Spatio-temporal analysis of wetland loss in the lower Mekong River Basin based on surface water detection datasets and machine learning

Spatio-temporal analysis of wetland loss in the lower Mekong River Basin based on surface water detection datasets and machine learning | 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 Research Article Spatio-temporal analysis of wetland loss in the lower Mekong River Basin based on surface water detection datasets and machine learning Yusuke Hiraga, Mayu Aoki, So Kazama This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-4781968/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 Wetland loss and degradation is a major global issue in which its detailed estimation of spatio-temporal distribution will be a key for understanding the dynamics and subsequent impact assessment studies. This study aimed to estimate the spatio-temporal distributions of wetland loss, analyze its geographical characteristics, and quantify the likelihood of loss occurrence for existing wetlands in the lower Mekong River Basin in Cambodia. Using the global surface water detection datasets, the spatiotemporal distribution of wetland loss with high resolution (30m) over the entire study area (140km×210km) during 1984–2021 was estimated. Statistically significant differences were found in the distance from urban areas and distance from river channels for the existing wetlands and lost wetlands as of 2021, in which the lost wetlands tend to locate closer to urban areas. Subsequent Land Use/Land Cover after the wetland loss was found to be mainly croplands (72.2%) in the study area. Though our estimate overall agrees with the recent global-scale estimate, our estimate resulted in notable ratio of rangelands (11.3%), which represents the unique characteristics of floodplain wetlands in the lower Mekong River Basin. The Random Forest and Light GBM algorithms-based wetland loss prediction models resulted in good statistical evaluation metrics. In both models, the distance from river channels was found to be the most important feature for classifying existing wetlands and lost wetlands. Application of the developed models successfully provided the map of likelihood of wetland loss for existing wetlands in the study area. Landfill Flood Floodplain Dams Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Figure 8 Figure 9 Figure 10 1. Introduction Wetland loss and degradation is a major global issue, drawing the attention of researchers, policymakers, and environmentalists around the world (Nicholls et al., 1999 ; Coleman et al., 2008 ; Davidson, 2014 ; Mitsch & Gosselink, 2015 ; Hu et al., 2017 ; Fluet-Chouinard et al., 2023 ). The rapid decline of wetlands due to anthropogenic activities such as urbanization, agriculture, and industrialization, and climate change poses severe threats to the critical functions of wetlands, including ecological habitats for diverse species, water purification processes, and natural buffers against floods and storms (Nicholls et al., 1999 ; Nicholls, 2004 ; Verhoeven et al., 2006 ; Erwin, 2009 ; Moreno-Mateos et al., 2012 ; Asselen et al., 2013 ; Davidson, 2014 ). The impacts of wetland loss are extensive and multifaceted. One of the most immediate consequences is the increased risk of flooding (Bullock & Acreman, 2003; Hiraga et al., 2018 ). Wetlands function as natural sponges, absorbing excess rainfall and releasing it slowly, thus reducing flood risks and securing water resources (Kazama et al., 2007 ; Acreman & Holden, 2013; Shrestha et al., 2016 ; Narayan et al., 2017 ). When wetlands are degraded or lost, this natural flood mitigation is lost, leading to more frequent and severe flooding events. Additionally, wetlands are crucial for maintaining water quality. They filter pollutants, sediments, and nutrients from water, contributing to cleaner rivers, lakes, and groundwater (Johnston, 1991 ; Vymazal, 2010 ; Hammer & Bastian, 2020 ). The degradation and reclamation of wetlands compromises this filtration capability, leading to poorer water quality and higher pollution levels (Hiraga et al., 2020 ), which can have serious implications for human health and biodiversity (Verhoeven & Meuleman, 1999). The lower Mekong River Basin in Cambodia is an example where wetlands play essential roles for locals in various aspects (Ringler and Cai, 2006 ; MacAlister & Mahaxay, 2009 ; Arias et al., 2014a ). Their floodplain wetlands, formed mainly by river flooding, are sources for agriculture, fisheries, water resources, and cultural development in the lower Mekong River Basin, one of the most important geographical characteristics of the region. Recent studies based on the numerical modeling and field surveys showed that this important wetland is facing the danger of being lost due to anthropogenic landfill (Hiraga et al., 2018 ; Hiraga et al., 2020 ), dam constructions and climate change (Arias et al., 2014b ). Thus, it is required to address the impacts of wetland loss for their potential mitigations in the region. To effectively address the impacts of wetland loss for their potential mitigations, a detailed estimation of the spatio-temporal distribution of wetland loss is essential. Such estimation can help researchers identify the locations and timings of significant wetland changes, providing critical data for accurate impact evaluations and predictions (Ozesmi & Bauer, 2002 ; Rebelo et al., 2009 ). Recent studies have utilized remote sensing techniques to estimate the wetland loss (Ozesmi and Bauer, 2002 ; Rebelo et al., 2009 ; Guo et al., 2017 ; Mahdavi et al., 2018 ), including the global-scale estimations (Davidson, 2014 ; Hu et al., 2017 ; Fluet-Chouinard et al., 2023 ) and regional-scale estimations (Zhang et al., 2009 ; Gong et al., 2010 ; Teheri et al., 2010; Chen et al., 2014 ; Dang et al., 2021 ). While the global-scale analyses offer a broad overview, regional-scale analyses with high resolution are crucial for understanding specific local conditions and impacts. Due to the limited number of studies estimating spatio-temporal distributions of wetland loss in the lower Mekong River Basin in Cambodia, this study aimed to estimate the spatio-temporal distributions of wetland loss, analyze its geographical characteristics, and quantify the likelihood of loss occurrence for existing wetlands in the region. The wetlands in the lower Mekong River Basin in Cambodia are largely dependent on seasonal river flooding through bank overtopping and water conveyance by many agricultural canals, called “colmatage”, making the wetland dynamics processes unique. To deal with such unique flood plain wetland changes, this study used the surface water detection dataset to estimate the wetland loss. Section 2 describes the study area and datasets. The methodology is described in Section 3. Section 4 demonstrates the estimated spatio-temporal distributions of wetland loss, analyzes the geographical characteristics of the lost wetlands, and discusses the machine learning model-based prediction results. Finally, Section 5 summarizes the findings. 2. Study area and datasets 2.1. Study area This study focuses on wetlands in the lower Mekong River Basin in Cambodia. The Mekong River basin can be broadly divided into the upper and lower reaches, with the upper reaches bordered by China and Laos. The lower basin encompasses Laos, Thailand, Cambodia, and Vietnam, covering approximately 80% of the total basin area (Wang et al., 2024 ). Geographically, the region is subdivided into four areas: the Northern Highlands, the Khorat Plateau, the Tonle Sap Basin, and the Mekong Delta (Gupta, 2009 ). The Mekong River flows through a broad valley east of the Khorat Plateau and into the Tonle Sap Basin. The Tonle Sap Basin is a vast alluvial plain surrounded by hills, with Tonle Sap Lake, the largest freshwater lake in Southeast Asia, situated in its western and central parts. Near Phnom Penh, the Mekong River divides, forming the Bassac River and the Mekong Delta. The Mekong and Bassac rivers branch into numerous smaller channels along their course, with the Mekong Delta covering an area of about 62,520 km². Our study area, which is an area approximately 140 km×210 km around Phnom Penh (Fig. 1 ), the capital of Cambodia, includes the Tonle Sap Basin, located in the southern part of the lower Mekong River basin, and a part of the Mekong Delta. The climate of the lower Mekong River basin is classified as a tropical monsoon climate, characterized by distinct wet and dry seasons. Consequently, the river's flow fluctuates significantly throughout the year (Li et al., 2017 ). The flow during the flood season, which extends from June to November, accounts for approximately 80% of the total annual flow (Try et al., 2020). During the flood season, large-scale inundations frequently occur in the low-lying floodplains in the lower Mekong River basin. Such inundations occur owing to not only bank overtopping or failure but also water conveyance through many agricultural canals, called “colmatage” (Fig. 2 a; Amano and Kazama, 2016; Hiraga et al., 2018 ). Such agricultural canals were mainly developed intentionally breaking a portion of the levee, so that river water can be provided to the surroundings, forming large floodplain wetlands during the flood season when river water level is high. The conveyed water from rivers includes rich nutrients, contributing to rice farming, aquatic lives, and vegetation growths in the floodplain wetlands (Fig. 2 b; Hiraga et al., 2018 ). Such wetlands are unique and important for not only regional perspective but also global perspective in food production and greenhouse gas emissions (Arai et al., 2018). Hence, the flood-inundation process is a vital water and nutrient source for agriculture, fisheries, and ecosystems in the region (Kummu and Sarkkula, 2008 ; Sok et al., 2022 ). This river flood-driven floodplain wetland is the dominant type of wetlands in the study area, which is our focus in this study. The inundated area typically covers more than 10,000 km², and in particularly large flood years, this area can expand to 30,000–40,000 km² with water depths exceeding 2 meters in the lower Mekong River Basin (MRC, 1997 ). In recent years, losing such important floodplain wetlands is a critical issue in the lower Mekong River Basin (Huu Nguyen et al., 2016 ; Fluet-Chouinard et al., 2023 ). Anthropogenic factors, such as construction of dams, dykes, and roads (Huu Nguyen et al., 2016 ), and wetland drainage and landfill (Hiraga et al., 2018 ; 2020 ) are frequently cited causes of wetland loss. Climate change also significantly altered the regional hydrology, influencing the wetland area (Pokhrel et al., 2018 ; Arias et al., 2019 ; Dang et al., 2021 ). Horton et al. ( 2022 ) found that the planned infrastructural development will cause substantial changes in the wetlands in the Cambodian Mekong floodplain, potentially impacting ecosystems and people's livelihoods. Given the dependence of local communities on agriculture and fisheries sustained by the natural resources of the wetlands, such changes are expected to have serious impacts, emphasizing the importance of better understanding the wetland loss dynamics. 2.2. Datasets This study used two types of Global Surface Water (Jean-Francois et al., 2016): (1) Monthly Water History dataset and (2) Occurrence dataset. The Monthly Water History from Global Surface Water (Jean-Francois et al., 2016) was used to estimate the area of wetland loss in the study area. This Monthly Water History dataset is based on the detection of surface water using spectral analysis of satellite images, with a spatial resolution of 30 m×30 m. The temporal coverage extends from March 1984 to December 2021. In the Monthly Water History dataset, the water surface detection is based on satellite observations conducted over an 8-day cycle, resulting in multiple observations in a month. This study defined a grid cell as water surface grid for each month when the grid cell is classified as water surface in any of the observations during the month. A no-data classification was assigned for periods when observations are not available mainly because of the cloud cover or equipment malfunctions. The Occurrence dataset was used to determine the permanent water areas in the target region. The Occurrence dataset includes the percentage of all monthly water history data for each grid cell that was classified as water bodies throughout the study period. In this study, a grid cell in which 100% of the data was identified as a water area was classified as a permanent water area, such as a river or a lake. This study used the Sentinel-2 10m land use/land cover (LULC) data to assess LULC changes following wetland loss. This LULC dataset consists of annual data from 2017 to 2021, derived from ESA Sentinel-2 images with a spatial resolution of 10 m × 10 m. In this study, the observation in the year 2021 was used to examine the land use/land cover after the wetland loss. The LULC data were spatially interpolated to a resolution of 30 m×30 m to align with the spatial resolution of the other datasets used in the analysis. During this process, the most frequently occurring LULC category within each 30 m × 30 m grid cell was assigned as the LULC for the target grid cell. Due to the differences in the definition of "water area" between the LULC data and the Global Surface Water data, any grid cell classified as a water area and identified as having lost wetlands in this study was excluded from the analyses in Chap. 4 and subsequent chapters. The ALOS Global Numerical Land Surface Model (ALOS World 3D 30m×30m) was used as the elevation data. This dataset was compiled using the Optical Sensor and Panchromatic Remote-sensing Instrument for Stereo Mapping (PRISM) onboard the Advanced Land Observing Satellite Daichi (ALOS), which was operational from January 2006 to April 2011. 