Description of New Soil Code:

We update the YL SNOx parameterization as follows: the soil moisture and temperature

dependence is decoupled allowing for a continuum of SNOx response rather than discrete wet or

dry states; pulsing length and strength is modified to depend on soil moisture history rather than

precipitation; N fertilizer emissions are updated to account for timing and distribution of N

fertilizer based on MODIS derived seasonality; N fertilizer emissions are now allowed to respond

to temperature/soil moisture and pulsing; and wet and dry deposition of ammonia (NH3),

ammonium (NH4+), nitric acid (HNO3), nitrate (NO3

-), nitrogen dioxide (NO2), and peroxyacetyl

nitrate (PAN) are calculated as an additional fertilization effect on SNOx.

1. Soil NOx Parameterization

The current implementation of the YL scheme in GEOS-Chem produces 6.2 Tg N yr-1

for the

year 2006. SNOx is computed as a function of vegetation type (from Olson [1992] map),

temperature, precipitation history, fertilizer use, and a canopy reduction factor:

( )( )[ ] ),()(, // SAILAICRFFertionprecipitatPbiomeATfFluxS dwdwNOx ×+×= (1)

where fw/d is a constant, linear, or exponential function of soil temperature (T) and Aw/d(biome) is

a coefficient to distinguish between vegetation type. The subscript w/d refers to ‘dry’ or ‘wet’

soils (Section 2.1). P(precipitation) is a scaling factor used to adjust the flux during pulsing events

(Section 2.2), and Fert is fertilizer emissions which are set to 2.5% of total fertilizer applied,

evenly emitted over the growing season. CRF(LAI,SAI) is a scaling factor to account for loss of

NOx to plant canopy based on Jacob and Bakwin [1991].

GOME 2000 a posteriori emissions from Jaegle et al., [2005] compared to the YL scheme as

implemented in GEOS-Chem from 2006. While we would expect some interannual variability,

the year to year global totals in GEOS-Chem do not change by more than 5%. Predicted SNOx is

too low over the fertilized midlatitudes and over seasonally wet grassland over Africa and the

southwest United States, which suggests discrepancy in the treatment of moisture and in N

fertilizer. Our new scheme, described below corrects much of these discrepancies and improves

the seasonal and pusling behavior of SNOx.

2.1 Soil Moisture / Soil Temperature Dependence

To account for soil moisture, YL label soil as either ‘dry’ or ‘wet’ based on the prior two

week precipitation and have separate soil temperature dependencies for each. A wet soil is one

that has received in excess of 10 mm of rainfall in the previous two weeks, otherwise, it is dry.

The 2m temperature is used along with experimentally derived coefficients to convert to a soil

temperature as described by Williams et al. [1992] and Johansson et al. [1988]. For wet soils,

emissions are described by a linearly increasing function (with zero intercept) for temperatures

between 0 and 10°C and an exponentially increasing function for temperatures between 10°C and

30°C. For dry soils, emissions are described by a linearly increasing function for temperatures

between 0 and 30°C. In both cases, emissions are scaled by the appropriate biome factor and do

not depend on temperature for temperatures greater than 30°C [Cardenas et al., 1993; Scholes et

al., 1997].

In our revised model we decouple the relationship of soil moisture and T. Ormeci et al.

[1999] and Otter et al. [1999] report that there exists an entirely exponential relationship between

T and SNOx in the range 0°C to 30°C. Thus, we use the wet biome factors (Aw) from Yienger and

Levy [1995] with an exponential dependence on soil temperature (T) between 0°C and 30°C

(constant at T > 30) (equation 2), which we then scale using soil moisture as discussed below.

( )( )[ ] scalingeAsngNmbiomeATfT

www ××=×− 103.012, (2)

We also modify the YL scheme to allow tropical forests to have temperature/soil moisture

dependence with a wet biome factor of 0.2 based on recent work of 12 field studies, whereas this

flux was held constant in YL scheme [Yan et al., 2005].

