Contributions of long-distance dust transport to atmospheric P inputs in the Yucatan Peninsula



[1] Atmospheric deposition is not typically considered in conceptual models of P cycling in terrestrial ecosystems, but in P-limited tropical forests that receive significant inputs of dust, it may play an important role in sustaining ecosystem productivity. We used models and observations to quantify total atmospheric P inputs and the contribution of long-distance dust transport to tropical dry forests in the Yucatan peninsula over a 10 year period. Total atmospheric P input was estimated from atmospheric bulk deposition sampling in the southern Yucatan peninsula in 2007 and 2010–2011, and P input from dust deposition was estimated using two independent methods: zonal dust flux divergence based on MODIS AOD retrievals, and MATCH atmospheric transport modeling. Total atmospheric P input was 265 ± 80 g P ha-1 yr-1, and dust P input was 46 ± 12 g P ha-1 yr-1. There was significant seasonal and interannual variation, with local biomass burning accounting for high P inputs in April and May, and dust transport from June through August. We found that MATCH underestimates P deposition from dust to the Yucatan because of underestimation of dust transport relative to MODIS remote sensing. Dust accounted for 25% of total atmospheric P inputs in the Yucatan, indicating local sources dominate atmospheric inputs. However, dust is still an important source of P for tropical dry forests in the Yucatan since it is a new input of P that occurs during the early growing season of the forest, and is large enough to offset leaching and erosional losses from soils.

1 Introduction

[2] Atmospheric deposition of dust and other aerosols is a source of many nutrients, including phosphorus (P) to terrestrial ecosystems [Newman, 1995; Bruijnzeel, 1991]. However, the established conceptual model of terrestrial P cycling considers parent material as the only source of P [Crews et al., 1995; Walker and Syers, 1976], and studies of terrestrial P cycling typically focus on bedrock-derived inputs of P [Vitousek and Sanford, 1986]. This paradigm of terrestrial P cycling postulates that as soils get older, the availability of mineral P declines through leaching, erosion and chemical occlusion, and through time reduced P availability limits ecosystem productivity [Walker and Syers, 1976]. Many tropical forests occur on old, highly weathered soils, and there is widespread evidence of P limitation in these ecosystems [Reed et al., 2011; Vitousek et al., 2010; Cross and Schlesinger, 1995]. Several studies have suggested that atmospheric deposition may be an important source of new P inputs, especially for P-limited tropical forests downwind of global dust source regions [Okin et al., 2004; Chadwick et al., 1999; Swap et al., 1992]. However, these inputs remain poorly quantified, especially in the tropics, and consequently there is uncertainty about how much they vary and whether they are important in sustaining tropical forest ecosystems. Here we use models and observations to quantify atmospheric P inputs from dust, over a 10 year period, to a tropical dry forest region in the Yucatan peninsula, assess their importance relative to other local sources of atmospheric P inputs, and discuss the significance of long-distance dust inputs to P cycling in these forests.

[3] Atmospheric aerosols play an important role in a number of planetary processes, altering the atmospheric radiative balance [Kaufman et al., 2005; Miller and Tegen, 1998], potentially affecting tropical storm and hurricane activity [Evan et al., 2009, 2011], and providing a variety of nutrients to terrestrial and marine regions [Mahowald et al., 2008; Okin et al., 2004]. Dust is the predominant source of P in the atmosphere [Mahowald et al., 2008], and the largest source globally are the Sahara and Sahel regions of North Africa, with an estimated 240 to 1600 Tg of dust emitted annually from the region, and 140 to 259 Tg transported and deposited across the Atlantic “dust belt” of 0–30°N [Engelstaedter et al., 2006; Prospero et al., 2002]. Since forests of the neotropics are found within this latitudinal range, Saharan dust has been suggested as an important source of P to forests in the Amazon basin and across the Caribbean [Pett-Ridge, 2009; Okin et al., 2004; Swap et al., 1992]. However, few multiyear measurements or estimates of atmospheric P inputs, and specifically contributions from dust, have been made in the neotropics [Pett-Ridge, 2009].

