Scalable Regulatory Machine-Learning Framework for U.S. Water Resources
Patent Status
| Country | Type | Number | Dated | Case |
| Patent Cooperation Treaty | Published Application | WO 2025/015313 | 01/16/2025 | 2023-146 |
Additional Patent Pending
Brief Description
Navigating the complex regulatory landscape of the 1972 Clean Water Act often requires labor-intensive site visits and inconsistent jurisdictional determinations. To streamline this process, researchers at UC Berkeley have developed a machine learning framework that assesses which water resources fall under federal protection. By training a model on 150,000 historical jurisdictional determinations made by the Army Corps of Engineers and integrating high-resolution aerial imagery with geophysical data, the system can predict regulatory status with high precision. This technology quantifies the impact of shifting legal definitions, such as illustrating how a 2020 White House rule deregulates approximately 608,000 stream miles and 32 million wetland acres compared to previous Supreme Court standards.
Suggested uses
Regulatory Compliance: Providing developers and land managers with rapid, automated assessments of whether specific water bodies require federal permits. Policy Impact Analysis: Helping government agencies and environmental groups simulate the real-world effects of proposed changes to environmental rules. Conservation Planning: Identifying critical wetlands and streams that have lost federal protection to prioritize state-level or private conservation efforts. Infrastructure Design: Supporting large-scale project planning by mapping jurisdictional waters across entire counties or states using remote sensing data. Hydrological Mapping: Enhancing geophysical databases with high-confidence predictions of surface water connectivity and environmental significance.
Advantages
High-Throughput Analysis: Replaces manual, case-by-case determinations with a scalable system capable of evaluating thousands of sites simultaneously. Objective Consistency: Reduces the variability inherent in human jurisdictional determinations by utilizing standardized aerial and geophysical inputs. Dynamic Modeling: Allows for the immediate recalculation of federal jurisdiction across the entire country following Supreme Court rulings or executive rule changes. Data-Driven Transparency: Provides a clear, quantifiable template for how machine learning can be applied to complex regulatory implementation problems. Resource Optimization: Frees up personnel from the Army Corps of Engineers to focus on more complex field validations while the model handles routine screening.
Related Materials
Contact
- Michael Cohen
- mcohen@berkeley.edu
- tel: View Phone Number.
Inventors
- Shapiro, Joseph S.
