Data-scarce Regions Research Articles

Flood impact assessment in remote areas using machine learning, SAR, and GIS: a case study of Ngabang District, Indonesia

ABSTRACT Flooding in remote regions presents significant challenges due to data scarcity, complicating impact assessment and mitigation efforts. This research delineates an integrated methodology for quantifying flood impacts in such contexts. By leveraging machine-learning algorithms, Sentinel-1 synthetic aperture radar (SAR) imagery was combined with digital elevation model data and river proximity metrics to predict and accurately demarcate flood extents. Geographic information systems overlay techniques were then employed for spatial analysis of the floods’ impacts on population and infrastructural assets. The methodology was applied in a case study in Ngabang District, Indonesia, demonstrating its utility. Analysis using decision tree, random forest (RF), and gradient boosting machine models provided critical insights into flood prediction factors. The RF model was chosen as the best, successfully identified flood-prone regions, achieving an accuracy of 0.94 and a Kappa of 0.87 on the testing data, demonstrating its robustness. The flood map showed significant impacts, affecting 373.81 hectares, 10,706 people, 1,500 buildings, and 15 km of roads. This study highlights the importance of proximity, elevation, SAR imagery, and iterative model improvements in flood prediction, offering valuable insights for flood management and mitigation efforts in data-scarce regions.

Tufas record significant imprints of climate and tectonic activity over the past 600 ka: Evidence from a multi-story wedge in Northwest Africa

The Quaternary Period climatic oscillations, typically those driven by Milankovitch cycles, have significantly left profound imprints in the geological records. However, the potential of terrestrial archives, particularly tufa deposits, as archives for Quaternary climate remain relatively underexplored. This study aims to contribute to filling the gap in a data-scarce region (Northwest Africa) through investigating Middle Pleistocene to Holocene groundwater-fed tufa deposits, using fieldwork-based, process-oriented facies analysis and advanced 230Th technique. Eight stratigraphic sections were measured, identifying thirteen sedimentary facies grouped into four facies associations representing specific depositional settings: (i) fluvio-lacustrine and palustrine tufa, (ii) barrage cascade tufa with buttresses, (iii) tufa in channels with pools, and (iv) tufa in shallow braided channels with free-flowing water. The depositional system comprises a single multi-story wedge of vertically and laterally stacked barrage-cascade dammed-area systems.Our Integrated analysis revealed significant tufa growth during MIS 14, with implications for a warm and humid MISs 15–13 ‘extra-long interglacial’, consistent with multiple paleoclimate records. Such conditions would have supported late Acheulean hominin occupation and the emergence of Early Modern Humans in the region around 300 ka. The findings also highlight the possible impact of the Mid-Brunhes Event (MBE) on tufa deposition patterns, likely through CO2-climate feedbacks, underscoring the sensitivity of tufa records to global climatic changes. Moreover, the study outcomes suggest significant regional tectonic activity during the late Middle Pleistocene, causing counter-clockwise rotation and northward tilting of MIS 14 deposits. The associated seismic events may have modified the hydrogeological budget and, by extension, altered the balance between tufa aggradation and degradation. The rotation about the vertical axis suggests a bookshelf faulting mechanism driven by regional left-lateral strike-slip tectonics. Differential karstic dissolution and tectonic forces would have further contributed to deposit tilting. The study highlights the potential of tufa as a valuable terrestrial archive for understanding Quaternary climate dynamics and tectonic processes. Future research should aim to expand the chronological framework and further investigate the climatic and tectonic influences on tufa deposition in this data-scarce region. This can be achieved through additional 230Th dating, high-resolution stable isotope measurements on microbialites, particularly focusing on the significance of the Mid-Brunhes Event (MBE) in Northwest Africa.