[4] The Yucatan peninsula of Mexico is located within the global “dust belt,” and although it is further downwind of the African dust source compared to regions like the Caribbean and Amazon basin, it is longitudinally close to Florida, where Saharan dust transport is well documented [Prospero et al., 2010; Prospero, 1996]. Despite being downwind of the largest global dust source, atmospheric P inputs to the Yucatan peninsula have never been estimated, and the contribution of dust to terrestrial P cycling in the region is unknown. Studies of soils and geology in the Yucatan have suggested the possible importance of Saharan dust and other eolian inputs to soil development in the region [Cabadas et al., 2010; Perry et al., 2003], but no estimates of dust inputs to the region exist. Soils in the Yucatan peninsula are similar to thin soils on carbonate platforms in Caribbean islands where Saharan dust inputs have been shown to be a major source of soil development [Borg and Banner, 1996; Muhs et al., 1990]. The Yucatan peninsula has large areas of tropical dry forest that are P-limited [Lawrence et al., 2007; Campo and Vazquez-Yanes, 2004; Read and Lawrence, 2003], and dust inputs may represent an important source of P to these ecosystems. Few measurements of atmospheric P inputs to tropical dry forests have been made, with the majority of tropical forests studied being located in the wet tropics [Das et al., 2011; Newman, 1995; Bruijnzeel, 1991]. In the Yucatan peninsula, atmospheric P inputs are likely from a number of different sources, with spatial and temporal variations in inputs from different sources. Rainfall in the region occurs between June and November, with the remaining part of the year being mostly dry. Widespread biomass burning for agriculture occurs across the region in April and May, and there is very little fire activity during the rest of the year, especially during the wet season. Saharan dust transport to the region would occur between June and August, and is determined primarily by the direction of the trade winds as they are influenced by the ITCZ [Prospero, 1996].

[5] Estimating atmospheric P inputs from dust is challenging for a number of reasons. Field sampling of atmospheric deposition is limited to small spatial and temporal scales, and does not adequately represent the complexity and heterogeneity of deposition processes at the landscape or regional scale. It is also difficult to separate the contribution of dust from other sources such as biomass burning, industrial/urban sources and biogenic particles that occur on regional scales [Pett-Ridge, 2009; Redfield, 2002]. Dust inputs may be spatially and temporally heterogeneous, with large proportions of deposition occurring in a small number of events with large variability between sampling points [Prospero et al., 1987; Kellman et al., 1982]. Thus obtaining sufficient sample sizes in space and time is always a challenge for field estimates. Remote sensing platforms like Terra and Aqua that have the Moderate Resolution Imaging Spectrometer (MODIS) [Kaufman et al., 2005] or ground-based sun photometers like the Aerosol Robotic Network (AERONET) [Holben et al., 1998] provide measurements of aerosol optical depth (AOD), which is related to the mass of dust and other atmospheric aerosols, and may be used to infer deposition. However, methods for estimating deposition directly from AOD retrievals have relatively large errors [Kaufman et al., 2005] that result from several uncertainties and assumptions in the calculation of dust deposition from AOD [Ben-Ami et al., 2010]. Atmospheric transport models are a third method of estimating global dust emission, transport and deposition that are currently being widely applied [e.g., Mahowald et al., 2003]. Models provide the advantage of explicitly simulating the movement and deposition of dust and other aerosols [Mahowald et al., 2008]. However, modeled estimates vary greatly and sometimes differ from field measurements of dust deposition [Prospero et al., 2010].

[6] Here we estimate atmospheric P deposition to the Yucatan peninsula using three methods—field measurements of atmospheric P deposition, MODIS remote sensing-based estimates of dust P inputs, and an atmospheric transport model-based estimate. We compare the three estimates and discuss the major differences between them, and conclude with the significance of atmospheric P inputs to tropical dry forests in the Yucatan.

