Interpretable baseflow segmentation and prediction based on numerical experiments and deep learning.

Baseflow is a crucial water source in the inland river basins of high-cold mountainous region, playing a significant role in maintaining runoff stability. It is challenging to select the most suitable baseflow separation method in data-scarce high-cold mountainous region and to evaluate effects of climate factors and underlying surface changes on baseflow variability and seasonal distribution characteristics. Here we attempt to address how meteorological factors and underlying surface changes affect baseflow using the Grey Wolf Optimizer Digital Filter Method (GWO-DFM) for rapid baseflow separation and the Long Short-Term Memory (LSTM) neural network model for baseflow prediction, clarifying interpretability of the LSTM model in baseflow forecasting. The proposed method was successfully implemented using a 63-year time series (1958-2020) of flow data from the Tai Lan River (TLR) basin in the high-cold mountainous region, along with 21 years of ERA5-land meteorological data and MODIS data (2000-2020). The results indicate that: (1) GWO-DFM can rapidly identify the optimal filtering parameters. It employs the arithmetic average of three methods, namely Chapman, Chapman-Maxwell and Eckhardt filter, as the best baseflow separation approach for the TLR basin. Additionally, the baseflow significantly increases after the second mutation of the baseflow rate. (2) Baseflow sources are mainly influenced by precipitation infiltration, glacier frozen soil layers, and seasonal ponding. (3) Solar radiation, temperature, precipitation, and NDVI are the primary factors influencing baseflow changes, with Nash-Sutcliffe efficiency coefficients exceeding 0.78 in both the LSTM model training and prediction periods. (4) Changes in baseflow are most influenced by solar radiation, temperature, and NDVI. This study systematically analyzes the changes in baseflow and response mechanisms in high-cold mountainous region, contributing to the management of water resources in mountainous basins under changing environmental conditions.

Characterising baseflow signature variability in the Yellow River Basin.

Baseflow is pivotal in maintaining catchment ecological health and improving sustainable economic development. The Yellow River Basin (YRB) is northern China's most important water supplier. However, it faces water shortage due to synergistic effects between natural conditions and anthropogenic activities. Investigating baseflow characteristics quantitively is, therefore, beneficial to promoting the sustainable development of the YRB. In this study, daily ensemble means baseflow data derived from four revised baseflow separation algorithms (i.e., the United Kingdom Institute of Hydrology (UKIH), Lyne-Hollick, Chapman-Maxwell, and Eckhardt methods) - was obtained from 2001 to 2020. Thirteen baseflow dynamics signatures were extracted to investigate baseflow spatiotemporal variations and their determinants across the YRB. The main findings were: (1) There were significant spatial distribution patterns of baseflow signatures, and most signatures had higher values in upstream and downstream reaches than in the middle reaches. There were also mixing patterns with higher values in middle and downstream reaches simultaneously. (2) The magnitude of temporal variation in baseflow signatures was most strongly correlated with catchment terrain (r = -0.4), vegetation growth (r > 0.3), and cropland coverage (r > 0.4). (3) There was a strong synergistic effect of multiple factors (e.g., soil textures, precipitation and vegetation conditions) on baseflow signature values. This study provided a heuristic evaluation of baseflow characteristics in the YRB, contributing to water resources management in the YRB and similar catchments.

Baseflow estimation based on a self-adaptive non-linear reservoir algorithm in a rainy watershed of eastern China.

Accurate baseflow estimation is critical for water resources evaluation and management, and non-point source pollution quantification. Nonlinear reservoir algorithm (NRA) has been increasingly applied to baseflow separation because of its good approximation to the real groundwater discharge (commonly dominated by the unconfined aquifer) in most watersheds. However, in the rainy regions, large uncertainties may remain in the traditional NRA-separated baseflow sequences due to its empirical transition function for the rising limb of discharge process, and the evident variations of baseflow recession in the initial period of the falling limb caused by the disturbance from surface flow or rainfall events. To improve the reliability of baseflow separation, a self-adaptive non-linear reservoir algorithm (SA-NRA) was developed in this study based on the NRA, a self-adaptive groundwater discharge modified parameter, and the Particle Swarm Optimization algorithm (PSO). The validation of SA-NRA in a rainy watershed of eastern China showed that SA-NRA could be the approach to provide a goodness-of-fit for baseflow recession behaviors in the rainy regions. The traditional NRA and Eckhardt's two-parameter recursive digital filter (ERDF), calibrated (or validated) only with the pure baseflow recession data, can hardly provide reliable baseflow predictions for the non-pure baseflow recession periods (including the rising limb and the falling limb with surface flow or rainfall disturbance) due to the apparent variations of baseflow recession behavior. Therefore, more attentions should be paid to the uncertainties of baseflow separation for the non-pure baseflow recession periods in the rainy regions.

