Cropland Carbon Uptake Delayed and Reduced by 2019 Midwest Floods

Cropland Carbon Uptake Delayed and Reduced

by 2019 Midwest Floods

Yi Yin

, Brendan Byrne

, Junjie Liu

3,1

, Paul O. Wennberg

1,4

, Kenneth J. Davis

5,6

Troy Magney

1,7

, Philipp Köhler

, Liyin He

, Rupesh Jeyaram

, Vincent Humphrey

Tobias Gerken

, Sha Feng

, Joshua P. Digangi

, and Christian Frankenberg

1,3

Division of Geological and Planetary Sciences, California Institute of Technology, Pasadena, CA, USA,

NASA

Postdoctoral Program Fellow, Jet Propulsion Laboratory, California Institute of Technology, Pasadena, CA, USA,

Jet

Propulsion Laboratory, California Institute of Technology, Pasadena, CA, USA,

Division of Engineering and Applied

Science, California Institute of Technology, Pasadena, CA, USA,

Department of Meteorology and Atmospheric Science,

Pennsylvania State University, University Park, PA, USA,

Earth and Environmental Sciences Institute, Pennsylvania

State University, University Park, PA, USA,

Department of Plant Sciences, Davis, CA, USA,

Langley Research Center,

National Aeronautics and Space Administration, Hampton, VA, USA

Abstract

While large

‐

scale

oods directly impact human lives and infrastructures, they also profoundly

impact agricultural productivity. New satellite observations of vegetation activity and atmospheric CO

offer the opportunity to quantify the effects of such extreme events on cropland carbon sequestration.

Widespread

ooding during spring and early summer 2019 induced conditions that delayed crop planting

across the U.S. Midwest. As a result, satellite observations of solar

‐

induced chlorophyll

uorescence from

TROPOspheric Monitoring Instrument and Orbiting Carbon Observatory reveal a 16

‐

day shift in the

seasonal cycle of photosynthesis relative to 2018, along with a 15% lower peak value. We estimate a reduction

of 0.21 PgC in cropland gross primary productivity in June and July, partially compensated in August and

September (+0.14 PgC). The extension of the 2019 growing season into late September is likely to have

bene

ted from increased water availability and late

‐

season temperature. Ultimately, this change is predicted

to reduce the crop productivity in the Midwest Corn/Soy belt by ~15% compared to 2018. Using an

atmospheric transport model, we show that a decline of ~0.1 PgC in the net carbon uptake during June and

July is consistent with observed CO

enhancements of up to 10 ppm in the midday boundary layer from

Atmospheric Carbon and Transport

‐

America aircraft and over 3 ppm in column

‐

averaged dry

‐

air mole

fractions from Orbiting Carbon Observatory. This study quanti

es the impact of

oods on cropland

productivity and demonstrates the potential of combining solar

‐

induced chlorophyll

uorescence with

atmospheric CO

observations to monitor regional carbon

ux anomalies.

Plain Language Summary

Widespread

ooding and inundation across the U.S. Midwest during

spring and early summer 2019 forced many farmers to delay crop planting. New satellite observations of

vegetation photosynthesis and atmospheric CO

offer the opportunity to quantify the effects of such events

on cropland carbon sequestration. We show that the delayed planting resulted in a shift of 16 days in the

seasonal cycle of the crop growth and a ~15% lower peak solar

‐

induced chlorophyll

uorescence value. We

estimate a reduction of 0.21 PgC in the gross primary production during June and July, partially

compensated in August and September (+0.14 PgC). The extension of the 2019 growing season into late

September is likely to have bene

ted from increased water availability and late

‐

season temperature.

Ultimately, this change is predicted to reduce the crop production over most of the Midwest Corn/Soy belt by

15%, based on the strong empirical correlation between 2018 growing season SIF and crop yield. The bottom

‐

up estimated net carbon uptake reduction of ~0.1 PgC in June and July is consistently supported by top

‐

down inferred CO

anomalies from both aircraft and satellite observations. We anticipate that such a rapid

event detection can bene

t agricultural and natural resource management and ecological forecasting efforts.

1. Introduction

Many studies have described the impacts of drought on the carbon cycle (Ciais et al., 2005; Humphrey et al.,

2018; J. Liu et al., 2018; Sun et al., 2015; Wolf et al., 2016); however, the impacts of extreme wetness (i.e.,

oods) have been less well documented. Floods are among the major climate

‐

related disasters that are

This is an open access article under the

terms of the Creative Commons

Attribution

‐

NonCommercial License,

which permits use, distribution and

reproduction in any medium, provided

the original work is properly cited and

is not used for commercial purposes.