Enhanced landslide susceptibility mapping in data-scarce regions via unsupervised few-shot learning

Given the critical need to assess landslide hazards, producing landslide susceptibility map (LSM) in regions with scarce historical landslide inventories poses significant challenges. This study introduces a novel landslide susceptibility assessment framework that combines unsupervised learning strategies with few-shot learning methods to increase the accuracy of LSM in these areas. The framework has been practically validated in a representative geological disaster-prone area along the West-East Gas Pipeline in Shaanxi Province, China. We employed three advanced few-shot learning models: a support vector machine, meta-learning, and transfer learning. These models implement feature representation learning for weakly correlated influencing factors through an unsupervised approach, thereby constructing an effective landslide susceptibility assessment model. We compared traditional learning methods and used the receiver operating characteristic (ROC) curve and SHAP values to quantify the effectiveness of the models. The results indicate that the meta-learning algorithm outperforms both the SVM and transfer learning in areas with limited landslide data. The integration of unsupervised strategies significantly improves performance, achieving area under the curve (AUC) values of 0.9385 and 0.9861, respectively. Compared with using meta-learning alone, incorporating unsupervised learning strategies increased the AUC by 4.76%, enhancing both the predictive power of the model and the interpretability of the features. Meta-learning under unsupervised conditions effectively mitigates the evaluation difficulties caused by insufficient landslide records, providing a viable path and empirical evidence for performance improvement in similar data- scarce regions worldwide.

Integrating machine learning and zoning-based techniques for bias correction in gridded precipitation data to improve hydrological estimation in the data-scarce region

Integrating machine learning and zoning-based techniques for bias correction in gridded precipitation data to improve hydrological estimation in the data-scarce region

Enhancing the performance of runoff prediction in data-scarce hydrological domains using advanced transfer learning

Accurate hydrological predictions are often hindered by the lack of stream gauges in data-scarce regions, where traditional transfer learning (TL) models like Long Short-Term Memory (LSTM) networks often face limitations due to reduced accuracy and adaptability. To enhance runoff prediction in such regions, we developed DAformer, a novel TL approach that integrates domain adversarial neural networks with the Informer model. Trained on comprehensive runoff data from U.S. basins, DAformer was applied to three basins in Chile and the Chaersen basin in China, demonstrating an effective transfer from data-rich to data-scarce environments. Results show that DAformer significantly outperforms LSTM-based models, improving forecast accuracy by 16.1% for 1-day lead time and by 100.5% for 5-day lead time. These improvements indicate that the DAformer model not only enhances prediction accuracy but also holds substantial practical implications for flood risk management and water resource planning in regions with limited data availability. By clustering basins based on Shuttle Radar Topography Mission (SRTM) and other geographical data, we found that relying on multiple source basins further enhances the performance. DAformer, therefore, serves as a robust and scalable method for enhancing runoff prediction for regions with limited data.

Evaluation of satellite-derived precipitation datasets: a case study in the Ten Tributaries region of the Yellow River Basin

ABSTRACT Satellite-derived precipitation datasets are essential components of hydrological simulations, particularly in data-scarce regions of western China. However, a comprehensive assessment of their accuracy and reliability is required. Here, the accuracy of two high-resolution satellite-derived precipitation datasets, Integrated Multi-satellite Retrievals for GPM – Final (IMERG-F) and Gauge-Adjusted Global Satellite Mapping of Precipitation (GSMaP-Gauge), was evaluated across the Ten Tributaries region of the Yellow River Basin in western China using four quantitative metrics and three categorical scoring indicators. This evaluation sought to ascertain the retrieval accuracy of these products on both the daily scale and hourly scale of heavy precipitation events, and investigated their inversion error characteristics across various spatiotemporal scales. Both datasets effectively captured the spatiotemporal patterns of annual average precipitation within the study area. Notably, the daily-scale accuracy of these satellite-derived precipitation products surpassed their hourly and half-hourly counterparts. Both GPM-IMERG and GSMaP-Gauge adeptly reproduced most precipitation events in the Ten Tributaries region, with peak detection performance observed in the central and southern zones, providing a reliable data source for drought monitoring and hydrological modeling. Overall, compared with GPM-IMERG, GSMaP-Gauge displayed superior inversion accuracy across diverse spatiotemporal scales.

