Addendum No. 1 - DEP RFI No. 2026002 Final.pdf

DEP Solicitation No. 2026002 Addendum No. 1 Page 1 of 7

ADDENDUM No. 1

FLORIDA DEPARTMENT OF ENVIRONMENTAL PROTECTION

DEP Procurement Section

3800 Commonwealth Boulevard, MS#93 Tallahassee, Florida 32399-3000

September 18, 2025

Addendum To: DEP RFI No. 2026002, entitled “Land Use Land Cover Automation”

The Department hereby answers questions posed by interested entities. This addendum does not need to be returned with the response. The Department hereby answers the following questions:

Question # Solicitation Section Solicitation Page # Question / Answer

1. - - I’m trying to find out where in Florida is the need for 82131601 Aerial photography services for this RFI?

Answer: See Section 2.04., paragraph 1.a.

2. - - What link do I download to provide the information (if the need is in South Florida)?

Answer: See Schedule of Events of the Department’s RFI.

3. - - Is the Department's current thinking to pursue a single, integrated solution for both imagery acquisition and data automation, or will these be separate procurements?

Understanding this will help us structure our response to address the most likely scenario.

Answer: The Department is open to either a single or multiple procurement whichever is in the best interest of the state.

4. - - Does DEP envision a future contract as a one-time statewide update, or a recurring service (e.g., an annual update cycle)?

Answer: The Department and districts are committed to continuing this data update. The Department encourages responses that explain the potential benefits and drawbacks of (1) complete updates at a defined interval; (2) recurring services annually; (3) cyclically-updating data (where regions or classifications are updated as updates become available).

5. - - The RFI states responses become public record. For any potential future solution involving proprietary algorithms or models, how would the Department envision handling licensing? Would the State seek to own the model outright, or license its use?

Addendum No. 1 Page 2 of 7

Question # Solicitation Section Solicitation Page # Question / Answer

Answer: The Department is interested in learning about the benefits and drawbacks of all approaches and does not have a set preference at this time between owning a model versus licensing for its use. However, also see answer to question #16.

See Section 1.06, Public Records which provides direction as to how confidential or trade secret information is to be identified if it is submitted as part of a response.

6. - - If a vendor provides enhanced data (e.g., impervious surface layers), what are the State's expectations regarding data ownership and the right to redistribute that data to other agencies and the public?

Answer: The Department’s Land Use Land Cover (LULC) dataset is designed to categorize data and the Department’s ultimate solicitation may include evaluation of impervious/pervious surfaces as a surrogate for land cover in certain classification codes. The Department will offer this valuable dataset to others who may also find it valuable. As such, the data outputs would be public record.

7. - - Beyond the technical requirements, what is the strategy for ensuring adoption and consistency across the Water Management Districts (WMDs) who have historically developed their own data? Will the future solution mandate a change in their processes?

Answer: The Department and WMDs coordinate on the development of the LULC dataset and collectively utilize the Florida land use, cover and forms classification system (FLUCCS). The WMDs participated in the development of this RFI and are similarly interested in automation information.

We acknowledge there are regional differences based on regional needs. The Department is interested in a statewide dataset that serves all needs, including regions where a classification code could be more narrowly defined by a district in need of that detail. For example, one district may elect to go to level 3 where other districts elect to go to level

2. Nevertheless, the Department would like to eliminate the potential for conflicting data classifications. Tools to assist the districts in obtaining more refined classifications, as well as tools to eliminate conflicting data classifications, would be of interest to the Department.

8. - - How does DEP plan to handle the transition from the current, manually interpreted basemap to a new, potentially AI-generated one? Is there an expectation of a one-time conversion or a gradual phased approach?

Addendum No. 1 Page 3 of 7

Question # Solicitation Section Solicitation Page # Question / Answer

Answer: The LULC dataset is not an interpreted basemap.

Following best practice in AI, a professional within the agency will review all material.

9. - - The RFI mentions 6-inch to 1-foot resolution. Beyond spatial resolution, what are the requirements for spectral resolution? Is 4-band (RGBI) imagery sufficient, or would there be significant value in 8-band or other multispectral data for improving classification accuracy of vegetation and water features?

Answer: The Department is interested in learning about the benefits and drawbacks of all approaches and does not have a set preference at this time. For example, responses could explain how different spectral resolutions would help the Department achieve its objectives.

10. - - The RFI specifies collection in Winter. Could the Department elaborate on the acceptable date range (e.g., Jan-Feb, leaf-off period) to ensure minimal tree canopy obstruction for certain analyses?