RESEARCH ARTICLE

10.1029/2019AV000140

Key Points:

•

Flood

‐

induced delay in crop

planting shifted the 2019 SIF

seasonal cycle by 16 days and

reduced the peak value by ~15%

compared to 2018

•

A ~0.1 PgC reduction in Midwest net

ecosystem uptake during June and

July is consistent with the observed

increase in atmospheric CO

•

The 2019

ood induced a 0.06 PgC

reduction in the annual GPP of

croplands (

−

4%) but a 0.04 PgC

increase for natural vegetation

(+3%)

Supporting Information:

•

Supporting Information S1

•

Original Version of Manuscript

•

Peer Review History

•

First Revision of Manuscript

[Accepted]

Correspondence to:

Y. Yin and B. Byrne

yiyin@caltech.edu

brendan.k.byrne@jpl.nasa.gov

Citation:

Yin, Y., Byrne, B., Liu, J., Wennberg, P.,

Davis, K. J., Magney, T., et al. (2020).

Cropland carbon uptake delayed and

reduced by 2019 Midwest

oods.

AGU

Advances

, e2019AV000140. https://

doi.org/10.1029/2019AV000140

Received 6 DEC 2019

Accepted 5 FEB 2020

Peer Review

The peer review history

for this article is available as a PDF in

the Supporting Information.

YIN ET AL.

1of15

projected to increase in a warmer climate (Hirabayashi et al., 2013); quantifying their impact on the terres-

trial carbon cycle is critical for the assessment of future climate change impacts. In 2019, the contiguous

United States recorded its wettest January to August in 125 years, with wetter

‐

than

‐

normal conditions from

the northern Plains to the Gulf Coast (NOAA, October 2019). Vast

ooding across the Midwest from March to

June forced farmers to signi

cantly delay the planting of crops in this region

—

known as the Corn Belt

—

which accounts for ~40% of world corn and soybean production (USDA, 2019). Previous studies have documen-

ted the highest peak in photosynthesis across the globe in this area (Guanter et al., 2014; Mueller et al., 2016).

Higher net carbon uptake by Midwest cropland compared to the nearby forest has also been shown by tower

‐

based atmospheric CO

measurements (Miles et al., 2012), resulting in a large regional carbon sink associated

with Midwestern agriculture (Lauvaux et al., 2012; Schuh et al., 2013). Thus, the

ood and associated delay in

timing of planting is expected to impact the regional carbon cycle; however, it remains unclear to what extent

crop growth was affected by the 2019

ood and its consequent impact on the regional carbon cycle.

Satellite observations of solar

‐

induced chlorophyll

uorescence (SIF), a by

‐

product of photosynthesis, have

been shown to be a useful proxy of gross primary productivity (GPP) (Frankenberg et al., 2011; Sun et al.,

2017; Yang et al., 2015) and crop yield (Guan et al., 2016; Guanter et al., 2014). SIF is mechanistically linked

with the light reactions of photosynthesis and has shown close correspondence to GPP across many ecosys-

tems (Frankenberg & Berry, 2018; Gu et al., 2019). As a result of the strong empirical and mechanistic rela-

tionship between SIF and GPP at the satellite scale, and con

rmation of this at smaller scales (Liu, Guan,

et al., 2017), we use SIF as an indicator of GPP (Byrne et al., 2018; Green et al., 2017, 2019; Parazoo et al.,

2014). In particular, in the context of

oods, the SIF signal is not impaired by surface spectral re

ectance

properties that are confounded by surface water inundation. The recently launched TROPOspheric

Monitoring Instrument (TROPOMI) provides SIF data at unprecedented spatial resolution (7 km × 3.5 km

at Nadir) with almost daily global coverage, allowing close monitoring of photosynthesis and carbon uptake

as associated events unfold.

From the top

‐

down, measurements of the atmospheric CO

can provide constraints on net ecosystem

exchange (NEE)

—

the net exchange of CO

between an ecosystem and the atmosphere, determined as auto-

trophic and heterotrophic respiration minus GPP (Bolin & Keeling, 1963; Gurney et al., 2002; Yin et al., 2018).

