Rain garden infiltration rate modeling using gradient boosting machine and deep learning techniques.

Area of Science:

• Environmental Engineering
• Hydrology
• Urban Planning

Background:

• Rain gardens effectively reduce stormwater runoff, but their design is limited by a lack of accurate infiltration modeling.
• Understanding infiltration rate differences between rain gardens and natural surfaces is crucial for optimizing performance.

Purpose of the Study:

• To evaluate the accuracy of soft computing tools (Gradient Boosting Machine and Deep Learning) in modeling rain garden infiltration rates.
• To compare the performance of these models against Philip's infiltration model, commonly used for natural surfaces.
• To assess the influence of vegetation type and density on infiltration rates.

Main Methods:

• Experimentally observed infiltration rates from rain gardens with diverse vegetation (grass, candytuft, marigold, daisy) and flow conditions were collected.
• Philip's infiltration model was applied for comparison.
• Gradient Boosting Machine (GBM) and Deep Learning (DL) models were developed and trained using the experimental data.

Main Results:

• GBM and DL models showed promising performance, with predictions largely within ±5% error margins.
• Philip's model exhibited significantly higher errors, with predictions often outside ±25% error lines.
• GBM outperformed DL, achieving higher correlation coefficients (CC) and Nash-Sutcliffe model efficiency (NSE), with the lowest root mean square error (RMSE).

Conclusions:

• Soft computing tools, particularly GBM, offer superior accuracy for modeling rain garden infiltration compared to traditional methods like Philip's model.
• The findings provide valuable insights for selecting optimal plant types and densities to enhance the design and efficiency of urban rain gardens.
• Accurate infiltration modeling is key to improving the effectiveness of green infrastructure in stormwater management.