3. Methodology This study first estimated the spatiotemporal distribution of wetland loss with high resolution (30m) over the entire study area (140km×210km) during 1984–2021. It should be noted that this study attempted to estimate the distribution of wetlands and their losses using the water surface detection data, not the LULC data. It is well known that different LULC datasets include large variations in the water-related classification (Nakaegawa, 2012 ). Furthermore, it is difficult to accurately capture the changes of river flood-driven wetlands using LULC datasets available with less frequency (i.e., usually annual or every five years). Thus, this study attempted to detect wetland changes purely based on the water surface detection datasets with high resolution in time and space (Jean-Francois et al., 2016). The factors influencing wetland loss and LULC after the wetland loss were then examined. Further, we conducted machine learning analysis using the identified factors as input values, constructing a model to predict the probability of wetland loss in the study area. Applying this model to the existing wetlands in the study area, the probability of future wetland loss was estimated. The following subsections describe the detailed procedure for each step: 3.1. Spatiotemporal distribution of Wetland loss We first used the Monthly Water History dataset to classify each 30m×30m grid cell as water and non-water grid cell for each month of August, September, and October during 1984–2021 in the study area. August, September, and October were selected as they are the latter part of the rainy season in the lower Mekong River Basin (MRC, 2005 ). When water/non-water classification is unavailable for all three months in a year due to missing data in the Monthly Water History dataset for a grid cell, we performed the interpolation assuming the same classification as the previous year for the grid cell. Figure 3 describes the process above. Using the water/non-water classification, the timing of the wetland loss was estimated identifying the year in which water classification transitioned to non-water classification for all the grid cells. Changes in water areas are influenced by various factors such as fluctuations in river levels and precipitation. Hence it is difficult to determine wetland loss based solely on water/non-water classification data at a specific time or over a short period. To accurately assess wetland loss, it is necessary to consider long-term trends and patterns. We used the following two conditions to define the loss of a wetland: The first condition is that a grid cell must have been classified as non-water for at least five consecutive years between 2017 and 2021. This condition was set to prevent the detection of a grid cell temporarily dried up as non-wetland due to drought-induced water level decrease. Figure 4 shows an example of applying the first condition. The second condition is that, for a grid cell, 50% or more of the classifications must be water in the time series of classifications from 1984 to the year in which the wetland was estimated to have been lost. This condition was set to prevent the detection of a grid cell temporally covered by water as wetland due to high rainfall season etc. Figure 5 shows an example of applying the second condition. We estimated the year of wetland loss for all grid cells in the study area, which led to estimating the spatiotemporal distribution of wetland loss with high resolution. Furthermore, based on the estimated spatiotemporal distribution of wetland loss, factors influencing wetland loss were examined based on previous studies. This study focused on the distance from the river channels and the distance from the urban areas as influential factors on the wetland loss, following Hiraga et al. ( 2018 ) who developed the wetland landfill prediction model using those variables. Here, we defined grid cells as urban areas when satisfying the following conditions: grid cells belonging to the urban category in the LULC dataset and not being wetlands at any time from 1984 to 2021 in the Monthly Water History dataset. We calculated the shortest distance from the river channel and from the urban areas for all the existing and lost wetlands in the study area. Then, we performed the Mann-Whitney U-test (McKnight and Najab, 2010 ) to evaluate the statistical significance in the differences in distances from river channels and distances from urban areas between existing wetlands and lost wetlands. Moreover, the LULC in 2021 (latest year in the study period) where wetland loss occurred was also examined to investigate the pattern of LULC at lost wetlands. 3.2. Machine learning analysis to develop wetland loss prediction models We used machine learning analysis to (1) understand the feature importance of each factor affecting wetland loss and (2) estimate the probability of wetland loss in the study area. The methods employed are supervised learning (Random Forest and Light GBM) to construct a binary classification model for estimating the probability of wetland loss. 3.2.1. Random Forest classifier Random Forest is an algorithm that combines two methods: decision trees and bagging, an ensemble learning technique (Breiman, 2001 ). A decision tree is an algorithm that classifies binary problems by organizing them in a hierarchical structure. Random forests create an accurate model by combining multiple decision trees. Bagging is the method used for combining these decision trees. It involves replicating and extracting data from the original dataset. A model is created for each replicated dataset, and the final prediction is made referring to a majority decision based on the predictions from these models. This study performed the Random Forest classification analysis using the scikit-learn in Python ver3.8 (Pedregosa et al., 2011 ). The used parameters in the Random Forest model were mainly its default values, including the followings: (1) n_estimators: the number of decision trees; (2) criterion: criterion for the split; (3) max_depth: maximum depth of the decision tree; (4) min_samples_split: minimum number of samples required to split a node; (5) min_samples_leaf: minimum number of samples required for a leaf node; (6) max_features: maximum number of features to be used in each decision tree split; (7) random_state: random number seed to ensure model reproducibility. The parameters used in this study are listed in Table 1 . Table 1 Parameters of Random Forest Parameters Applied values in this study n_estimators 100 criterion gini max_depth None min_sample_split 2 min_sample_leaf 1 max_features auto random_state 42 3.2.2. Light GBM classifier Light GBM utilizes gradient boosting as its ensemble learning technique (Ke et al., 2017 ). In gradient boosting, a decision tree model is constructed from the original data, and the difference between the predicted value by this model and the actual value is calculated. Light GBM is one of the fastest algorithms for processing large amounts of data in machine learning using gradient boosting. This study performed the Light GBM classification analysis using the scikit-learn in Python ver3.8 (Pedregosa et al., 2011 ). The used parameters in the Light GBM model were mainly its default values, including the followings: (1) boosting_type: applied boosting method; (2) num_leaves: the number of leaves in the decision tree; (3) learning_rate: controls the importance of each decision tree on the prediction; (4) n_estimators: the number of decision trees to be trained; (5) max_depth: maximum depth of the decision tree; (6) min_child_samples: minimum number of samples required for a leaf node, preventing overfitting; (7) subsample: the percentage of subsampling of the training data. (8) colsample_bytree: the ratio of features used to train each decision tree. The parameters used in this study are listed in Table 2 . Table 2 Parameters of LightGBM Parameters Applied values in this study boosting_type gbdt num_leaves 400 learning_rate 0.1 n_estimators 800 max_depth -50 min_child_sample 20 subsample 1.0 colsample_bytree 1.0 3.2.3. Development of the machine learning model For the development of the machine learning models, we first divided all grid cells representing existing wetlands (total of 12,300,000 cells) into two groups in half, one for train and test of the model, the other for predicting wetland loss. As the objective variables of the model, we used the existing wetland grid cells in one group (labelled as “existing wetland”) and the lost wetland grid cells (labelled as “lost wetlands”). The objective variables were divided as 8:2 for training and testing the model. The explanatory variables include the elevation, distance from the river channel, and distance from urban areas. Hiraga et al. ( 2018 ) showed that the wetland landfill can be modelled using those variables for the region. The accuracy of the developed model was evaluated using the statistic metrics described in the following section. Subsequently, the developed model was applied to the other group of grid cells representing existing wetlands that were not used to construct the model, to estimate the probability of loss of existing wetlands. 3.2.4. Model evaluation metrics We used statistical metrics to quantify the prediction accuracy of the developed machine learning model. Accuracy, Precision, Recall, and F1-score are commonly used evaluation metrics for machine learning models. Accuracy is the percentage of correct predictions made by the model. Precision represents the percentage of data that is actually “positive” out of the test data that the model predicts to be “positive”. Recall represents the percentage of test data that the model was able to predict to be “positive” out of the test data that were actually “positive”. F1-score is an index that simultaneously evaluates both the precision and recall. These metrics were calculated as follows: $$\:\begin{array}{c}Accuracy=\frac{\text{N}\text{u}\text{m}\text{b}\text{e}\text{r}\:\text{o}\text{f}\:\text{c}\text{o}\text{r}\text{r}\text{e}\text{c}\text{t}\:\text{p}\text{r}\text{e}\text{d}\text{i}\text{c}\text{t}\text{i}\text{o}\text{n}\text{s}}{\text{T}\text{o}\text{t}\text{a}\text{l}\:\text{n}\text{u}\text{m}\text{b}\text{e}\text{r}\:\text{o}\text{f}\:\text{p}\text{r}\text{e}\text{d}\text{i}\text{c}\text{t}\text{i}\text{o}\text{n}\text{s}}\:\#\left(1\right)\end{array}$$ $$\:\begin{array}{c}Precision=\frac{\text{T}\text{r}\text{u}\text{e}\:\text{P}\text{o}\text{s}\text{i}\text{t}\text{i}\text{v}\text{e}\text{s}}{\text{T}\text{r}\text{u}\text{e}\:\text{P}\text{o}\text{s}\text{i}\text{t}\text{i}\text{v}\text{e}\text{s}\:+\:\text{F}\text{a}\text{l}\text{s}\text{e}\:\text{P}\text{o}\text{s}\text{i}\text{t}\text{i}\text{v}\text{e}\text{s}}\#\left(2\right)\end{array}$$ $$\:\begin{array}{c}Recall=\frac{\text{T}\text{r}\text{u}\text{e}\:\text{P}\text{o}\text{s}\text{i}\text{t}\text{i}\text{v}\text{e}\text{s}}{\text{T}\text{r}\text{u}\text{e}\:\text{P}\text{o}\text{s}\text{i}\text{t}\text{i}\text{v}\text{e}\text{s}\:+\:\text{F}\text{a}\text{l}\text{s}\text{e}\:\text{N}\text{e}\text{g}\text{a}\text{t}\text{i}\text{v}\text{e}\text{s}}\#\left(3\right)\end{array}$$ $$\:\begin{array}{c}F1=2\times\:\frac{\text{P}\text{r}\text{e}\text{c}\text{i}\text{s}\text{i}\text{o}\text{n}\times\:\text{R}\text{e}\text{c}\text{a}\text{l}\text{l}}{\text{P}\text{r}\text{e}\text{c}\text{i}\text{s}\text{i}\text{o}\text{n}+\text{R}\text{e}\text{c}\text{a}\text{l}\text{l}}\#\left(4\right)\end{array}$$ where positive represents existing wetlands and negative represents lost wetlands in this study. 4. Results and Discussion 4.1. Spatiotemporal distribution of Wetland loss and its analysis Figure 6 Spatial distribution of lost wetlands with its lost timing (color bar) and still existing wetlands as of 2021 (light blue). Dark blue shows the permanent water area. The area of wetlands loss shows a gradually increasing trend from 1993 to 2010, with a significant area recorded during 2012–2014. Although economic development and population growth in Cambodia may have contributed to the increase in lost wetlands, no significant relationship was observed (Figure A1 ). GDP and population growth alone may not account for the factors influencing wetland loss, such as wetland development and dam construction in the upper reaches. This study also examined the time series of the river water level in the study area based on the assumption that changes in floodplain wetlands may be strongly related to variations in river levels (Figure A2 ). Though the small area of wetland loss coincided in the year 2011, the historical flood year showing notably high-water level, it did not clearly explain the changes in wetland areas partially due to the limited consideration of water levels at only three points in the study area. Though accurately capturing local landfilling pattern or local population dynamics requires deep analysis as the information is usually not available in public, such analysis will help better explaining how the timing and location of wetland loss is determined in the region. Such studies are left for future studies. This study further analyzed the factors influencing wetland loss based on our estimates shown in Fig. 6 . Figure 7 shows the box plots of the (a) distances from the urban areas and (b) distances from the river channels for all the existing wetlands and lost wetlands as of 2021 in the study area. Figure 7 implies that the lost wetlands tend to locate close to the urban areas and far from the river channels, which is consistent with the findings of Hiraga et al. ( 2018 ) who examined the geographical characteristics of the limited number (~ 100) of landfilled wetlands in the lower Mekong River Basin. Table 3 summarizes the distances from urban areas and river channels for existing wetlands and lost wetlands with the Mann-Whitney U-test results. The Mann-Whitney U-test results ( p < 0.01) suggest that the differences in the distances between existing wetlands and lost wetlands are statistically significant. Wetland loss tends to occur at the locations close to urban areas potentially because of the gradual and spatially continuous expansion of urban areas and convenience in carrying necessary resources for urban developments. Meanwhile, wetland loss tends to occur at locations far from river channels potentially because of low water level in wetlands which is more vulnerable to hydrological regime changes and easier to landfill. Table 3 Summary of the statistics of the distances from urban areas and river channels for existing wetlands and lost wetlands with the Mann-Whitney U-test results Median [km] Mean [km] p -values Shortest distance from urban areas Existing wetlands 1.00 1.26 < 0.01 Lost wetlands 0.60 0.86 Shortest distance from river channels Existing wetlands 5.88 10.2 < 0.01 Lost wetlands 8.61 12.6 This study investigated the subsequent LULC after the wetland loss at all the grid cells classified as wetland loss. Figure 8 shows the spatial distribution of the subsequent LULC after the wetland loss at all the grid cells classified as wetland loss in the study area. Figure 8 shows that most of the wetlands were found to be turned into croplands, especially at the downstream (i.e., south) of the study area. Meanwhile, Fig. 8 shows some of the rangelands (i.e., grassland, shrubland, etc.) after the wetland loss. Table 3 summarizes the LULC at the wetland loss area. Table 3 Subsequent LULC (2021) after the wetland loss at all the grid cells classified as wetland loss in the study area This study Fluet-Chouinard et al. ( 2023 ) Classification Ratio [%] Classification Ratio [%] Crops 72.2 Upland croplands 61.7 Rangeland 11.3 Flooded rice 18.2 Built (Urban) 7.5 Urban areas 8.0 Trees 5.1 Forestry 4.7 Flooded vegetation 3.5 Wetland cultivation 4.3 Bare ground 0.4 Pasture 2.0 Peat extraction 0.9 Table 3 also shows the percentage of LULC at the lost inland wetlands worldwide, estimated by Fluet-Chouinard et al. ( 2023 ), to compare our estimates with the global scale analysis. Table 3 shows that crops (i.e., agricultural lands) account for more than 70% of the total wetland loss area, which is a similar tendency to Fluet-Chouinard et al. ( 2023 ). Though the LULC percentage basically shows similar results with Fluet-Chouinard et al. ( 2023 ), our estimate for the lower Mekong River Basin resulted in 11.3% of rangelands (Table 3 ). This may be a unique feature for seasonal floodplain wetlands, such as the lower Mekong River Basin. The extent of floodplain wetlands at the lower Mekong River Basin is largely affected by seasonal river flooding, thus the wetland loss due to the hydrological regime changes can drastically occur. For instance, the upstream dam constructions or climate change can cause more severe and frequent droughts (Pokhrel et al., 2018 ; Yang et al., 2019 ), which can result in significant amount of downstream wetland loss. Such hydrological regime change-induced wetland loss rather than urbanizations or agriculturalizations may explain the wetlands being turned into rangelands or other natural LULC. Other inland wetlands in the world are not necessarily sustained by river flood, potentially explaining the difference in Table 3 . 