Space

‐

based measurements of column

‐

averaged dry

‐

air mole fractions of CO

CO2

) have been shown to

provide information on NEE anomalies at large subcontinental scales (Byrne et al., 2017; Byrne et al.,

2019; Guerlet et al., 2013; Ishizawa et al., 2016; Liu, Bowman, et al., 2017; Liu et al., 2018; Yin et al., 2016)

and for point sources (Nassar et al., 2017; Schwandner et al., 2017) but are untested on smaller regional scales,

such as the Midwest croplands. The Orbiting Carbon Observatory 2 (OCO

‐

2), launched in 2014, provides

CO2

to monitor the atmospheric signal of the event. Boundary layer CO

measurements have been shown

to provide strong constraints on regional carbon

uxes over the Midwest croplands (Lauvaux et al., 2012;

Schuh et al., 2013). The Atmospheric Carbon and Transport (ACT)

‐

America aircraft campaign over the

Midwest during summer 2019 provides spatially extensive measurements designed to capture regional atmo-

spheric CO

signals.

The combination of these newly available observations offers a unique opportunity to monitor the impacts of

2019

oods on the regional carbon cycle. Speci

cally, we aim to ask: How does the delayed planting impact

the seasonal cycle of the crop growth? What are the implications for the crop productivity of this year? And

how does the

ooding and inundation impact the overall carbon uptake of these cropping systems? To that

end, we

rst quantify the impact of 2019

oods on the photosynthetic carbon uptake throughout the growing

season based on SIF observations relative to the previous years and evaluate potential impacts of this

ooding

and inundation on crop productivity over the Midwest. Then, we estimate associated atmospheric CO

anom-

aly using a bottom

‐

up approach based on the SIF anomaly and a top

‐

down approach based on satellite and

aircraft CO

measurements. A schematic for the overview of the methods is shown in Figure S1 in the

supporting information.

2. Materials and Methods

2.1. Data

2.1.1. Satellite

‐

Based SIF Observations From TROPOMI and OCO

‐

We use satellite

‐

based SIF retrievals to track the progress of photosynthesis. The TROPOMI instrument on

‐

board of the Sentinel 5 Precursor (S

‐

5P) satellite was launched on 13 October 2017. The S

‐

5P satellite

ies in

10.1029/2019AV000140

AGU Advances

YIN ET AL.

2of15

a near

‐

polar Sun

‐

synchronous orbit with an equatorial crossing time at 13:30 local solar time. The

TROPOMI instrument has a wide swath of ~2,600 km, yielding almost daily global coverage that allows

an unprecedented temporal resolution to track the change in photosynthesis at a given location (Köhler

et al., 2018). The spectrometer has a spatial resolution of 7 km along track and 3.5

–

15 km across track, a spa-

tial resolution that allows us to more appropriately link with county

‐

level agricultural census data. Hence,

individual retrievals were aggregated at the county level to align with agricultural census data, as introduced

in the section below. A daily correction factor accounting for the diurnal and seasonal variations of solar

zenith angle (SZA) is applied to convert an instantaneous SIF signal to a daily average as detailed in

Köhler et al., 2018

Additionally, we use SIF retrievals from the Orbiting Carbon Observatory (OCO

‐

2) to complement the

TROPOMI SIF records. The OCO

‐

2 instrument has a longer temporal record (since September 2014), provid-

ing a multiyear reference of a typical seasonal cycle. Compared to TROPOMI, it has a higher spectral and

spatial resolution (1.3 × 2.25 km), and as a trade

‐

off, it has a much sparser sampling coverage in both space

(swath width of up to 10 km) and time (with a global revisit time of 16 days; Sun et al., 2017). The two instru-

ments have been shown to be in close agreement for overlapping retrievals (Köhler et al., 2018).

2.1.2. Agricultural Statistics

We obtain county

‐

level crop statistics of 2018 from the United States Department of Agriculture (USDA)

National Agricultural Statistics Service (NASS) Quick Stats Database (quickstats.nass.usda.gov), including

the planted/harvest area and crop yield of individual crop types. Note that reported crop yield refers to

the amount of crop produced per harvested area, whereas crop production refers to the total amount of har-

vest, which is the sum of crop yield multiplied by harvested area. As satellite observes vegetation productiv-

ity per unit area

—

not necessarily harvested area

—

we introduce a term

“

crop productivity

”

here, speci

cally

referring to the amount of all crops produced per unit area. We add up all reported crops for

individual county.

