Agriculture Financial Services Corporation - Request for Proposal - RFP 0709 Digital Field Borders Detection and Analysis

2.0 2.1 Introduction Through this RFP, AFSC is seeking to do an experimental pilot with a selected vendor for the creation (mapping) and the evaluation (fit, quality, accuracy) of field borders for crop insurance and risk management. Interested Vendors are requested to submit proposals, quotes, and provide a free test data sample for evaluation as outlined in this RFP. For the experimental pilot, AFSC has identified risk area (RA) pair (Crop RA # 13 and Forage RA # 26) where these RAs overlap (as specified in Appendix E), consistently has the highest number of fields across the years, has an excellent mix of annual and perennial fields (total insured acres ~1.91 M; ~11.23% annuals and ~8.53% perennials), and offers mapping challenges (irregular field edges, shrubs, trees). For the free test data sample required as part of the RFP evaluation, all interested bidders must perform field border mapping for all fields for the specified 3*3 township grid within the experimental pilot area. 2.2 Background Agriculture Financial Services Corporation (AFSC) is exploring integrating imagery-delineated (machine/deep learning) digital field borders to help advance AFSC’s digital-first ambitions. AFSC’s entire business and financial services is centered around land and crop production, and every year AFSC insures >25,000 clients across ~30M acres or ~205,000 crop fields. AFSC believes that digital field border polygons will be useful to readily locate fields, identify field edges or parts, estimate cropped and/or unseeded acres, identify if cropped or planted acres change across years, for regional or provincial reporting, and enable the application of remote sensing solutions on insured fields. AFSC also believes that digital field borders are foundational dataset for any crop reporting and crucial precursor for future adjusting augmentation and desk releases. AFSC hopes that integrating field borders could help advance AFSC’s digital-first ambitions towards a future-state driven by business optimization and business transformation goals for enhanced client experience and transparency, process improvements, operational efficiency, and data-driven actions to redefine industry norms and benefit a broad spectrum of stakeholders. AFSC believes that integrating the field-level geospatial component (“spatial context”) as digital field borders will be the starting point of such initiatives. 2.3 Objectives AFSC is looking for the vendor to provide a scalable automated (ideally) or semi-automated methodology or workflow to map field borders from imagery at very high accuracy using machine learning (ML) or deep learning (DL) approaches. The vendor may use any imagery (public, super-resolved public, commercial, or super-resolved commercial) or image processing/enhancement as they see fit. Note that AFSC’s expectation in the operational phase will be that field borders for a crop year will be available from the vendor in May or early June (seeding/planting), refreshed and confirmed in July (flowering), and finalized in late August or early September (maturity/harvest). AFSC will evaluate the accuracy (border/edges, planted/exclusions, splits/parts, mapped versus reported/verified acres) of mapped field borders using both geospatial analyses and field verification against the Alberta cadastral grid, GPS-mapped fields, and cloudless basemaps. Minimum accuracy of delineated field borders must be within the AFSC thresholds (within 5% and 20 acres). Ideally AFSC is looking for a field bordering/mapping solution that exceeds this threshold, and preference will be given to vendors that demonstrate this ability. The vendor may have the option (at AFSC’s discretion) to perform model re-training. The end goal of this experimental pilot will be the evaluation of integration of digital field borders into AFSC's Enterprise Architecture framework and Business Intelligence technology stack (AFSC Connect, Salesforce, Envoy, SAS Enterprise Server, ESRI ArcGIS, Microsoft Azure). AFSC recognizes that to successfully enhance our systems, to which this competitive bid is for, we require a strategic long-term business ‘partner’ to work in a visionary/collaborative way with our staff. 2.4 Duration The initial term of engagement is intended to be for one (1) year. The contract may also allow for multiple year extensions, and the successful Vendor may be AFSC’s preferred Vendor for these services until such time as AFSC’s business needs change. Expansion Capabilities – In the future, AFSC may wish to expand its services based on the changing needs of the organization. AFSC reserves the right to engage the successful Vendor in the future, for integrated expansion capabilities/modules for other Business areas of AFSC. This may or may not include an assessment/discovery phase in which we will be required to move into an implementation phase. 