Earth Engine for cropland mapping Tyler Erickson, [email protected] Noel Gorelick Google Earth Engine Team
Mar 12, 2020
Earth Engine for cropland mappingTyler Erickson, [email protected]
Noel Gorelick
Google Earth Engine Team
Background
"To organize the world's information and make it universally accessible and useful."
https://www.google.com/about/company/
Google Mission Statement
-Jim Gray (1944-2007)
“Often it turns out to be more efficient to move the questions than to move the data.”
Reshaping Big Data Analysis Workflows
Create Hypothesis
Download Data
Code Algorithm
Process Data
Test Hypothesis
Setup Hardware
Download Data
Setup Hardware
Code Algorithm
Process Data
Test Hypothesis
Create Hypothesis
iterateiterate
Before Earth Engine
Three Major Vertical Efforts
● Forests● Water● Emerging Infectious Diseases
June 2015 - Vertical strategy meetings were conducted at Google HQ.
Supported projects in many other areas:Disaster risk management, biodiversity, urban growth monitoring, supply chain monitoring, agriculture and food security, air quality and pollution, climate change adaptation, etc....
Forests
Goal: Conserve the world’s forests.
Needs: Plot-scale global monitoring at <= monthly cadence, comparing new data to long-term trends.
Key metric: Fraction of world’s forests actively monitored with EE.
Progress: • Currently: 62% of tropical rainforest actively monitored.• Global map of forest change (Science 2013).• 2015: EE-generated alerts for 100% of tropical rainforest.
Water
Goal: Save lives by monitoring water quality and quantity.
Needs: Monthly tracking of location, quantity and quality of surface water & plot-scale water use (evapotranspiration).
Key metric: Fraction of high-risk area actively monitored with EE.
Progress: • Global map of permanent and transient water with EU/JRC.• Evapotranspiration algorithms close to deployment.• US drought monitoring system in development.
Emerging Infectious Diseases
Goal: Save lives by eliminating malaria, dengue, and other EIDs.
Needs: Bringing together a diverse range of data sources at near-real-time to support disease risk mapping and operations.
Key metric: Fraction of high-risk area actively monitored with EE.
Progress: • Malaria risk mapping tools in development with UCSF, Oxford.• Strong relationships formed with Gates Foundation, Wellcome
Trust, and others.• Invited to present at White House expert panel on public
health.
Data Catalog
> 200 public datasets
MODISDaily, NBAR, LST, ...
TerrainSRTM, GTOPO, NED, ...
AtmosphericNOAA NCEP, OMI, ...
Land CoverGlobCover, NLCD, ...
The Earth Engine Public Data Catalog
> 4000 new images every day
> 5 million images > 5 petabytes of data
Landsat 4, 5, 7, 8Raw, TOA, SR, ...
... and many more, updating daily!
Sentinel
Sentinel-1
API
Data Types and Geospatial Processing Functions
● Image - band math, clip, convolution, neighborhood, selection ...● Image Collection - map, aggregate, filter, mosaic, sort ...● Feature - buffer, centroid, intersection, union, transform ...● Feature Collection - aggregate, filter, flatten, merge, sort …● Filter - by bounds, within distance, date, day-of-year, metadata ...● Reducer - mean, linearRegression, percentile, histogram ….● Join - simple, inner, outer, inverted ...● Kernel - square, circle, gaussian, sobel, kirsch …● Machine Learning - CART, random forests, bayes, SVM, kmeans, cobweb …● Projection - transform, translate, scale …
over 1000 data types and operators, and growing!
GeospatialDatasets
AlgorithmicPrimitives
add
focal_min
filter
reduce
join
distancemosaic
convolve
Results
Storage and Compute
Requests
Relevant Datasets and Applications
Training Data
Surface water occurrence - Pekel et al., JRC
南昌
Simplified surface energy balance, reference evapotranspiration
Modesto
Allen, Kilic, Huntington (UIdaho, UNL, DRI)
Prasad's Requests
(a) Earth Engine set up(b) coding(c) Landsat time-series EVI, NDVI etc. mosaics of the world(d) k-means algorithm(e) supervised vs. unsupervised classification
Request Access at earthengine.google.com
Python API Installation
Code Editor Example - Computing NDVI Mosaic
Classification
Supervised
Unsupervised
Classification
Supervised
Unsupervised
Classification
Supervised
Unsupervised
Supervised Classification using a Sampled Training Raster
Regression - Fitting models to time series
Cloud/Shadow Masking
Radar Data - Sentinel 1 Temporal Mosaic
NAIP 4-band Imagery
Other Topics
• Datasets• Sentinel 1 / 2• Hyperion• Proba-V ?• VIIRS ?
• Tools• Export to Google Cloud Storage• Render computed products as Earth Engine assets• Exploration of Ommision/Commision maps
• Distribution of Products• Data viewers• Earth Engine public data layers
GFSAD v1.0
Summary
• Earth Engine has datasets and algorithms that are relevant for global cropland mapping.
• Cropland mapping is of interest to Google Earth Engine, because of alignment with the Water Vertical.
• We are interested in feedback on datasets, algorithms, and tools for interacting with the data.
Other Topics
• Uploading data (docs)• Render• IP / open data
How should Google engage with the GFSDA30 project?
● Access to reference data / image products○ Distribution using mobile app (trendy lights)
● Classification Analysis○ Adding/improving EE primitives
● Reference Data○ Helping to share generated data○ Improving tools for calculating/characterizing/exploring errors
■ Tools for characterizing the sampling balance■ https://code.earthengine.google.com/0600d7c2b140cda25f644cad627f4529
○ Generating additional reference data ■ Streetview Service■ FAO Collect Earth■ Crowd Compute on VHR images