TL;DR
This study evaluates the AnnAGNPS model's ability to predict runoff and nutrient (nitrogen and phosphorus) export in a small watershed near China's Taihu Lake. The model demonstrated satisfactory performance for runoff and total nitrogen (TN) prediction, but showed weaker results for total phosphorus (TP), highlighting the need for further refinement in nutrient modeling, especially for TP.
What Problem Does This Paper Address?
Non-point source pollution, originating from diffuse sources like agricultural runoff, is a major contributor to water quality degradation, leading to issues like algal blooms and oxygen depletion. Understanding and predicting these pollutant loads is crucial for effective watershed management. While various hydrological and water quality models exist, their accuracy and applicability can vary significantly depending on the watershed's characteristics (e.g., size, land use, climate). This paper specifically addresses the need to evaluate the AnnAGNPS model's performance in a typical small, hilly watershed in the Taihu Lake region of China, a region facing significant water quality challenges. The study aims to validate the model's capability for both hydrological (runoff) and nutrient (TN and TP) predictions, and to identify the most influential parameters for nutrient modeling in this specific context.
Methods
The researchers applied the AnnAGNPS (Annualized Agricultural Non-Point Source Pollution Model) version 5.1. The study involved several key steps:
- Watershed Delineation: The Wucun watershed (1.81 km²) was delineated using a Digital Elevation Model (DEM) and GIS tools, dividing it into 78 cells and 36 reaches.
- Data Acquisition: Various datasets were collected, including:
- Climate Data: Nine years (2005-2013) of temperature, precipitation, dew point, and solar radiation data.
- Topographic Data: A 1:50,000 relief map digitized into a 30m DEM.
- Soil Data: Properties of soil layers were obtained or calculated using the SPAW model.
- Land Use Data: A 2009 aerial image was used to classify land use into six types.
- Agricultural Practices: Schedules of cultivation and fertilization were gathered through farmer interviews.
- Hydrologic and Nutrient Data: Runoff data was estimated using an empirical formula for annual calibration/validation, and monthly runoff and nutrient (TN, TP) concentrations were measured at three monitoring sites.
- Model Calibration and Validation:
- Runoff: The model was calibrated for the period 2005-2009 and validated for 2010-2013. Calibration involved adjusting Curve Number (CN) values for different land uses.
- Nutrients: A parameter sensitivity analysis was conducted first to identify key parameters influencing TN and TP outputs. Calibration and validation for nutrients were then performed using the identified sensitive parameters.
- Parameter Sensitivity Analysis: Differential Sensitivity Analysis (DSA) was employed to assess the impact of various input parameters (e.g., fertilizer rates, canopy cover) on TN and TP outputs.
- Model Performance Evaluation: Statistical measures like Coefficient of Determination (R²), Coefficient of Efficiency (E), Modified Coefficient of Efficiency (E'), Root Mean Square Error (RMSE), and Coefficient of Residual Mass (CRM) were used to assess the "goodness-of-fit" between simulated and observed data.
Datasets
- Climate Data: Nine years (January 2005 - December 2013) of meteorological data from the Liyang national climate station and Nanjing Institute of Geography and Limnology.
- Topographic Data: A 1:50,000 relief map digitized into a 30m x 30m Digital Elevation Model (DEM).
- Soil Data: Properties of soil layers for the Wucun watershed, obtained from the Soil Survey Office in Jiangsu Province and supplemented by the SPAW model.
- Land Use Data: A 2009 aerial image (0.5m x 0.5m resolution) of the Wucun watershed.
- Agricultural Practices Data: Schedules of annual cultivation and agricultural practices for rice and rape, obtained from local farmers.
- Hydrologic and Nutrient Data:
- Estimated annual runoff data (2005-2009 for calibration, 2010-2013 for validation).
- Monthly runoff and nutrient (TN, TP) concentration data collected from December 2012 to December 2013 at three monitoring sites.
Key Insights and Ideas
- Runoff is Predictable: The AnnAGNPS model demonstrated a strong ability to simulate runoff at both annual and monthly scales in this watershed, suggesting that the model's hydrological component is robust for this region.
- Nutrient Sensitivity Varies: The study identified distinct sets of sensitive parameters for TN and TP. For TN, fertilizer-related parameters and canopy cover were most influential. For TP, residue mass ratio, fertilizer rate, and canopy cover were key. This highlights that different factors drive nitrogen and phosphorus losses.
- TN Prediction is Promising: The model performed well in predicting total nitrogen (TN) loading during both calibration and validation, indicating its potential as a tool for managing nitrogen pollution.
- TP Prediction Needs Improvement: The model's performance for total phosphorus (TP) loading was less satisfactory. This suggests that the model's current representation of phosphorus cycling, or the input data related to it, may not fully capture the complexities of the watershed.
- Parameterization is Crucial: The sensitivity analysis underscores the importance of accurately defining parameters related to fertilizer application, crop cover, and residue management for accurate nutrient predictions.
Key Results
- Runoff Calibration/Validation:
- Annual scale: R² = 0.93, E = 0.81, E' = 0.65.
- Monthly scale: R² = 0.86, E = 0.59, E' = 0.39.
- Validation results were also strong, with annual R² = 0.88 and monthly R² = 0.84.
- Nutrient Calibration:
- TN loading: R² = 0.91, E = 0.86, E' = 0.71.
- TP loading: R² = 0.66, E = 0.38, E' = 0.46.
- Nutrient Validation:
- TN loading at Site 2: R² = 0.92, E = 0.71, E' = 0.58.
- TP loading at Site 2: R² = 0.624, E = 0.18, E' = 0.37.
- Parameter Sensitivity:
- Highly sensitive parameters for TN: Fertilizer rate (0.94), Fertilizer organic (0.69), Fertilizer inorganic (0.25).
- Highly sensitive parameters for TP: Residue mass ratio (0.39), Fertilizer rate (0.30), Fertilizer inorganic (0.26).
Limitations and Future Work
- TP Prediction Weakness: The study explicitly notes that TP loading performance was "slightly poor" and "not satisfactory." This is attributed to potential issues with model mechanisms (mass conservation assumption) and a lack of reliable input data for phosphorus cycling (e.g., plant uptake).
- Data Limitations: The paper mentions that some nutrient cycling parameters were taken from unofficial sources, and that a lack of reliable information is common. This can significantly impact nutrient prediction accuracy.
- Site 3 Unsuitability: Site 3, located in an urban area, was deemed unsuitable for calibration or validation due to the strong influence of residential activities on TN and TP concentrations, leading to high observed values.
- Future Work Suggestions: The authors suggest that more detailed data monitoring is needed to obtain more realistic watershed simulations, particularly for nutrient prediction. Further refinement of the model's representation of phosphorus dynamics and exploring alternative modeling approaches for nutrient transport could also be beneficial.
Summary updated