in the seven Indonesian peatland units (Peatland Hydrological Unit/PHU or Kesatuan Hidrologi Gambut/KHG) in Sumatra and Kalimantan (Supplementary Information 3). They developed logistic general and land-use-based response curves after compiling 25 chamber-based emission studies conducted in Southeast Asian peatlands (organic layer \u0026gt; 50 cm), representing total soil respiration (Rs; autotrophic + heterotrophic respirations). The first authors applied a general response curve, whereas the latter employed a combination of general and land-use-based response curves to generate biweekly Rs maps from XGB-predicted GWL maps with the same spatiotemporal resolutions. Specifically for the response curve in the primary forest, bi-logistic functions derived from \u003csup\u003e\u003cspan class=\"CitationRef\"\u003e52\u003c/span\u003e\u003c/sup\u003e were selected. Moreover, the upscaled results were compared to state-of-the-art similar compilations and Rs-GWL responses developed previously by \u003csup\u003e37,38\u003c/sup\u003e, as well as Rh-based EFs from Tier-1 IPCC \u003csup\u003e\u003cspan class=\"CitationRef\"\u003e11\u003c/span\u003e\u003c/sup\u003e and Indonesian forest reference emission level/FREL\u003csup\u003e13\u003c/sup\u003e. To further visualize the response curves, this study simulated the relative change in peat Rs (reported as a percentage) in response to a 10-cm change in GWL (from the current annual GWL to the new target annual GWL), following \u003csup\u003e\u003cspan class=\"CitationRef\"\u003e48\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eAcknowledgements\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors gratefully thanked all office and field staff in helping during field campaign. Special acknowledgement to the IMPLI Project in supporting funds for this study.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthor contributions statement\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eConceptualization, W.Y.U., B.B.; methodology, data collection, formal analyses, and visualization W.Y.U.; writing, original draft preparation, W.Y.U.; review and editing, W.Y.U., B.B., S.A., S.D.T., F.K.A.; supervision, S.A., S.D.T., W.Y.U., F.K.A., M.A.; project administration, W.Y.U., F.K.A., M.A.; funding acquisition, W.Y.U, F.K.A., M.A.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting Interests\u003c/strong\u003e\u003cbr\u003e\u0026nbsp;The authors declare no competing interests.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis research was funded by Integrated Management of Peatland Landscapes in Indonesia/IMPLI Grant Number No. 2P1GN48A, by GEF-6 (Global Environment Facility-6) 2021–2026.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003ePage, S. E., Rieley, J. O. \u0026amp; Banks, C. J. Global and regional importance of the tropical peatland carbon pool. \u003cem\u003eGlob Chang. Biol.\u003c/em\u003e \u003cb\u003e17\u003c/b\u003e, 798\u0026ndash;818 (2011).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eXu, J., Morris, P. J., Liu, J. \u0026amp; Holden, J. P. E. A. T. M. A. P. Refining estimates of global peatland distribution based on a meta-analysis. \u003cem\u003eCatena (Amst)\u003c/em\u003e. \u003cb\u003e160\u003c/b\u003e, 134\u0026ndash;140 (2018).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAnda, M. et al. Revisiting tropical peatlands in Indonesia: Semi-detailed mapping, extent and depth distribution assessment. \u003cem\u003eGeoderma\u003c/em\u003e \u003cb\u003e402\u003c/b\u003e, (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMiettinen, J., Hooijer, A., Vernimmen, R., Liew, S. C. \u0026amp; Page, S. E. From carbon sink to carbon source: extensive peat oxidation in insular Southeast Asia since 1990. \u003cem\u003eEnviron. Res. Lett.\u003c/em\u003e \u003cb\u003e12\u003c/b\u003e, 24014 (2017).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGirkin, N. T. et al. Tropical peatlands in the Anthropocene: The present and the future. \u003cem\u003eAnthropocene\u003c/em\u003e \u003cb\u003e40\u003c/b\u003e, 100354 (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAustin, K. G. et al. Mismatch Between Global Importance of Peatlands and the Extent of Their Protection. \u003cem\u003eConserv Lett\u003c/em\u003e \u003cb\u003e18\u003c/b\u003e, (2025).