Response of the runoff process to meteorological drought: Baseflow index as an important indicator.

Runoff and baseflow are two hydrological elements most closely involved in water-resource management. Defining the response of runoff/baseflow to meteorological drought (MD) is helpful for designing precise drought resisting measures. Thus, Pearson correlation coefficients and mutual information scores between runoff/baseflow and MD in five sub-basins of the Weihe River Basin (WRB) were estimated on a weekly scale, and the best response times of runoff/baseflow to MD on annual and calendar months were determined according to the maximum degree of response. Furthermore, the spatial and seasonal differences in response characteristics in the WRB were discussed and the baseflow index (BFI) was introduced to further explain the propagation process of MD to runoff/baseflow. The results showed that (1) in addition to the response time, the transition sequences of MD propagating to runoff and baseflow varied across basins due to the specific basin properties; (2) Response time of runoff to MD was related to BFI value and showed significant seasonality and hydrological periodicity. In summer and autumn (wet season), the response was faster and stronger, whereas the opposite occurred in winter and spring (normal/dry season); (3) BFI values indicated the main path of drought propagation, explaining the variation in response time between basins and seasons; hence, it can be used to simply and effectively determine the propagation speed of MD to runoff. This study clarified the response characteristics of the runoff process to MD and enhanced our understanding of the drought propagation process, which is crucial for mitigating and managing drought-related hazards.

Contribution of nonpoint source pollution from baseflow of a typical agriculture-intensive basin in northern China.

Baseflow is an essential component of total surface runoff that is widely considered one of the most influential factors regarding water quality via nonpoint source (NPS) pollution. Previously, many researchers and policy makers have directed their efforts toward surface runoff pollution, largely ignoring nutrient delivery via baseflow. Taking a typical agriculture-intensive basin of northern China as an example, this study explored the spatiotemporal characteristics of baseflow and pollution load in relation to NPS pollution. Baseflow was quantified using digital filtering techniques, and the results together with observed pollution data were used to validate a physically based hydrological model, i.e., the Soil and Water Assessment Tool. Then, the spatial and temporal distribution characteristics of NPS and baseflow pollution were investigated using the modeling results. Results indicated that baseflow contribution to total runoff accounted for more than 70% during the studied years (2016-2018), and 84.15% of the basin area showed non-point source pollution dominated by baseflow pollution; both baseflow and its pollution load were greater in the nonflood seasons (spring, autumn, and winter) than in the flood season (summer); the spatial distribution of baseflow total nitrogen and total phosphorus pollution intensity showed higher values in the east and lower values in the west; the scaling effects of baseflow and its pollution load was found with increasing basin area. The results of our study highlighted the necessity for management of pollution load via baseflow in the river basin and provided reference information for improvement of NPS pollution management in other similar basins.

Modeling seasonal water yield for landscape management: Applications in Peru and Myanmar.

A common objective of watershed management programs is to secure water supply, especially during the dry season. To develop such programs in contexts of low data and resource availability, program managers need tools to understand the effect of landscape management on the seasonal water balance. However, the performance of simple, parsimonious models is poorly understood. Here, we examine the behavior of a geospatial tool, developed to map monthly water budgets and baseflow contributions and forming part of the InVEST (integrated valuation of ecosystem services and trade-offs) software suite. The model uses monthly climate, topography, and land-use data to compute spatial indices of groundwater recharge, baseflow, and quickflow. We illustrate the model application in two large basins in Peru and Myanmar, where we compare results with observed data and alternative hydrologic models. We show that the spatial distribution of baseflow contributions correlated well with an established model in the Peruvian basin (r2 = 0.81 at the parcel scale). In Myanmar, the model shows an overall satisfactory performance for representing month to month variation (Nash-Sutcliffe-Efficiency 0.6-0.8); however, errors are scale dependent highlighting limitations in representing processes in large basins. Our study highlights modeling challenges, in particular trade-offs between model complexity and accuracy, and illustrates the role that parsimonious models can play to support watershed management programs.

Challenges Calibrating Hydrology for Groundwater-Fed Wetlands: a Headwater Wetland Case Study.