4.2. Machine learning analysis to develop wetland loss prediction models This study used the Random Forest and Light GBM algorithms to develop wetland loss prediction models for the study area. Table 4 summarizes the statistical metrics of the developed wetland loss prediction models. Overall, both algorithms resulted in developing good models as all the metrics reached above 80%. Though the Accuracy metric was lower, the Random Forest-based model resulted in notably better performances in the other metrics, especially F1-score, which can consider the trade-off (i.e., balance) of Precision and Recall. It should be noted that, though this study judged the statistical metrics in Table 4 good enough for understanding the wetland changes in large extent, more efforts in tuning the model parameters may help developing more accurate models. Such analysis will be a trade-off issue of the computational cost and model accuracy, which can be considered in the future. Table 5 shows the feature importance of the developed wetland loss prediction models. Both models show that the distance from river channels was found to be the most important feature for classifying existing wetlands and lost wetlands, which is considered to represent the characteristic of the floodplain wetlands in the lower Mekong River Basin well. In the Random Forest model, the importance of the distance from river channels was notably high (69.6%), while the importance of elevation was found to be less (4.9%). In contrast, in the Light GBM model, the importance of the elevation was found to be much higher (26.1%), resulting in mild differences among the considered features. The feature importance differed greatly depending on the algorithm used, suggesting large uncertainty due to the model algorithm. It is well established that employing multiple algorithms can lead to more accurate prediction with quantifying uncertainty (Abdar et al., 2021 ), which can be confirmed in our analysis too. Figure 9 shows the application of the developed models to the existing wetlands as of 2021 (not used for the model development), showing the predictions of the existing wetlands becoming the lost wetlands. Thus, the values in Fig. 9 can be interpreted as the likelihood of existing wetlands being lost. The estimated likelihoods using the two different models were generally consistent in their spatial distribution. The tendency of the likelihood of being higher at the edge of the floodplain can be found, which represents the strong effect of distance from river channels. The large extent of the wetlands with higher likelihood of being lost is distributed downstream (i.e., south) of the study area, especially in the Light GBM model-based result (Fig. 9 b), possibly due to the lower elevation over the area. Though Fig. 9 does not provide probability of wetland loss with its timing (e.g., 20% in 2050s etc.), which is a remaining important research issue, it can still be useful for an interregional comparison in the likelihood of wetland loss for the study area. Our proposed method can provide the wetland loss likelihood with very high spatial resolution, demonstrating the potential to be considered in further numerical model-based analysis to assess the impact of wetland loss on regional hydrology, climatology, ecology, and agricultural/fishery productivity. Such studies can quantify the value of the regional wetlands on the local and global, contributing to emphasizing the importance of the wetlands. This is beyond the scope of this study and will be addressed in the future. This study includes some of the limitations that need to be addressed in the future. First, the methodology to define wetland loss, especially for floodplain wetlands, can be further improved. Testing various criteria in durations being covered by water or not will be ideal to accurately capture the regional wetland change dynamics. Also, the other explanatory variables can be considered for developing a more accurate wetland loss prediction model. Examples are population and distance from local roads. However, data availability and spatial and temporal resolution of the datasets will be a key issue on whether such inputs can be used to develop a model or not. Lastly, the subsequent LULC prediction after the wetland loss will be an interesting research issue. In fact, this study already attempted to predict the subsequent LULC after the wetland loss using similar inputs in the Light GBM algorithm (not shown), which did not result in high accuracy models (less than 30% of all the statistical metrics). Predicting future LULC requires various inputs and considering LULC change dynamics, which can be included in the future analysis. Table 4 Statistical metrics of the developed wetland loss prediction models Accuracy Precision Recall F1-score Random forest 82.8% 87.8% 92.5% 90.1% Light GBM 86.2% 84.3% 86.0% 82.5% Table 5 Feature importance of the developed wetland loss prediction models Distance from urban areas Distance from river channels Elevation Random forest 25.5% 69.6% 4.9% Light GBM 36.6% 37.3% 26.1% 5. Conclusion This study aimed to estimate the spatio-temporal distributions of wetland loss, analyze its geographical characteristics, and quantify the likelihood of loss occurrence for existing wetlands in the lower Mekong River Basin in Cambodia. The main findings of this study are summarized below: Using the Global Surface Water data (Jean-Francois et al., 2016), this study estimated the spatiotemporal distribution of wetland loss with high resolution (30m) over the entire study area (140km×210km) during 1984–2021. The statistically significant differences were found in the distance from urban areas and distance from river channels for the existing wetlands and lost wetlands as of 2021, in which the lost wetlands tend to locate closer to urban areas (existing wetlands: mean of 0.86 km; lost wetlands: mean of 1.26 km) and more far to river channels (existing wetlands: mean of 5.88 km; lost wetlands: mean of 8.61 km). Subsequent LULC after the wetland loss was found to be mainly croplands (72.2%) in the study area. Though our estimate overall agrees with the recent global-scale estimate (Fluet-Chouinard et al., 2023 ), our estimate resulted in notable ratio of rangelands (11.3%), which represents the unique characteristics of floodplain wetlands in the lower Mekong River Basin. The Random Forest and Light GBM algorithms-based wetland loss prediction models resulted in good statistical evaluation metrics (> 80%). In both models, the distance from river channels was found to be the most important feature for classifying existing wetlands and lost wetlands. Application of the developed models successfully provided the map of likelihood of wetland loss for existing wetlands in the study area. This study provides useful information and insights on better understanding the spatio-temporal distribution of wetland loss in one of the most important floodplain wetland areas, which can be used for impact assessment studies for their potential mitigations in the region. Declarations Competing Interests The authors have no relevant financial or non-financial interests to disclose. Funding The author(s) received no financial support for the research, authorship, and/or publication of this article. Author Contributions Yusuke Hiraga and So Kazama contributed to the study conception and design. The first draft of the manuscript was written by Yusuke Hiraga and all authors commented on previous versions of the manuscript. Material preparation, data collection and analysis were performed by Yusuke Hiraga and Mayu Aoki. The review and edit of the first draft were done by So Kazama. All authors read and approved the final manuscript. Acknowledgement This study was partially supported by the JSPS KAKENHI Grant number 24H00329. Data Availability The datasets generated during and/or analysed during the current study are available from the corresponding author on reasonable request. References Abdar M, Pourpanah F, Hussain S, Rezazadegan D, Liu L, Ghavamzadeh M, Nahavandi S (2021) A review of uncertainty quantification in deep learning: Techniques, applications and challenges. Inform fusion 76:243–297 Arias ME, Cochrane TA, Elliott V (2014a) Modelling future changes of habitat and fauna in the Tonle Sap wetland of the Mekong. Environ Conserv 41(2):165–175 Arias ME, Cochrane TA, Kummu M, Lauri H, Holtgrieve GW, Koponen J, Piman T (2014b) Impacts of hydropower and climate change on drivers of ecological productivity of Southeast Asia's most important wetland. Ecol Model 272:252–263 Arias ME, Holtgrieve GW, Ngor PB, Dang TD, Piman T (2019) Maintaining perspective of ongoing environmental change in the Mekong floodplains. Curr Opin Environ Sustain 37:1–7 Asselen SV, Verburg PH, Vermaat JE, Janse JH (2013) Drivers of wetland conversion: a global meta-analysis. PLoS ONE, 8(11), e81292 Breiman L (2001) Random forests. Mach Learn 45:5–32 Chen L, Jin Z, Michishita R, Cai J, Yue T, Chen B, Xu B (2014) Dynamic monitoring of wetland cover changes using time-series remote sensing imagery. Ecol Inf 24:17–26 Coleman JM, Huh OK, Braud D Jr (2008) Wetland loss in world deltas. J Coastal Res 24:1–14 Dang AT, Kumar L, Reid M, Nguyen H (2021) Remote sensing approach for monitoring coastal wetland in the Mekong Delta, Vietnam: Change trends and their driving forces. Remote Sens 13(17):3359 Davidson NC (2014) How much wetland has the world lost? Long-term and recent trends in global wetland area. Mar Freshw Res 65(10):934–941 Erwin KL (2009) Wetlands and global climate change: the role of wetland restoration in a changing world. Wetlands Ecol Manage 17(1):71–84 Fluet-Chouinard E, Stocker BD, Zhang Z, Malhotra A, Melton JR, Poulter B, McIntyre PB (2023) Extensive global wetland loss over the past three centuries. Nature 614(7947):281–286 Gong P, Niu Z, Cheng X, Zhao K, Zhou D, Guo J, Yan J (2010) China’s wetland change (1990–2000) determined by remote sensing. Sci China Earth Sci 53:1036–1042 Guo M, Li J, Sheng C, Xu J, Wu L (2017) A review of wetland remote sensing. Sensors 17(4):777 Gupta A (2009) Geology and landforms of the Mekong Basin. The Mekong. Academic, pp 29–51 Hammer DA, Bastian RK (2020) Wetlands ecosystems: natural water purifiers? Constructed wetlands for wastewater treatment. CRC, pp 5–19 Hiraga Y, Kazama S, Ekkawatpanit C, Touge Y (2018) Impact of reclamation on the environment of the lower Mekong River basin. J Hydrology: Reg Stud 18:143–155 Hiraga Y, Kazama S, Ekkawatpanit C, Yamada K (2020) Evaluation of the impact of landfill on floodplain water quality in a tropical monsoon region. Hydrol Res Lett 14(1):48–55 Horton AJ, Triet NV, Hoang LP, Heng S, Hok P, Chung S, Kummu M (2022) The Cambodian Mekong floodplain under future development plans and climate change. Nat Hazards Earth Syst Sci 22(3):967–983 Hu S, Niu Z, Chen Y, Li L, Zhang H (2017) Global wetlands: Potential distribution, wetland loss, and status. Sci Total Environ 586:319–327 Huu Nguyen H, Dargusch P, Moss P, Tran DB (2016) A review of the drivers of 200 years of wetland degradation in the Mekong Delta of Vietnam. Reg Envriron Chang 16(8):2303–2315 Johnston CA (1991) Sediment and nutrient retention by freshwater wetlands: effects on surface water quality. Crit Rev Environ Sci Technol 21(5–6):491–565 Kazama S, Hagiwara T, Ranjan P, Sawamoto M (2007) Evaluation of groundwater resources in wide inundation areas of the Mekong River basin. J Hydrol 340(3–4):233–243 Ke G, Meng Q, Finley T, Wang T, Chen W, Ma W, Liu TY (2017) Lightgbm: A highly efficient gradient boosting decision tree. Advances in neural information processing systems , 30 Kummu M, Sarkkula J (2008) Impact of the Mekong River flow alteration on the Tonle Sap flood pulse. AMBIO: J Hum Environ 37(3):185–192 Li D, Long D, Zhao J, Lu H, Hong Y (2017) Observed changes in flow regimes in the Mekong River basin. J Hydrol 551:217–232 MacAlister C, Mahaxay M (2009) Mapping wetlands in the Lower Mekong Basin for wetland resource and conservation management using Landsat ETM images and field survey data. J Environ Manage 90(7):2130–2137 Mahdavi S, Salehi B, Granger J, Amani M, Brisco B, Huang W (2018) Remote sensing for wetland classification: A comprehensive review. GIScience remote Sens 55(5):623–658 McKnight PE, Najab J (2010) Mann-Whitney U Test. The Corsini Encyclopedia of Psychology, 1–1 Mitsch WJ, Gosselink JG (2015) Wetlands. Wiley Moreno-Mateos D, Power ME, Comín FA, Yockteng R (2012) Structural and functional loss in restored wetland ecosystems. PLoS Biol, 10(1), e1001247 MRC (1997) Mekong River Basin Diagnostic Study. Final Report. MKG/R. 97010. Thailand, Bangkok MRC (2005) Overview of the hydrology of the Mekong Basin vol, vol 82. Mekong River Commission, Vientiane, Lao PDR Narayan S, Beck MW, Wilson P, Thomas CJ, Guerrero A, Shepard CC, Trespalacios D (2017) The value of coastal wetlands for flood damage reduction in the northeastern USA. Sci Rep 7(1):1–12 Nicholls RJ, Hoozemans FM, Marchand M (1999) Increasing flood risk and wetland losses due to global sea-level rise: regional and global analyses. Glob Environ Change 9:S69–S87 Nicholls RJ (2004) Coastal flooding and wetland loss in the 21st century: changes under the SRES climate and socio-economic scenarios. Glob Environ Change 14(1):69–86 Nakaegawa T (2012) Comparison of water-related land cover types in six 1-km global land cover datasets. J Hydrometeorol 13(2):649–664 Ozesmi SL, Bauer ME (2002) Satellite remote sensing of wetlands. Wetlands Ecol Manage 10:381–402 Pekel JF, Cottam A, Gorelick N, Belward AS (2016) High-resolution mapping of global surface water and its long-term changes. Nature 540(7633):418–422 Pedregosa F, Varoquaux G, Gramfort A, Michel V, Thirion B, Grisel O, Duchesnay É (2011) Scikit-learn: Machine learning in Python. J Mach Learn Res 12:2825–2830 Pokhrel Y, Burbano M, Roush J, Kang H, Sridhar V, Hyndman DW (2018) A review of the integrated effects of changing climate, land use, and dams on Mekong river hydrology. Water 10(3):266 Rebelo LM, Finlayson CM, Nagabhatla N (2009) Remote sensing and GIS for wetland inventory, mapping and change analysis. J Environ Manage 90(7):2144–2153 Ringler C, Cai X (2006) Valuing fisheries and wetlands using integrated economic-hydrologic modeling—Mekong River Basin. J Water Resour Plan Manag 132(6):480–487 Shrestha S, Bach TV, Pandey VP (2016) Climate change impacts on groundwater resources in Mekong Delta under representative concentration pathways (RCPs) scenarios, vol 61. Environmental science & policy, pp 1–13 Sok T, Oeurng C, Kaing V, Sauvage S, Lu X, Pérez JMS (2022) Nutrient transport and exchange between the Mekong River and Tonle Sap Lake in Cambodia. Ecol Eng 176:106527 Teferi E, Uhlenbrook S, Bewket W, Wenninger J, Simane B (2010) The use of remote sensing to quantify wetland loss in the Choke Mountain range, Upper Blue Nile basin, Ethiopia. Hydrol Earth Syst Sci 14(12):2415–2428 Verhoeven JT, Arheimer B, Yin C, Hefting MM (2006) Regional and global concerns over wetlands and water quality. Trends Ecol Evol 21(2):96–103 Vymazal J (2010) Constructed wetlands for wastewater treatment. Water 2(3):530–549 Wang C, Leisz S, Li L, Shi X, Mao J, Zheng Y, Chen A (2024) Historical and projected future runoff over the Mekong River basin. Earth Sys Dyn 15(1):75–90 Yang J, Yang YE, Chang J, Zhang J, Yao J (2019) Impact of dam development and climate change on hydroecological conditions and natural hazard risk in the Mekong River Basin. J Hydrol 579:124177 Zhang S, Na X, Kong B, Wang Z, Jiang H, Yu H, Dale P (2009) Identifying wetland change in China’s Sanjiang Plain using remote sensing. Wetlands 29:302–313 Supplementary Files Appendix.