For each county, we calculate the percentage of cropland area and the ratio of C4 to C3 crops. The latter is

important to consider because C3 and C4 plants use different photosynthetic pathways, which show a differ-

ent response of SIF to GPP (Liu, Guan, & Liu, 2017). C4 plants inhibit photorespiration, which results in

higher GPP to SIF ratio, as SIF is most sensitive to the light reactions of photosynthesis (Gu et al., 2019).

Typical C3 crops include soybeans, wheat, barley, oats, rice, and tree crops, whereas typical C4 crops include

corn, sugarcane, and sorghum. Here, we focus on corn and soybeans, which are predominately planted in

the Midwest. As the county

‐

level statistics of 2019 are not yet available, we use state

‐

level

planted/emerged areas of corn and soybean for 2019 from weekly USDA reports (USDA, 2019); hence, the

uncertainty of those reported dates is around 1 week.

2.1.3. Atmospheric CO

Observations

To analyze signals of atmospheric CO

, we use CO

observations from both satellites and aircraft. We down-

loaded Version 9 of the Atmospheric CO

Observations from Space OCO

‐

2 lite X

CO2

retrieval

les from the

Virtual Science Data Environment (https://co2.jpl.nasa.gov/; Crisp et al., 2012). We include land nadir

and land glint measurements.

We also use airborne CO

measurements from the NASA ACT

‐

America campaign conducted during the

summer 2019 over the eastern United States (Digangi et al., 2017). We use observations from a total of 37

campaign

ight days over 11 June to 27 July 2019; the campaign schedule is available online (at https://

act

‐

america.larc.nasa.gov/). For each campaign, CO

was measured from two aircraft in coordinated regio-

nal

ight patterns using PICARRO G2401

‐

m monitors at 0.4 Hz. The measurements are averaged into a 5

‐

product, which approximately corresponds to 500

‐

m segments as the plane speed is ~100 m/s. The instru-

ment is calibrated hourly using standards traceable to the WMO X2007 scale (Tans et al., 2017). For analysis,

we gridded the raw data to 2° × 2.5° horizontally and 47 layers vertically to match the resolution of the che-

mical transport model. To reduce the impact of remote sources, we only include data within the atmospheric

boundary layer (between 300 and 1,500 m above ground level).

2.1.4. Environmental Variables

We use terrestrial water storage (TWS) anomalies derived from the Gravity Recovery and Climate

Experiment (GRACE) and GRACE Follow

‐

On (GRACE

‐

FO) missions (Flechtner et al., 2014; Tapley et al.,

2004). TWS anomalies over the U.S. Midwest re

ect changes in soil moisture, groundwater, snowpack,

and surface waters, mainly in response to climate variability (Humphrey et al., 2016). We use the RL06

10.1029/2019AV000140

AGU Advances

YIN ET AL.

3of15

monthly mass grids by NASA's Jet Propulsion Laboratory, apply scaling factors (Landerer & Swenson, 2012),

and retrieve the regional average TWS anomalies over all regions with crop area fractions larger than 25%

over the period April 2002 to August 2019. We note that there is a gap of 12 months between GRACE and

GRACE

‐

FO which is still under investigation; however, preliminary analyses indicate that there is no sys-

tematic bias between the two missions.

Temperature, vapor pressure de

cit and precipitation data are obtained from ERA5 (Copernicus Climate

Change Service (C3S), 2019), the

fth

‐

generation atmospheric reanalysis of the European Centre for

Medium

‐

Range Weather Forecasts. This reanalysis assimilates multiple data streams from satellite and in

situ measurements and provides hourly data at a spatial resolution of 30 km. We compute daily regional

averages over all Midwest states with crop area fractions larger than 25% (see mask of the 17 selected states

in Figure S4).