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHooijer, A. et al. Current and future CO2 emissions from drained peatlands in Southeast Asia. \u003cem\u003eBiogeosciences\u003c/em\u003e \u003cb\u003e7\u003c/b\u003e, 1505\u0026ndash;1514 (2010).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHodgkins, S. B. et al. Tropical peatland carbon storage linked to global latitudinal trends in peat recalcitrance. \u003cem\u003eNat Commun\u003c/em\u003e \u003cb\u003e9\u003c/b\u003e, (2018).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMander, \u0026Uuml;., \u0026Ouml;pik, M. \u0026amp; Espenberg, M. Global peatland greenhouse gas dynamics: state of the art, processes, and perspectives. \u003cem\u003eNew Phytologist\u003c/em\u003e vol. 246 94\u0026ndash;102 Preprint at (2025). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1111/nph.20436\u003c/span\u003e\u003cspan address=\"10.1111/nph.20436\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDadap, N. C. et al. Climate change-induced peatland drying in Southeast Asia. \u003cem\u003eEnvironmental Res. Letters\u003c/em\u003e \u003cb\u003e17\u003c/b\u003e, (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHiraishi, T. et al. \u003cem\u003eSupplement to the 2006 IPCC Guidelines for National Greenhouse Gas Inventories: Wetlands : Methodological Guidance on Lands with Wet and Drained Soils, and Constructed Wetlands for Wastewater Treatment\u003c/em\u003e. (IPCC, Intergovernmental Panel on Climate Change, 2014). (IPCC, Intergovernmental Panel on Climate Change, 2014). (2013).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSasmito, S. D. et al. Half of land use carbon emissions in Southeast Asia can be mitigated through peat swamp forest and mangrove conservation and restoration. \u003cem\u003eNature Communications\u003c/em\u003e \u003cb\u003e16\u003c/b\u003e, (2025).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBudiharto \u003cem\u003eet al. National Forest Reference Level for Deforestation, Forest Degradation and Enhancement of Forest Carbon Stock. In the Context of Decision 12/CP.17 Para 12 UNFCCC\u003c/em\u003e. (2022). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://redd.unfccc.int/files/2nd_frl_indonesia_final_submit.pdf\u003c/span\u003e\u003cspan address=\"https://redd.unfccc.int/files/2nd_frl_indonesia_final_submit.pdf\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAini, F. K. et al. AsiaFlux2025, Pangkalan Kerinci, Riau, Indonesia,. A Synthesis of Groundwater Level-Driven CO2 Dynamics: A Review of Different Land-Use Systems in Sumatra and Borneo, Indonesia. in \u003cem\u003eAsiaFlux Conference 2025\u003c/em\u003e (2025).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePulunggono, H. B. et al. AsiaFlux2025, Pangkalan Kerinci, Riau, Indonesia,. CO2 Emission Estimation from Modelled Groundwater Table in Indonesian Tropical Peatland Ecosystems. in \u003cem\u003eAsiaFlux Conference 2025\u003c/em\u003e (2025).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHirano, T. et al. Impact of Land Use Change and Drought on the Net Emissions of Carbon Dioxide and Methane From Tropical Peatlands in Southeast Asia. \u003cem\u003eAGU Advances\u003c/em\u003e \u003cb\u003e6\u003c/b\u003e, (2025).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTay, C. et al. Satellite radar advances carbon emissions accountability over tropical peat. \u003cem\u003eCommun Earth Environ\u003c/em\u003e \u003cb\u003e6\u003c/b\u003e, (2025).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eApers, S. et al. Tropical Peatland Hydrology Simulated With a Global Land Surface Model. \u003cem\u003eJ Adv. Model. Earth Syst\u003c/em\u003e \u003cb\u003e14\u003c/b\u003e, (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eUrzainki, I. et al. A process-based model for quantifying the effects of canal blocking on water table and CO2 emissions in tropical peatlands. \u003cem\u003eBiogeosciences\u003c/em\u003e \u003cb\u003e20\u003c/b\u003e, 2099\u0026ndash;2116 (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCochrane, M. A. Spatiotemporal Variations in Hydrology Drive Greenhouse Gas Emissions in Tropical Peatlands. \u003cem\u003eAGU Advances\u003c/em\u003e \u003cb\u003e7\u003c/b\u003e, (2026).