CCTV image‐based classification of blocked trash screens

AbstractThis study introduces image‐based classification techniques to identify whether trash screens in urban rivers are blocked. The study obtained 755 images from a CCTV camera surveying a trash screen located on an urban river at Tongwynlais in Cardiff. Manual quality control reduced the dataset to 577 images, labelled as either blocked (80%) or unblocked (20%). The performance of a logistic regression for classification of images was investigated using three different subsets of the labelled images: (1) the original dataset, (2) a balanced but under‐sampled dataset with equal number of blocked and unblocked images, and (3) an augmented dataset with an equal number of blocked and unblocked images using Gaussian noise augmentation to increase the number of unblocked images. Results show that our data‐augmentation method enhanced model accuracy by 8%, successfully classifying images as blocked or unblocked with an accuracy of 88%; by overcoming the bias in the dataset these results also highlight potential solutions to overcome the challenges of operating this methodology across a network of cameras. This enables authorities in both data rich and data scarce regions the ability to take advantage of machine learning to open up the next generation of a distributed, data‐driven flood warning systems, protecting people, infrastructure and the environment.

Advancing streamflow prediction in data-scarce regions through vegetation-constrained distributed hybrid ecohydrological models

Hybrid models that combine deep learning with physical principles have recently shown significant promise in improving streamflow prediction in data-scarce regions, achieving generally greater accuracy and reliability than either method alone. However, existing hybrid hydrological models underutilize the extensive and readily available remote sensing data on vegetation, which plays a critical role in hydrologic processes. Incorporating vegetation information can better constrain model processes by accounting for the interactions and feedbacks between vegetation and hydrology, thereby enabling more robust predictions. In this context, we present a novel distributed hybrid ecohydrological model that integrates dynamic vegetation processes into a distributed hybrid hydrological model, capable of simultaneously simulating leaf area index (LAI) and streamflow through multi-task learning. Our results indicate that in scenarios with limited available streamflow data, the combined constraint of LAI spatiotemporal patterns and streamflow significantly enhances the model’s ability to generalize streamflow simulation across space and time compared to models trained solely with either LAI or streamflow. Even with the increased availability of streamflow data, models utilizing both LAI and streamflow continue to exhibit superior robustness in streamflow prediction. The distributed ecohydrological model, which employs multi-process coupling and multi-source data constraints, accurately simulates the target variables (i.e., streamflow and LAI) while reliably capturing dynamic changes in other ecohydrological processes, such as evapotranspiration. This capability highlights its potential for multi-process diagnostics. Overall, our approach offers a new perspective for streamflow prediction in data-scarce regions and represents an important step towards integrating observational constraints from multiple subsystems into hybrid Earth system modeling

Recent advances in electrical engineering: exploring graph neural networks for weather prediction in data-scarce environments

Abstract In regions like Nepal, characterized by diverse geography, missing weather data poses a significant challenge for traditional imputation methods. These methods often struggle to capture the complexities of dynamic environments adequately. To overcome this challenge, our study explores the application of graph neural networks for weather prediction in data-scarce environments. Our approach entails the development of specialized models tailored to accommodate the non-Euclidean topology inherent in weather data. This framework encompasses preprocessing, graph representation, feature selection, and imputation techniques to predict missing atmospheric variables. The adaptability of our models to intricate geography ensures more precise representations of weather conditions. Our research demonstrates the efficacy of these models through rigorous testing on a substantial dataset spanning four decades since 1981. By harnessing state-of-the-art graph neural network technology, our study aims to address existing gaps in weather data prediction, leading to improved historical weather forecasting accuracy. Ultimately, this advancement contributes to enhanced meteorological understanding and forecasting precision in data-scarce regions.

How do gridded meteorological datasets perform in a typical data-scarce cryospheric basin?

