Shaping the Deep Tech Revolution in Agriculture 2025

Integration of satellite-enabled remote sensing: Before and after FIGURE 8 Drivers and barriers to the use of satellite-enabled remote sensing in agriculture FIGURE 9 Source: Consultations with AI4AI community experts AFTER Agribusinesses including input suppliers and food companies rely on fragmented, infr equent and localized farm surveys for input demand and pr oduction estimates. This limits their capacity to adjust pr ocur ement, storage or transport plans in real time. Similarly , insurance firms struggle to verify losses quickly after weather events. Financial institutions use generic risk pr ofiles, unable to accurately monitor cr op performance on small dispersed plots. Overall, business decisions throughout the value chain ar e based on outdated or incomplete data, and data collection is generally expensive.High-r esolution satellite imagery pr ovides near r eal-time, consistent and large-scale monitoring of cr op acr eage and health, along with water use and weather impacts. Agribusinesses can ther efore adjust pr ocur ement and logistics weeks earlier , with impr oved yield for ecasts. Insur ers use satellite data for automated claim verification, r educing fraud and payout times. Banks combine r emote sensing with AI to cr eate field-level cr edit-risk models and design innovative loan pr oducts. The entir e value chain gains transpar ency , agility and r esilience thr ough mor e precise, frequent and af fordable farm data.BEFORE DRI VERS Enhanced spectral and spatial capabilities: Satellites now of fer finer gr ound sampling and richer spectral bands, enabling better detection of cr op stress, soil pr operties and moistur e anomalies at the field level, driving their applicability . Increasing data frequency and revisit rates: Several satellites or their constellations can pr ovide daily or even hourly data. This has impr oved their usability for near r eal-time action in agricultur e.Applicability in fragmented land parcels: In emerging economies with smallholder agricultur e, land par cels may often be small and fragmented growing a mix of cr ops. Getting quality data may be challenging when land par cels ar e small. Limited ground truthing data: Training r emote sensing models r equir es the validation of data captur ed thr ough satellites with actual on-the- ground data (r eferr ed to as gr ound truthing data). Such data is siloed, with limited centralization. This cr eates duplication and an incr eased cost of gr ound truthing and hampers accuracy of use cases. Presence of cover crops, trees and cloud cover: Agricultural landscapes may have cover cr ops, which can pr event the collection of accurate data. Similarly , the transition to agr o ecological models with intercropping limits efficacy . Additionally , most satellites used for agricultur e rely on optical sensing (and not synthetic apertur e radar systems), which ar e unable to collect data in settings of cloud cover . BARRIER S Democratization and affordability of satellite data: The pr oliferation of both fr ee (Sentinel, Landsat) and competitively priced commer cial imagery has dramatically lower ed data costs, making high- frequency monitoring economically viable for farms of all sizes. Furthermor e, services such as Google Earth Engine, A WS Open Data and Earth Observation as a service (EOaaS) application pr ogramming interfaces (APIs) of fer ready-to-use pr ocessing pipelines, reducing the overhead costs for pr eprocessing and use case development. Shaping the Deep-Tech Revolution in Agriculture 17