RFI_140G0325Q0183_1.pdf

Sources Sought 140G0325Q0183: AI Model Calibration Services

Department of the Interior, United States Geological Survey (USGS) Oregon Water Science Center (ORWSC)

Request for Information (RFI): AI Model Calibration Services All information submitted in response to this announcement is voluntary. The Government will not pay for information requested nor will it compensate any respondent for any cost incurred in developing or submitting information provided to the Government. Materials submitted to the Government for review will not be returned.

NAICS Code: 541715 – Research and Development in the Physical, Engineering, and Life Sciences (except Nanotechnology and Biotechnology) PSC Code: B529 – Special Studies Analysis Scientific Data

Introduction The Department of the Interior, United States Geological Survey (USGS), Oregon Water Science Center (ORWSC), is conducting market research to identify qualified vendors capable of providing AI model calibration and geospatial image classification services to support efforts in mapping large woody debris in the Willamette River Basin. This effort includes adapting deep learning image classification models to local riverine environments using government-provided training datasets.

This RFI is issued under the Federal Acquisition Regulation (FAR) for market research purposes only and does not constitute a solicitation for proposals or quotes Responses should be concise, relevant, and directly address the qualifications and capabilities outlined in this notice. Submissions should avoid general marketing materials unrelated to the specific scope of work described. Responses that fail to address the technical requirements or do not adhere to specified guidelines will not be considered.

Scope of Work Summary The contractor will:

- Adapt and calibrate a deep learning image classification model originally used in the Elwha River study.

- Train the model using aerial imagery of the Santiam, McKenzie, and Willamette Rivers.

- Identify an optimal training strategy and assist in validation.

- Document the methodology in a format suitable for peer-reviewed publication.

- Support interpretation of classification results and participate in review sessions.

The contractor must possess:

- Expertise in machine learning model calibration for geospatial imagery.

- Experience with transfer learning and image classification for riverine environments.

- A PhD in geosciences or a closely related discipline.

- A demonstrated record of peer-reviewed publications on related topics.

Requested Vendor Information (3 PAGE MAXIMUM) Interested vendors are requested to submit a capability statement that includes the following:

- Demonstrated experience calibrating AI models on geospatial riverine imagery.

- Expertise in deep learning and transfer learning approaches.

- Qualifications and experience of scientific personnel assigned to the project.

- Description of computational tools and infrastructure available to support training.

- Examples of prior peer-reviewed publications relevant to this scope.

- UEI number, SAM registration status

Please do not respond if your organization does not have proven expertise in AI model development for geospatial image analysis and river science applications.

Submission Instructions Responses must be submitted electronically by email to:

Joel Berberena Email: jberberena@usgs.gov Subject Line: "RFI Response – AI Model Calibration Services"

Disclaimer This announcement is not a request for quote or proposal, and the Government is not committed to awarding a purchase order or contract pursuant to this announcement. The information resulting from this market research will assist the Government in determining its acquisition strategy.

Performance-Based Statement of Work (PWS)

Willamette Wood Mapping

1. Introduction Trees and other large woody debris in rivers pose hazards to people and infrastructure, such as bridges and dams. To help identify and monitor these hazards, the Oregon Water Science Center (ORWSC), with funding from the U.S. Army Corps of Engineers, is developing maps and other datasets to monitor the current and historical distribution of large wood in the Willamette River Basin. These datasets will help inform river and infrastructure management within the basin and serve as a template for other watersheds.

Geospatial artificial intelligence tools which identify in-stream wood debris from aerial imagery are critical to this effort, as they facilitate efficient and accurate analysis.

ORWSC seeks contractor assistance in adapting artificial intelligence models developed in other watersheds to apply in the Willamette River Basin.

2. Background Large wood in rivers is a known hazard, particularly after wildfires. The 2020 wildfires in the Willamette Basin increased concern regarding wood-loading in rivers and reservoirs.

Aerial imagery can be used to monitor these changes, but manual mapping is inefficient.

ORWSC has identified a pre-existing deep learning model used in the Elwha River study to classify wood, sediment, and vegetation, and seeks assistance in calibrating this model to the Willamette Basin. The model identified for adaptation is documented in the journal article “Remote sensing large-wood storage downstream of reservoirs during and after dam removal: Elwha River, Washington, USA.” This model utilizes deep learning techniques to classify wood, sediment, and vegetation from aerial imagery.

3. Scope The purpose of this PWS is to acquire contractor support to adapt and calibrate existing artificial intelligence models for application in the Willamette River Basin, specifically the Santiam, McKenzie, and Willamette Rivers. ORWSC will provide the training datasets.

4. Performance Objectives Task Performance Standard Acceptable Quality Level (AQL)

Develop and calibrate image classification models using local datasets

Models are trained and calibrated with local data and meet ORWSC accuracy benchmarks

Model results achieve ≥90% classification accuracy

Identify optimal training strategy and assist in model validation

Propose strategy and validation approach consistent with best practices

Strategy approved by ORWSC technical team

Document methods, inputs, and results suitable for publication

Provide documentation of methods, inputs, and results in formats and styles suitable for academic publication

Documentation meets USGS and academic publication standards

Support interpretation of results and review of final outputs

Participate in review meetings and provide technical input

Review feedback incorporated within agreed timelines

5. Deliverables

- Calibrated image classification models for the Santiam, McKenzie, and Willamette Rivers.

- Documented methodology and findings in publication-ready format.

- Final report including results and implications for river management.

- Participation in review sessions and model evaluation with ORWSC.

6. Place and Period of Performance Work will be conducted using government-furnished virtual infrastructure. Regular meetings (weekly to bi-weekly) will be held between 9:00am – 4:00pm PDT. The period of performance is from September 1, 2025, to July 31, 2026, with an estimated workload of 10 hours per week.

7. Contractor Requirements

- Expertise in machine learning models for riverine environments

- Proficiency in deep learning and image classification

- Experience training river scientists on AI tools

- Peer-reviewed publications on fluvial AI applications

- PhD in geosciences or a related field