Gridded meteorological datasets have been widely used in meteorological and hydrological studies, especially in data-scarce regions. However, assessments of these datasets in cryospheric basins, which are highly sensitive to global warming, remain inadequate. Therefore, three representative gridded meteorological datasets, i.e., the China meteorological forcing dataset (CMFD), CN05.1, and the meteorological forcing dataset for the third-pole region (TPMFD), were systematically evaluated in the Nachitai Basin based on comparisons with limited gauge observations, pairwise intercomparisons, and hydrological simulations. All gridded precipitation and temperature datasets were in good agreement with gauge observations on an intra-annual scale, while substantial differences existed among them in terms of precipitation and temperature on an annual scale. Additionally, pairwise correlation analyses indicated that the temporal consistency between the datasets was higher than their spatial consistency. Furthermore, all three datasets achieved satisfactory and reasonable results in simulating glacier area and streamflow variations when used as inputs for the hydrological−hydrodynamic model, exhibiting the robustness of the model in this data-scarce basin. This study provides a more profound understanding of the significance of gridded datasets in data-scarce areas situated on the Tibetan Plateau.

Urban flooding damage prediction in matrix scenarios of extreme rainfall using a convolutional neural network

The pivotal role of quantitative risk assessment in managing urban stormwater is underscored by the high spatial heterogeneity and complex non-stationarity complex of urban flooding. Conventional methods, reliant on measurable data, often fall short in accurately mapping spatial variations and gauging the full impacts of urban flooding. Addressing this gap, this study proposed a robust convolutional neural networks (CNN)-based tool, specifically designed for urban flooding risk and economic losses estimation under extreme design rainfall scenarios. Furthermore, the study validates the transferability of CNN models trained in data-abundant regions to similar but data-scarce regions, using Guangzhou as a case study for urban flooding damage prediction. The findings reveal that the most affected areas, particularly the old, densely built-up urban areas in the south-central Guangzhou, are susceptible to significant economic losses during extreme rainfall events. Notably, under the most severe scenario (Scenario 5), estimated economic losses amount to approximately $6359.91 million, with industrial and residential sectors bearing the brunt, accounting for 28.29 % and 39.94 % of the total losses, respectively. These insights are crucial for prioritizing mitigation efforts and formulating effective evacuation strategies in high-risk areas, ultimately aiding in the reduction of economic losses.

Assessment of Systematic Errors in Mapping Electricity Access Using Night-Time Lights: A Case Study of Rwanda and Kenya

Remotely sensed nighttime light data have become vital for electrification mapping in data-scarce regions. However, uncertainty persists regarding the veracity of these electrification maps. This study investigates how characteristics of electrified areas influence their detectability using nighttime lights. Utilizing a dataset comprising the locations, installation date, and electricity purchase history of thousands of electric meters and transformers from utilities in Rwanda and Kenya, we present a systematic error assessment of electrification maps produced with nighttime lights. Descriptive analysis is employed to offer empirical evidence that the likelihood of successfully identifying an electrified nighttime light pixel increases as characteristics including the time since electrification, the number of meters within a pixel, and the total annual electricity purchase of meters in a pixel increase. The performance of models trained on various temporal aggregations of nighttime light data (annual, quarterly, monthly, and daily) was compared, and it was determined that aggregation at the monthly level yielded the best results. Additionally, we investigate the transferability of electrification models across locations. Our findings reveal that models trained on data from Rwanda demonstrate strong transferability to Kenya, and vice versa, as indicated by balanced accuracies differing by less than 5% when additional data from the test location are included in the training set. Also, models developed with data from the centralized grid in East Africa were found to be useful for detecting areas electrified with off-grid systems in West Africa. This research provides valuable insight into the characterization of sources of nighttime lights and their utility for mapping electrification.

