Research on the estimation method of crop net primary productivity based on improved CASA model

Area of Science:

• Ecology
• Remote Sensing
• Agricultural Science

Background:

• Net Primary Productivity (NPP) is crucial for ecosystem carbon balance and agricultural productivity assessments.
• Accurate NPP estimation in agriculture is challenging, particularly at large scales.
• Existing models often struggle with precise estimation due to limitations in key parameter calculations.

Purpose of the Study:

• To refine the estimation of Fraction of Photosynthetically Active Radiation (FPAR) within the CASA model for improved NPP accuracy.
• To develop a novel methodology for large-scale crop NPP monitoring using remote sensing data.
• To enhance the reliability of NPP assessments for agricultural decision-making and carbon cycle studies.

Main Methods:

• Utilized high-resolution Sentinel-2 satellite imagery.
• Employed Recursive Feature Elimination algorithm to identify FPAR-relevant vegetation indices.
• Estimated FPAR using a Convolutional Neural Network (CNN).
• Integrated the improved FPAR estimation into the Coupled Agriculture and Environment (CASA) model.

Main Results:

• Achieved a significant reduction in FPAR estimation Root Mean Square Error (RMSE) from 0.2040 to 0.0020.
• Demonstrated prediction errors for FPAR ranging from 0.0001 to 0.0092, with Mean Absolute Error (MAE) below 0.01.
• Reduced the Mean Absolute Percentage Error (MAPE) for NPP estimation from 28.92% to 20.31% compared to field data.
• Achieved MAPE values between 15% and 25% across test samples, indicating enhanced reliability.

Conclusions:

• The novel FPAR estimation methodology significantly improves NPP accuracy in the CASA model.
• The optimized model shows strong potential for large-scale crop NPP monitoring using remote sensing.
• This advancement provides robust support for agricultural management and carbon cycle research.