2 Methods and Data

[7] Since there are limitations associated with any one method of estimating atmospheric P input and the contributions from dust, we used three independent methods for estimating seasonal and annual atmospheric P input to the Yucatan peninsula from 2000 to 2010: (1) field sampling and measurement of total P (inorganic + organic) in atmospheric bulk deposition at a site in the southern Yucatan peninsula, (2) MODIS remote sensing of dust and other aerosol optical depth, and (3) simulation of dust transport and deposition using the MATCH atmospheric chemical transport model. Each of the three approaches is described in detail below.

2.1 Field Sampling Estimate

[8] Atmospheric bulk deposition (wet + dry) was sampled from May–November 2007 and March 2010 to May 2011 in the village of El Refugio, Quintana Roo, in the southern Yucatan peninsula region of Mexico (18.80°N, 89.38°W; 250 m elevation) (Figure 1). Mean annual rainfall is 892 mm, most of which occurs during the wet season (May–October) rainfall, and very scant rainfall during the dry season (November–April) [Lawrence, 2005]. Five collectors, 15 l LDPE buckets with a top surface area of 700 cm2, were placed in a circle in an open field, 30 cm above the ground and covered with a single layer of 1 mm nylon mesh screen to exclude litter and debris. After each rainfall event, the volume of water collected in each bucket was measured, and a representative sample was collected after stirring each bucket completely. Samples were not filtered, and frozen soon after collection and brought back to the University of Virginia for analysis. Samples were analyzed for dissolved inorganic P (Pi) and total P (Pt) by colorimetric analysis [Murphy and Riley, 1962] on a Lachat Quikchem 8500. For the Pt measurement, samples were digested with potassium persulfate [Hosomi and Sudo, 1986]. Organic P (Po) was determined as the difference between Pt and Pi concentrations. Pi and Po concentrations were multiplied by the volumes for the respective collectors and rain events to obtain inputs for each collector and event, and inputs were scaled to units of g P.ha-1. For 2007, the wet season (May–November) estimate was scaled up to an annual estimate using a proportional scaling based on the 2010 observations, using the ratio of wet season to dry season total deposition.

Figure 1.

Map of the western Caribbean and Yucatan peninsula. Locations of field sampling of bulk deposition was conducted at El Refugio, and the AERONET station at Tuxtla-Gutierrez, are indicated. Spatial extent of MODIS AOD products used to estimate dust transport is indicated with the dashed box. MATCH model output points over the Yucatan peninsula only were used to estimate dust deposition. MODIS products have a resolution of 1° × 1°, and MATCH dust deposition estimates have a resolution of 1.8° × 1.8°.

2.2 Remote Sensing Estimate Using MODIS AOD

[9] Moderate Resolution Imaging Spectrometer (MODIS) instruments aboard the Terra and Aqua satellites provide daily global retrievals of aerosol optical depth (AOD) [Remer et al., 2005] from 2000 through the present. MODIS AOD can be subdivided into the contribution from fine aerosols like smoke from biomass burning and sulfate aerosols from urban areas (effective radius 0.1–0.25 µm) and coarse aerosols like maritime and dust (effective radius 1–2.5 µm), using retrievals of so-called fine mode fraction [Kaufman et al., 2005]. We used Collection 5 Level 3 MODIS AOD products (MOD08/MYD08) that are available over land and over ocean at daily and monthly temporal resolutions, and at 1° × 1°spatial resolution, which are currently the highest quality, most reliable MODIS AOD retrievals available. Collection 5 Level 3 MODIS products were produced with the Giovanni online data system, developed and maintained by NASA GES DISC (see