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHirano, T. et al. Large variation in carbon dioxide emissions from tropical peat swamp forests due to disturbances. \u003cem\u003eCommun Earth Environ\u003c/em\u003e \u003cb\u003e5\u003c/b\u003e, (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDeshmukh, C. S. et al. Net greenhouse gas balance of fibre wood plantation on peat in Indonesia. \u003cem\u003eNature\u003c/em\u003e \u003cb\u003e616\u003c/b\u003e, 740\u0026ndash;746 (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRankin, T., Roulet, N., Humphreys, E., Peichl, M. \u0026amp; Jӓrveoja, J. Partitioning autotrophic and heterotrophic respiration in an ombrotrophic bog. \u003cem\u003eFront Earth Sci. (Lausanne)\u003c/em\u003e \u003cb\u003e11\u003c/b\u003e, (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRankin, T. E., Roulet, N. T. \u0026amp; Moore, T. R. Controls on autotrophic and heterotrophic respiration in an ombrotrophic bog. \u003cem\u003eBiogeosciences\u003c/em\u003e \u003cb\u003e19\u003c/b\u003e, 3285\u0026ndash;3303 (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGirkin, N. T., Turner, B. L., Ostle, N., Craigon, J. \u0026amp; Sj\u0026ouml;gersten, S. Root exudate analogues accelerate CO2 and CH4 production in tropical peat. \u003cem\u003eSoil. Biol. Biochem.\u003c/em\u003e \u003cb\u003e117\u003c/b\u003e, 48\u0026ndash;55 (2018).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eUrzainki, I. et al. Canal blocking optimization in restoration of drained peatlands. \u003cem\u003eBiogeosciences\u003c/em\u003e \u003cb\u003e17\u003c/b\u003e, 4769\u0026ndash;4784 (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMukhaiyar, U. et al. The generalized STAR modelling with three-dimensional of spatial weight matrix in predicting the Indonesia peatland\u0026rsquo;s water level. \u003cem\u003eEnviron Sci. Eur\u003c/em\u003e \u003cb\u003e36\u003c/b\u003e, (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWidiarso, B. et al. Predicting peatland groundwater table and soil moisture dynamics affected by drainage level. \u003cem\u003eSains Tanah\u003c/em\u003e. \u003cb\u003e17\u003c/b\u003e, 42\u0026ndash;49 (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHikouei, I. S. et al. Using machine learning algorithms to predict groundwater levels in Indonesian tropical peatlands. \u003cem\u003eSci. Total Environ.\u003c/em\u003e \u003cb\u003e857\u003c/b\u003e, 159701 (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHikouei, I. S. et al. Machine-learning based spatiotemporal performance analysis of degraded tropical peatland groundwater level numerical model. \u003cem\u003eGroundw. Sustain. Dev.\u003c/em\u003e \u003cb\u003e29\u003c/b\u003e, 101413 (2025).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLi, L. et al. Estimation of Ground Water Level (GWL) for Tropical Peatland Forest Using Machine Learning. \u003cem\u003eIEEE Access.\u003c/em\u003e \u003cb\u003e10\u003c/b\u003e, 126180\u0026ndash;126187 (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYonekura, K. et al. Prediction of groundwater level in Indonesian tropical peatland forest plantations using machine learning. \u003cem\u003eArtificial Intell. Geosciences\u003c/em\u003e \u003cb\u003e6\u003c/b\u003e, (2025).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKoch, J. et al. Water-table-driven greenhouse gas emission estimates guide peatland restoration at national scale. \u003cem\u003eBiogeosciences\u003c/em\u003e \u003cb\u003e20\u003c/b\u003e, 2387\u0026ndash;2403 (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBechtold, M. et al. Large-scale regionalization of water table depth in peatlands optimized for greenhouse gas emission upscaling. \u003cem\u003eHydrol. Earth Syst. Sci.\u003c/em\u003e \u003cb\u003e18\u003c/b\u003e, 3319\u0026ndash;3339 (2014).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWidyatmanti, W. et al. Codification to secure Indonesian peatlands: From policy to practices as revealed by remote sensing analysis. \u003cem\u003eSoil. Secur.\u003c/em\u003e \u003cb\u003e9\u003c/b\u003e, 100080 (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRegulation No 57/2016. \u003cem\u003ePerlindungan Dan Pengelolaan Ekosistem Gambut\u003c/em\u003e (Indonesian Government, 2014).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePrananto, J. A., Minasny, B., Comeau, L., Rudiyanto, R. \u0026amp; Grace, P. Drainage increases CO2 and N2O emissions from tropical peat soils. \u003cem\u003eGlob Chang. Biol.