Soil moisture data is now available within the meteorological fields of the GEOS model. We

use this as the explicit parameter in place of rainfall. Soil moisture is best described in terms of

water filled pore space (WFPS). WFPS is defined as the ratio of the volumetric soil moisture

content to the porosity [Linn and Doran, 1984]. Dividing by porosity acts as a normalizing step

that makes WFPS (θ) satisfy 0 ≤ θ ≤1, allowing comparison between soils of different textures

[Otter et al., 1999]. In the GEOS meteorological fields, WFPS is available for the top 2 cm of soil,

where the majority of SNOx originate [Pierce and Aneja, 2000], as a 3-hrly time/area-mean.

The response of SNOx is not monotonic to WFPS. SNOx are low for the extreme values of

WFPS (0 and 1). For low values, emissions are substrate-limited. For high values, emissions are

trapped and cannot diffuse to the surface [Yan et al., 2005]. SNOx dependence on soil moisture is

best described as a Poisson function [Parsons et al., 1996; Otter et al., 1999; Pierce and Aneja,

2000; Kirkman et al., 2001; van Dijk and Meixner, 2001; van Dijk et al., 2002] (see equation 2):

2θ

θ beaScaling

−= (3)

where the values of a and b are chosen such that the maximum value (unity) occurs for θ=0.3,

which laboratory and field measurements have found to be the optimal value for emissions in

most soils. The typical range of values are 0.2 (arid) up to 0.45 (floodplain) [Yang and Meixner,

1997; Ormeci et al., 1999].

The implementation of the soil temperature/soil moisture treatment leads to some

pronounced differences. The changes result in a decrease of 1 Tg N yr-1

to 5.2 TgN/yr versus 6.2

TgN/yr in the former inventory (Figure 2). The largest uncertainty associated with this

implementation is the choice of the optimal value of θ. Although the range is from 0.2 to 0.45, the

median is closer to the lower extreme value. Floodplains are the only soil type that reaches the

optimal value of 0.45 and it represents a small amount of global soils, so emissions are relatively

insensitive to varying the optimal value of θ over 0.2-0.45 (<5%). The biggest effect of this

treatment is the accounting for very wet and inundated soils (large values of θ), which during the

wet season drastically reduces the SNOx and improves comparison to observations.

Largest changes take place in the tropics where WFPS span a large-range due to the

migration of the Intertropical Convergence Zone (ITCZ). Unlike the former wet/dry treatment,

emissions have a continuous dependence on soil moisture. This causes an increase in emissions

during the wet season for lands near the wet/dry transition in the former treatment and a decrease

in emissions for lands with high values of θ. Overall, we see that the latter effect is more

dominant, causing a decrease in the emissions total. In the northern tropics, differences are seen

over the African Sahel changing seasonality and over the Amazon forests (±100%). Decreases in

emissions over the Amazon, now subject to fluctuations based on temperature and soil moisture,

are consistent with decreases predicted by GOME a posteriori. In the southern hemisphere,

largest differences are seen over the grasslands of South America, Africa, and Australia. Over

North America, increases are predicted over the monsoonal southwest and Great Plains consistent

with GOME constraints.

2.2 Pulsing

Pulsed SNOx emissions occur when very dry soil is wetted resulting in a reactivation of water

stressed bacteria or over freshly fertilized fields. Pulsing is characterized by heightened NOx

emissions following the wetting event which subsequently decay. YL assume the magnitude and

duration of a pulse to be a function of rainfall rate over dry soils. Land is defined as either ‘dry’

or ‘wet’ (as described in section 2.1) to determine if pulsing can occur. The value of the scaling

factor (P, see equation 1) increases with amount of rainfall received. YL use three rainfall

scenarios where P is either 5, 10, or 15 and last 3 days, 7 days, and 14 days, respectively. The

value of P decays back to 1 over the pulse duration. More recent studies, however, suggest that P

may be stronger and of shorter duration [Yan et al., 2005; Hudman et al., 2010]. Here, we follow

the parameterization implemented by Yan et al. [2005]:

bt

peak ePP−

×= (6)

where Ppeak is the magnitude of the peak flux relative to the pre-wetting flux, b is a rate

constant (b = 0.068h−1

), and t is time (h) since initiation of the pulse. The value of b is an average

of four calculated values that range from 0.050 h−1

to 0.110 h−1

. Ppeak is given by equation 7:

( ) 6.53ln01.13 −= drypeak tP (7)

where tdry is the antecedent dry period in hours. The two main differences between this treatment

and that used by Yienger and Levy [1995] is that Ppeak depends logarithmically on the antecedent

dry period and the condition for a pulse is a change in soil moisture rather than rainfall. The

advantage of the latter is that soil moisture is a more relevant parameter since it describes the

environment of the NOx-producing microbial biomass. We use the two-part condition described

in Yan et al. [2005] to check for pulsing potential. The dry period is defined as time since

volumetric soil moisture content decreased to less than 17.5% (v/v). A pulse occurs when there is

a soil moisture increase of 0.5% (v/v). Assuming soil bulk density of 1.4 Mg m-3

(typical of

seasonally dry savannahs), this is equivalent to WFPS ~ 0.3 and a ∆WFPS > 0.01, which we use

here.

Using this scheme increases SNOx over seasonally wet grass lands. Because the new

pulsing treatment is coupled to changes in the soil moisture/ soil temperature fluxes calculated in

section 2.1, we compare changes to those fluxes. The new pulsing treatment increases SNOx by +1

Tg N yr-1

to 6.2 TgN/yr. Figure 3 shows seasonal changes in SNOx due to the new treatment of

pulsing. Largest increases are seen over the African Sahel during MAM and JJA, which

corresponds to the onset of the wet season. First rains reactivate bacteria water-stressed from the

long dry season, releasing NO as a byproduct [Davidson et al., 1992]. As the excess N is

consumed, NO emissions remain high compared to dry season [Serca et al., 1998]. A similar

response is seen over the savannahs/grasslands of South America and Australia. Because the

original GEOS-Chem parameterization used climatological precipitation to determine the ‘wet’

and ‘dry’ criteria of soils, the new parameterization has a more realistic timing of the onset of the

dry and wet season as well as allowing for drying out of soils within the wet season, which

improves both magnitude and timing of pulsing events compared with observations.

2.3 N Fertilizer Treatment

We use a new spatially explicit (native resolution 0.5°x0.5°) chemical fertilizer (70 Tg N yr-1

)

and manure (128 Tg N yr-1

) dataset from Potter et al., [2010] valid for the year 2000 (available at

http://www.geog.mcgill.ca/~nramankutty/Datasets/Datasets.html). For manure emissions we

assume 37% of manure N, 47 Tg N yr-1

, returns to the soil as N input [Sheldrick et al., 2003]. To

introduce timing, the satellite instruments MODIS (Moderate Resolution Imaging Spectrometer)

and TRMM (Tropical Rainfall Measuring Mission) are used to give information regarding the

beginning (green-up) and end (brown-down) of the growing season of each model grid square.

Huete et al. [2002] describe deriving green-up and brown-down dates using a timeseries of

enhanced vegetation index (EVI). We use a global data set of green-up and brown-down dates

averaged over 2001 to 2004 regridded to the GEOS-Chem model to define the beginning and end

of the growing season respectively (Figure 4) [Mark Friedl, unpublished results]. Fertilizer is no

longer applied evenly over the growing season [Yeinger and Levy, 1995]. Instead, 75% of the

yearly fertilization amount is applied over the first month as a Gaussian distribution around the

green-up day and the remaining 25% is applied evenly over the remaining time in the growing

season. This 75/25 treatment is the most typical global farming practice [Matson et al., 1998].

The need for the distribution over the first month is due to varying farming practices.

To determine the dynamic N fertilizer available in the soil, we solve the mass-balance

equation

)1()0()( ττ τ ×−×−−××+=

tt

availavail eSeNtN (4)

where Navail is the mass of available nitrogen in the soil (ng N m2), S is the application rate, and τ

is a decay constant. Based on measurements within the top 10 cm of soil, τ is chosen as 4 months,

with values in the literature for agricultural soils ranging from 2 months to 7 months [Matson et

al., 1998; Chen et al., 2004; Russell et al., 2006]. The value of S varies over the growing seasonas

described above. Upon the brown-down day, the value of S is zero and the remaining N fertilizer

in the soil is left to decay.

