Landsat as the wall-to-wall phase of a hierarchical global

forest monitoring system

Sean Healey, Svetlana Saarela, Zhiqiang Yang, Warren Cohen, Göran Ståhl, Paul Patterson, GEDI Science Team

Landsat Science Team meeting August 8, 2018 Boulder, CO

G E D I L I D A R

G l o b a l E c o s y s t e m s D y n a m i c s I n v e s t i g a t i o n

GEDI (Global Ecosystem Dynamics Investigation)

S P A C E - X

D R A G O N

C A P S U L EExposed Facility

Investigator-led Venture-Class mission, 3-laser waveform lidar(PI: Dubayah)

Launch Readiness Date: 29 November

Range: 51 ̊S to 51°N

Primary science deliverable is a 1-km grid of forest biomass prediction, but, approximately 10% of cells will have no clear returns

OBI-WAN (Online Biomass Inference using Waveforms And iNventory)

• The user identifies an area of interest in a Google Earth Engine app, and gets back a forest carbon storage report

• Potential users: forest reserves, individual companies, municipalities

• User-defined areas may also have zero or few clear GEDI returns

Forest Reserve

Unprotected Forest

Traditional Inventory

No modeling

Model-Based Inference

Uses a field-to-Landsat model

Estimate variance is a function of the sample

design

Estimate variance is a function of model

parameter uncertainty

Hybrid Inference

Field-to-lidar model

Hierarchical Model-Based Inference

Uses a field-to-lidar and a lidar-to-Landsat model

Estimate variance is a function of BOTH sample

design and model parameter uncertainty

Estimate variance is a function of parameter

uncertainty of two models

(McRoberts et al., 2014, RSE)

(Ståhl et al., 2016) (Saarela et al., 2016)

Field plots

Lidar tracks

Landsat grid

With disturbance history as a predictor (with reflectance) Overall r2 = -.7

Without disturbance history as a predictor (with reflectance) Overall r2 = -.63

MSS Tangent

Splicing MSS and TM/ETM+/OLI ensemble-based disturbance maps

MSS Tangent

MSS ensemble

1985 1995 20181972

TM ensemble

Harmonize during overlap

MSS Cloud Mask:A random scene: LANDSAT/LM01/C01/T2/LM01_048030_19730809

GEE code available, based on Braaten et al., 2015 RSE

DN DN with BQA mask TOA with MSSmask

MSS Tangent

MSS data status on August 2• From USGS

• Globally there are 1,318,867 MSS images in collection 1

• From GEE

Global

Sensor Tier 1 Tier 2

Landsat 1 2029 136825

Landsat 2 1765 245393

Landsat 3 156 123527

Landsat 4 416 171230

Landsat 5 348 531961

CONUS

Sensor Tier 1 Tier 2

Landsat 1 122 32295

Landsat 2 143 53717

Landsat 3 15 24608

Landsat 4 21 21757

Landsat 5 22 70056

MSS Tangent

Take-home: The free access and high level of processing of the MSS archive enable meaningful improvements in prediction of current biomass

Generalized Hierarchical Model-Based Estimation (Saarela et al., in review)

• Now accounts for clustered lidar data and heteroskedastic models

Saarela et al., In review (Case A) (Case B)

Using Landsat and GEDI together will:

1. Help us fill in areas with no GEDI shots

2. In some cases give us the most precise estimates of biomass

3. Produce a really good Landsat-resolution biomass map