\u003c/em\u003e \u003cb\u003e26\u003c/b\u003e, 4583\u0026ndash;4600 (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNovita, N. et al. Geographic Setting and Groundwater Table Control Carbon Emission from Indonesian Peatland: A Meta-Analysis. \u003cem\u003eForests\u003c/em\u003e \u003cb\u003e12\u003c/b\u003e, 832 (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCobb, A. R. \u0026amp; Harvey, C. F. Scalar Simulation and Parameterization of Water Table Dynamics in Tropical Peatlands. \u003cem\u003eWater Resour. Res.\u003c/em\u003e \u003cb\u003e55\u003c/b\u003e, 9351\u0026ndash;9377 (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFauziah, Prasetyo, L. B., Saribanon, N. \u0026amp; Hayati, N. Vulnerability of peatland fires in bengkalis regency during the ENSO El nino phase using a machine learning approach. \u003cem\u003eMethodsX\u003c/em\u003e \u003cb\u003e14\u003c/b\u003e, 103128 (2025).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eUtomo, B., Oktavia, M., Susilo, Y. \u0026amp; Putri, M. K. Identification of drought endemic areas in Musi Banyuasin regency. \u003cem\u003eKuwait J. Sci.\u003c/em\u003e \u003cb\u003e50\u003c/b\u003e, 168\u0026ndash;173 (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAstiani, D., Burhanuddin, B., Curran, L. M., Mujiman, M. \u0026amp; Salim, R. Effects of Drainage Ditches on Water Table Level, Soil Conditions and Tree Growth of Degraded Peatland Forests in West Kalimantan. \u003cem\u003eIndonesian J. Forestry Res.\u003c/em\u003e \u003cb\u003e4\u003c/b\u003e, 15\u0026ndash;25 (2017).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eEvans, C. D. et al. Rates and spatial variability of peat subsidence in Acacia plantation and forest landscapes in Sumatra, Indonesia. \u003cem\u003eGeoderma\u003c/em\u003e \u003cb\u003e338\u003c/b\u003e, 410\u0026ndash;421 (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eParish, F., Afham, A. \u0026amp; Lew, S. Y. (Serena). Role of the Roundtable on Sustainable Palm Oil (RSPO) in Tropical Peatland Management. in \u003cem\u003eTropical Peatland Eco-management\u003c/em\u003e 509\u0026ndash;533Springer Singapore, (2021). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/978-981-33-4654-3_18\u003c/span\u003e\u003cspan address=\"10.1007/978-981-33-4654-3_18\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHooijer, A. et al. Subsidence and carbon loss in drained tropical peatlands. \u003cem\u003eBiogeosciences\u003c/em\u003e \u003cb\u003e9\u003c/b\u003e, 1053\u0026ndash;1071 (2012).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJauhiainen, J., Hooijer, A. \u0026amp; Page, S. E. Carbon dioxide emissions from an Acacia plantation on peatland in Sumatra, Indonesia. \u003cem\u003eBiogeosciences\u003c/em\u003e \u003cb\u003e9\u003c/b\u003e, 617\u0026ndash;630 (2012).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHirano, T. et al. Effects of disturbances on the carbon balance of tropical peat swamp forests. \u003cem\u003eGlob Chang. Biol.\u003c/em\u003e \u003cb\u003e18\u003c/b\u003e, 3410\u0026ndash;3422 (2012).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMcCalmont, J. et al. Short- and long-term carbon emissions from oil palm plantations converted from logged tropical peat swamp forest. \u003cem\u003eGlob Chang. Biol.\u003c/em\u003e \u003cb\u003e27\u003c/b\u003e, 2361\u0026ndash;2376 (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWeaver, M. M., Garner, A. J., Samanta, D., Mann, M. E. \u0026amp; Horton, B. P. Seasonal variations of tropical cyclone genesis and landfall patterns impacting Southeast Asia in a warmer climate. \u003cem\u003eCommun. Earth Environ.\u003c/em\u003e \u003cb\u003e6\u003c/b\u003e, 866 (2025).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLoisel, J. et al. Expert assessment of future vulnerability of the global peatland carbon sink. \u003cem\u003eNat. Clim. Chang.\u003c/em\u003e \u003cb\u003e11\u003c/b\u003e, 70\u0026ndash;77 (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDadap, N. C. et al. Drainage Canals in Southeast Asian Peatlands Increase Carbon Emissions. \u003cem\u003eAGU Advances\u003c/em\u003e \u003cb\u003e2\u003c/b\u003e, (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eIshikura, K. et al. Carbon Dioxide and Methane Emissions from Peat Soil in an Undrained Tropical Peat Swamp Forest. \u003cem\u003eEcosystems\u003c/em\u003e \u003cb\u003e22\u003c/b\u003e, 1852\u0026ndash;1868 (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePage, S., Hooijer, A., Rieley, J., Banks, C. \u0026amp; Hoscilo, A. The tropical peat swamps of Southeast Asia: in Biotic Evolution and Environmental Change in Southeast Asia 406\u0026ndash;433 (Cambridge University Press, doi:\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1017/CBO9780511735882.018\u003c/span\u003e\u003cspan address=\"10.1017/CBO9780511735882.018\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. (2012).