In GEOS-Chem, fertilizer emissions were emitted instantaneously as 2.5% of applied

fertilizer, independent of soil moisture/soil temperature, so that they were constant over the

growing season. Similar to the YL parameterization, we now treat fertilizer emissions as part of

the Aw in equation 1. If we treat the wet biome coefficient as a measure of available N multiplied

by a mean emission rate, we can treat fertilizer N in the same manner.

[ ] ×+=−

availww NbiomeAsngNmtotalA )(12mean emission rate (5)

Instead of choosing an emission rate for each box equivalent to 2.5% of applied N yearly as done

in the YL scheme, we chose the mean emission rate so that the total global above canopy SNOx

due to fertilizer matches observed estimates of fertilizer emissions of 1.8 Tg N yr-1

from Stehfest

and Bouman [2006], which in GEOS-Chem is 0.62% of available N. This treatment allows for

interannual and daily variability in the strength of response to temperature and precipitation.

2.4 Dry/Wet Deposition

Online wet and dry deposition rates of NH3, NH4, HNO3, NO3-, NO2, and PAN are archived

each dynamic timestep [Liu et al., 2001]. We assume, 60% of this deposited N enters the soil,

with continental values in literature ranging from 55% to 80% [Gleick , 1993]. The remainder is

lost to runoff into waterways. Available N in soil is then calculated as with fertilizer (Equation 4).

The decay constant (τ) is chosen to be 6 months based on measurements made over lands

with natural vegetation, with measurements ranging from 4 months to 1 year [Hart et al.,

1993; Nadelhoffer et al., 1995].

Soil Model Update

GOME EMISSIONS LARGER THAN ORIGINAL MODEL

GOME Total: 8.9 Tg N yr-1

Orig Model:6.2 Tg N yr-1

Red regions are where GOME emissions > original model

Particularly at N. Midlatitudes and N. Tropical Africa

GOME – ORIGINAL YIENGER AND LEVY [1995] MODEL

ENOx = f( T, biome, w/d) x Pulse (dryspell) x canopy uptake + FERT

AN UPDATED GLOBAL MODEL OF SOIL NOx

IMPROVEMENTS:

•Update Fertilizer: new maps (include N deposition), MODIS EVI seasonality and treat like other N (impact = +1.3 Tg N/yr)

•Update Pulsing Scheme: Yan et al., [2005] (shorter, stronger pulses) (impact = +1. Tg N/yr)

•Update moisture treatment: soil moisture as a continuous variable (impact = -1. Tg N/yr)

ENOx = f( T, biome, WFPS, Fert) x Pulse (dryspell) x canopy uptake

Overall 6.2 ���� 7.5 Tg N/yr

OLD

NEW

GOME Total: 8.9 Tg N yr-1

New Model:7.5 Tg N yr-1

NEW MODEL MATCHES WITH GOME OBSERVATIONS IN A BROAD SENSE

Red regions are increases with updated model

NEW MODEL – ORIGINAL YIENGER AND LEVY [1995] MODEL

INCREASES EMISSIONS BY 50% IN SUMMER & BETTER CAPTURES SEASONALITY

•Update Fertilizer: new maps (include N deposition), MODIS EVI seasonality and treat like other N

NEW MODEL COMPARES WELL WITH PREVIOUS US CONSTRAINTS

[Hudman et al., 2010] This work

COMPARISON TO SELECT SURFACE OBSERVATIONS New model better matches surface obs

0.6-136

(min/max)

*6-60

(compilation

of obs)

African Sahel (rainy season

onset mid-May – June)

7-35*8-36AMMA (Sahel, August 6,

2006)

138African Sahel (max monthly

mean)

0-2

7

10

6-8

Old Model

(ng N m2/s)

0-30.2-5Colorado grass (natural) JJA

1010South Dakota (fertilized) JJA

1718 Iowa (fertilized) JJA

11-1512-43Texas Grass (fertilized) JJA

New Model

(ng N m2/s)

Obs(ng N m2/s)

Location/Timing

MODEL IMPROVES MAGNITUDE BUT SEASONALITY SHIFTED , PEAKS EARLIER

•Update moisture treatment: soil moisture as a continuous variable (impact = -1. Tg N/yr)