2.2. Model

2.2.1. Atmospheric Transport and Flux Inversion Model (GHGF

‐

Flux)

We use an atmospheric transport model to simulate expected signals in the atmospheric CO

due to the

ood

induced anomaly in cropland carbon uptake. We use the forward and adjoint components of the

Greenhouse gas framework

‐

Flux model (GHGF

‐

Flux) for atmospheric chemical transport and

ux inversion

analysis. GHGF

‐

Flux is a

ux inversion system developed within NASA's Carbon Monitoring System Flux

project (Bowman et al., 2017; J. Liu et al., 2014). GHGF

‐

Flux inherits the chemistry transport model from

the GEOS

‐

Chem and the adjoint analysis methods from the GEOS

‐

Chem

‐

adjoint. Chemical transport is dri-

ven by the Modern

‐

Era Retrospective Analysis for Research and Applications, Version 2 (MERRA

‐

meteorology produced with Version 5.12.4 of the GEOS atmospheric data assimilation system (Gelaro

et al., 2017).

In this study, we use GHGF

‐

Flux to perform forward tracer transport at 2° × 2.5° spatial resolution for the

year 2019. We also use GHGF

‐

Flux to perform inversions to derive the 2018 baseline NEE at a 4° × 5° spatial

resolution. To estimate 2018 NEE, we use the average of three inversions that employ different prior NEE

uxes: SiB3 (Baker et al., 2008), CASA (Potter et al., 1993; James T Randerson et al., 1996; van der Werf

et al., 2006), and FLUXCOM (Jung et al., 2017; Tramontana et al., 2016). All versions assimilate OCO

‐

2 land

nadir and land glint data from October 2017 to April 2019 to optimize 14

‐

day scaling factors for gridded NEE

and ocean

uxes using the approach of Byrne et al. (2019); see Appendix 1 for details. The optimized 2018

NEE was then used to simulate a theoretical baseline for CO

mole fractions at 2° × 2.5° driven by 2019

meteorological reanalysis, which represents an ideal case when 2019 carbon

uxes are identical to 2018

while the transport pattern being different. The mismatch between the baseline CO

and measured CO

pro-

vides a measure of the difference in

uxes between years (Figure S6).

2.3. Methods

2.3.1. SIF

‐

Based GPP and NEE Estimates

SIF has been shown to be a robust proxy for GPP (Frankenberg & Berry, 2018; van der Tol et al., 2014). While

nonlinear relationships between GPP and SIF have been observed at leaf and

‐

tower scales under certain

conditions, for example, strong incoming light (Magney et al., 2019; Verma et al., 2017), linear relationships

have been generally observed at ecosystem and regional scales (Frankenberg et al., 2011; Li et al., 2018;

Magney et al., 2019; Sun et al., 2017). The robust linear relationship at increasing scales is likely because

satellite measurements are primarily measuring an integrated canopy average of low Photosynthetically

Active Radiation (PAR). At higher light levels, GPP saturates while SIF continues to increase (Magney,

Frankenberg, et al., 2019), but it is unlikely that satellite measurements over large pixels are measuring in

this light regime. Based on this, and results from previous studies, we quantify GPP using SIF observations

(MacBean et al., 2018; Parazoo et al., 2014). Due to the different GPP to SIF slopes between C3 and C4 plants

(He et al., in review; Gu et al., 2019; Li et al., 2018; Liu, Guan, & Liu, 2017; Miao et al., 2018; Pérez

‐

Priego

et al., 2015), we use different scaling factors for C3 and C4 crops following the linear GPP:SIF ratios for daily

averages as documented by Li et al. (2018) (19.8 and 29.4 g C·m

‐

·day

‐

/Wm

‐

·sr

‐

for C3 and C4,

respectively). Thus, for each county the scaling factor is C3/C4 growing area weighted.

For the noncrop area, we use 2018 MODIS MCD12Q1 land cover map (https://doi.org/10.5067/MODIS/

MCD12Q1.006) to identify the land cover type at 0.083° spatial resolution into seven different vegetation

types, namely, evergreen broadleaf forest, deciduous broadleaf and mixed forest, needleleaf forest, shrub,

10.1029/2019AV000140

AGU Advances

YIN ET AL.

4of15

woody savanna and savanna, and grasslands. We then apply corresponding GPP:SIF ratios for each vegeta-

tion type following (Li et al., 2018), where the ratios are derived from

ux tower GPP observations and OCO

‐

2 SIF.

Based on diurnal

‐

cycle

‐

corrected TROPOMI SIF, we estimate daily county

‐

level GPP for every 8

‐

day interval

accounting for its planted areas of C3/C4 as documented by USDA. A conversion factor of 0.64 is used to con-

vert TROPOMI SIF at 740 nm to SIF at 757 nm, to account for spectral scaling differences at different wave-

lengths in the SIF signal retrieval (Köehler et al., 2018). The differences in GPP between 2019 and 2018

(noted as

GPP) at the same time of the year are used to estimate the 2019 GPP anomaly.