2.3 MODIS AOD Validation

[10] We validated MODIS AOD and the contribution by coarse mode aerosols over the Yucatan peninsula by comparing daily MODIS data to AERONET retrievals. AERONET is a global network of ground-based sun photometer instruments that provide direct measurements of AOD [Holben et al., 1998]. AERONET Level 2 (quality controlled) retrievals of AOD and fine mode AOD at the station of Tuxtla-Gutierrez (AERONET-TGZ) (17°N, 93°W; Figure 1) were matched with MOD08/MYD08 AOD retrievals over the AERONET TGZ station at the time of satellite overpass. Multiple AERONET observations within a 30 min window of the MODIS overpass were averaged for the validation. AERONET-TGZ AOD and fine mode fraction retrievals from 2005 to 2009 were used for the validation, and a total of 465 matched AERONET and MODIS daily retrievals during this period were used. MODIS and AERONET AOD observations were highly positively correlated (r2 = 0.72), and the slope of the linear regression of MODIS against AERONET AOD retrievals was 0.82, suggesting that MODIS retrievals are biased low (Figure 2). This result is in agreement with other methods of validating global MODIS AOD over ocean [Zhang and Reid, 2006] and over land [Hyer et al., 2010].

Figure 2.

Plot of MODIS daily AOD retrievals over AERONET station Tuxtla-Gutierrez against AERONET AOD at the same station (17°N, 93°W). The linear correlation is indicated by the solid line and the 1:1 line is indicated by the dashed line.

2.4 MODIS Monthly P Deposition Calculation

[11] We used MODIS monthly AOD products to calculate dust aerosol transport and deposition across a region with dimensions 17–22°N and 80–92°W, corresponding to the western Caribbean and the Yucatan peninsula (Figure 1). Level 3 MODIS monthly AOD is produced by averaging L3 daily retrievals of AOD, and represents the average monthly values of AOD for a location for 1 month. We used monthly average MODIS products because daily retrievals are limited to approximately one half of the cross-track scan area due to scattering geometry. We used the method described in Kaufman et al. [2005] to calculate dust transport and deposition. MODIS monthly average AOD and fine mode fraction for the period February 2000 to May 2011 were produced using the Giovanni online data system. Monthly average Dust AOD over water (AODD) was calculated via Kaufman et al. [2005],

display math(1)

where AOD is the monthly average total AOD, fmf is fine mode fraction, and AODM is the contribution to total AOD by marine aerosols, which is defined as

display math(2)

where W is the monthly average 10 m wind speed (m/s) [Kaufman et al., 2005]. Wind speed data was obtained from National Center for Environmental Prediction/National Center for Atmospheric Research (NNRP) Reanalysis 1 [Kalnay et al., 1996]. Dust mass path (DMP), which is the mass of dust (g) per unit of atmospheric column (m2) was calculated as

display math(3)

where the ratio of 2.7 ± 0.4 g/m2 is an empirically derived estimate of the ratio of dust mass path to dust AOD [Kaufman et al., 2005]. This was used to determine the dust load in the atmosphere at different longitudes. A zonal dust flux (F) was calculated along meridional transects at 80°W, 86°W and 92°W by multiplying the monthly dust mass path by the monthly average westward wind speed at 700 hPa (W700) and the longitudinal length of the meridional segment (L),

display math(4)

where W700 was obtained from NNRP wind data. Finally, monthly average dust deposition rates were estimated by calculating zonal flux divergence, assuming that decreasing flux in the westward direction is due to deposition to the surface, by continuity. We only considered divergence in zonal direction because meridional wind movement is minor, especially between May and August, when African dust transport to the region occurs. The overall uncertainty in this method for estimating deposition is reported to be 35%, based on the combined uncertainties in dust column concentrations and wind speed calculations [Kaufman et al., 2005]. The flux divergence from 86°W to 92°W correspond to the deposition over the Yucatan peninsula.

[12] Average monthly dust deposition rates were converted into monthly cumulative dust deposition amounts and multiplied by the average P concentration of mineral dust, 700 ppm [Mahowald et al., 2008; Okin et al., 2004], to obtain the monthly P inputs from dust.