This study aims to adapt the Soil and Watershed Assessment Tool (SWAT), a ubiquitously used watershed model, for ground-water dominated surface waterbodies by accounting for recharge from the aquifers. Using measured flow to a headwater slope wetland in Alabama's coastal plain region as a case study, we present challenges and relatively simple approaches in using the SWAT model to predict flows from the draining watershed and relatively simple approaches to model groundwater upwelling. SWAT-simulated flow at the study watershed was limited by precipitation, and consequently, simulated flows were several times smaller in magnitude than observed flows. Thus, our first approach involved a separate stormflow and baseflow calibration which included the use of a regression relationship between observed and simulated baseflow (E NASH = 0.67). Our next approach involved adapting SWAT to simulate upwelling groundwater discharge instead of deep aquifer losses by constraining the range of deep losses, ß deep parameter, to negative values (E NASH = 0.75). Finally, we also investigated the use of artificial neural networks (ANN) in conjunction with SWAT to further improve calibration performance. This approach used SWAT-calibrated flow, evapotranspiration, and precipitation as inputs to ANN (E NASH = 0.88). The methods investigated in this study can be used to navigate similar flow calibration challenges in other groundwater dominant watersheds which can be very useful tool for managers and modelers alike.

Baseflow estimation for catchments in the Loess Plateau, China.

Baseflow is the portion of streamflow that originates from groundwater. It is pivotal to groundwater supply as well as the interactions between groundwater and surface water. In the Loess Plateau of China, the baseflow plays a fundamental role in sustaining the aquatic ecosystem. Rational estimation of baseflow is of critical importance for understanding the functioning of the groundwater system and informing management of regional water resources. In this study, a two-well parameterised digital filter was used to separate baseflow from observed daily total streamflow on the Weihe River Basin. Two parameters (maximum baseflow index (BFImax) and the recession constant) of this filter were estimated using UKIH and a recession analysis with physical meaning, respectively. The results show that the baseflow index increased from the upstream to downstream (0.27-0.32) of the Weihe River, and the baseflow of the river is summer dominant. In general, the baseflow has been increasing gradually owing to the implementation of soil conservation measures. This study provides an understanding of baseflow response to seasonal variability and water environmental management in basin scale.

Hydrograph separation of subsurface tile discharge.

Baseflow is an important component of streamflow and watershed hydrologic budgets, yet quantifying the baseflow fraction of tile drainage has rarely been reported. In this study, we used two common hydrograph separation methods (local minimum method, recursive digital filter) to separate the discharge hydrographs from three drainage district tiles located in Iowa. Based on data collected from 2009 to 2013, annual baseflow ranged from 116 to 162 mm and comprised approximately 60% of the annual discharge. Baseflow was greatest during June (average of 34% of annual baseflow) and the March through August period produced 86% of the total annual baseflow. We found that the two methods of hydrograph separation produced similar results but the digital filter method was less erratic in estimating baseflow fraction. Study results can be used to better quantify hydrologic pathways in tiled landscapes and improve the design, implementation, and evaluation of nutrient reduction strategies.

Linking fecal bacteria in rivers to landscape, geochemical, and hydrologic factors and sources at the basin scale.

Linking fecal indicator bacteria concentrations in large mixed-use watersheds back to diffuse human sources, such as septic systems, has met limited success. In this study, 64 rivers that drain 84% of Michigan's Lower Peninsula were sampled under baseflow conditions for Escherichia coli, Bacteroides thetaiotaomicron (a human source-tracking marker), landscape characteristics, and geochemical and hydrologic variables. E. coli and B. thetaiotaomicron were routinely detected in sampled rivers and an E. coli reference level was defined (1.4 log10 most probable number⋅100 mL(-1)). Using classification and regression tree analysis and demographic estimates of wastewater treatments per watershed, septic systems seem to be the primary driver of fecal bacteria levels. In particular, watersheds with more than 1,621 septic systems exhibited significantly higher concentrations of B. thetaiotaomicron. This information is vital for evaluating water quality and health implications, determining the impacts of septic systems on watersheds, and improving management decisions for locating, constructing, and maintaining on-site wastewater treatment systems.

Projected impacts of urbanisation on hydrological resource flows: A case study within the uMngeni Catchment, South Africa.