To estimate the impact on the net carbon exchanges (

NEE), we assume that the anomaly is solely induced

by a reduction in net primary productivity (NPP), which accounts for about half of GPP following Randerson

et al. (1996). Due to the lack of direct observational evidence, we assume no signi

cant changes in hetero-

trophic respiration between the 2 years. Hence, for further analysis,

NEE =

−

0.5 ×

GPP. To isolate the

impact from Midwest cropland on NEE, we only include counties in the 17 states of the Midwest and south-

ern United States whose cropland coverages are larger than 25% for this estimate. We tested that the results

are not sensitive to the choice of this threshold as shown in Table S1. Resultant

NEE between 2019 and

2018 are aggregated into gridded data at 2° × 2.5° resolution for atmospheric transport model simulation

as detailed below.

2.3.2. Flood

‐

Induced Atmospheric CO

Signal Estimates

A reduction in NEE is expected to increase atmospheric CO

downwind of the croplands. We estimate the

anomaly in atmospheric CO

by calculating differences in NEE between 2019 and 2018 (denoted as

). The

signal from this event is calculated using both top

‐

down and bottom

‐

up approaches,

and then we evaluate the consistency between the two approaches.

Bottom

‐

up expected

are simulated from the SIF

‐

inferred

NEE using an atmospheric transport model

(GHGF

‐

Flux). We use estimated

NEE surface

uxes at an 8

‐

day temporal resolution as input and sample

modeled CO

at the time and location of the ACT

‐

America and OCO

‐

2 measurements. The prior pro

les

and the averaging kernels of the X

CO2

measurements are applied.

Top

‐

down inferred

are calculated as the difference between OCO

‐

2 or ACT

‐

America measurements

and the baseline CO

simulated with posterior 2018 NEE and 2019 meteorology (as described in

section 2.2.1).

3. Delayed Cropland Growing Season Seen by SIF

The monthly distributions of county

‐

level instantaneous SIF values from TROPOMI during the growing sea-

son of 2018 and the differences between 2019 and 2018 are shown in Figure 1. For both years, the highest SIF

values across the United States occurred in the Midwest in July, demonstrating the highest peak productivity

of cropland relative to natural ecosystems. This large

‐

scale signal is consistent with previous space

‐

borne SIF

observations from a different instrument (Guanter et al., 2014), while the high

‐

resolution of TROPOMI data

provides a new opportunity to look into county

‐

level details at a higher temporal resolution. In the Midwest,

the 2019 SIF values are much lower than 2018 in June and July; however, they surpass the 2018 levels in

August and September.

The distribution of areas showing signi

cant differences in SIF between the 2 years resembles that of the

cropland density as shown on the lower left of Figure 1. The largest differences occurred in the Midwest

states that have the longest delay in crop planting as shown on the lower right (Figure 1). For regions with

crop area fractions larger than 25%, SIF values in 2019 are 30% and 15% lower than 2018 in June and July,

respectively. When looking speci

cally at areas with higher cropland fractions (>50%), SIF decreased by

47% and 20% for June and July. In contrast, 2019 SIF surpasses the 2018 level slightly in August (+8%)

and markedly in September (+50% for counties with crop area >25% and +75% for areas with cropland frac-

tion >50%). The increased growth during late growing season partially compensates for the reduction in the

early growing season.

The seasonal cycle of daily TROPOMI SIF during 2018 and 2019 is shown in Figure 2a. By shifting the 2019

SIF time series 16 days ahead, we can see that the growing season lengths of the 2 years are similar but the

2019 seasonal peak values are 15% lower. A large part of the decline could be attributed to changes in solar

10.1029/2019AV000140

AGU Advances

YIN ET AL.

5of15

Figure 1.

Spatial distribution of instantaneous SIF retrievals from TROPOMI during the main growing seasons in 2018 (the left column) and the differences in SI

between 2019 and 2018 (the right column). The bottom panel shows the cropland area ratio for each county (on the left), and the USDA reported delay in pla

nting

75% of the total planted corn area relative to 2018 (on the right).

10.1029/2019AV000140

AGU Advances

YIN ET AL.