2.5 MATCH Transport Modeling

[13] We used data from a published simulation of the chemical transport model, the Model of Atmospheric Transport and Chemistry (MATCH, v.4.2), driven by offline NNRP winds [Mahowald et al., 2008]. The simulation of dust and mineral aerosols within MATCH was done using the Dust Entrainment and Deposition model [Mahowald et al., 2008; Zender et al., 2003]. The horizontal spatial resolution of the model is T62 (~1.8°), and monthly average dust deposition rates over the Yucatan for the January 2000 – December 2010 from the model output were converted into cumulative dust deposition amounts. Monthly dust deposition from the model was multiplied by 700 ppm as with the MODIS estimate to obtain monthly atmospheric P inputs. Detailed description of the model setup may be found in Mahowald et al. [2008] and Mahowald and Luo [2003].

3 Results

[14] Mean annual atmospheric P deposition from the field estimate is 265 ± 80 g P.ha-1 yr-1, of which 173 ± 73 g P.ha-1 yr-1 is inorganic P (Pi) and 92 ± 7 g P ha-1 yr-1 is organic P (Po) input. The MODIS estimate of mean annual P input from dust deposition is 46 ± 12 g P ha-1 yr-1 and the MATCH estimate of mean annual P input from dust deposition is 12 ± 2 g P ha-1 yr-1 (Figure 3). Dust inputs are assumed to be inorganic P only. Field observations suggest mean Pi deposition rates that are 2 to 7 times greater than the MODIS estimate, accounting for standard error, and 7 to 24 times greater than the MATCH estimate of dust P deposition rates.

Figure 3.

Estimates of annual mean atmospheric P input for the Yucatan peninsula. Plotted (from L to R) are the field estimate of atmospheric bulk deposition with inorganic P (Pi) and organic P (Po) shown separately, and MODIS and MATCH estimates of inorganic P inputs from dust. Error bars indicate standard error.

[15] The field estimate of P input showed more interannual variability than the MODIS or MATCH estimates. The field estimate is 185 g P ha-1 yr-1 for 2007–2008 and 345 g P ha-1 yr-1 for 2010–2011, with inorganic P input being more variable (100 – 246 g P ha-1 yr-1) than organic P (85 – 99 g P ha−1 yr−1). The MODIS estimate of annual dust P input is 33 – 75 g P ha-1 yr-1, and the MATCH estimate of dust P input varies from 4 – 21 g P ha-1 yr-1 (Figure 4). The MODIS estimate of dust P input is consistently higher than the MATCH estimate, ranging from two-fold to almost an order of magnitude difference between satellite and model-based estimates. Although the absolute values of P deposition from the satellite and model estimates are different, there is some consistency between the two trends; 2008 was the year with the highest annual P input for both estimates, and 2004 was the year with the lowest P input for both estimates, and the two estimates were significantly positively correlated (r2 = 0.47, p < 0.05).

Figure 4.

Annual estimates of P inputs from field sampling of bulk deposition, and dust P inputs based on MODIS and MATCH methods. Field estimates are broken down into inorganic (Pi) and organic (Po) inputs. MATCH estimate for 2010 was not included because of a problem with the simulation. Error bars indicate uncertainty in MODIS estimate.

[16] We consider the discrepancies between the different estimates of atmospheric P deposition at length in the discussion, but it is worth noting here that the field estimate includes all sources of atmospheric P inputs, remote and local. Local sources of atmospheric P include local soil dust, regional biomass burning which occurs in April and May, and biogenic particles such as pollen, microbes and spores that are actively generated and deposited through the wet season, from June through November. During the remaining 5 months of the year we expect low atmospheric P inputs because of reduced sources of P and lower removal rates. These patterns are apparent in the seasonal variation in different types of aerosol optical depth from MODIS data (Figure 5). Fine aerosols such as smoke from biomass burning peak annually in April or May, while coarse aerosols such as dust peak in July and August. Interannual variation in atmospheric P deposition is likely driven by variation in biomass burning and other regional sources as well as dust transport dynamics.