Significant land use changes from natural/agricultural to urban land uses have been proposed within the Mpushini/Mkhondeni sub-catchments of the uMngeni Catchment in South Africa. A better understanding of the influences which such land use changes are likely to have on hydrological flows, is required, in order to make informed land use decisions for a sustainable future. As a point of departure, an overview of linkages between urbanisation and hydrological flow responses within this sub-humid study area is given. The urban characteristics of increased impervious areas and the potential return flows from transfers of potable water from outside the catchment were identified as being important in regard to hydrological flow responses. A methodology was developed to model urban response scenarios with urban characteristics as variables, using the daily time-step process based ACRU model. This is a hydrological multi-process model and not an urban hydraulic model and it addresses the landscape as well as the channel components of a catchment, and in addition to runoff components includes evaporation and transpiration losses as outputs. For the study area strong links between proposed urbanisation and hydrological resource flow responses were found, with increases in stormflows, together with increased and more regulated baseflows, and with impacts varying markedly between dry or wet years and by season. The impacts will depend on the fractions of impervious areas, whether or not these are connected to permeable areas, the amount of imported water and water system leaks. Furthermore, the urban hydrological impacts were found to be relatively greater in more arid than humid areas because of changes in the rainfall to runoff conversion. Flow changes due to urbanisation are considered to have important environmental impacts, requiring mitigation. The methodology used in this paper could be used for other urbanising areas.

Bundling ecosystem services in the Panama Canal watershed.

Land cover change in watersheds affects the supply of a number of ecosystem services, including water supply, the production of timber and nontimber forest products, the provision of habitat for forest species, and climate regulation through carbon sequestration. The Panama Canal watershed is currently being reforested to protect the dry-season flows needed for Canal operations. Whether reforestation of the watershed is desirable depends on its impacts on all services. We develop a spatially explicit model to evaluate the implications of reforestation both for water flows and for other services. We find that reforestation does not necessarily increase water supply, but does increase carbon sequestration and timber production.

A Watershed Analysis of Seasonal Concentration- and Loading-based Results for Escherichia coli in Inland Waters.

Fecal indicator bacteria, such as Escherichia coli, are frequently monitored in recreational waterbodies as indicators of potential fecal pathogen presence and exposure. The present watershed analysis investigated the influence of season on E. coli concentration (MPN/100 mL) and loading (MPN/day) measurements for inland waters at baseflow conditions. The master dataset collected during 2007-2012 for three watersheds included 896 E. coli (Colilert) samples with simultaneous flow taken for a subset (39 %) of samples. The outcomes for grouped watersheds were reflected in most cases for individual watersheds. Concentration- and loading-based results were highest in summer and spring, and lowest in the winter and fall, respectively. The comparison of these two measurement techniques (concentration and loading) highlight the impact of flow data during baseflow conditions for inland waters and reveal that caution should be used when inferring one method's results from another.

Enhancing hydrological insight: isotopic methods revealing groundwater-surface water interactions in the Lower Quang Tri River Group, Vietnam.

The Lower Quang Tri River Group, situated in central Vietnam, faces a myriad of challenges, notably the decline in groundwater levels and the salinisation of both groundwater and surface water, significantly impacting water availability for domestic, agricultural, and industrial purposes. To address these pressing concerns, this study adopts a comprehensive methodology integrating hydrogeological measurements, isotopic techniques, and chemical analyses of various water sources, including local precipitation, surface water bodies, reservoirs, and groundwater samples. Utilising the deuterium and oxygen-18 signatures (δ2H and δ18O) in water molecules as environmental tracers for the assessment of base flow and water sources enables a nuanced understanding of the intricate interaction between surface water and groundwater. Research findings elucidate that during the dry season, groundwater recharge primarily stems from water in the reservoirs over approximately seven months. Base flow contributes between 80 and 85 % of streamflow during the rainy season, escalating to 100 % during the dry season. The mean travelling time of the base flow is estimated at 120 ± 10 days using the sine curve model developed by Rodgers et al. The insights gleaned from this study are poised to play a pivotal role in guiding the local water resources managers in licensing for the exploitation of a right quantities of groundwater as sustainable management strategies in the region.

Pesticides and pharmaceuticals data collected during two consecutive years in a Mediterranean micro-estuary.

