PLPeijun Li

Peijun Li

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article

Previous Organizations / Institutes

The organizations / institutes from the scholar's other articles in the past ten years


Peijun Li has a total of 18 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Systematic Discrepancies Should Be Addressed Before Effective Parameter Learning: Insights From Soil Moisture Simulations Using Differentiable Process‐Based Models

    by Peijun Li, Chaopeng Shen, Jiangtao Liu, Farshid Rahmani, Ming Pan, Kathryn Lawson

    Differentiable hydrologic models combining neural networks with process‐based backbones—fully implemented or approximated by surrogate models—offer highly competitive accuracy and improved interpretability compared to either model type used alone. However, the impacts of surrogate model fidelity (accurate sensitivity of relationships compared to the original model) and systematic discrepancies between models and observations on parameter learning effectiveness remain underexplored. This study evaluates optimization performance of the differentiable HBV model ( δ HBV) and its surrogate variants in simulating soil moisture across the contiguous United States. Results show that high‐fidelity surrogate models can support effective differentiable parameter learning once systematic discrepancies are addressed. In contrast, unresolved systematic discrepancies degrade optimization performance, with particularly severe deterioration when low‐fidelity surrogates are used. These findings provide one possible explanation for the modest improvements reported in recent surrogate‐based differentiable parameter learning (dPL) applications. Linear correction partially mitigates these systematic discrepancies, improves optimization performance, and reduces compensatory parameter adjustment, although it does not resolve the underlying model–observation mismatch. Structural refinement remains a physically grounded long‐term direction, while pragmatic correction modules can provide useful partial mitigation when such refinement is not yet available. In addition, fully implementing differentiable process‐based models or using a rigorously trained surrogate can improve gradient fidelity and reduce amplified deterioration associated with surrogate infidelity.

  • Open Access

    Reply on RC2

    by Kamlesh Sawadekar, Seth McGinnis, Peijun Li, Kathryn Lawson, Chaopeng Shen

    Systematic biases in General Circulation Model (GCM) outputs limit their direct applicability in regional planning, making bias correction a technically demanding but necessary step for both short-term and long-term impact assessment. Correcting precipitation is particularly challenging due to its non-Gaussian distribution, intermittent nature, and heavy-tailed extremes. However, traditional statistical bias-correction methods have limited ability to learn systematic patterns from large datasets or generalize to new locations. While machine learning (ML) provides greater flexibility, it can produce unpredictable and difficult-to-interpret results, limiting generalization across GCMs and locations. In this study, we propose a differentiable bias-adjustment framework called δCLIMBA, or dCLIMBA, that learns a spatiotemporally adaptive parametric bias-adjustment procedure, rather than corrected precipitation directly, between historical CMIP6 model outputs and a gridded observation-based dataset, Livneh. Results demonstrate that the proposed method corrects the magnitude and distribution of extreme precipitation with particularly strong performance in the upper tail. The quantile distribution of precipitation was well reproduced across diverse U.S. cities, and spatial patterns were comparable to those from the widely used LOCA2 statistical downscaling product. In addition, the framework showed partial future trend preservation and promising attenuation of marginal biases in unseen regions. This work presents a modular and efficient bias-correction approach. The differentiable approach provides an easy-to-use option for connecting atmospheric-model outputs to on-the-ground impacts.