Figure 5.

Monthly mean time series plot of MODIS fine mode and coarse mode AOD over the Yucatan peninsula for 2000–2010. Fine AOD indicates biomass burning and anthropogenic aerosols, and coarse AOD indicates marine and dust aerosols. Fine AOD reflects fire activity, and high values in April 2003, April 2005, and April 2008 indicate especially active fire years in the Yucatan.

[17] Total atmospheric P inputs and dust P inputs show distinct seasonality (Figure 6). The field estimate of total atmospheric P had a bimodal distribution, with peaks in April and September, and inputs larger than 25 g P/ha/month from May through September. Eighty percent of the annual P input in atmospheric bulk deposition occurs between April and October, which are the months immediately preceding and including the wet season in the region. The MATCH and MODIS estimates have single peaks in June (MODIS) or July (MATCH), and significant P deposition (2–15 g P.ha−−1) from June until August. For the remaining months of the year, both MATCH and MODIS estimates of dust P inputs were negligible (<1 g P.ha−−1). This timing of the peak in satellite and model-based estimates of P inputs from dust is consistent with the seasonality of long-distance dust transport to the region, which is characteristically from May through August (Prospero et al., 1996). The MODIS estimate is about half the size of the field estimate of inorganic P input for June and July, but for all other months the satellite estimates of dust P input are an order of magnitude smaller than the field estimate. The MATCH estimates are 15–30% of the MODIS estimates of dust P input for all months. The ratio of inorganic P to organic P also appeared to vary seasonally (Figure 6). From March until June, organic P was a larger proportion of total P in atmospheric bulk deposition. During these months, organic P was ≥ 40% of total P, while it was 25 – 35% for the other months.

Figure 6.

Seasonal variability of estimates of atmospheric P inputs from field sampling of bulk deposition, and MODIS and MATCH estimates of dust P inputs. The field estimate is broken down into inorganic (Pi) and organic (Po) inputs.

4 Discussion

[18] We expected to see the observed differences between the three estimates of atmospheric P deposition to the Yucatan for several reasons. There is a wide range in the field estimate, and this uncertainty is partly due to the limited spatial and temporal nature of the field sampling effort. The field estimate also includes P deposition from all sources, local as well as long distance, and consequently is greater than the MATCH and MODIS estimates, which are of P inputs from dust transport only. Atmospheric bulk deposition sampling with passive collectors tends to have higher estimates of P inputs from atmospheric deposition compared to micrometeorological sampling and atmospheric transport modeling methods, particularly because large particles (>2 µm) such as soils and biogenic debris and giant particles (>10 µm) tend to get deposited at higher rates to bulk deposition collectors [Redfield, 2002]. These larger particles tend to convey larger amounts of P in dry deposition fluxes [Redfield, 2002; Ruijrok et al., 1995]. We did not implement any of the precautions to reduce fluxes of large particles from local sources [Tsukuda et al., 2006], because the objective of the field sampling campaign was to assess total atmospheric P deposition from all sources. The samples were also not filtered, and so the measurement of P in the field estimate includes the contribution of large particles. The MATCH and MODIS estimates are of deposition of dust particles <10 µm [Mahowald et al., 2008; Kaufman et al., 2005], which convey a smaller proportion of the total atmospheric deposition, and this may account for the order of magnitude difference between these types of estimates.