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-4781968","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":354058065,"identity":"1b38eb9b-0d6e-47cb-894c-d600178c4a31","order_by":0,"name":"Yusuke Hiraga","email":"data:image/png;base64,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","orcid":"https://orcid.org/0000-0002-7791-5431","institution":"Tohoku University: Tohoku Daigaku","correspondingAuthor":true,"prefix":"","firstName":"Yusuke","middleName":"","lastName":"Hiraga","suffix":""},{"id":354058066,"identity":"92c3b8cb-9050-4a84-98f9-bd4ad45f5409","order_by":1,"name":"Mayu Aoki","email":"","orcid":"","institution":"Kyoto University: Kyoto Daigaku","correspondingAuthor":false,"prefix":"","firstName":"Mayu","middleName":"","lastName":"Aoki","suffix":""},{"id":354058067,"identity":"18aa8a7d-8907-4257-b4cf-d6c8688fa37d","order_by":2,"name":"So Kazama","email":"","orcid":"","institution":"Tohoku University: Tohoku Daigaku","correspondingAuthor":false,"prefix":"","firstName":"So","middleName":"","lastName":"Kazama","suffix":""}],"badges":[],"createdAt":"2024-07-22 12:44:58","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-4781968/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-4781968/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":73663571,"identity":"ae3e44b4-ccf0-4ee1-92bc-1296c5e07508","added_by":"auto","created_at":"2025-01-13 11:37:07","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":335084,"visible":true,"origin":"","legend":"\u003cp\u003eStudy area (140km×210km) with (a) land use/land cover obtained from Sentinel-2 10m land use/land cover (LULC) data and (b) elevation obtained from ALOS Global Numerical Land Surface Model.\u003c/p\u003e","description":"","filename":"Picture1.png","url":"https://assets-eu.researchsquare.com/files/rs-4781968/v1/9fbce45a1938187062f7e7af.png"},{"id":73663351,"identity":"3a8f7b60-fb09-4f82-aa0a-1c3cb55d3caf","added_by":"auto","created_at":"2025-01-13 11:29:07","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":529651,"visible":true,"origin":"","legend":"\u003cp\u003e(a) Agricultural canal “Colmatage”. This side is river and the other side is wetlands. (b) Wetland over the lower Mekong River Basin. Photos were taken by the first author at the study area in January 2016.\u003c/p\u003e","description":"","filename":"Picture2.png","url":"https://assets-eu.researchsquare.com/files/rs-4781968/v1/60a2535eef7a800c49511973.png"},{"id":73663360,"identity":"b1d7e9be-8587-48f3-8bd1-cdf5772c29d6","added_by":"auto","created_at":"2025-01-13 11:29:11","extension":"jpg","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":45597,"visible":true,"origin":"","legend":"\u003cp\u003eWater/non-water classification for a grid cell in the study area.\u003c/p\u003e","description":"","filename":"Picture3.jpg","url":"https://assets-eu.researchsquare.com/files/rs-4781968/v1/7746fef794ab4eda583da060.jpg"},{"id":73663352,"identity":"9c163575-22e0-4abd-aef9-2f0fe3fe847e","added_by":"auto","created_at":"2025-01-13 11:29:07","extension":"jpg","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":63273,"visible":true,"origin":"","legend":"\u003cp\u003eExample of determining the timing of the wetland loss, applying the first condition.\u003c/p\u003e","description":"","filename":"Picture4.jpg","url":"https://assets-eu.researchsquare.com/files/rs-4781968/v1/5d000b1f5a5752565f5c81b9.jpg"},{"id":73663355,"identity":"53635167-6dc7-470a-91d0-cf921b81a530","added_by":"auto","created_at":"2025-01-13 11:29:07","extension":"jpg","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":84228,"visible":true,"origin":"","legend":"\u003cp\u003eExample of determining the timing of the wetland loss, applying the second condition.\u003c/p\u003e","description":"","filename":"Picture5.jpg","url":"https://assets-eu.researchsquare.com/files/rs-4781968/v1/d1201d29b2c362336bf51acf.jpg"},{"id":73663361,"identity":"536b74c5-ae51-444d-a229-befaa5265b17","added_by":"auto","created_at":"2025-01-13 11:29:13","extension":"jpg","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":84023,"visible":true,"origin":"","legend":"\u003cp\u003eSpatial distribution of lost wetlands with its lost timing (color bar) and still existing wetlands as of 2021 (light blue). Dark blue shows the permanent water area.\u003c/p\u003e","description":"","filename":"Picture6.jpg","url":"https://assets-eu.researchsquare.com/files/rs-4781968/v1/c51628aa97c4e84cc6d9d124.jpg"},{"id":73663570,"identity":"0c784869-637a-457b-a9cd-33d7f5df6ce0","added_by":"auto","created_at":"2025-01-13 11:37:07","extension":"jpg","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":37278,"visible":true,"origin":"","legend":"\u003cp\u003eTime series of the area of wetland loss in the study area.\u003c/p\u003e","description":"","filename":"Picture7.jpg","url":"https://assets-eu.researchsquare.com/files/rs-4781968/v1/e6f50a2cb674e03baf311fff.jpg"},{"id":73663357,"identity":"79f76aea-1255-4835-9023-03d515ad22cc","added_by":"auto","created_at":"2025-01-13 11:29:07","extension":"jpg","order_by":8,"title":"Figure 8","display":"","copyAsset":false,"role":"figure","size":73090,"visible":true,"origin":"","legend":"\u003cp\u003e(a) Distances from urban areas for existing wetlands and lost wetlands as of 2021 in the study area; (b) Distances from river channels for existing wetlands and lost wetlands as of 2021 in the study area.\u003c/p\u003e","description":"","filename":"Picture8.jpg","url":"https://assets-eu.researchsquare.com/files/rs-4781968/v1/6a65071f818857aef3f6d080.jpg"},{"id":73663356,"identity":"ef06c591-2f55-4381-ae16-a93467ce70de","added_by":"auto","created_at":"2025-01-13 11:29:07","extension":"jpg","order_by":9,"title":"Figure 9","display":"","copyAsset":false,"role":"figure","size":90127,"visible":true,"origin":"","legend":"\u003cp\u003eSpatial distribution of the subsequent LULC after the wetland loss in the study area. Dark blue shows the permanent water area.\u003c/p\u003e","description":"","filename":"Picture9.jpg","url":"https://assets-eu.researchsquare.com/files/rs-4781968/v1/c05c1964980c24cb69ea6e9e.jpg"},{"id":73663358,"identity":"c2ba2ffe-50b6-4d50-8612-91be12b3eec1","added_by":"auto","created_at":"2025-01-13 11:29:08","extension":"jpg","order_by":10,"title":"Figure 10","display":"","copyAsset":false,"role":"figure","size":158905,"visible":true,"origin":"","legend":"\u003cp\u003eResults of applying the developed (a) Random Forest-based and (b) Light GBM-based wetland loss prediction models to the existing wetlands as of 2021 (not used for the model development). Dark blue shows the permanent water area.\u003c/p\u003e","description":"","filename":"Picture10.jpg","url":"https://assets-eu.researchsquare.com/files/rs-4781968/v1/3b8fc77be135cb32e9b94278.jpg"},{"id":79465747,"identity":"5a9c54c4-a45b-4a96-bad5-d2ff4f8fe75a","added_by":"auto","created_at":"2025-03-28 18:55:49","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":2453472,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-4781968/v1/3dcdf10d-6ad1-4345-9d29-892bac19d4d7.pdf"},{"id":73663569,"identity":"9cefc033-75e5-4382-9aad-c4032e8efaab","added_by":"auto","created_at":"2025-01-13 11:37:07","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":130083,"visible":true,"origin":"","legend":"","description":"","filename":"Appendix.docx","url":"https://assets-eu.researchsquare.com/files/rs-4781968/v1/10d2b6293381603cadb9fbed.docx"}],"financialInterests":"","formattedTitle":"Spatio-temporal analysis of wetland loss in the lower Mekong River Basin based on surface water detection datasets and machine learning","fulltext":[{"header":"1. Introduction","content":"\u003cp\u003eWetland loss and degradation is a major global issue, drawing the attention of researchers, policymakers, and environmentalists around the world (Nicholls et al., \u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e1999\u003c/span\u003e; Coleman et al., \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e2008\u003c/span\u003e; Davidson, \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e2014\u003c/span\u003e; Mitsch \u0026amp; Gosselink, \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e2015\u003c/span\u003e; Hu et al., \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e2017\u003c/span\u003e; Fluet-Chouinard et al., \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). The rapid decline of wetlands due to anthropogenic activities such as urbanization, agriculture, and industrialization, and climate change poses severe threats to the critical functions of wetlands, including ecological habitats for diverse species, water purification processes, and natural buffers against floods and storms (Nicholls et al., \u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e1999\u003c/span\u003e; Nicholls, \u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e2004\u003c/span\u003e; Verhoeven et al., \u003cspan citationid=\"CR47\" class=\"CitationRef\"\u003e2006\u003c/span\u003e; Erwin, \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e2009\u003c/span\u003e; Moreno-Mateos et al., \u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e2012\u003c/span\u003e; Asselen et al., \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e2013\u003c/span\u003e; Davidson, \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e2014\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eThe impacts of wetland loss are extensive and multifaceted. One of the most immediate consequences is the increased risk of flooding (Bullock \u0026amp; Acreman, 2003; Hiraga et al., \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2018\u003c/span\u003e). Wetlands function as natural sponges, absorbing excess rainfall and releasing it slowly, thus reducing flood risks and securing water resources (Kazama et al., \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2007\u003c/span\u003e; Acreman \u0026amp; Holden, 2013; Shrestha et al., \u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e2016\u003c/span\u003e; Narayan et al., \u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e2017\u003c/span\u003e). When wetlands are degraded or lost, this natural flood mitigation is lost, leading to more frequent and severe flooding events. Additionally, wetlands are crucial for maintaining water quality. They filter pollutants, sediments, and nutrients from water, contributing to cleaner rivers, lakes, and groundwater (Johnston, \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e1991\u003c/span\u003e; Vymazal, \u003cspan citationid=\"CR48\" class=\"CitationRef\"\u003e2010\u003c/span\u003e; Hammer \u0026amp; Bastian, \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e2020\u003c/span\u003e). The degradation and reclamation of wetlands compromises this filtration capability, leading to poorer water quality and higher pollution levels (Hiraga et al., \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e2020\u003c/span\u003e), which can have serious implications for human health and biodiversity (Verhoeven \u0026amp; Meuleman, 1999). The lower Mekong River Basin in Cambodia is an example where wetlands play essential roles for locals in various aspects (Ringler and Cai, \u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e2006\u003c/span\u003e; MacAlister \u0026amp; Mahaxay, \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e2009\u003c/span\u003e; Arias et al., \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2014a\u003c/span\u003e). Their floodplain wetlands, formed mainly by river flooding, are sources for agriculture, fisheries, water resources, and cultural development in the lower Mekong River Basin, one of the most important geographical characteristics of the region. Recent studies based on the numerical modeling and field surveys showed that this important wetland is facing the danger of being lost due to anthropogenic landfill (Hiraga et al., \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2018\u003c/span\u003e; Hiraga et al., \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e2020\u003c/span\u003e), dam constructions and climate change (Arias et al., \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e2014b\u003c/span\u003e). Thus, it is required to address the impacts of wetland loss for their potential mitigations in the region.\u003c/p\u003e \u003cp\u003eTo effectively address the impacts of wetland loss for their potential mitigations, a detailed estimation of the spatio-temporal distribution of wetland loss is essential. Such estimation can help researchers identify the locations and timings of significant wetland changes, providing critical data for accurate impact evaluations and predictions (Ozesmi \u0026amp; Bauer, \u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e2002\u003c/span\u003e; Rebelo et al., \u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e2009\u003c/span\u003e). Recent studies have utilized remote sensing techniques to estimate the wetland loss (Ozesmi and Bauer, \u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e2002\u003c/span\u003e; Rebelo et al., \u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e2009\u003c/span\u003e; Guo et al., \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e2017\u003c/span\u003e; Mahdavi et al., \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e2018\u003c/span\u003e), including the global-scale estimations (Davidson, \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e2014\u003c/span\u003e; Hu et al., \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e2017\u003c/span\u003e; Fluet-Chouinard et al., \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e2023\u003c/span\u003e) and regional-scale estimations (Zhang et al., \u003cspan citationid=\"CR51\" class=\"CitationRef\"\u003e2009\u003c/span\u003e; Gong et al., \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e2010\u003c/span\u003e; Teheri et al., 2010; Chen et al., \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e2014\u003c/span\u003e; Dang et al., \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). While the global-scale analyses offer a broad overview, regional-scale analyses with high resolution are crucial for understanding specific local conditions and impacts. Due to the limited number of studies estimating spatio-temporal distributions of wetland loss in the lower Mekong River Basin in Cambodia, this study aimed to estimate the spatio-temporal distributions of wetland loss, analyze its geographical characteristics, and quantify the likelihood of loss occurrence for existing wetlands in the region. The wetlands in the lower Mekong River Basin in Cambodia are largely dependent on seasonal river flooding through bank overtopping and water conveyance by many agricultural canals, called \u0026ldquo;colmatage\u0026rdquo;, making the wetland dynamics processes unique. To deal with such unique flood plain wetland changes, this study used the surface water detection dataset to estimate the wetland loss.