2.5 Requirements Proposals should be no longer than 40 pages. References may be additional appendices. Mandatory Requirements: All submissions require the inclusion of a 3*3 test data sample as outlined in 2.5.1. 2.5.1 Specifications 25% The vendor will supply the mapped digital field borders in these specifications: 1) Identify and map all agriculture fields in the experimental pilot area. 2) The test data sample for a 3*3 township grid (attached Appendix E) is to be submitted as part of the RFP evaluation process, and this sample will be free of cost to AFSC. Test data sample is required for one year (Crop year 2024) for July (derived from prior imagery, e.g. May and June, 2024 - see Requirement #11). Test data sample should be zipped and shared as download links. 3) The accuracy of mapped digital field borders should exceed the minimum requirements. 4) Digital field border vector polygons must be in one of the following ESRI compatible file formats: geodatabase, shapefile, or GeoJSON. Note that service-way (e.g. API) might be the preferred way of accessing field borders in the operational phase. The vendor should mention in their proposal all their capabilities or possibilities for sharing mapped field borders with AFSC. AFSC requires the field borders in the below custom coordinate system (“nad”): Projection: Transverse_Mercator False_Easting: 500000.0 False_Northing: 0.0 Central_Meridian: -115.0 Scale_Factor: 0.9992 Latitude_Of_Origin: 0.0 Linear Unit: Meter (1.0) Geographic Coordinate System: GCS_North_American_1983 Angular Unit: Degree (0.0174532925199433) Prime Meridian: Greenwich (0.0) Datum: D_North_American_1983 Spheroid: GRS_1980 Semimajor Axis: 6378137.0 Semiminor Axis: 6356752.314140356 Inverse Flattening: 298.257222101 5) Attributes of each/every field border polygon will include Legal Land Description (centroid of LLD that has majority of acres), mapped acres, accuracy/confidence, date/period of imagery used, date of mapping, and imagery source(s). 6) Digital field border file will have prefix fb_ in the file name and as an attribute. 7) Field borders will be single or multi-part (disconnected areas that are part of a single field) polygons and topologically correct. 8) Document that outlines: • high-level of methodology or workflow (include proposed accuracy, ML/DL algorithm used, approximate time taken to map the 3*3 township grid). • source imagery used and any image processing. • explanation and calculation and use of error metrics. • advantages or merits of the proposed methodology/workflow. • describe any limitations/challenges/improvements/opportunities. 9) The selected vendor to provide option to remap any agriculture field in the experimental test area for years 2024-2025 or the last 3 years. 10) For the contract – One year (crop year 2024 whole experimental pilot area and three years (crop year 2024, 2023, and 2022) for the select township grid. 11) It is critical that the field border mapping methodology or workflow developed for the experimental pilot is in-season. Later imagery must not be used to map field borders for an earlier time. For example, May 2023 field borders cannot be mapped from July/August 2023 imagery, or August 2023 field borders should be mapped from May to July 2023 imagery. Current research suggests that vendors use different mapping methodologies for delineating field borders from imagery, therefore a comparative evaluation of test data samples must be done to confirm their quality (applicability, accuracy) for AFSC’s crop insurance. One dependency is that the accuracy (e.g. edges, acres, exclusions) of field borders may not be sufficient for complex fields or within AFSC’s thresholds, and some model re-training or using finer imagery might be required. Field borders may also change across years and in-season, and therefore 2-3 refresh of field borders per cropping season is recommended. Note that it might be possible that the field border accuracy (border/edges, planted/exclusions, splits/parts, acres) may not be within AFSC’s thresholds or sufficient for complex fields, and if so, some model re-training or using finer imagery might be required. Field borders may also change across years and in-season (further justification for 2-3 refresh per cropping season). AFSC will: 1) Complete the RFP evaluation (based on evaluation and scoring of the proposal and test data sample), vendor selection, and engage into a purchasing agreement with a selected vendor. 2) Evaluate accuracy of field borders using geospatial analyses and field verification. 3) Complete evaluation of the applicability and suitability - consultation of findings with the AFSC areas against their requirements and use cases and suitability for ongoing remote sensing solution applications and geospatial and AI-ML endeavors. 4) Share findings with the vendor. At AFSC’s discretion; further model re-training with crop insurance training data if not within AFSC thresholds. If retraining is decided upon, complete a data sharing agreement and share insurance data to improve the model or methodology.