Forecasting Crop Productivity with High-Resolution Satellite Data - - PowerPoint PPT Presentation

Forecasting Crop Productivity with High-Resolution Satellite Data

Scaling Up to the Whole US Corn Belt

PI: Kaiyu Guan Co-PI: Jian Peng Team members: Bin Peng, Yunan Luo, Sibo Wang Presented by: Sibo Wang

Department of Natural Resources and Environmental Sciences (NRES) National Center for Supercomputing Applications (NCSA) University of Illinois at Urbana-Champaign June 6, 2018 Blue Waters Symposium 2018

Objective

Improve our predicting skills for regional/global crop yield by integrating advanced remote sensing observations and process-based modeling.

2

Satellite Remote Sensing for Agriculture

• Wide coverage provides large data volume

○ “Big data” ○ Data with high resolution in both space and time is needed

• Uniform, standardized metric

○ Overcomes regional limitations ○ Allows large-scale (eg. planetary) application

3

US Corn Belt

• Maize (corn) is the most important staple food and feed crop in the

world (according to the total production).

• The US Midwest Corn Belt produces over 45% of global maize

production.

4

Why Blue Waters

• High-resolution satellite images are huge
• Blue Waters provides great computation power and storage capacity

○ Allows scaling up at fine resolution

5

• PC. Cray Inc.

Outline

Part I: High-Resolution Satellite Image Processing

• STAIR: fusing datasets at different resolution and frequency
• PlanetScope: making use of high-resolution CubeSAT data

Part II: Process-Based Crop Modeling

• CLM-APSIM: improving maize growth process models

6

High-Resolution Satellite Image Processing

• STAIR Fusion
• CubeSAT Processing
• PC. NASA Earth Observatory

https://earthobservatory.nasa.gov/IOTD/view.php?id=5772

7

The Dilemma

• Satellite platforms face trade-offs between spatial resolution and

temporal frequency

○ High resolution and high revisiting frequency can’t be achieved by the same sensor ○ Agricultural analysis requires continuous time series data at fine resolution

8

Satellite Platforms

9

MODIS, NASA

Moderate Resolution Imaging Spectroradiometer

Landsat, USGS

US Geological Survey

Sentinel 2, ESA

European Space Agency

CubeSAT, Planet Lab

PlanetScope

High Resolution

3.125m 10m 30m 500m

Satellite Platforms

10

MODIS, NASA

Moderate Resolution Imaging Spectroradiometer

Landsat, USGS

US Geological Survey

Sentinel 2, ESA

European Space Agency

CubeSAT, Planet Lab

PlanetScope

High Frequency

daily every ~3 days ~ weekly every ~2 weeks

11 MODIS (500m) Landsat (30m) Planet CubeSat (3m) Sentinel 2 (10m)

STAIR Fusion

• A generic and fully automated method for fusing multi-spectral

satellite data

• Achieves a high-resolution, high-frequency, cloud-free, gap-free

composite dataset

Luo, Y., Guan, K., Peng, J. 2018. “STAIR: A generic and fully-automated method to fuse multiple sources of optical satellite data to generate a high-resolution, daily and cloud-/gap-free surface reflectance product.” Remote Sensing of Environment, 214.

12

Additional Challenges

• Gaps due to clouds
• Landsat 7 Scan Line Corrector failure
• Inter-tile consistency

13

Sample Landsat 7 image with strips after SLC malfunction Sample Landsat 5 image with clouds and cloud shadows

14

https://www.youtube.com/wat ch?v=ISZ7MZrG8nM

PlanetScope CubeSAT

• Owned by private company Planet Labs
• A fleet of 175+ small CubeSAT satellite platforms
• High resolution (3m)
• Also has more potential issues

○ Inconsistency in spectral response among platforms ○ Quality assessment (mainly cloud detection) ○ Quality of surface reflectance product ○ Cost

15

• PC. Planet Labs

A Complete Pipeline

• Goal:

○ Generate a more usable PlanetScope-derived dataset ready for agricultural analyses

• What we have:

○ Fusion product with reliable time series (from MODIS) and 30m-resolution spatial details (from Landsat) ○ Large volume of raw PlanetScope data

• Workflow:

○ Fusion-based spectral correction ○ Fusion-based outlier/cloud detection ○ Time series smoothing

16

17

18

Atmospheric Correction

19

Land-Cover-Specific Outlier Detection

Spectral Correction

20

21

Raw Planet Lab Data Processed Planet Lab Data

Process-Based Modeling

• CLM-APSIM

Public Domain

CLM-APSIM

• Agronomy crop models (eg. Agricultural Production System Simulator (APSIM), Keating et al.,

2003; Holzworth et al., 2014) simulate field-level crop growth. Phenology stages are considered.

• Earth system models (ELMs) such as the Community Land Model (CLM, Oleson et al., 2013;

Lawrence et al., 2011) numerically and explicitly solve surface water, energy and carbon balances,

but are much simpler and usually do not consider phenology-stage-dependent stresses.

• CLM-APSIM integrates the strengths of both families of crop models.

○ Improved the representation of maize phenological development. ○ Corrected the deficiencies in carbon allocation scheme.

Peng, B. et al., 2018. “Improving maize growth processes in the community land model: Implementation and Evaluation.” Agricultural and Forest Meteorology, 250-251 (2018).

23

24

25

← harvest index at the three sites ↑ canopy height ↓ biomass pools

Crop Modeling: Moving Forward

1) Simulate the whole US Corn belt 2) Explicit calibration at the grid level 3) Ingest satellite data for data assimilation 4) Real-time forecasting of crop yield

26

Questions?

27