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDeshmukh, C. S. et al. Conservation slows down emission increase from a tropical peatland in Indonesia. \u003cem\u003eNat. Geosci.\u003c/em\u003e \u003cb\u003e14\u003c/b\u003e, 484\u0026ndash;490 (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKiew, F. et al. Carbon dioxide balance of an oil palm plantation established on tropical peat. \u003cem\u003eAgric Meteorol\u003c/em\u003e \u003cb\u003e295\u003c/b\u003e, (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRePPProT \u003cem\u003eRegional Physical Planning Programme for Transmigration (Reppprot)\u003c/em\u003e. (1987).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHe, H. \u0026amp; Roulet, N. T. Improved estimates of carbon dioxide emissions from drained peatlands support a reduction in emission factor. \u003cem\u003eCommun Earth Environ\u003c/em\u003e \u003cb\u003e4\u003c/b\u003e, (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLestari, I., Murdiyarso, D. \u0026amp; Taufik, M. Rewetting Tropical Peatlands Reduced Net Greenhouse Gas Emissions in Riau Province, Indonesia. \u003cem\u003eForests\u003c/em\u003e \u003cb\u003e13\u003c/b\u003e, 505 (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHooijer, A. et al. Benefits of tropical peatland rewetting for subsidence reduction and forest regrowth: results from a large-scale restoration trial. \u003cem\u003eSci Rep\u003c/em\u003e \u003cb\u003e14\u003c/b\u003e, (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNovita, N. et al. Strong climate mitigation potential of rewetting oil palm plantations on tropical peatlands. \u003cem\u003eSci. Total Environ.\u003c/em\u003e \u003cb\u003e952\u003c/b\u003e, 175829 (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFunk, C. et al. The climate hazards infrared precipitation with stations\u0026mdash;a new environmental record for monitoring extremes. \u003cem\u003eSci Data\u003c/em\u003e \u003cb\u003e2\u003c/b\u003e, (2015).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eQiaozhen Mu, M., Zhao \u0026amp; Steven, W. Running. \u003cem\u003eMODIS Global Terrestrial Evapotranspiration (ET) Product (NASA MOD16A2/A3) Algorithm Theoretical Basis Document Collection 5\u003c/em\u003e. (2013).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRCoreTeam., R. A Language and Environment for Statistical Computing. Preprint at (2026).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRobert, J. et al. terra: Spatial Data Analysis, version 1.7\u0026ndash;82. Preprint at (2026). (2026).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHadley Wickham, M. \u0026amp; Girlich Kirill M\u0026uuml;ller \u0026amp; Davis Vaughan. tidyverse: Easily Install and Load the \u0026lsquo;Tidyverse\u0026rsquo;. Preprint at (2025).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChen, T., Guestrin, C. \u0026amp; XGBoost: A Scalable Tree Boosting System. in \u003cem\u003eProceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining\u003c/em\u003e 785\u0026ndash;794ACM, (2016). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1145/2939672.2939785\u003c/span\u003e\u003cspan address=\"10.1145/2939672.2939785\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHikouei, I. S. et al. Using machine learning algorithms to predict groundwater levels in Indonesian tropical peatlands. \u003cem\u003eSci. Total Environ.\u003c/em\u003e \u003cb\u003e857\u003c/b\u003e, 159701 (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMax et al. caret: Classification and Regression Training. Preprint at (2025).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eShapley, L. S. Cambridge University Press,. A value for n-person games. in \u003cem\u003eThe Shapley Value\u003c/em\u003e 31\u0026ndash;40 (1988). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1017/CBO9780511528446.003\u003c/span\u003e\u003cspan address=\"10.1017/CBO9780511528446.003\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLundberg, S. M. \u0026amp; Lee, S. I. A unified approach to interpreting model predictions. in \u003cem\u003eNIPS\u0026rsquo;17: Proceedings of the 31st International Conference on Neural Information Processing Syste\u003c/em\u003e 4768\u0026ndash;4777Curran Associates Inc., New York, United States, (2017).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLundberg, S. M. et al. Explainable machine-learning predictions for the prevention of hypoxaemia during surgery. \u003cem\u003eNat. Biomed. Eng.\u003c/em\u003e \u003cb\u003e2\u003c/b\u003e, 749\u0026ndash;760 (2018).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLundberg, S. M. et al. From local explanations to global understanding with explainable AI for trees. \u003cem\u003eNat. Mach. Intell.\u003c/em\u003e \u003cb\u003e2\u003c/b\u003e, 56\u0026ndash;67 (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBrandon Greenwell. fastshap: Fast Approximate Shapley Values. Preprint at. (2025).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMichael Mayer \u0026amp; Adrian Stando. shapviz: SHAP Visualizations. Preprint at (2025).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiu, Y., Just, A. \u0026amp; Mayer, M. SHAPforxgboost: SHAP Plots for \u0026lsquo;XGBoost\u0026rsquo;.\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"CO2 emission, groundwater level, machine learning, peatland hydrological unit, soil respiration, tropical peatlands","lastPublishedDoi":"10.21203/rs.3.rs-9002157/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-9002157/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eTropical peatlands store a disproportionately large fraction of global soil carbon and emit large amounts of soil CO\u003csub\u003e2\u003c/sub\u003e from peat decomposition and root respiration. Accurate quantification of peatland carbon emissions requires a prediction that is sensitive to the space-time dynamics of the environmental factors. This study estimated a biweekly peat CO\u003csub\u003e2\u003c/sub\u003e emission representing total soil respiration (Rs) representing the whole peatland ecosystem in Rupat Island, Indonesia. The Rs was estimated using nonlinear CO\u003csub\u003e2\u003c/sub\u003e to groundwater level/GWL response curves reviewed from Southeast Asian chamber studies. The GWL dataset was predicted using Extreme Gradient Boosting (XGBoost), integrating dense GWL measurements (Jan-2019 to Apr-2025) against dynamic and static predictors. Spatial upscaling of predicted GWL (\u0026minus;\u0026thinsp;32.19\u0026thinsp;\u0026plusmn;\u0026thinsp;5.58 cm) resulting in estimated annual Rs of 4.02 to 5.67 Mt CO\u003csub\u003e2\u003c/sub\u003e and totaling 28.15 to 39.66 Mt CO\u003csub\u003e2\u003c/sub\u003e cumulative, using general and land use-based response curves. Our study reported relatively low interannual and seasonal variations (\u0026lt;\u0026thinsp;1 Mt CO\u003csub\u003e2\u003c/sub\u003e y-1), although cultivated and drained shallow peats consistently act as emission hotspots during dry years. Our Rs estimates exceeded the global and national emission factors (EFs) and other Rs-GWL prediction methods by 1.5 to 6 times, suggesting potential for implementation in CO2 modelling at the ecosystem scale.\u003c/p\u003e","manuscriptTitle":"Machine Learning-Aided Groundwater Level and CO 2 Emission Estimations in Tropical Peatlands using Global and Regional Scenarios","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-03-10 07:06:11","doi":"10.21203/rs.3.rs-9002157/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"ddfa8b0c-0f56-407b-8ae1-1dcf00b95e07","owner":[],"postedDate":"March 10th, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":64211122,"name":"Earth and environmental sciences/Biogeochemistry"},{"id":64211123,"name":"Earth and environmental sciences/Climate sciences"},{"id":64211124,"name":"Biological sciences/Ecology"},{"id":64211125,"name":"Earth and environmental sciences/Ecology"},{"id":64211126,"name":"Earth and environmental sciences/Environmental sciences"}],"tags":[],"updatedAt":"2026-04-23T07:26:49+00:00","versionOfRecord":[],"versionCreatedAt":"2026-03-10 07:06:11","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-9002157","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-9002157","identity":"rs-9002157","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}