PyEt v1.3.1: a Python package for the estimation of potential evapotranspiration

Abstract. Evapotranspiration (ET) is a crucial flux of the hydrological water balance, commonly estimated using (semi-)empirical formulas. The estimated flux may strongly depend on the formula used, adding uncertainty to the outcomes of environmental studies using ET. Climate change may cause additional uncertainty, as the ET estimated by each formula may respond differently to changes in meteorological input data. To include the effects of model uncertainty and climate change and facilitate the use of these formulas in a consistent, tested, and reproducible workflow, we present PyEt. PyEt is an open-source Python package for the estimation of daily potential evapotranspiration (PET) using available meteorological data. It allows the application of 20 different PET methods on both time series and gridded datasets. The majority of the implemented methods are benchmarked against literature values and tested with continuous integration to ensure the correctness of the implementation. This article provides an overview of PyEt's capabilities, including the estimation of PET with 20 PET methods for station and gridded data, a simple procedure for calibrating the empirical coefficients in the alternative PET methods, and estimation of PET under warming and elevated atmospheric CO2 concentration. Further discussion on the advantages of using PyEt estimates as input for hydrological models, sensitivity and uncertainty analyses, and hindcasting and forecasting studies (especially in data-scarce regions) is provided.

Comparative analysis of rainfall-runoff simulation using a long short-term memory (LSTM) deep learning model and a conceptual hydrological model, HEC-HMS: a case study of the mountainous river basin of Nepal

Accurate simulation and stream flow prediction are essential for effective water resources planning and management. Hydrological models have long been used for rainfall-runoff modeling; in recent years, with the availability of large datasets, machine learning algorithms, and deep learning techniques have become powerful tools to explore. This study compares rainfall-runoff simulation between the Long and Short-Term Memory (LSTM) (a deep learning model) and the Hydrologic Engineering Centre Hydrologic Modeling System (HEC-HMS) (a conceptual model) for a steep mountainous catchment in Nepal. During calibration (1989–2006) and validation (2007–2016), the HEC-HMS model demonstrated a root mean square error (RMSE) and Nash Sutcliffe Efficiency (NSE) of 18.3 and 0.73 and 22.93 and 0.66, respectively, for the calibration and validation periods. During the same periods, the LSTM model showed values of 7.25 and 0.87 and 19.59 and 0.64, respectively, for the calibration and validation periods. The performance indices show that the LSTM model has performed better than the HEC-HMS model. The outcomes of this study suggest that the LSTM deep learning model can simulate the discharge successfully across steep mountainous catchments characterized by monsoon and snowmelt runoff, even with a wide range of precipitation variation, making it an asset for hydrological modeling and water resources management in the data-scarce region.

Synthetic libraries of urban landslide simulations to identify slope failure hotspots and drivers across spatial scales and landscapes

AbstractRainfall-triggered landslides are most deadly in developing countries, and future urban sprawl and climate change could intensify existing risks. In these regions, enhancing efforts in urban landslide risk mitigation and climate change adaptation is crucial. Current landslide probability assessment methodologies struggle to support effective mitigation because they fail to represent local anthropogenic factors (e.g. informal housing) across space and time scales. To meet this challenge, we demonstrated in previous work that hillslope-scale mechanistic models representing such localised changes can be used to create synthetic libraries of urban landslides that account for both data and future scenario uncertainty. Here, we show how these libraries can become an explorative tool for researchers and stakeholders, allowing them to investigate slope stability variations across spatial scales and landscapes. Results highlight, for example, how the main slope instability drivers change according to the location (e.g., upper vs lower catchment), the landcover (e.g. forest vs urban) and the spatial scale analysed (e.g. at hillslope scale slope stability was mostly controlled by water table height, whereas at regional scale by slope geometry). Ultimately, we demonstrate that stochastic analyses can lead to a greater understanding of the system interactions and they can support the identification of mitigation strategies that perform well across spatial scales and uncertain scenarios. These strategies should be prioritised even if future conditions are unknown. This reasoning is shown on a data-scarce region with expanding informal housing. However, the same methodology can be applied to any urban context and with any mechanistic-based model.