\u003c/p\u003e \u003cp\u003eSection 2 describes the study area and datasets. The methodology is described in Section 3. Section 4 demonstrates the estimated spatio-temporal distributions of wetland loss, analyzes the geographical characteristics of the lost wetlands, and discusses the machine learning model-based prediction results. Finally, Section 5 summarizes the findings.\u003c/p\u003e"},{"header":"2. Study area and datasets","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003e2.1. Study area\u003c/h2\u003e \u003cp\u003eThis study focuses on wetlands in the lower Mekong River Basin in Cambodia. The Mekong River basin can be broadly divided into the upper and lower reaches, with the upper reaches bordered by China and Laos. The lower basin encompasses Laos, Thailand, Cambodia, and Vietnam, covering approximately 80% of the total basin area (Wang et al., \u003cspan citationid=\"CR49\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). Geographically, the region is subdivided into four areas: the Northern Highlands, the Khorat Plateau, the Tonle Sap Basin, and the Mekong Delta (Gupta, \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e2009\u003c/span\u003e). The Mekong River flows through a broad valley east of the Khorat Plateau and into the Tonle Sap Basin. The Tonle Sap Basin is a vast alluvial plain surrounded by hills, with Tonle Sap Lake, the largest freshwater lake in Southeast Asia, situated in its western and central parts. Near Phnom Penh, the Mekong River divides, forming the Bassac River and the Mekong Delta. The Mekong and Bassac rivers branch into numerous smaller channels along their course, with the Mekong Delta covering an area of about 62,520 km\u0026sup2;. Our study area, which is an area approximately 140 km\u0026times;210 km around Phnom Penh (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e), the capital of Cambodia, includes the Tonle Sap Basin, located in the southern part of the lower Mekong River basin, and a part of the Mekong Delta.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThe climate of the lower Mekong River basin is classified as a tropical monsoon climate, characterized by distinct wet and dry seasons. Consequently, the river's flow fluctuates significantly throughout the year (Li et al., \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e2017\u003c/span\u003e). The flow during the flood season, which extends from June to November, accounts for approximately 80% of the total annual flow (Try et al., 2020). During the flood season, large-scale inundations frequently occur in the low-lying floodplains in the lower Mekong River basin. Such inundations occur owing to not only bank overtopping or failure but also water conveyance through many agricultural canals, called \u0026ldquo;colmatage\u0026rdquo; (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003ea; Amano and Kazama, 2016; Hiraga et al., \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2018\u003c/span\u003e). Such agricultural canals were mainly developed intentionally breaking a portion of the levee, so that river water can be provided to the surroundings, forming large floodplain wetlands during the flood season when river water level is high. The conveyed water from rivers includes rich nutrients, contributing to rice farming, aquatic lives, and vegetation growths in the floodplain wetlands (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eb; Hiraga et al., \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2018\u003c/span\u003e). Such wetlands are unique and important for not only regional perspective but also global perspective in food production and greenhouse gas emissions (Arai et al., 2018). Hence, the flood-inundation process is a vital water and nutrient source for agriculture, fisheries, and ecosystems in the region (Kummu and Sarkkula, \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2008\u003c/span\u003e; Sok et al., \u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). This river flood-driven floodplain wetland is the dominant type of wetlands in the study area, which is our focus in this study. The inundated area typically covers more than 10,000 km\u0026sup2;, and in particularly large flood years, this area can expand to 30,000\u0026ndash;40,000 km\u0026sup2; with water depths exceeding 2 meters in the lower Mekong River Basin (MRC, \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e1997\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eIn recent years, losing such important floodplain wetlands is a critical issue in the lower Mekong River Basin (Huu Nguyen et al., \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e2016\u003c/span\u003e; Fluet-Chouinard et al., \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). Anthropogenic factors, such as construction of dams, dykes, and roads (Huu Nguyen et al., \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e2016\u003c/span\u003e), and wetland drainage and landfill (Hiraga et al., \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2018\u003c/span\u003e; \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e2020\u003c/span\u003e) are frequently cited causes of wetland loss. Climate change also significantly altered the regional hydrology, influencing the wetland area (Pokhrel et al., \u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e2018\u003c/span\u003e; Arias et al., \u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e2019\u003c/span\u003e; Dang et al., \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). Horton et al. (\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2022\u003c/span\u003e) found that the planned infrastructural development will cause substantial changes in the wetlands in the Cambodian Mekong floodplain, potentially impacting ecosystems and people's livelihoods. Given the dependence of local communities on agriculture and fisheries sustained by the natural resources of the wetlands, such changes are expected to have serious impacts, emphasizing the importance of better understanding the wetland loss dynamics.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec4\" class=\"Section2\"\u003e \u003ch2\u003e2.2. Datasets\u003c/h2\u003e \u003cp\u003eThis study used two types of Global Surface Water (Jean-Francois et al., 2016): (1) Monthly Water History dataset and (2) Occurrence dataset. The Monthly Water History from Global Surface Water (Jean-Francois et al., 2016) was used to estimate the area of wetland loss in the study area. This Monthly Water History dataset is based on the detection of surface water using spectral analysis of satellite images, with a spatial resolution of 30 m\u0026times;30 m. The temporal coverage extends from March 1984 to December 2021. In the Monthly Water History dataset, the water surface detection is based on satellite observations conducted over an 8-day cycle, resulting in multiple observations in a month. This study defined a grid cell as water surface grid for each month when the grid cell is classified as water surface in any of the observations during the month. A no-data classification was assigned for periods when observations are not available mainly because of the cloud cover or equipment malfunctions. The Occurrence dataset was used to determine the permanent water areas in the target region. The Occurrence dataset includes the percentage of all monthly water history data for each grid cell that was classified as water bodies throughout the study period. In this study, a grid cell in which 100% of the data was identified as a water area was classified as a permanent water area, such as a river or a lake.\u003c/p\u003e \u003cp\u003eThis study used the Sentinel-2 10m land use/land cover (LULC) data to assess LULC changes following wetland loss. This LULC dataset consists of annual data from 2017 to 2021, derived from ESA Sentinel-2 images with a spatial resolution of 10 m \u0026times; 10 m. In this study, the observation in the year 2021 was used to examine the land use/land cover after the wetland loss. The LULC data were spatially interpolated to a resolution of 30 m\u0026times;30 m to align with the spatial resolution of the other datasets used in the analysis. During this process, the most frequently occurring LULC category within each 30 m \u0026times; 30 m grid cell was assigned as the LULC for the target grid cell. Due to the differences in the definition of \"water area\" between the LULC data and the Global Surface Water data, any grid cell classified as a water area and identified as having lost wetlands in this study was excluded from the analyses in Chap.\u0026nbsp;4 and subsequent chapters.\u003c/p\u003e \u003cp\u003eThe ALOS Global Numerical Land Surface Model (ALOS World 3D 30m\u0026times;30m) was used as the elevation data. This dataset was compiled using the Optical Sensor and Panchromatic Remote-sensing Instrument for Stereo Mapping (PRISM) onboard the Advanced Land Observing Satellite Daichi (ALOS), which was operational from January 2006 to April 2011.\u003c/p\u003e \u003c/div\u003e"},{"header":"3. Methodology","content":"\u003cp\u003eThis study first estimated the spatiotemporal distribution of wetland loss with high resolution (30m) over the entire study area (140km\u0026times;210km) during 1984\u0026ndash;2021. It should be noted that this study attempted to estimate the distribution of wetlands and their losses using the water surface detection data, not the LULC data. It is well known that different LULC datasets include large variations in the water-related classification (Nakaegawa, \u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e2012\u003c/span\u003e). Furthermore, it is difficult to accurately capture the changes of river flood-driven wetlands using LULC datasets available with less frequency (i.e., usually annual or every five years). Thus, this study attempted to detect wetland changes purely based on the water surface detection datasets with high resolution in time and space (Jean-Francois et al., 2016). The factors influencing wetland loss and LULC after the wetland loss were then examined. Further, we conducted machine learning analysis using the identified factors as input values, constructing a model to predict the probability of wetland loss in the study area. Applying this model to the existing wetlands in the study area, the probability of future wetland loss was estimated. The following subsections describe the detailed procedure for each step:\u003c/p\u003e \u003cdiv id=\"Sec6\" class=\"Section2\"\u003e \u003ch2\u003e3.1. Spatiotemporal distribution of Wetland loss\u003c/h2\u003e \u003cp\u003eWe first used the Monthly Water History dataset to classify each 30m\u0026times;30m grid cell as water and non-water grid cell for each month of August, September, and October during 1984\u0026ndash;2021 in the study area. August, September, and October were selected as they are the latter part of the rainy season in the lower Mekong River Basin (MRC, \u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e2005\u003c/span\u003e). When water/non-water classification is unavailable for all three months in a year due to missing data in the Monthly Water History dataset for a grid cell, we performed the interpolation assuming the same classification as the previous year for the grid cell. Figure\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e describes the process above.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eUsing the water/non-water classification, the timing of the wetland loss was estimated identifying the year in which water classification transitioned to non-water classification for all the grid cells. Changes in water areas are influenced by various factors such as fluctuations in river levels and precipitation. Hence it is difficult to determine wetland loss based solely on water/non-water classification data at a specific time or over a short period. To accurately assess wetland loss, it is necessary to consider long-term trends and patterns. We used the following two conditions to define the loss of a wetland:\u003c/p\u003e \u003cp\u003eThe first condition is that a grid cell must have been classified as non-water for at least five consecutive years between 2017 and 2021. This condition was set to prevent the detection of a grid cell temporarily dried up as non-wetland due to drought-induced water level decrease. Figure\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e shows an example of applying the first condition.\u003c/p\u003e \u003cp\u003eThe second condition is that, for a grid cell, 50% or more of the classifications must be water in the time series of classifications from 1984 to the year in which the wetland was estimated to have been lost. This condition was set to prevent the detection of a grid cell temporally covered by water as wetland due to high rainfall season etc. Figure\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e shows an example of applying the second condition. We estimated the year of wetland loss for all grid cells in the study area, which led to estimating the spatiotemporal distribution of wetland loss with high resolution.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eFurthermore, based on the estimated spatiotemporal distribution of wetland loss, factors influencing wetland loss were examined based on previous studies. This study focused on the distance from the river channels and the distance from the urban areas as influential factors on the wetland loss, following Hiraga et al. (\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2018\u003c/span\u003e) who developed the wetland landfill prediction model using those variables. Here, we defined grid cells as urban areas when satisfying the following conditions: grid cells belonging to the urban category in the LULC dataset and not being wetlands at any time from 1984 to 2021 in the Monthly Water History dataset. We calculated the shortest distance from the river channel and from the urban areas for all the existing and lost wetlands in the study area. Then, we performed the Mann-Whitney U-test (McKnight and Najab, \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e2010\u003c/span\u003e) to evaluate the statistical significance in the differences in distances from river channels and distances from urban areas between existing wetlands and lost wetlands. Moreover, the LULC in 2021 (latest year in the study period) where wetland loss occurred was also examined to investigate the pattern of LULC at lost wetlands.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec7\" class=\"Section2\"\u003e \u003ch2\u003e3.2. Machine learning analysis to develop wetland loss prediction models\u003c/h2\u003e \u003cp\u003eWe used machine learning analysis to (1) understand the feature importance of each factor affecting wetland loss and (2) estimate the probability of wetland loss in the study area. The methods employed are supervised learning (Random Forest and Light GBM) to construct a binary classification model for estimating the probability of wetland loss.