6of15

illumination, as the SZA and daytime length decline after the summer sol-

stice. This effect is illustrated by a proxy for seasonal changes in the poten-

tial daily PAR under cloud

‐

free conditions using the daily integral of

cosine (SZA) at 10

‐

min time steps (Figure 2a). Although the anomalies

in TROPOMI SIF shown here are just based on 2 years (2019

–

2018), ana-

lysis of OCO

‐

2 SIF covering the last 5 years (2015

–

2019) shows that the

seasonality of SIF in 2018 is similar to the preceding 3 years but that

2019 is indeed anomalous (Figure 2b). Both data sets reveal a shift in

the seasonal cycle and a lower peak SIF value in the growing season of

2019, even though the two instruments have a different temporal and

spatial sampling.

Such a delay in the seasonal cycle of crop growth is induced by the

late planting of crops as a result of the

ood and associated wet soils,

which are not conducive to seed germination (anoxic conditions) and

are challenging for farm machinery to drive on (Figure 2c). The accu-

mulation of water over the Midwest is re

ected in the gradual increase

of TWS since the beginning of 2019 as observed by GRACE

‐

FO from

space (Figure 3a). The TWS anomaly in June 2019 is around 10 cm

higher than that of June 2018, more than two times the standard

deviations above the decadal mean as monitored by GRACE.

Accumulated precipitation in 2019 is more than one standard deviation

above the mean beginning in February and more than two standard

deviations above the mean from mid

‐

May onward (Figure S2). By the

end of June, the cumulative precipitation is ~13 cm higher than the

‐

year average (46 ± 6 cm), a magnitude comparable to the increase

in TWS. As a result, the time by which 90% of the corn and soy was

planted was delayed by approximately 3 weeks compared to 2018

(Figure 2c). Accordingly, the time by which 90% of the crop had

emerged from the soil was delayed by ~3 weeks for corn and ~2 weeks

for soybean (Figure S3), consistent with the SIF observations.

The extended 2019 growing season into late September might have been

favored by increased water availability as suggested by TWS (+10 cm

the decadal mean, Figure 3a) and accumulative precipitation

(Figure S2a), as well as a warmer late growing season (+2.5 °C in

September, Figure S2b).

4. Estimated Reduction in GPP, Crop Productivity Versus Observed

Enhancement

Regions with higher cropland ratio show the largest differences in per

‐

area GPP

uxes relative to 2018

(Figure 3b). For the 17 states located in the Midwest and southern United States along the watershed of

Missouri and Mississippi rivers (see state mask in Figure S4), GPP reduction is noted from May to July

for counties with cropland coverage larger than 10%, with the peak de

cit in late June. In contrast,

there is a recovery in GPP since early August, with a peak compensation occurring in mid

‐

September. Much smaller differences in SIF

‐

based per

‐

area GPP estimates are observed for the natural

vegetation (here de

ned as regions with cropland coverage <10%), suggesting that unmanaged ecosys-

tems are less sensitive to waterlogged soils that have prevented the planting of crops. Given the large

area of noncrop lands over the 17 states, the total GPP anomaly from April to September amounts to

a net gain of 0.04 PgC (+3%). As for the croplands (counties in the 17 states with cropland area

>10%), we estimate the 2019 anomaly to have led to a reduction of 0.21 PgC in June and July GPP,

partially compensated in August and September (+0.14 PgC). Those changes are primarily contributed

by areas with cropland coverage >50% (Table S1). The net effect results in a 0.06 PgC reduction in

the growing season GPP of croplands (

−

4%).

Figure 2.

Seasonal cycles of SIF and crop planting date for the Midwest. (a)

Four

‐

day running average of 2018 and 2019 daily

‐

corrected TROPOMI SIF

over Midwest counties with cropland fraction larger than 50%. The shaded

areas represent 1

‐

sigma standard deviations (

) of the spatial variation

across those counties. The seasonal variation in potential PAR due to change

in solar zenith angle (SZA) is represented by daily integral of cos (SZA) at 10

‐

min time steps. (b) OCO

‐

2 daily

‐

corrected SIF from 2015 to 2019. Note OCO

‐

2 has sparser spatial and temporal sampling relative to TROPOMI. (c)

Percentages of planted area relative to the 2018 total planted area for corn

and soy as reported by USDA.

10.1029/2019AV000140

AGU Advances

YIN ET AL.