Developing an open-source flood forecasting system adapted to data-scarce regions: A digital twin coupled with hydrologic-hydrodynamic simulations

Economic and human losses from flooding have had a significant global impact. Undeveloped nations often require extended periods to recover from flood-related damage, exacerbating the climate poverty trap, specifically in flood-prone regions. To address this issue, early warning systems (EWS) provide lead time for preparedness and measures to reduce vulnerability. However, EWS are mainly empirical at large scales and often do not incorporate hydrodynamic behaviors in flood forecasting, at least in developing regions with a lack of information. This study presents an open-source system integrating a hydrodynamic model with satellite rainfall data (PERSIANN PDIR-Now) and weather prediction data (GFS). It functions as a near real-time Digital Twin (DT) and Early Warning System for high-resolution flood forecasting. Simulated data can be compared with gauge stations in real-time through the model monitoring interface. A proof-of-concept was made by assessing the model capabilities in two case studies. First, the system simulated two consecutive extreme events (hurricanes ETA and IOTA) over the Sula Valley, Honduras, showing fidelity in streamflow responses. Second, the system worked as a DT and EWS to monitor the current and future hydrological states for two periods in 2022 and 2023. Results indicate that satellite data coupled with DT can provide up-to-date system conditions for flood forecasts for regions of lack of data for extreme rainfall events. This tool offered insights to enhance civil protection and societal engagement through warning dissemination against extreme events to build resilience to cope with the increasing magnitude and frequency of disasters in regions with data scarcity.

Turbidity assessment in coastal regions combining machine learning, numerical modeling, and remote sensing

ABSTRACT Machine learning models for water quality prediction often face challenges due to insufficient data and uneven spatial-temporal distributions. To address these issues, we introduce a framework combining machine learning, numerical modeling, and remote sensing imagery to predict coastal water turbidity, a key water quality proxy. This approach was tested in the Great Lakes region, specifically Cleveland Harbor, Lake Erie. We trained models using observed and synthetic data from 3D numerical models and tested them against in situ and remote sensing data from PlanetLabs' Dove satellites. High-resolution (HR) data improved prediction accuracy, with RMSE values of 0.154 and 0.146 log10(FNU) and R2 values of 0.92 and 0.93 for validation and test datasets, respectively. Our study highlights the importance of unified turbidity measures for data comparability. The machine learning model demonstrated skill in predicting turbidity through transfer learning, indicating applicability in diverse, data-scarce regions. This approach can enhance decision support systems for coastal environments by providing accurate, timely predictions of water quality variables. Our methodology offers robust strategies for turbidity and water quality monitoring and holds significant potential for improving input data quality for numerical models and developing predictive models from remote sensing data.

Evaluating input data sources for isotope‐enabled rainfall‐runoff models

AbstractIsotope‐enabled models provide a means to generate robust hydrological simulations. However, daily isotope‐enabled rainfall‐runoff models applied to larger spatial scales (>100 km2) require more input data than conventional non‐isotope models in the form of precipitation isotope time series, which are difficult to generate even with point station measurements. Spatially distributed isotope data can be circumvented by isotope‐enabled climate models. Here, we evaluate the hydrological simulations of the J2000‐isotope enabled hydrological model driven with data from corrected and un‐corrected isotope‐enabled global and regional climate models (isotope‐enabled global spectral model [IsoGSM] and isotope‐enabled regional spectral model [IsoRSM], respectively) compared with 1 year of measured reference station and a yearly average precipitation isotope input for a pilot site, the data‐scarce sub‐humid Eerste River catchment in South Africa. The models driven by all input products performed well for upstream and downstream discharge gauges with Nash Sutcliffe efficiency (NSE) from 0.58 to 0.85 and LogNSE of 0.66 to 0.93. The simulated δ2H stream isotopes using the reference J2000‐iso and J2000‐isoRSM were good for the main river with a stream Kling Gupta efficiency (KGE) of between 0.4–0.9 and the top 100 Monte Carlo simulations varying by around 5‰ for δ2H. For smaller tributaries the model was unable to capture the measured stream isotopes due to biased precipitation isotope inputs. Adjusting the J2000‐iso with a bias corrected IsoRSM improved the stream and groundwater isotope simulation and outperformed the model driven by an average yearly precipitation isotope input. Differences in simulated hydrological processes were only evident between the models when evaluating percolation with unrealistic simulations for the standard J2000 model. While the regional climate model is computationally more intensive than its global counterpart, it provided better stream isotope simulations and improvements to simulated percolation. Our results indicate that isotope‐enabled climate models can provide useful input data in data scarce regions for hydrological models, where improved water management to address climate change impacts is needed.