\u003c/p\u003e \u003cdiv id=\"Sec8\" class=\"Section3\"\u003e \u003ch2\u003e3.2.1. Random Forest classifier\u003c/h2\u003e \u003cp\u003eRandom Forest is an algorithm that combines two methods: decision trees and bagging, an ensemble learning technique (Breiman, \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e2001\u003c/span\u003e). A decision tree is an algorithm that classifies binary problems by organizing them in a hierarchical structure. Random forests create an accurate model by combining multiple decision trees. Bagging is the method used for combining these decision trees. It involves replicating and extracting data from the original dataset. A model is created for each replicated dataset, and the final prediction is made referring to a majority decision based on the predictions from these models. This study performed the Random Forest classification analysis using the scikit-learn in Python ver3.8 (Pedregosa et al., \u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e2011\u003c/span\u003e). The used parameters in the Random Forest model were mainly its default values, including the followings: (1) n_estimators: the number of decision trees; (2) criterion: criterion for the split; (3) max_depth: maximum depth of the decision tree; (4) min_samples_split: minimum number of samples required to split a node; (5) min_samples_leaf: minimum number of samples required for a leaf node; (6) max_features: maximum number of features to be used in each decision tree split; (7) random_state: random number seed to ensure model reproducibility. The parameters used in this study are listed in Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\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\u003eParameters of Random Forest\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"2\"\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 \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eParameters\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eApplied values in this study\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003en_estimators\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e100\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ecriterion\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003egini\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003emax_depth\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eNone\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003emin_sample_split\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e2\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003emin_sample_leaf\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003emax_features\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eauto\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003erandom_state\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e42\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec9\" class=\"Section3\"\u003e \u003ch2\u003e3.2.2. Light GBM classifier\u003c/h2\u003e \u003cp\u003eLight GBM utilizes gradient boosting as its ensemble learning technique (Ke et al., \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2017\u003c/span\u003e). In gradient boosting, a decision tree model is constructed from the original data, and the difference between the predicted value by this model and the actual value is calculated. Light GBM is one of the fastest algorithms for processing large amounts of data in machine learning using gradient boosting. This study performed the Light GBM classification analysis using the scikit-learn in Python ver3.8 (Pedregosa et al., \u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e2011\u003c/span\u003e). The used parameters in the Light GBM model were mainly its default values, including the followings: (1) boosting_type: applied boosting method; (2) num_leaves: the number of leaves in the decision tree; (3) learning_rate: controls the importance of each decision tree on the prediction; (4) n_estimators: the number of decision trees to be trained; (5) max_depth: maximum depth of the decision tree; (6) min_child_samples: minimum number of samples required for a leaf node, preventing overfitting; (7) subsample: the percentage of subsampling of the training data. (8) colsample_bytree: the ratio of features used to train each decision tree.\u003c/p\u003e \u003cp\u003eThe parameters used in this study are listed in Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e.\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\u003eParameters of LightGBM\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"2\"\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 \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eParameters\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eApplied values in this study\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eboosting_type\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003egbdt\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003enum_leaves\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e400\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003elearning_rate\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.1\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003en_estimators\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e800\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003emax_depth\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e-50\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003emin_child_sample\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e20\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003esubsample\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1.0\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ecolsample_bytree\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1.0\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec10\" class=\"Section3\"\u003e \u003ch2\u003e3.2.3. Development of the machine learning model\u003c/h2\u003e \u003cp\u003eFor the development of the machine learning models, we first divided all grid cells representing existing wetlands (total of 12,300,000 cells) into two groups in half, one for train and test of the model, the other for predicting wetland loss. As the objective variables of the model, we used the existing wetland grid cells in one group (labelled as \u0026ldquo;existing wetland\u0026rdquo;) and the lost wetland grid cells (labelled as \u0026ldquo;lost wetlands\u0026rdquo;). The objective variables were divided as 8:2 for training and testing the model. The explanatory variables include the elevation, distance from the river channel, and distance from urban areas. Hiraga et al. (\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2018\u003c/span\u003e) showed that the wetland landfill can be modelled using those variables for the region. The accuracy of the developed model was evaluated using the statistic metrics described in the following section. Subsequently, the developed model was applied to the other group of grid cells representing existing wetlands that were not used to construct the model, to estimate the probability of loss of existing wetlands.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec11\" class=\"Section3\"\u003e \u003ch2\u003e3.2.4. Model evaluation metrics\u003c/h2\u003e \u003cp\u003eWe used statistical metrics to quantify the prediction accuracy of the developed machine learning model. Accuracy, Precision, Recall, and F1-score are commonly used evaluation metrics for machine learning models. Accuracy is the percentage of correct predictions made by the model. Precision represents the percentage of data that is actually \u0026ldquo;positive\u0026rdquo; out of the test data that the model predicts to be \u0026ldquo;positive\u0026rdquo;. Recall represents the percentage of test data that the model was able to predict to be \u0026ldquo;positive\u0026rdquo; out of the test data that were actually \u0026ldquo;positive\u0026rdquo;. F1-score is an index that simultaneously evaluates both the precision and recall. These metrics were calculated as follows:\u003cdiv id=\"Equa\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equa\" name=\"EquationSource\"\u003e\n$$\\:\\begin{array}{c}Accuracy=\\frac{\\text{N}\\text{u}\\text{m}\\text{b}\\text{e}\\text{r}\\:\\text{o}\\text{f}\\:\\text{c}\\text{o}\\text{r}\\text{r}\\text{e}\\text{c}\\text{t}\\:\\text{p}\\text{r}\\text{e}\\text{d}\\text{i}\\text{c}\\text{t}\\text{i}\\text{o}\\text{n}\\text{s}}{\\text{T}\\text{o}\\text{t}\\text{a}\\text{l}\\:\\text{n}\\text{u}\\text{m}\\text{b}\\text{e}\\text{r}\\:\\text{o}\\text{f}\\:\\text{p}\\text{r}\\text{e}\\text{d}\\text{i}\\text{c}\\text{t}\\text{i}\\text{o}\\text{n}\\text{s}}\\:\\#\\left(1\\right)\\end{array}$$\u003c/div\u003e\u003c/div\u003e\u003cdiv id=\"Equb\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equb\" name=\"EquationSource\"\u003e\n$$\\:\\begin{array}{c}Precision=\\frac{\\text{T}\\text{r}\\text{u}\\text{e}\\:\\text{P}\\text{o}\\text{s}\\text{i}\\text{t}\\text{i}\\text{v}\\text{e}\\text{s}}{\\text{T}\\text{r}\\text{u}\\text{e}\\:\\text{P}\\text{o}\\text{s}\\text{i}\\text{t}\\text{i}\\text{v}\\text{e}\\text{s}\\:+\\:\\text{F}\\text{a}\\text{l}\\text{s}\\text{e}\\:\\text{P}\\text{o}\\text{s}\\text{i}\\text{t}\\text{i}\\text{v}\\text{e}\\text{s}}\\#\\left(2\\right)\\end{array}$$\u003c/div\u003e\u003c/div\u003e\u003cdiv id=\"Equc\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equc\" name=\"EquationSource\"\u003e\n$$\\:\\begin{array}{c}Recall=\\frac{\\text{T}\\text{r}\\text{u}\\text{e}\\:\\text{P}\\text{o}\\text{s}\\text{i}\\text{t}\\text{i}\\text{v}\\text{e}\\text{s}}{\\text{T}\\text{r}\\text{u}\\text{e}\\:\\text{P}\\text{o}\\text{s}\\text{i}\\text{t}\\text{i}\\text{v}\\text{e}\\text{s}\\:+\\:\\text{F}\\text{a}\\text{l}\\text{s}\\text{e}\\:\\text{N}\\text{e}\\text{g}\\text{a}\\text{t}\\text{i}\\text{v}\\text{e}\\text{s}}\\#\\left(3\\right)\\end{array}$$\u003c/div\u003e\u003c/div\u003e\u003cdiv id=\"Equd\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equd\" name=\"EquationSource\"\u003e\n$$\\:\\begin{array}{c}F1=2\\times\\:\\frac{\\text{P}\\text{r}\\text{e}\\text{c}\\text{i}\\text{s}\\text{i}\\text{o}\\text{n}\\times\\:\\text{R}\\text{e}\\text{c}\\text{a}\\text{l}\\text{l}}{\\text{P}\\text{r}\\text{e}\\text{c}\\text{i}\\text{s}\\text{i}\\text{o}\\text{n}+\\text{R}\\text{e}\\text{c}\\text{a}\\text{l}\\text{l}}\\#\\left(4\\right)\\end{array}$$\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003ewhere positive represents existing wetlands and negative represents lost wetlands in this study.\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e"},{"header":"4. Results and Discussion","content":"\u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003e4.1. Spatiotemporal distribution of Wetland loss and its analysis\u003c/h2\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eFigure \u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e Spatial distribution of lost wetlands with its lost timing (color bar) and still existing wetlands as of 2021 (light blue). Dark blue shows the permanent water area.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThe area of wetlands loss shows a gradually increasing trend from 1993 to 2010, with a significant area recorded during 2012\u0026ndash;2014. Although economic development and population growth in Cambodia may have contributed to the increase in lost wetlands, no significant relationship was observed (Figure \u003cspan refid=\"Fig11\" class=\"InternalRef\"\u003eA1\u003c/span\u003e). GDP and population growth alone may not account for the factors influencing wetland loss, such as wetland development and dam construction in the upper reaches. This study also examined the time series of the river water level in the study area based on the assumption that changes in floodplain wetlands may be strongly related to variations in river levels (Figure \u003cspan refid=\"Fig12\" class=\"InternalRef\"\u003eA2\u003c/span\u003e). Though the small area of wetland loss coincided in the year 2011, the historical flood year showing notably high-water level, it did not clearly explain the changes in wetland areas partially due to the limited consideration of water levels at only three points in the study area. Though accurately capturing local landfilling pattern or local population dynamics requires deep analysis as the information is usually not available in public, such analysis will help better explaining how the timing and location of wetland loss is determined in the region. Such studies are left for future studies.\u003c/p\u003e \u003cp\u003eThis study further analyzed the factors influencing wetland loss based on our estimates shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e. Figure\u0026nbsp;\u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e7\u003c/span\u003e shows the box plots of the (a) distances from the urban areas and (b) distances from the river channels for all the existing wetlands and lost wetlands as of 2021 in the study area. Figure\u0026nbsp;\u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e7\u003c/span\u003e implies that the lost wetlands tend to locate close to the urban areas and far from the river channels, which is consistent with the findings of Hiraga et al. (\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2018\u003c/span\u003e) who examined the geographical characteristics of the limited number (~\u0026thinsp;100) of landfilled wetlands in the lower Mekong River Basin. Table\u0026nbsp;\u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e3\u003c/span\u003e summarizes the distances from urban areas and river channels for existing wetlands and lost wetlands with the Mann-Whitney U-test results. The Mann-Whitney U-test results (\u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.01) suggest that the differences in the distances between existing wetlands and lost wetlands are statistically significant. Wetland loss tends to occur at the locations close to urban areas potentially because of the gradual and spatially continuous expansion of urban areas and convenience in carrying necessary resources for urban developments. Meanwhile, wetland loss tends to occur at locations far from river channels potentially because of low water level in wetlands which is more vulnerable to hydrological regime changes and easier to landfill.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab3\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eSummary of the statistics of the distances from urban areas and river channels for existing wetlands and lost wetlands with the Mann-Whitney U-test results\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"5\"\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 \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eMedian [km]\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eMean [km]\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003e\u003cem\u003ep\u003c/em\u003e-values\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eShortest distance from urban areas\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eExisting wetlands\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e1.