7of15

In a separate study, we show strong correlations between county

‐

level

TROPOMI SIF during the growing season (

, mW·m

‐

·nm sr

‐

) and

USDA reported crop productivity (

, g/m

) in 2018, where

= 968x

−

131 (

= 0.72; He et al., in review). The correlation could be further

improved when accounting for planted area and C3/C4 contributions

(

= 0.86). Based on this empirical SIF

‐

crop productivity relationship,

we estimate that the 2019

ood results in ~15% decline of crop yield in

the Midwest counties with cropland fraction larger than 50%

(Figure S5); part of this decline is the result of a decrease in planted crop

area (Table S2). Smaller reductions are noted for counties with less dense

crop distribution; however, as the relative contribution of crops to the

observed SIF signal over a pixel declines, the uncertainty of such estimates

increases. We note that this estimate remains speculative as many other

factors that are not accounted for in the growing season SIF could contri-

bute to the crop yield at the end of the season. For instance, delayed har-

vesting can expose crops to unfavorable weather conditions and reduce

crop yield (Thomison et al., 2011). In addition, the ratios between the

weight of a harvest product and the above ground biomass of the entire

plant (commonly noted as harvest index) also change depending on the

seed and the environment conditions (Hay, 1995). The allocation of car-

bon between the below

‐

ground and above

‐

ground biomass may also

change given the condition of water and nutrient availability (Hay,

1995). Combining SIF observations with crop models may improve the

estimates and forecasting of future crop yield productivity (Guan et al.,

2015; Guanter et al., 2014; Somkuti et al., 2020; Zhang et al., 2014).

Unlike the spatially explicit SIF anomalies, the atmosphere is constantly

being mixed with a zonal mixing on the timescale of around 2 weeks

(Keppel

‐

Aleks et al., 2011, 2012). Hence, we look at differences in the

detrended OCO

‐

CO2

between 2019 and 2018 (noted as Dif_X

CO2

) for

both the zonal mean over the 35

‐

50°N latitudinal band and for the eastern

US domain including the croplands and their downwind area (75

‐

95°W)

(Figure 3c). Coincident with the time when negative SIF anomalies

emerge, an abrupt increase of ~1 ppm is

rst observed in the Dif_X

CO2

over the eastern US domain in late May. Dif_X

CO2

continues to increase

through June and July for both the eastern US and zonal mean domains,

with larger anomaly over the eastern US domain than the zonal mean.

The regional X

CO2

enhancement (i.e. the difference between the U.S.

domain and the zonal mean) reaches 0.86±0.83 ppm during June.

Dif_X

CO2

start to decline in August when

GPP becomes positive. The

alignment in the timing of anomalies of X

CO2

and SIF suggests that

reduced uptake over cropland regions is re

ected in increased X

CO2

downwind. However, with such a sim-

ple comparison, changes in NEE from other regions between 2019 and 2018 could also contribute to the

observed difference in atmospheric CO

. In addition, differences in atmospheric transport between the

two years, as well as in sampling time and location of the observations, could also impact these observed dif-

ferences in regionally averaged X

CO2

5. Consistent Estimates of Crop Anomalies Between SIF and Atmospheric CO

The spatial distribution of the top

‐

down and bottom

‐

up estimates of

for ACT

‐

America and OCO

‐

measurements between 9 June to 18 July 2019 are shown in Figure 4. Bottom

‐

estimates are mostly

positive because

NEE used for the bottom

‐

up simulation only accounts for Midwest counties with cropland

fractions greater than 25%, resulting in reduced net carbon uptake across the region. In contrast, top

‐

down

estimated

are impacted by differences in surface

uxes globally. Nevertheless, positive

values

Figure 3.

(a) Terrestrial water storage from GRACE

‐

FO (dots, June 2018 to

august 2019) and GRACE climatology (2003

–

2014). The shaded gray area

shows the 1

‐

of the decadal variations. Note a data gap from August to

October 2018. (b) Estimated differences in per area GPP between 2019 and

2018 for regions with different cropland area portions. (c) Differences in

detrended OCO

‐

CO2

for the zonal mean between 35°

–

50°N and the

subdomain containing the Corn Belt and downwind areas (de

ned as 75°

–

95°W). The solid lines show the 24

‐

day running mean of each 4

‐

day average,

and the shaded areas show the 1

‐

of the 4

‐

day variations within the 24

‐

day

window.

10.1029/2019AV000140

AGU Advances

YIN ET AL.

8of15