  • Open Access

    Reply on RC1

    by Kamlesh Sawadekar, Seth McGinnis, Peijun Li, Kathryn Lawson, Chaopeng Shen

    Systematic biases in General Circulation Model (GCM) outputs limit their direct applicability in regional planning, making bias correction a technically demanding but necessary step for both short-term and long-term impact assessment. Correcting precipitation is particularly challenging due to its non-Gaussian distribution, intermittent nature, and heavy-tailed extremes. However, traditional statistical bias-correction methods have limited ability to learn systematic patterns from large datasets or generalize to new locations. While machine learning (ML) provides greater flexibility, it can produce unpredictable and difficult-to-interpret results, limiting generalization across GCMs and locations. In this study, we propose a differentiable bias-adjustment framework called δCLIMBA, or dCLIMBA, that learns a spatiotemporally adaptive parametric bias-adjustment procedure, rather than corrected precipitation directly, between historical CMIP6 model outputs and a gridded observation-based dataset, Livneh. Results demonstrate that the proposed method corrects the magnitude and distribution of extreme precipitation with particularly strong performance in the upper tail. The quantile distribution of precipitation was well reproduced across diverse U.S. cities, and spatial patterns were comparable to those from the widely used LOCA2 statistical downscaling product. In addition, the framework showed partial future trend preservation and promising attenuation of marginal biases in unseen regions. This work presents a modular and efficient bias-correction approach. The differentiable approach provides an easy-to-use option for connecting atmospheric-model outputs to on-the-ground impacts.

  • Open Access

    Comment on egusphere-2026-2546

    by Kamlesh Sawadekar, Seth McGinnis, Peijun Li, Kathryn Lawson, Chaopeng Shen

    Systematic biases in General Circulation Model (GCM) outputs limit their direct applicability in regional planning, making bias correction a technically demanding but necessary step for both short-term and long-term impact assessment. Correcting precipitation is particularly challenging due to its non-Gaussian distribution, intermittent nature, and heavy-tailed extremes. However, traditional statistical bias-correction methods have limited ability to learn systematic patterns from large datasets or generalize to new locations. While machine learning (ML) provides greater flexibility, it can produce unpredictable and difficult-to-interpret results, limiting generalization across GCMs and locations. In this study, we propose a differentiable bias-adjustment framework called δCLIMBA, or dCLIMBA, that learns a spatiotemporally adaptive parametric bias-adjustment procedure, rather than corrected precipitation directly, between historical CMIP6 model outputs and a gridded observation-based dataset, Livneh. Results demonstrate that the proposed method corrects the magnitude and distribution of extreme precipitation with particularly strong performance in the upper tail. The quantile distribution of precipitation was well reproduced across diverse U.S. cities, and spatial patterns were comparable to those from the widely used LOCA2 statistical downscaling product. In addition, the framework showed partial future trend preservation and promising attenuation of marginal biases in unseen regions. This work presents a modular and efficient bias-correction approach. The differentiable approach provides an easy-to-use option for connecting atmospheric-model outputs to on-the-ground impacts.

  • Open Access

    Comment on egusphere-2026-2546

    by Kamlesh Sawadekar, Seth McGinnis, Peijun Li, Kathryn Lawson, Chaopeng Shen

    Systematic biases in General Circulation Model (GCM) outputs limit their direct applicability in regional planning, making bias correction a technically demanding but necessary step for both short-term and long-term impact assessment. Correcting precipitation is particularly challenging due to its non-Gaussian distribution, intermittent nature, and heavy-tailed extremes. However, traditional statistical bias-correction methods have limited ability to learn systematic patterns from large datasets or generalize to new locations. While machine learning (ML) provides greater flexibility, it can produce unpredictable and difficult-to-interpret results, limiting generalization across GCMs and locations. In this study, we propose a differentiable bias-adjustment framework called δCLIMBA, or dCLIMBA, that learns a spatiotemporally adaptive parametric bias-adjustment procedure, rather than corrected precipitation directly, between historical CMIP6 model outputs and a gridded observation-based dataset, Livneh. Results demonstrate that the proposed method corrects the magnitude and distribution of extreme precipitation with particularly strong performance in the upper tail. The quantile distribution of precipitation was well reproduced across diverse U.S. cities, and spatial patterns were comparable to those from the widely used LOCA2 statistical downscaling product. In addition, the framework showed partial future trend preservation and promising attenuation of marginal biases in unseen regions. This work presents a modular and efficient bias-correction approach. The differentiable approach provides an easy-to-use option for connecting atmospheric-model outputs to on-the-ground impacts.