The Alexander micro-estuary, located at the eastern edge of the Mediterranean Sea, is a typical example of small water bodies that suffer from a combination of urban and agricultural pollution, and overuse of its natural water sources. It is∼6.5 km long, with maximum depth of 3 m and maximum width of 45 m. To evaluate the anthropogenic stress on the system and its ability to mitigate pollution, water samples were collected within the framework of Ruppin's Estuarine and Coastal Observatory [1]. Water samples were collected from the estuary head, which drains about 510 km2, and at a point 300 m upstream from the estuary mouth before water flows into the Mediterranean Sea. A total of 236 stormwater and 44 base-flow water samples between December 2016 and December 2018. Stormwater samples were collected every 0.25 - 4 h along the entire course of the flow events using an automated samplers (Sigma 900, Hach Company, Loveland CO, USA; and ISCO 3700 Full-Size Portable Sampler, Teledyne, Lincoln, NE, USA). Base-flow samples were taken once a month using a horizontal grab sampler (5 L, model 110B, OceanTest Equipment, Fort Lauderdale, FL, USA). All samples were filtered using 90mmGF/F filters (nominal pore size of 0.7 µm, MGF, Sartorius, Göttingen, Germany) and immediately frozen (-20°C) before chemical analysis. Chemical analysis was performed using liquid chromatography with high-resolution mass spectrometry (LC-HRMS) analysis using a QExactive Plus hybrid FT mass spectrometer coupled with a Dionex Ultimate 3000 RS UPLC (Thermo Fisher Scientific, Waltham, MA, USA). The targeted analysis, which included 15 fungicides, 25 herbicides, 18 Insecticides, and 19 pharmaceuticals, concluded with a total of 21,142 entries. All entries are organized in a worksheet, along with location, date, flood section duration, discharge rate, and the total water volume discharged during the relevant period. The provided data offers an opportunity to explore the sources, transport, and impact of a large mixture of organic pollutants in a confined aquatic system located in an urbanized coastal environment.

Increasing volatility of reconstructed Morava River warm-season flow, Czech Republic.

Study region: The Morava River basin, Czech Republic, Danube Basin, Central Europe. Study focus: Hydrological summer extremes represent a prominent natural hazard in Central Europe. River low flows constrain transport and water supply for agriculture, industry and society, and flood events are known to cause material damage and human loss. However, understanding changes in the frequency and magnitude of hydrological extremes is associated with great uncertainty due to the limited number of gauge observations. Here, we compile a tree-ring network to reconstruct the July-September baseflow variability of the Morava River from 1745 to 2018 CE. An ensemble of reconstructions was produced to assess the impact of calibration period length and trend on the long-term mean of reconstruction estimates. The final estimates represent the first baseflow reconstruction based on tree rings from the European continent. Simulated flows and historical documentation provide quantitative and qualitative validation of estimates prior to the 20th century. New hydrological insights for the region: The reconstructions indicate an increased variability of warm-season flow during the past 100 years, with the most extreme high and low flows occurring after the start of instrumental observations. When analyzing the entire reconstruction, the negative trend in baseflow displayed by gauges across the basin after 1960 is not unprecedented. We conjecture that even lower flows could likely occur in the future considering that pre-instrumental trends were not primarily driven by rising temperature (and the evaporative demand) in contrast to the recent trends.

Attenuation of organic pollutants and the effects of salinity and seasonality in a Mediterranean micro-estuary.

Micro-estuaries are small ubiquitous transitional water bodies that are often located in semi-arid zones. Unlike the vastly studied large estuaries, micro-estuaries lack the ability to dilute and contain pollution from point and non-point sources due to low natural water discharges. Therefore, these diverse ecological systems are susceptible to pollutant loads due to prolonged water residence time and complex geochemical dynamics. Although this elevated anthropogenic stress limits their potential to provide ecological and recreational services, micro-estuaries have some traits similar to those found in wetlands, which provide a natural potential to retain and mitigate organic pollutants. A two consecutive years study conducted at the Alexander micro-estuary tracked the influx and outflux of a large organic pollutant mixture during base-flow and flood events. During the research period, 165 kg of active ingredients entered the micro-estuary and 160 kg flowed out to the Mediterranean Sea, suggesting negligible net attenuation. However, this broad picture conceals inner shifts in pollutant mixture loads, which contained 46 pesticides and 19 pharmaceuticals. Only a handful of pollutants were actually balanced, whereas most compounds were either removed or added to the flow, with no observed correlation to chemical properties. A prominent observation was the load increase along the flow for some pollutants during base-flow conditions. This trend, which was correlated with salinity elevation and was verified in lab experiments, suggests that seawater intrusion to the bottom of the estuary may increase desorption rates of pollutants from the estuary bed, creating an estuarine desorption magnification effect. The combination of strong anthropogenic stress with increased desorption rates severely limits the estuary's potential to mitigate pollutants, frequently transforming it into a pollution source rather than a sink.

