a prototype of global cropping intensity mapping using ......mapping using google earth engine...
TRANSCRIPT
Jun Xiong | Research Scientist
United States Geological SurveyWestern Geographic Science Center
A Prototype of Global Cropping Intensity
Mapping using Google Earth Engine
Global Food Security-Support Analysis Data at 30 m (GFSAD30) Workshop
Reston, VA from August 15-17, 2017
U.S. Department of the Interior
U.S. Geological Survey
Definition of “Cropping Intensity”
Overview
Cropping intensity (whether a field is single, double, or
continuous cropped in a given year)
➢ Cropping intensity can greatly affect net production
➢ Help to identify crop fallow, crop type information partly
“crop intensity / cropping intensification”, technically
defined as an increase in agricultural production per unit
of inputs (which may be labour, land, time, fertilizer, seed,
feed or cash).
U.S. Department of the Interior
U.S. Geological Survey
Challenges of Cropping intensity MappingOverview
➢ Uncertainities in the NDVI Times series data
➢ Modules of pre-processing: gap-filling, fitting, smooth
➢ Algorithm for single and double crops
➢ Variance of fallow-land / continuous crops in large area
➢ Validation & evaluation approach
U.S. Department of the Interior
U.S. Geological Survey
Overview
➢ GEE scripts of processing MODIS or Landsat/Sentinel-2
NDVI Times series on the fly
➢ Time-series-analyzed based algorithm to map single &
double crops
➢ Alternatives algorithm when time-series fails
➢ Setting of thresholds to identify fallow-land and continuous
crops
Global GEE Algorithm Components
U.S. Department of the Interior
U.S. Geological Survey
GEE assets
➢ Source:
MODIS/006/MOD13Q1 and
MODIS/006/MYD13Q1
➢ Data Period: since Feb 2000
➢ Resolution: 250m
➢ Process: n/a
➢ Status: ready to use
ImageCollection: MODIS 16-day Time series
VHRI
MODIS NDVI
U.S. Department of the Interior
U.S. Geological Survey
Modules
➢ Source:
LANDSAT/LC8_L1T_TOA
➢ Data Period: since Apr 2013
➢ Resolution: 30 m
➢ Process:
1. Cloud Mask (FMASK)
2. Gap-filling (16-day)
➢ Status: running on the flow,
moderate speed, scalable
ImageCollection: Landsat-8 16-day Time series
Landsat
VHRI
U.S. Department of the Interior
U.S. Geological Survey
Modules
➢ Source:
LANDSAT/L*_L1T_TOA
➢ Data Period: since Jan 1984
➢ Resolution: 30 m
➢ Process:
1. Cloud Mask (FMASK)
2. Gap-filling (16-day)
➢ Status: running on the flow,
moderate speed, scalable,
time-out error
ImageCollection: Landsat-all-sensors 16-day Time series
U.S. Department of the Interior
U.S. Geological Survey
Modules
➢ Source: Sentinel-2
➢ Data Period: since Jun 2015
➢ Resolution: 10 m
➢ Process:
1. Cloud Mask (QA60)
2. Gap-filling (10-day)
➢ Status: running on the flow,
slow speed, scalable, time-
out error
ImageCollection: Sentinel-2 10-day Time series
South Africa (25.637, -33.463)
U.S. Department of the Interior
U.S. Geological Survey
Modules
➢ Data Period: since Jun
2015
➢ Resolution: 10-30 m
➢ Process:
1. Cloud Mask (QA60)
2. Gap-filling (16-day)
3. Harmonizer required
➢ Status: export-required,
very slow speed, scalable,
time-out error
ImageCollection: Landsat-Sentinel 16-day Time series
Landsat
Sentinel
South Africa
May 2016
25.637, -33.463
U.S. Department of the Interior
U.S. Geological Survey
Modules
➢ Pro
very fast and stable; easy to use
➢ Con
fail when too much noise
Smoother: moving average
U.S. Department of the Interior
U.S. Geological Survey
Modules
Savitzky Golay: polynomial-weighted moving average
➢ Pro
smooth without greatly distorting the signal
➢ Con
parameters tuning; time-consuming
Smoother: Savitzky Golay
U.S. Department of the Interior
U.S. Geological Survey
Modules
implement by Justin
➢ Pro
more controls and powerful
➢ Con
expensive; time-out error
Smoother: 4253H twice smoothing
U.S. Department of the Interior
U.S. Geological Survey
Modules
➢ Pro
good balance between
performance and speed
➢ Con
introduce pattern which
might not exists
Smoother: Harmonic model
Shumway and Stoffer (2017)
U.S. Department of the Interior
U.S. Geological Survey
Methodology
pro
➢ no need of prior knowledge
➢ fast to implement
con
➢ depend on time series data quality
➢ uncertainty of smoother
1. Remove abnormal points
2. Smooth Time Series data
3. Locate the local max value (peak)
4. Filter out global low-value peaks
5. Counting the number of Peaks
Method 1: Peak Counting
U.S. Department of the Interior
U.S. Geological Survey
Method #2: Time Series MatchingMethodology
Euclidean
MaySep
Jan
Feb
DTW
Sep May
➢ nonlinear matching
Dynamic time warping➢ linear matching
• Reference samples required; smoother required
• Linear approach: fast, fail when time series shifting
• Non-linear approach: accept given variance, very time-
consuming computing O(N2), PrunedDTW, SparseDTW,
FastDTW, MultiscaleDTW Demo
U.S. Department of the Interior
U.S. Geological Survey
To identify if a pixel is cropped or
uncropped during each growing
season.
pro
easy to scale in large area, high
resolution
no need to smooth
computing for every season
con
need regional parameter
Methodology
Method #3: Growing Season Threshold
U.S. Department of the Interior
U.S. Geological Survey
Harvested
Date
\
Methodology
Harvest Date: Sage Crop Calendar Dataset
U.S. Department of the Interior
U.S. Geological Survey
Methodology
Flowchart of Cropping Intensity Mapping
U.S. Department of the Interior
U.S. Geological Survey
Methodology
Global Cropping Intensity Map (MODIS, 2016, trial)
Xiong J, Thenkabail PS, Teluguntla, P., et al. Mapping Cropping Intensity of
Smallholder Farms Globally using Google Earth Engine. Journal TBD. 2017;(In
Preparation).
U.S. Department of the Interior
U.S. Geological Survey
Methodology
Demo: MODIS vs. Sentinel-2
U.S. Department of the Interior
U.S. Geological Survey
Future WorkTo-do
➢ Thresholds of mapping of fallow-land & continuous
crops
➢ Gap-filling in 30-m satellite time series
➢ More testing in typical sites globally
➢ Sampling of reference signatures
➢ Finalize global cropping intensity map
➢ Validation & area evaluation
U.S. Department of the Interior
U.S. Geological Survey
Thank you!