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e1.26\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.01\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eLost wetlands\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.60\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.86\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eShortest distance from river channels\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eExisting wetlands\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e5.88\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e10.2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.01\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eLost wetlands\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e8.61\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e12.6\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\u003eThis study investigated the subsequent LULC after the wetland loss at all the grid cells classified as wetland loss. Figure\u0026nbsp;\u003cspan refid=\"Fig9\" class=\"InternalRef\"\u003e8\u003c/span\u003e shows the spatial distribution of the subsequent LULC after the wetland loss at all the grid cells classified as wetland loss in the study area. Figure\u0026nbsp;\u003cspan refid=\"Fig9\" class=\"InternalRef\"\u003e8\u003c/span\u003e shows that most of the wetlands were found to be turned into croplands, especially at the downstream (i.e., south) of the study area. Meanwhile, Fig.\u0026nbsp;\u003cspan refid=\"Fig9\" class=\"InternalRef\"\u003e8\u003c/span\u003e shows some of the rangelands (i.e., grassland, shrubland, etc.) after the wetland loss. Table\u0026nbsp;\u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e3\u003c/span\u003e summarizes the LULC at the wetland loss area.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab4\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eSubsequent LULC (2021) after the wetland loss at all the grid cells classified as wetland loss in the study area\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"4\"\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 \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e \u003cp\u003eThis study\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"2\" nameend=\"c4\" namest=\"c3\"\u003e \u003cp\u003eFluet-Chouinard et al. (\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e2023\u003c/span\u003e)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eClassification\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eRatio [%]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eClassification\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eRatio [%]\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCrops\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e72.2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eUpland croplands\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e61.7\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRangeland\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e11.3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eFlooded rice\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e18.2\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBuilt (Urban)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e7.5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eUrban areas\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e8.0\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTrees\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e5.1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eForestry\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e4.7\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFlooded vegetation\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e3.5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eWetland cultivation\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e4.3\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBare ground\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003ePasture\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e2.0\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003ePeat extraction\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.9\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\u003eTable\u0026nbsp;\u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e3\u003c/span\u003e also shows the percentage of LULC at the lost inland wetlands worldwide, estimated by Fluet-Chouinard et al. (\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e2023\u003c/span\u003e), to compare our estimates with the global scale analysis. Table\u0026nbsp;\u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e3\u003c/span\u003e shows that crops (i.e., agricultural lands) account for more than 70% of the total wetland loss area, which is a similar tendency to Fluet-Chouinard et al. (\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). Though the LULC percentage basically shows similar results with Fluet-Chouinard et al. (\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e2023\u003c/span\u003e), our estimate for the lower Mekong River Basin resulted in 11.3% of rangelands (Table\u0026nbsp;\u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e3\u003c/span\u003e). This may be a unique feature for seasonal floodplain wetlands, such as the lower Mekong River Basin. The extent of floodplain wetlands at the lower Mekong River Basin is largely affected by seasonal river flooding, thus the wetland loss due to the hydrological regime changes can drastically occur. For instance, the upstream dam constructions or climate change can cause more severe and frequent droughts (Pokhrel et al., \u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e2018\u003c/span\u003e; Yang et al., \u003cspan citationid=\"CR50\" class=\"CitationRef\"\u003e2019\u003c/span\u003e), which can result in significant amount of downstream wetland loss. Such hydrological regime change-induced wetland loss rather than urbanizations or agriculturalizations may explain the wetlands being turned into rangelands or other natural LULC. Other inland wetlands in the world are not necessarily sustained by river flood, potentially explaining the difference in Table\u0026nbsp;\u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e3\u003c/span\u003e.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec14\" class=\"Section2\"\u003e \u003ch2\u003e4.2. Machine learning analysis to develop wetland loss prediction models\u003c/h2\u003e \u003cp\u003eThis study used the Random Forest and Light GBM algorithms to develop wetland loss prediction models for the study area. Table\u0026nbsp;\u003cspan refid=\"Tab5\" class=\"InternalRef\"\u003e4\u003c/span\u003e summarizes the statistical metrics of the developed wetland loss prediction models. Overall, both algorithms resulted in developing good models as all the metrics reached above 80%. Though the Accuracy metric was lower, the Random Forest-based model resulted in notably better performances in the other metrics, especially F1-score, which can consider the trade-off (i.e., balance) of Precision and Recall. It should be noted that, though this study judged the statistical metrics in Table\u0026nbsp;\u003cspan refid=\"Tab5\" class=\"InternalRef\"\u003e4\u003c/span\u003e good enough for understanding the wetland changes in large extent, more efforts in tuning the model parameters may help developing more accurate models. Such analysis will be a trade-off issue of the computational cost and model accuracy, which can be considered in the future.\u003c/p\u003e \u003cp\u003eTable\u0026nbsp;\u003cspan refid=\"Tab6\" class=\"InternalRef\"\u003e5\u003c/span\u003e shows the feature importance of the developed wetland loss prediction models. Both models show that the distance from river channels was found to be the most important feature for classifying existing wetlands and lost wetlands, which is considered to represent the characteristic of the floodplain wetlands in the lower Mekong River Basin well. In the Random Forest model, the importance of the distance from river channels was notably high (69.6%), while the importance of elevation was found to be less (4.9%). In contrast, in the Light GBM model, the importance of the elevation was found to be much higher (26.1%), resulting in mild differences among the considered features. The feature importance differed greatly depending on the algorithm used, suggesting large uncertainty due to the model algorithm. It is well established that employing multiple algorithms can lead to more accurate prediction with quantifying uncertainty (Abdar et al., \u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e2021\u003c/span\u003e), which can be confirmed in our analysis too.\u003c/p\u003e \u003cp\u003eFigure \u003cspan refid=\"Fig10\" class=\"InternalRef\"\u003e9\u003c/span\u003e shows the application of the developed models to the existing wetlands as of 2021 (not used for the model development), showing the predictions of the existing wetlands becoming the lost wetlands. Thus, the values in Fig.\u0026nbsp;\u003cspan refid=\"Fig10\" class=\"InternalRef\"\u003e9\u003c/span\u003e can be interpreted as the likelihood of existing wetlands being lost. The estimated likelihoods using the two different models were generally consistent in their spatial distribution. The tendency of the likelihood of being higher at the edge of the floodplain can be found, which represents the strong effect of distance from river channels. The large extent of the wetlands with higher likelihood of being lost is distributed downstream (i.e., south) of the study area, especially in the Light GBM model-based result (Fig.\u0026nbsp;\u003cspan refid=\"Fig10\" class=\"InternalRef\"\u003e9\u003c/span\u003eb), possibly due to the lower elevation over the area. Though Fig.\u0026nbsp;\u003cspan refid=\"Fig10\" class=\"InternalRef\"\u003e9\u003c/span\u003e does not provide probability of wetland loss with its timing (e.g., 20% in 2050s etc.), which is a remaining important research issue, it can still be useful for an interregional comparison in the likelihood of wetland loss for the study area. Our proposed method can provide the wetland loss likelihood with very high spatial resolution, demonstrating the potential to be considered in further numerical model-based analysis to assess the impact of wetland loss on regional hydrology, climatology, ecology, and agricultural/fishery productivity. Such studies can quantify the value of the regional wetlands on the local and global, contributing to emphasizing the importance of the wetlands. This is beyond the scope of this study and will be addressed in the future.\u003c/p\u003e \u003cp\u003eThis study includes some of the limitations that need to be addressed in the future. First, the methodology to define wetland loss, especially for floodplain wetlands, can be further improved. Testing various criteria in durations being covered by water or not will be ideal to accurately capture the regional wetland change dynamics. Also, the other explanatory variables can be considered for developing a more accurate wetland loss prediction model. Examples are population and distance from local roads. However, data availability and spatial and temporal resolution of the datasets will be a key issue on whether such inputs can be used to develop a model or not. Lastly, the subsequent LULC prediction after the wetland loss will be an interesting research issue. In fact, this study already attempted to predict the subsequent LULC after the wetland loss using similar inputs in the Light GBM algorithm (not shown), which did not result in high accuracy models (less than 30% of all the statistical metrics). Predicting future LULC requires various inputs and considering LULC change dynamics, which can be included in the future analysis.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab5\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 4\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eStatistical metrics of the developed wetland loss prediction models\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"5\"\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 \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAccuracy\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003ePrecision\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eRecall\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eF1-score\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRandom forest\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e82.8%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e87.8%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e92.5%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e90.1%\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLight GBM\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e86.2%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e84.3%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e86.0%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e82.5%\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 \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab6\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 5\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eFeature importance of the developed wetland loss prediction models\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"4\"\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 \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eDistance from urban areas\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eDistance from river channels\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eElevation\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRandom forest\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e25.5%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e69.6%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e4.9%\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLight GBM\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e36.6%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e37.3%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e26.1%\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e"},{"header":"5. Conclusion","content":"\u003cp\u003eThis study aimed to estimate the spatio-temporal distributions of wetland loss, analyze its geographical characteristics, and quantify the likelihood of loss occurrence for existing wetlands in the lower Mekong River Basin in Cambodia. The main findings of this study are summarized below:\u003c/p\u003e \u003cp\u003e \u003cul\u003e \u003cli\u003e \u003cp\u003eUsing the Global Surface Water data (Jean-Francois et al., 2016), this study estimated the spatiotemporal distribution of wetland loss with high resolution (30m) over the entire study area (140km\u0026times;210km) during 1984\u0026ndash;2021.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eThe statistically significant differences were found in the distance from urban areas and distance from river channels for the existing wetlands and lost wetlands as of 2021, in which the lost wetlands tend to locate closer to urban areas (existing wetlands: mean of 0.86 km; lost wetlands: mean of 1.26 km) and more far to river channels (existing wetlands: mean of 5.88 km; lost wetlands: mean of 8.61 km).\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eSubsequent LULC after the wetland loss was found to be mainly croplands (72.2%) in the study area. Though our estimate overall agrees with the recent global-scale estimate (Fluet-Chouinard et al., \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e2023\u003c/span\u003e), our estimate resulted in notable ratio of rangelands (11.3%), which represents the unique characteristics of floodplain wetlands in the lower Mekong River Basin.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eThe Random Forest and Light GBM algorithms-based wetland loss prediction models resulted in good statistical evaluation metrics (\u0026gt;\u0026thinsp;80%). In both models, the distance from river channels was found to be the most important feature for classifying existing wetlands and lost wetlands. Application of the developed models successfully provided the map of likelihood of wetland loss for existing wetlands in the study area.