Answer: Current data collection occurs October through March with each county flown over a short time frame (typically less than five days). The Department welcomes input on what different timeframes may help achieve its objective. The Department recognizes that leaf-off periods can help identify building features in treed regions, but that should not be the only consideration. For example, early or mid-spring may be beneficial for agricultural classifications.

Further, the imagery/remote sensing method or technology may vary responses of acceptable date ranges.

11. - - What are the absolute accuracy requirements (e.g., RMSE) for the georeferenced imagery to ensure it aligns properly with other state base maps?

Answer: Pursuant to Florida Department of Transportation (FDOT), horizontal accuracy of 6-inch orthoimagery products shall be tested to meet 2.5 feet at the 95% confidence interval. The American Society for Photogrammetry and Remote Sensing (ASPRS) standards require direct comparison to checkpoints from an independent source of at least three times greater accuracy.

The minimum number of well distributed check points should meet the ASPRS recommendations for 6-inch orthoimagery products. Please refer to the FDOT Surveying and Mapping Documents and Publications, pages 5-7.

Also, see answer to Question #9.

12. - - The RFI mentions using other datasets for QC. Could the Department provide a definitive list of all 'authoritative https://www.fdot.gov/Geospatial/doc-pubs.shtm https://www.fdot.gov/Geospatial/doc-pubs.shtm

Addendum No. 1 Page 4 of 7

Question # Solicitation Section Solicitation Page # Question / Answer datasets' (e.g., FDOT roads, FHD water, FDACS FSAID, Critical Assets points) that a solution would be required to integrate with or cross-reference?

Answer: The LULC dataset is robust, identifying hundreds of different land features. The Department recognizes that there are dozens of authoritative datasets that may be utilized to evaluate the accuracy of the LULC dataset and invites responses on tools that may it do so. Example datasets would include FDOT roads, FHD/NHD waters, and Cadastral, as well as internal DEP data, such as Nation Pollutant Discharge Elimination Systems (NPDES) Wastewater locations or mines, currently spatially identified by latitude/longitude.

Key considerations that may help in response:

• How can the Department overcome the challenge of an authoritative dataset made of lines and points being used to quality check a dataset (such as the LULC) based completely on polygons (including when there may be no direct intersect)? E.g., for wastewater facilities, the LULC wastewater polygon may not intersect with the wastewater data’s latitude/longitude as the latter may be at the street of the plant.

• The Department recognizes that unique rules may need to support each dataset, such as buffers, filtering out data, and matching certain attributes.

What are the most effective tools the Department can implement dataset-specific rules and avoid manually inputting these rules for each run?

• If the Department has previously checked a point/polygon and found it to be correct in the LULC dataset, how can we avoid repeating the same check each run?

13. - - What are the versions and timelines of these supporting datasets? Should an automated system be designed to align with the most recent LULC cycle or a future cycle?

Answer: Supporting datasets would have various refresh cycles—some live data, such as internal permitting datasets; some annual; and some less frequent. Ideally the Department would look to run its LULC Quality Assurance check prior to publishing a new statewide LULC, making the refresh cycle of the LULC data the controlling timeline. The Department is looking to maintain at a minimum a 3-year update cycle, though is interested in whether automation may make more regular (such as annual) updates possible.

Also, see answer to question #4.

Addendum No. 1 Page 5 of 7

Question # Solicitation Section Solicitation Page # Question / Answer

14. 2.04 - Section 2.04 mentions 'performance measures.' What are the Department's target key performance indicators (KPIs) for accuracy? For example, are there minimum targets for Overall Accuracy, F1 Scores, or IoU (Intersection over Union) for specific critical classes (e.g., wetlands, wastewater facilities)?

Answer: One goal for this RFI is to understand what those targets should be for a solicitation. Ultimately, the Department would want to see an improvement upon the data quality from the current dataset developed primarily with photo interpretation. We recognize a suite of tools may be necessary to achieve that improvement.

15. - - How will 'ground truth' data be established for training and validating an AI model? Will the Department provide a subset of the existing manually interpreted data for this purpose?

Answer: The Department is interested in learning about the benefits and drawbacks of all approaches and does not have a set preference at this time. The existing statewide LULC is available for this purpose, as well as other authoritative datasets, such as Cadastral, and as discussed in the answer to question #12. Other means of ground-truthing, such as site visits, may also be possible if necessary and selected as the preferred approach.

The Department and each WMD utilizes a Photo Interpretation Key that would be made available as needed in any future solicitation should that information be beneficial. Most are available on the open data portal for each agency.