  • Open Access

    Ensembling differentiable process-based and data-driven models with diverse meteorological forcing datasets to advance streamflow simulation

    by Peijun Li, Yalan Song, Ming Pan, Kathryn Lawson, Chaopeng Shen

    Streamflow simulations produced by different hydrological models exhibit distinct characteristics and can provide valuable information when ensembled. However, few studies have focused on ensembling simulations from models with significant structural differences and evaluating them under both temporal and spatial tests. Here we systematically evaluated and utilized the simulations from two highly different models with great performances: a purely data-driven long short-term memory (LSTM) network and a physics-informed machine learning (“differentiable”) HBV (Hydrologiska Byråns Vattenbalansavdelning) model (δHBV). To effectively display the features of the two models, multiple forcing datasets are employed. The results show that the simulations of LSTM and δHBV have distinct features and complement each other well, leading to better Nash-Sutcliffe model efficiency coefficients (NSE) and improved high-flow and low-flow metrics across all spatiotemporal tests, compared to within-class ensembles. Ensembling models trained on a single forcing outperformed a single model using fused forcings, challenging the paradigm of feeding all available data into a single data-driven model. Most notably, δHBV significantly enhanced spatial interpolation when incorporated into LSTM, and provided even more prominent benefits for spatial extrapolation where the LSTM-only ensembles degraded significantly, attesting to the value of the structural constraints in δHBV. These advances set new benchmark records on the well-known CAMELS (Catchment Attributes and Meteorology for Large-sample Studies) hydrological dataset, reaching median NSE values of ∼ 0.83 for the temporal test (densely trained scenario), ∼ 0.79 for the ungauged basin test (PUB, Prediction in Ungauged Basins), and ∼ 0.70 for the ungauged region test (PUR, Prediction in Ungauged Regions). This study advances our understanding of how various model types, each with distinct mechanisms, can be effectively leveraged alongside multi-source datasets across diverse scenarios.


  • Open Access

    Structural bias should be addressed before effective parameter learning - Insights from SMAP soil moisture simulations using differentiable process-based models

    by Peijun Li, Chaopeng Shen, Jiangtao Liu, Farshid Rahmani, Kathryn Elizabeth Lawson

    Differentiable hydrologic models combining neural networks with process-based backbones—fully implemented or approximated by surrogate models—offer highly competitive accuracy and improved interpretability. However, the impacts of surrogate model fidelity and backbone structural biases on parameter learning effectiveness remain underexplored. This study evaluates optimization performances of the differentiable HBV model (δHBV) and its surrogate variants in simulating soil moisture across the contiguous United States. Results show that surrogate models often degrade optimization performance, raising domain-averaged RMSE from 0.054 to 0.064, whereas systematic structural bias further worsens it to 0.098 and induces parameter over-compensation. Consequently, recent studies that neglected these issues have shown suboptimal results. To preserve physical meaning and robustness, surrogates must address these two issues, potentially via structural revisions or, suboptimally, post-simulation corrections. Fully implementing differentiable process-based models and improving structural representations remains the preferred strategy, while the surrogate path requires careful designs to mitigate structural bias.

  • Open Access

    Reply on RC2

    by Peijun Li, Yalan Song, Ming Pan, Kathryn Lawson, Chaopeng Shen

    Streamflow simulations via different hydrological models have different features and can provide valuable information after being ensembled. While few studies have focused on ensembling simulations via models with significant structural differences and evaluating them under both temporal and spatial tests. Here we systematically evaluated and utilized the simulations from two highly different models with great performances: a purely data-driven long short-term memory (LSTM) network and a physics-informed machine learning (“differentiable”) HBV (Hydrologiska Byråns Vattenavdelning) model (δHBV). To effectively display the features of the two models, multiple forcing datasets are employed and utilized in two ways. The results show that the simulations of LSTM and δHBV have distinct features and complement each other well, leading to better Nash-Sutcliffe model efficiency coefficients (NSE) and improved high-flow and low-flow metrics across all spatiotemporal tests, compared to within-class ensembles. Ensembling models trained on a single forcing outperformed a single model using fused forcings, challenging the paradigm of feeding all available data into a single data-driven model. Most notably, δHBV significantly enhanced spatial interpolation when incorporated into LSTM, and even more prominent benefits for spatial extrapolation where the LSTM-only ensembles degraded significantly, attesting to the value of the structural constraints in δHBV. These advances set new benchmark records on the well-known CAMELS (Catchment Attributes and Meteorology for Large-sample Studies) hydrological dataset, reaching median NSE values of ~0.83 for the temporal test (densely trained scenario), ~0.79 for the ungauged basin test (PUB, Prediction in Ungauged Basins), and ~0.70 for the ungauged region test (PUR, Prediction in Ungauged Regions). This study advances our understanding of how various model types, each with distinct mechanisms, can be effectively leveraged alongside multi-source datasets across diverse scenarios.