\u003c/p\u003e \u003c/li\u003e \u003c/ul\u003e \u003c/p\u003e \u003cp\u003eThis study provides useful information and insights on better understanding the spatio-temporal distribution of wetland loss in one of the most important floodplain wetland areas, which can be used for impact assessment studies for their potential mitigations in the region.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e \u003ch2\u003eCompeting Interests\u003c/h2\u003e \u003cp\u003eThe authors have no relevant financial or non-financial interests to disclose.\u003c/p\u003e \u003c/p\u003e\u003ch2\u003eFunding\u003c/h2\u003e \u003cp\u003eThe author(s) received no financial support for the research, authorship, and/or publication of this article.\u003c/p\u003e\u003ch2\u003eAuthor Contributions\u003c/h2\u003e \u003cp\u003eYusuke Hiraga and So Kazama contributed to the study conception and design. The first draft of the manuscript was written by Yusuke Hiraga and all authors commented on previous versions of the manuscript. Material preparation, data collection and analysis were performed by Yusuke Hiraga and Mayu Aoki. The review and edit of the first draft were done by So Kazama. All authors read and approved the final manuscript.\u003c/p\u003e\u003ch2\u003eAcknowledgement\u003c/h2\u003e \u003cp\u003eThis study was partially supported by the JSPS KAKENHI Grant number 24H00329.\u003c/p\u003e\u003ch2\u003eData Availability\u003c/h2\u003e \u003cp\u003eThe datasets generated during and/or analysed during the current study are available from the corresponding author on reasonable request.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eAbdar M, Pourpanah F, Hussain S, Rezazadegan D, Liu L, Ghavamzadeh M, Nahavandi S (2021) A review of uncertainty quantification in deep learning: Techniques, applications and challenges. Inform fusion 76:243\u0026ndash;297\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eArias ME, Cochrane TA, Elliott V (2014a) Modelling future changes of habitat and fauna in the Tonle Sap wetland of the Mekong. Environ Conserv 41(2):165\u0026ndash;175\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eArias ME, Cochrane TA, Kummu M, Lauri H, Holtgrieve GW, Koponen J, Piman T (2014b) Impacts of hydropower and climate change on drivers of ecological productivity of Southeast Asia's most important wetland. Ecol Model 272:252\u0026ndash;263\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eArias ME, Holtgrieve GW, Ngor PB, Dang TD, Piman T (2019) Maintaining perspective of ongoing environmental change in the Mekong floodplains. Curr Opin Environ Sustain 37:1\u0026ndash;7\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAsselen SV, Verburg PH, Vermaat JE, Janse JH (2013) Drivers of wetland conversion: a global meta-analysis. PLoS ONE, 8(11), e81292\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBreiman L (2001) Random forests. Mach Learn 45:5\u0026ndash;32\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChen L, Jin Z, Michishita R, Cai J, Yue T, Chen B, Xu B (2014) Dynamic monitoring of wetland cover changes using time-series remote sensing imagery. Ecol Inf 24:17\u0026ndash;26\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eColeman JM, Huh OK, Braud D Jr (2008) Wetland loss in world deltas. J Coastal Res 24:1\u0026ndash;14\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDang AT, Kumar L, Reid M, Nguyen H (2021) Remote sensing approach for monitoring coastal wetland in the Mekong Delta, Vietnam: Change trends and their driving forces. Remote Sens 13(17):3359\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDavidson NC (2014) How much wetland has the world lost? Long-term and recent trends in global wetland area. Mar Freshw Res 65(10):934\u0026ndash;941\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eErwin KL (2009) Wetlands and global climate change: the role of wetland restoration in a changing world. Wetlands Ecol Manage 17(1):71\u0026ndash;84\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFluet-Chouinard E, Stocker BD, Zhang Z, Malhotra A, Melton JR, Poulter B, McIntyre PB (2023) Extensive global wetland loss over the past three centuries. Nature 614(7947):281\u0026ndash;286\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGong P, Niu Z, Cheng X, Zhao K, Zhou D, Guo J, Yan J (2010) China\u0026rsquo;s wetland change (1990\u0026ndash;2000) determined by remote sensing. Sci China Earth Sci 53:1036\u0026ndash;1042\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGuo M, Li J, Sheng C, Xu J, Wu L (2017) A review of wetland remote sensing. Sensors 17(4):777\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGupta A (2009) Geology and landforms of the Mekong Basin. The Mekong. Academic, pp 29\u0026ndash;51\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHammer DA, Bastian RK (2020) Wetlands ecosystems: natural water purifiers? Constructed wetlands for wastewater treatment. CRC, pp 5\u0026ndash;19\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHiraga Y, Kazama S, Ekkawatpanit C, Touge Y (2018) Impact of reclamation on the environment of the lower Mekong River basin. J Hydrology: Reg Stud 18:143\u0026ndash;155\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHiraga Y, Kazama S, Ekkawatpanit C, Yamada K (2020) Evaluation of the impact of landfill on floodplain water quality in a tropical monsoon region. Hydrol Res Lett 14(1):48\u0026ndash;55\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHorton AJ, Triet NV, Hoang LP, Heng S, Hok P, Chung S, Kummu M (2022) The Cambodian Mekong floodplain under future development plans and climate change. Nat Hazards Earth Syst Sci 22(3):967\u0026ndash;983\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHu S, Niu Z, Chen Y, Li L, Zhang H (2017) Global wetlands: Potential distribution, wetland loss, and status. Sci Total Environ 586:319\u0026ndash;327\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHuu Nguyen H, Dargusch P, Moss P, Tran DB (2016) A review of the drivers of 200 years of wetland degradation in the Mekong Delta of Vietnam. Reg Envriron Chang 16(8):2303\u0026ndash;2315\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJohnston CA (1991) Sediment and nutrient retention by freshwater wetlands: effects on surface water quality. Crit Rev Environ Sci Technol 21(5\u0026ndash;6):491\u0026ndash;565\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKazama S, Hagiwara T, Ranjan P, Sawamoto M (2007) Evaluation of groundwater resources in wide inundation areas of the Mekong River basin. J Hydrol 340(3\u0026ndash;4):233\u0026ndash;243\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKe G, Meng Q, Finley T, Wang T, Chen W, Ma W, Liu TY (2017) Lightgbm: A highly efficient gradient boosting decision tree. \u003cem\u003eAdvances in neural information processing systems\u003c/em\u003e, \u003cem\u003e30\u003c/em\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKummu M, Sarkkula J (2008) Impact of the Mekong River flow alteration on the Tonle Sap flood pulse. AMBIO: J Hum Environ 37(3):185\u0026ndash;192\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLi D, Long D, Zhao J, Lu H, Hong Y (2017) Observed changes in flow regimes in the Mekong River basin. J Hydrol 551:217\u0026ndash;232\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMacAlister C, Mahaxay M (2009) Mapping wetlands in the Lower Mekong Basin for wetland resource and conservation management using Landsat ETM images and field survey data. J Environ Manage 90(7):2130\u0026ndash;2137\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMahdavi S, Salehi B, Granger J, Amani M, Brisco B, Huang W (2018) Remote sensing for wetland classification: A comprehensive review. GIScience remote Sens 55(5):623\u0026ndash;658\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMcKnight PE, Najab J (2010) Mann-Whitney U Test. The Corsini Encyclopedia of Psychology, 1\u0026ndash;1\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMitsch WJ, Gosselink JG (2015) Wetlands. Wiley\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMoreno-Mateos D, Power ME, Com\u0026iacute;n FA, Yockteng R (2012) Structural and functional loss in restored wetland ecosystems. PLoS Biol, 10(1), e1001247\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMRC (1997) Mekong River Basin Diagnostic Study. Final Report. MKG/R. 97010. Thailand, Bangkok\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMRC (2005) Overview of the hydrology of the Mekong Basin vol, vol 82. Mekong River Commission, Vientiane, Lao PDR\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNarayan S, Beck MW, Wilson P, Thomas CJ, Guerrero A, Shepard CC, Trespalacios D (2017) The value of coastal wetlands for flood damage reduction in the northeastern USA. Sci Rep 7(1):1\u0026ndash;12\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNicholls RJ, Hoozemans FM, Marchand M (1999) Increasing flood risk and wetland losses due to global sea-level rise: regional and global analyses. Glob Environ Change 9:S69\u0026ndash;S87\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNicholls RJ (2004) Coastal flooding and wetland loss in the 21st century: changes under the SRES climate and socio-economic scenarios. Glob Environ Change 14(1):69\u0026ndash;86\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNakaegawa T (2012) Comparison of water-related land cover types in six 1-km global land cover datasets. J Hydrometeorol 13(2):649\u0026ndash;664\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eOzesmi SL, Bauer ME (2002) Satellite remote sensing of wetlands. Wetlands Ecol Manage 10:381\u0026ndash;402\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePekel JF, Cottam A, Gorelick N, Belward AS (2016) High-resolution mapping of global surface water and its long-term changes. Nature 540(7633):418\u0026ndash;422\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePedregosa F, Varoquaux G, Gramfort A, Michel V, Thirion B, Grisel O, Duchesnay \u0026Eacute; (2011) Scikit-learn: Machine learning in Python. J Mach Learn Res 12:2825\u0026ndash;2830\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePokhrel Y, Burbano M, Roush J, Kang H, Sridhar V, Hyndman DW (2018) A review of the integrated effects of changing climate, land use, and dams on Mekong river hydrology. Water 10(3):266\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRebelo LM, Finlayson CM, Nagabhatla N (2009) Remote sensing and GIS for wetland inventory, mapping and change analysis. J Environ Manage 90(7):2144\u0026ndash;2153\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRingler C, Cai X (2006) Valuing fisheries and wetlands using integrated economic-hydrologic modeling\u0026mdash;Mekong River Basin. J Water Resour Plan Manag 132(6):480\u0026ndash;487\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eShrestha S, Bach TV, Pandey VP (2016) Climate change impacts on groundwater resources in Mekong Delta under representative concentration pathways (RCPs) scenarios, vol 61. Environmental science \u0026amp; policy, pp 1\u0026ndash;13\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSok T, Oeurng C, Kaing V, Sauvage S, Lu X, P\u0026eacute;rez JMS (2022) Nutrient transport and exchange between the Mekong River and Tonle Sap Lake in Cambodia. Ecol Eng 176:106527\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTeferi E, Uhlenbrook S, Bewket W, Wenninger J, Simane B (2010) The use of remote sensing to quantify wetland loss in the Choke Mountain range, Upper Blue Nile basin, Ethiopia. Hydrol Earth Syst Sci 14(12):2415\u0026ndash;2428\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eVerhoeven JT, Arheimer B, Yin C, Hefting MM (2006) Regional and global concerns over wetlands and water quality. Trends Ecol Evol 21(2):96\u0026ndash;103\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eVymazal J (2010) Constructed wetlands for wastewater treatment. Water 2(3):530\u0026ndash;549\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWang C, Leisz S, Li L, Shi X, Mao J, Zheng Y, Chen A (2024) Historical and projected future runoff over the Mekong River basin. Earth Sys Dyn 15(1):75\u0026ndash;90\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYang J, Yang YE, Chang J, Zhang J, Yao J (2019) Impact of dam development and climate change on hydroecological conditions and natural hazard risk in the Mekong River Basin. J Hydrol 579:124177\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhang S, Na X, Kong B, Wang Z, Jiang H, Yu H, Dale P (2009) Identifying wetland change in China\u0026rsquo;s Sanjiang Plain using remote sensing. Wetlands 29:302\u0026ndash;313\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"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":"Landfill, Flood, Floodplain, Dams","lastPublishedDoi":"10.21203/rs.3.rs-4781968/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-4781968/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eWetland loss and degradation is a major global issue in which its detailed estimation of spatio-temporal distribution will be a key for understanding the dynamics and subsequent impact assessment studies. This study aimed to estimate the spatio-temporal distributions of wetland loss, analyze its geographical characteristics, and quantify the likelihood of loss occurrence for existing wetlands in the lower Mekong River Basin in Cambodia. Using the global surface water detection datasets, the spatiotemporal distribution of wetland loss with high resolution (30m) over the entire study area (140km\u0026times;210km) during 1984\u0026ndash;2021 was estimated. Statistically significant differences were found in the distance from urban areas and distance from river channels for the existing wetlands and lost wetlands as of 2021, in which the lost wetlands tend to locate closer to urban areas. Subsequent Land Use/Land Cover after the wetland loss was found to be mainly croplands (72.2%) in the study area. Though our estimate overall agrees with the recent global-scale estimate, our estimate resulted in notable ratio of rangelands (11.3%), which represents the unique characteristics of floodplain wetlands in the lower Mekong River Basin. The Random Forest and Light GBM algorithms-based wetland loss prediction models resulted in good statistical evaluation metrics. In both models, the distance from river channels was found to be the most important feature for classifying existing wetlands and lost wetlands. Application of the developed models successfully provided the map of likelihood of wetland loss for existing wetlands in the study area.\u003c/p\u003e","manuscriptTitle":"Spatio-temporal analysis of wetland loss in the lower Mekong River Basin based on surface water detection datasets and machine learning","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-01-13 11:28:59","doi":"10.21203/rs.3.rs-4781968/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":"013fd0dd-0b92-42a5-aebf-979337edbaf4","owner":[],"postedDate":"January 13th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2025-03-31T19:57:39+00:00","versionOfRecord":[],"versionCreatedAt":"2025-01-13 11:28:59","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-4781968","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-4781968","identity":"rs-4781968","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}