Performing Department
(N/A)
Non Technical Summary
Satellite data availability is a key component in modern agriculture. However, current access to satellite data and tools for farmers is limited and unequal, and numerous adoption barriers, such as the high cost of processing software and informational gaps (e.g., limited knowledge of effective interpretation and utilization of satellite data) prevent this promising technology from being integrated into farmers' decision-making and operational processes. This project aims to prototype the first farm-scale agriculture monitoring system using a satellite constellation product that will result in a new web-based portal (CropVista) that tracks crop growth and meteorological conditions to support U.S. farmers across 17 states. This research and design will use complex data science concepts to automate satellite image processing and deliver readily interpretable crop condition insights, such as the evolution of the growing season and crop phenology metrics. The project team will: (1) implement a crop field boundary detection from enhanced 2.5-m Sentinel imagery using deep learning, (2) extract field-scale crop phenology time series from NASA harmonized Landsat-Sentinel-2 images, (3) validate the crop phenology and derived metrics in different crop systems at MSU farms, (4) derive data insights from crop phenology and meteorological variability, and (5) provide a new web-based portal with field-scale crop phenology and meteorological conditions for agricultural management. Increasing farmers' adoption of these resources can significantly strengthen U.S. agricultural resilience using satellite observations.
Animal Health Component
50%
Research Effort Categories
Basic
(N/A)
Applied
50%
Developmental
50%
Goals / Objectives
The major goal of this project is to prototype a new field-scale agriculture monitoring system using satellite constellation products comprising major crop production areas in the Midwest and Southern U.S. states (17 states).The objectives of this project are:Objective 1: Implement a crop field boundary detection from enhanced Sentinel imagery using deep learningObjective 2:Extract field-scale crop phenology time series from NASA harmonized Landsat-Sentinel-2 (HLS) imagesObjective 3:Validate the crop phenology and derived metrics in crop systems at MSU farmsObjective4: Derive data insights from crop phenology and meteorological variabilityObjective5: Provide a new web-based portal with field-scale crop phenology and meteorological conditions for agricultural management
Project Methods
Objective 1: Implement a crop field boundary detection from enhanced Sentinel imagery using deep learningDeep learning super-resolution algorithm (Enhanced Super-Resolution Generative Adversarial Network - ESRGAN) and segmentation model (Segment Anything Model - SAM) will be used to derive crop field boundaries from Sentinel-2 annual temporal composites, downscaled from 10 to 2.5m, across 17 states. Each field boundarywill be used as a spatial investigation unit for crop phenology analysis. The procedures are described as follows: (1) All 10m Sentinel-2 images in 2023 will be stacked, and temporal composites with median value of all spectral bands will be generated; (2) a super-resolution deep learning algorithm, Enhanced Super-Resolution Generative Adversarial Network (ESRGAN), will generate enhanced 2.5m resolution median composite images; (3) Computer vision filters will be applied to highlight the crop field edges; and (4) field boundaries will be extracted with Segment Anything Model.Objective2: Extract field-scale crop phenology time series from NASA harmonized Landsat-Sentinel-2 (HLS) imagesA high volume of HLS images will be acquired and processedfrom 2020 to the present in 17 U.S. states. Wewill use the 30m HLS images to derive the field-scale crop phenology and its metrics via the Enhanced Vegetation Index (EVI) time series.The proceduresare as follows: (1) create a data cube of HLS images and use the quality assessment band (i.e., auxiliary raster file that comes with each image date) to mask and remove all "bad pixels" including clouds, cloud shadows, snow, water, high aerosol optical depth, (2) evaluate four filling gap techniques to derive a synthetic cloud-free time series and then calculate the EVI time series, and lastly (3) compute the per-field crop phenology and its metrics using the median value for each derived parameter within field boundary.Objective3: Validate the crop phenology and derived metrics in crop systems at MSU farmsWe will validate the HLS-derived crop phenology via EVI time series with in-situ radiometric measurements during the growing season. Surveys will be performed every summer period in at least six crop fields at MSU's North Farm, near the main campus, and basic information (crop type, phenological stage [emergence, flowering, maturity, senescence], plant population density in row, leaf area index, canopy cover per m2, and estimated yield near harvest) and radiometric data using Trios-RAMSES radiometer will be collected.?Objective4: Derive data insights from crop phenology and meteorological variabilityWe will use the generated EVI crop phenology and climate datasets to extract field-specific information, enabling us to derive actionable insights into individual farmers' needs.We will create a machine learning model, such as extreme gradient boosting and artificial neural network, to forecast future EVI values, and we will automatically identify when the vegetation index falls below its expected values, defined based on a 5-year average value.Objective5: Provide a new web-based portal with field-scale crop phenology and meteorological conditions for agricultural managementA new portal (CropVista) will be developed with customized tools for field-scale diagnostics, including current and previous phenological metrics and agrometeorological conditions. The portal will be developed in the ArcGIS Experience Builder service, a web-based application development platform ESRI created. This platform allowsus to create custom web applications with GIS data and mapping. We will prototype the desktop and mobile operational versions of CropVista.