  • Open Access

    Reply on RC1

    by Peijun Li, Yalan Song, Ming Pan, Kathryn Lawson, Chaopeng Shen

    Streamflow simulations via different hydrological models have different features and can provide valuable information after being ensembled. While few studies have focused on ensembling simulations via models with significant structural differences and evaluating them under both temporal and spatial tests. Here we systematically evaluated and utilized the simulations from two highly different models with great performances: a purely data-driven long short-term memory (LSTM) network and a physics-informed machine learning (“differentiable”) HBV (Hydrologiska Byråns Vattenavdelning) model (δHBV). To effectively display the features of the two models, multiple forcing datasets are employed and utilized in two ways. The results show that the simulations of LSTM and δHBV have distinct features and complement each other well, leading to better Nash-Sutcliffe model efficiency coefficients (NSE) and improved high-flow and low-flow metrics across all spatiotemporal tests, compared to within-class ensembles. Ensembling models trained on a single forcing outperformed a single model using fused forcings, challenging the paradigm of feeding all available data into a single data-driven model. Most notably, δHBV significantly enhanced spatial interpolation when incorporated into LSTM, and even more prominent benefits for spatial extrapolation where the LSTM-only ensembles degraded significantly, attesting to the value of the structural constraints in δHBV. These advances set new benchmark records on the well-known CAMELS (Catchment Attributes and Meteorology for Large-sample Studies) hydrological dataset, reaching median NSE values of ~0.83 for the temporal test (densely trained scenario), ~0.79 for the ungauged basin test (PUB, Prediction in Ungauged Basins), and ~0.70 for the ungauged region test (PUR, Prediction in Ungauged Regions). This study advances our understanding of how various model types, each with distinct mechanisms, can be effectively leveraged alongside multi-source datasets across diverse scenarios.

  • Open Access

    Comment on egusphere-2025-483

    by Peijun Li, Yalan Song, Ming Pan, Kathryn Lawson, Chaopeng Shen

    Streamflow simulations via different hydrological models have different features and can provide valuable information after being ensembled. While few studies have focused on ensembling simulations via models with significant structural differences and evaluating them under both temporal and spatial tests. Here we systematically evaluated and utilized the simulations from two highly different models with great performances: a purely data-driven long short-term memory (LSTM) network and a physics-informed machine learning (“differentiable”) HBV (Hydrologiska Byråns Vattenavdelning) model (δHBV). To effectively display the features of the two models, multiple forcing datasets are employed and utilized in two ways. The results show that the simulations of LSTM and δHBV have distinct features and complement each other well, leading to better Nash-Sutcliffe model efficiency coefficients (NSE) and improved high-flow and low-flow metrics across all spatiotemporal tests, compared to within-class ensembles. Ensembling models trained on a single forcing outperformed a single model using fused forcings, challenging the paradigm of feeding all available data into a single data-driven model. Most notably, δHBV significantly enhanced spatial interpolation when incorporated into LSTM, and even more prominent benefits for spatial extrapolation where the LSTM-only ensembles degraded significantly, attesting to the value of the structural constraints in δHBV. These advances set new benchmark records on the well-known CAMELS (Catchment Attributes and Meteorology for Large-sample Studies) hydrological dataset, reaching median NSE values of ~0.83 for the temporal test (densely trained scenario), ~0.79 for the ungauged basin test (PUB, Prediction in Ungauged Basins), and ~0.70 for the ungauged region test (PUR, Prediction in Ungauged Regions). This study advances our understanding of how various model types, each with distinct mechanisms, can be effectively leveraged alongside multi-source datasets across diverse scenarios.