High-resolution population maps derived from Sentinel-1 and Sentinel-2

Detailed population maps play an important role in diverse fields ranging from humanitarian action to urban planning. Generating such maps in a timely and scalable manner presents a challenge, especially in data-scarce regions. To address it we have developed Popcorn, a population mapping method whose only inputs are free, globally available satellite images from Sentinel-1 and Sentinel-2; and a small number of aggregate population counts over coarse census districts for calibration. Despite the minimal data requirements our approach surpasses the mapping accuracy of existing schemes, including several that rely on building footprints derived from high-resolution imagery. E.g., we were able to produce population maps for Rwanda with 100m GSD based on less than 400 regional census counts. In Kigali, those maps reach an R2 score of 66% w.r.t. a ground truth reference map, with an average error of only ±10 inhabitants/ha. Conveniently, Popcorn retrieves explicit maps of built-up areas and local building occupancy rates, making the mapping process interpretable and offering additional insights, for instance about the distribution of built-up, but unpopulated areas, e.g., industrial warehouses. With our work we aim to democratize access to up-to-date and high-resolution population maps, recognizing that some regions faced with particularly strong population dynamics may lack the resources for costly micro-census campaigns. Project page: https://popcorn-population.github.io/.

Stream salinity prediction in data-scarce regions: Application of transfer learning and uncertainty quantification

Scarcity of stream salinity data poses a challenge to understanding salinity dynamics and its implications for water supply management in water-scarce salt-prone regions around the world. This paper introduces a framework for generating continuous daily stream salinity estimates using instance-based transfer learning (TL) and assessing the reliability of the synthetic salinity data through uncertainty quantification via prediction intervals (PIs). The framework was developed using two temporally distinct specific conductance (SC) datasets from the Upper Red River Basin (URRB) located in southwestern Oklahoma and Texas Panhandle, United States. The instance-based TL approach was implemented by calibrating Feedforward Neural Networks (FFNNs) on a source SC dataset of around 1200 instantaneous grab samples collected by United States Geological Survey (USGS) from 1959 to 1993. The trained FFNNs were subsequently tested on a target dataset (1998-present) of 220 instantaneous grab samples collected by the Oklahoma Water Resources Board (OWRB). The framework's generalizability was assessed in the data-rich Bird Creek watershed in Oklahoma by manipulating continuous SC data to simulate data-scarce conditions for training the models and using the complete Bird Creek dataset for model evaluation. The Lower Upper Bound Estimation (LUBE) method was used with FFNNs to estimate PIs for uncertainty quantification. Autoregressive SC prediction methods via FFNN were found to be reliable with Nash Sutcliffe Efficiency (NSE) values of 0.65 and 0.45 on in-sample and out-of-sample test data, respectively. The same modeling scenario resulted in an NSE of 0.54 for the Bird Creek data using a similar missing data ratio, whereas a higher ratio of observed data increased the accuracy (NSE = 0.84). The relatively narrow estimated PIs for the North Fork Red River in the URRB indicated satisfactory stream salinity predictions, showing an average width equivalent to 25 % of the observed range and a confidence level of 70 %.