16. - - To ensure data security, would the Department require a fully on-premise deployment of the AI training and inference tools, or is a FedRAMP-approved cloud environment acceptable?

Answer: The inputs for the LULC area are largely publicly available data. While the Department is open to all solutions, we are cautious on taking on full deployment and future support of an AI algorithm and related infrastructure needed to update subsequent updates internally.

17. - - Does the State have a preference for the type of AI model?

For instance, is the focus on supervised learning (using existing LULC for training), unsupervised learning (for change detection), or a hybrid approach?

Answer: The Department is interested in learning about the benefits and drawbacks of all approaches and does not have a set preference at this time.

Addendum No. 1 Page 6 of 7

Question # Solicitation Section Solicitation Page # Question / Answer

18. - - The RFI discusses a 'semi-automated' process. What is the envisioned division of labor? Should the system simply 'flag' features for human review, or should it provide a confidence score and a suggested classification for a human to verify/edit?

Answer: See answer to question #17.

Notwithstanding, the Department prefers approaches that reduce time for staff review. Notwithstanding, DEP or WMD staff will be responsible for making final decisions and judgements about AI generated information.

19. - - What is the preferred output format? Are there specific schema requirements for the GIS attributes that must be maintained from the old system to the new?

Answer: The Department requires the continued use of the FLUCCS, but is willing to discuss adaptations to it based on the issues described in the RFI.

20. - - The Balmoral Report notes significant differences in feature sizes and classification levels between districts. For the future state, is the goal to enforce a single, statewide MMU and classification schema, even if it means re-digitizing some existing areas to a new standard?

Answer: See answer to question #7.

Further, the Department welcomes information on tools that assist in delineating features consistently statewide.

21. - - The report suggests using DEP's Critical Assets point data to improve LULC consistency. Should a proposed solution include a specific workflow to snap to, or create polygons from, these point features to ensure critical facilities are accurately represented in the LULC?

Answer: The LULC data will remain polygons only. The critical asset data includes records submitted by local governments as well as authoritative data sets at various temporal reference. The Department’s preference would be to utilize the authoritative data directly. Notwithstanding where the two datasets can be used to quality check one another, that is a service the Department would be interested in.

22. - - The report highlights the discrepancy between LULC and FDACS's FSAID data, especially for citrus. Should a future solution prioritize aligning the agricultural LULC classifications directly with the FSAID data to resolve this inconsistency?

Answer: The Department’s objectives with the LULC dataset are not identical to the objectives of FDACS’ FSAID dataset.

Addendum No. 1 Page 7 of 7

Question # Solicitation Section Solicitation Page # Question / Answer

Differences can include crop dormancy, crop rotations and delineation of features. That said, where FDACS’ FSAID may be valuable in meeting the Department’s objectives, we would be interested in tools that would allow us to utilize it in FLUCCS classifications. It is not anticipated that the two datasets, which serve different purposes, would be identical.

23. - - The report details the challenge of overlapping features like transmission lines. The RFI asks for suggestions on data structure. Would the Department be open to a solution that uses a separate linear feature class for things like transmission lines, while maintaining the polygon-based system for areal features, to better represent reality?

Answer: Yes, the Department is interested in those types of ideas.

In accordance with Section 1.04 – Addenda to the RFI, the Department hereby formally amends the RFI. The amendments are indicated in blue with strikethrough for deletions and underlines for additions.

The following modifications have been to the Schedule of Events.

– Remainder of Page Intentionally Left Blank –

DATES EVENTS METHOD

August 15, 2025 RFI Advertised Vendor Information Portal https://vendor.myfloridamarketplace.com/

August 29, 2025, 5 pm EST

Questions Submitted in Writing

Submit via email:

Attn: Chris Lowery, Procurement Officer E-mail: Christina.Lowery@FloridaDEP.gov

RFI NUMBER MUST BE IN SUBJECT LINE OF EMAIL

On or about, September 12, 2025

Answers to Questions Posted

Vendor Information Portal https://vendor.myfloridamarketplace.com/

September 29 26, 2025, 5 pm EST

Response Due

Submit via email:

Attn: Chris Lowery, Procurement Officer E-mail: Christina.Lowery@FloridaDEP.gov

RFI NUMBER MUST BE IN SUBJECT LINE OF

EMAIL

https://vendor.myfloridamarketplace.com/ mailto:Christina.Lowery@FloridaDEP.gov https://vendor.myfloridamarketplace.com/ mailto:Christina.Lowery@FloridaDEP.gov