AgriGEE.lite is a high-performance Google Earth Engine (GEE) wrapper designed to simplify and accelerate the download of Analysis Ready Multimodal Data (ARD) for agricultural and vegetation monitoring. The library focuses on making satellite data as accessible as reading a CSV file with pandas, removing the complexity typically associated with Earth Engine programming.

• Simplified API: Download satellite time series with just a few lines of code
• High Performance: Utilizes aria2 downloader under the hood, achieving 16-22 time series downloads per second
• Multimodal Support: Seamlessly integrates optical satellites, radar sensors, and derived products
• Vegetation/Agricultural Focus: Optimized for crop monitoring, vegetation analysis, and land use applications
• GeoPandas Integration: Built to work natively with spatial geodataframes

To download and view a cloud-free Sentinel-2 time series for a specific field and date range:

import agrigee_lite as agl
import geopandas as gpd
import ee

ee.Initialize()

# Load your area of interest
gdf = gpd.read_parquet("data/sample.parquet")
geometry = gdf.iloc[0].geometry

# Define satellite and download time series
satellite = agl.sat.Sentinel2(bands=["red", "green", "blue"])
time_series = agl.get.sits(geometry, "2022-10-01", "2023-10-01", satellite)

This example demonstrates the library's core philosophy: spatial data analysis should be simple and fast. The entire library is designed to work seamlessly with GeoPandas, making it essential to have basic knowledge of this framework.

You can also download temporal aggregations, such as spatial median aggregations of vegetation indices from multiple satellites:

For synchronized multi-satellite analysis, you can use the TwoSatelliteFusion class:

# Combine Landsat 8 and Sentinel-2 data for synchronized analysis
landsat8 = agl.sat.Landsat8(bands=["red", "green", "blue", "nir"])
sentinel2 = agl.sat.Sentinel2(bands=["red", "green", "blue", "nir"])
fusion_sat = agl.sat.TwoSatelliteFusion(landsat8, sentinel2)

# Download synchronized time series with both satellites' data
fused_time_series = agl.get.sits(geometry, "2022-01-01", "2022-12-31", fusion_sat)

For more comprehensive examples, see the examples folder.

One of AgriGEE.lite's key features is its use of aria2, a lightweight multi-protocol & multi-source command-line download utility. This integration provides:

• Parallel Downloads: Multiple simultaneous connections for faster data retrieval
• Resume Capability: Automatic resumption of interrupted downloads
• Optimized Performance: Achieving 16-22 time series per second (for 1-year cloud-free Sentinel-2 BOA series)
• Reliability: Robust error handling and retry mechanisms

The aria2 integration runs transparently behind the scenes, requiring no additional configuration from users while dramatically improving download speeds compared to traditional sequential downloading methods.

AgriGEE.lite is organized into three main modules:

• agl.sat = Data sources, usually coming from satellites/sensors. When defining a sensor, it is possible to choose which bands you want to view/download, or whether you want to use atmospheric corrections or not. By default, all bands are downloaded, and all atmospheric corrections and harmonizations are used.

• agl.vis = Module that allows you to view data, either through time series or images.

• agl.get = Module that allows you to download data on a large scale.

• *Landsat 7 images began to have artifacts caused by a sensor problem from 2003-05-31.
• **Pixel size/spatial resolution for active sensors (or models that use active sensors) often lacks a clear value, as it depends on the angle of incidence. Here, the GEE value itself is explained, representing the highest resolution captured.
• ***Analysis Ready Data (ARD) is an advanced post-processing method applied to a SAR. However, it is quite costly, and its usefulness must be evaluated on a case-by-case basis.
• ****Sentinel 1 was a twin satellite, one of which went out of service due to a malfunction. Therefore, the revisit time varies greatly depending on the desired geolocation.
• *****Two Satellite Fusion is a meta-satellite that combines data from exactly two optical satellites (e.g., Landsat 8 + Sentinel-2). It automatically finds common observation dates, harmonizes the datasets, and creates synchronized time series with bands from both satellites distinguished by prefixes.
• ******HLS (Harmonized Landsat Sentinel-2) provides atmospherically corrected surface reflectance data harmonized across Landsat-8/9 and Sentinel-2 missions. The 2-3 day revisit time is achieved by combining observations from both satellite constellations.