Improving the representation of stream water sources in surrogate nutrient models with water isotope data.

An important part of meeting nutrient reduction goals in the lower Great Lakes basin and assessing the success of different land management strategies is modeling nutrient losses from agricultural land. This study aimed to improve the representation of water source contributions to streamflow in generalized additive models for predicting nutrient fluxes from three headwater agricultural streams in southern Ontario monitored during the Multi-Watershed Nutrient Study (MWNS). The previous development of these models represented baseflow contributions to streamflow using the baseflow proportion derived using an uncalibrated recursive digital filter. Recursive digital filters are commonly used to partition stream discharge into separate components from slower and faster pathways. In this study, we calibrated the recursive digital filter using stream water source information from stable isotopes of oxygen in water. Across sites, optimization of the filter parameters reduced bias in baseflow estimates by as much as 68 %. In most cases, calibrating the filter also improved agreement between filter-derived baseflow and baseflow calculated from isotope and streamflow data: the average Kling-Gupta Efficiencies using default and calibrated parameters were 0.44 and 0.82, respectively. When incorporated into the generalized additive models, the revised baseflow proportion predictor was more often statistically significant, improved model parsimony, and reduced prediction uncertainty. Moreover, this information allowed for a more rigorous interpretation of how different stream water sources influence nutrient losses from the agricultural MWNS watersheds.

An index for inferring dominant transport pathways of solutes and sediment: Assessing land use impacts with high-frequency conductivity and turbidity sensor data.

Land use change threatens aquatic ecosystems through freshwater salinization and sediment pollution. Effective river management requires an understanding of the dominant hydrologic pathways of sediment and solute delivery. To address this, we applied hysteresis analysis, hydrograph separation, and linear regression to hundreds of events across a decade of specific conductance and turbidity data from three streams along a rural-to-urban gradient. Thereafter, we developed an index (ßrunoff') to quantify the relative influence of surface runoff to event-scale suspended sediment generation, where a value of '1' indicates complete alignment of suspended sediment generation with the temporal structure of runoff whereas '0' indicates total alignment with baseflow. Solute hysteresis results showed a predominance of dilution for the rural and mixed-use streams irrespective of road salt presence. On the other hand, urban stream behavior shifted from dilution to flushing following salt application, which was largely driven by greater runoff coefficients and the connectivity of distal solutes to the stream corridor. The newly developed index (ßrunoff') indicated that suspended sediment dynamics were more aligned with runoff in all three streams: rural stream (ßrunoff' = 0.70), mixed stream (ßrunoff' = 0.57), and urban stream (ßrunoff' = 0.64). The relative importance of baseflow to sediment generation grows slightly in urbanizing streams, as impervious surfaces disconnect upland sediment, which would otherwise transport with runoff, while piston-flow baseflow erodes exposed streambanks. Our findings emphasize the need to consider the impact of human modification of the landscape on solute and sediment transport in freshwater systems for effective water quality management. Further, our ßrunoff' index provides a useful tool for assessing the relative influence of surface runoff on event-scale solute or sediment generation in streams, supporting river management and conservation efforts.

River ecological flow early warning forecasting using baseflow separation and machine learning in the Jiaojiang River Basin, Southeast China.

Ecological flow early warning is crucial for the rational management of watershed water resources. However, determining of accurate ecological flow threshold and choosing the appropriate forecasting model are challenging tasks. In this study, we initially developed a baseflow separation and Tennant method-based technique for calculating ecological river flow. Then an ecological flow early warning model was created using the machine learning technique based on distributed gradient enhancement framework (LightGBM). Finally, we utilized the framework of Shapley Additive Planning (SHAP) to explain how various hydrometeorological factors affect the variations in ecological flow conditions. The Jiaojiang River basin in southeast China is selected as the study area, and the hydrological stations in upstream of Baizhiao (BZA) and Shaduan (SD) are chosen for key analysis. The results of these applications show that the monthly baseflow frequency of the river ecological flow conditions of the two stations in the dry season is 20 % (7.49 m3/s) and 30 % (4.79 m3/s), respectively. The ecological flow level early warning forecasting accuracy is close to 90 % in the BZA and SD stations during dry and wet seasons. The variations of ecological flow are most affected by evaporation and base flow index. The results of this study can serve as a strong basis for the effective allocation and utilization of locally available water resources.