My journey with Google Earth Engine began two and a half years ago. While GEE is an incredibly powerful platform that provides access to petabytes of satellite data, it comes with significant complexity challenges:

• Steep Learning Curve: GEE requires extensive boilerplate code and deep understanding of its functional programming paradigms
• Cryptic Error Messages: Server-side execution often produces confusing errors that are difficult to debug
• Harmonization Complexity: Each satellite has different value ranges, cloud masking approaches, and preprocessing requirements
• Inconsistent APIs: Different sensors require different coding approaches, making it hard to switch between data sources

During my master's degree, I found myself constantly rewriting similar code patterns for different projects. This frustration led to a simple but ambitious goal: making satellite data as easy to use as reading a CSV with pandas, without requiring users to be remote sensing experts.

AgriGEE.lite aims to be a simple, direct, and high-performance solution for downloading satellite data, serving both academic research and commercial applications. The library is designed for:

• Data Scientists who need satellite data but don't want to become GEE experts
• Agricultural Researchers studying crop monitoring and vegetation dynamics
• Environmental Consultants requiring rapid access to earth observation data
• Students and Educators learning remote sensing applications
• Commercial Users developing scalable earth observation solutions

Found this project helpful? ⭐ Give it a star! Want to contribute? 🚀 Open a Pull Request - we welcome all contributions!

While STAC (SpatioTemporal Asset Catalog) is an excellent approach that makes sense for large-scale infrastructure projects, processing satellite data locally can quickly become overwhelming, especially for regions with vast agricultural areas like Australia, the United States, or Brazil.

Key considerations:

• Infrastructure Costs: Building and maintaining your own processing infrastructure vs. GEE usage costs
• Processing Complexity: GEE handles complex atmospheric corrections, cloud masking, and data harmonization automatically
• Scalability: GEE's planetary-scale computing capabilities vs. local processing limitations
• Free Tier: GEE is completely free for academic research and non-commercial projects

The choice depends on your specific use case, scale, and budget requirements.

Remote sensing experts often note that terms like "satellite" for radar data or "bands" for SAR aren't technically precise. However, we've chosen standardized terminology to:

• Simplify the API: Consistent naming across all data sources reduces cognitive load
• Improve Accessibility: Non-experts can work with different sensors using the same patterns
• Maintain Code Consistency: Unified interfaces make the library easier to maintain and extend

Even excellent projects like MapBiomas (which uses models and algorithms) are treated as "satellites" in our framework. We welcome technical accuracy improvements, but prioritize usability and consistency. Your contributions to improve both aspects are very welcome!

The AgriGEE.lite mascot was created using AI tools, with artwork inspired by the "Odd-Eyes Venom Dragon" from Yu-Gi-Oh. The symbolism represents:

• 🌱 Plant Dragon: Agricultural focus
• 🔀 Fusion Card: Multimodal data integration
• 👀 Odd-Eyes: Multiple satellite perspectives on Earth

If you're an artist interested in creating a new mascot design, we'd love to make it official!

• QuadTree clustering functions produce absurd results when there are very uneven geographic density distributions, for example, thousands of points in one country and a few dozen in another. Some prior geospatial partitioning is recommended.

• Optical Satellites: Sentinel-2, Landsat 5/7/8/9 support
• Radar Sensors: Sentinel-1 GRD, ALOS-2 PALSAR-2
• Derived Products: MapBiomas Brazil, MODIS Terra/Aqua
• Time Series: Satellite Image Time Series (SITS) with aggregations
• Downloads: Online and task-based download methods with aria2 integration
• Visualizations: matplotlib-based plotting for images and time series
• Cloud Recovery: smart_open[gcs] integration for automatic data recovery
• Advanced Processing: Configurable cloud masking, Landsat pan-sharpening

• Enhanced Visualizations: plotly-based interactive plotting
• Expanded Coverage: All MapBiomas collections (Amazon, Cerrado, etc.)
• Advanced Radar: Sentinel-1 ARD (Analysis Ready Data)
• Ocean Monitoring: Sentinel-3 OLCI/SLSTR sensors
• Historical Data: Landsat 1-4 for long-term analysis