[19] The MODIS and MATCH estimates of annual dust P inputs also differ significantly, with MATCH estimates being about 30% of MODIS estimates on average. We found the satellite-based estimate to be larger primarily because the dust AOD retrieved by MODIS is higher than the dust AOD values in MATCH, rather than the differences in methods of calculating dust fluxes and deposition. We applied the Kaufman et al. [2005] method for calculating dust deposition using dust AOD values generated by MATCH and found strong correlation between these values and the model results for dust deposition (r2 = 0.91, p < 0.001) (Figure 7). This suggests that the difference between the satellite and model estimates is mainly because the model underestimates the amount of dust reaching the Yucatan peninsula. The Yucatan is over 7000 km from Saharan dust sources, and recent comparisons of various atmospheric transport models have found that they tend to underestimate dust deposition with increasing distance from dust source regions [Huneeus et al., 2010]. Another comparison of field observations of dust deposition with nine dust transport models in Florida found that while the models reproduced seasonal patterns well, they underestimated dust deposition by 60 – 80%, especially during the summer months when dust transport was greatest [Prospero et al., 2010]. The Yucatan is at a similar distance to Florida relative to African dust sources, and there appears to be similar discrepancy between the satellite and model estimates of dust deposition for the Yucatan.

Figure 7.

Comparison of MATCH model results for monthly dust deposition in the Yucatan with a monthly dust deposition calculation using the dust AOD values from MATCH with the flux divergence method of Kaufman et al. [2005].

[20] The field estimate of total atmospheric P deposition is generally within the range of global estimates of P in atmospheric deposition (Table 1). The MODIS and MATCH estimates of dust P inputs also fall within the global range of estimates for P inputs from dust deposition (Table 1). MATCH underestimates dust transport, as is evident in years like 2004 and 2010, when annual dust P inputs were less than 4 g P ha-1 yr-1, which is lower than estimates for Hawaii, which is relatively more remote and expected to have much lower dust deposition [Kurtz et al., 2001]. Estimates of dust P inputs for Puerto Rico, Barbados and the eastern Caribbean range from 11 to 300 g ha-1 yr-1 [Pett-Ridge, 2009; Mahowald et al., 2008; Prospero et al., 1996]. The western Caribbean and Yucatan are 20°–30° west, and downwind of African dust sources, and we expect less dust to reach the region. P inputs from dust in Florida are 16 g P ha-1 yr-1 [Prospero et al., 2010], and we expect the Yucatan to receive slightly higher inputs since it is close in longitude but further south in latitude.

Table 1. Comparison of Total Atmospheric P Deposition and P From Dust Deposition From This Study With Global Estimatesa
LocationAnnual Total P Deposition (g P ha-1 yr-1)Annual Dust P Deposition (g P ha-1 yr-1)Citation
  • a

    Numbers are all in g P ha-1 yr-1, and are mean values reported in studies, with ranges indicated below in parentheses, where available. Single means and ranges were calculated for multiple locations in Puerto Rico, Florida and the Amazon basin.

  • *

    no estimate.

Puerto Rico350230Pett-Ridge [2009]
(150 – 310)Heartsill-Scalley et al. [2007]
Amazon basin16111 – 50Artaxo et al. [2002]
(80 – 260)Mahowald et al. [2005]
Florida52716Redfield [2002]
(113 – 960)(11 – 20)Prospero et al. [2010]
Central Japan24090Tsukuda et al. [2006]
(220 – 500)(80 – 100)
Spain25032Avila et al. [1998]
Hawaii*9Kurtz et al. [2001]
Mexico – Pacific Coast156*Campo et al. [2001]
(69 – 228)
Mexico – Yucatan26546Das et al. [2011], this study
(185 – 345)(33 – 75)

[21] Based on the absolute magnitude of variation in annual total atmospheric P deposition between 2007 and 2010, from 185 to 345 g/ha/yr, we believe that local sources rather than long-distance dust transport exert a stronger influence on total atmospheric P inputs. This is consistent with modeled estimates of P deposition to the region that suggest that dust inputs provide < 30% of total atmospheric P inputs [Mahowald et al., 2008]. Local sources include biomass burning and biogenic sources such as pollen, fungal spores and organic debris. Biomass burning varies greatly, and 2003 and 2008 were active fire years, which are evident from MODIS fire data from the Yucatan peninsula for this period (J. Rogan, personal communication, 2010). Especially high peaks are also evident in MODIS fine mode AOD over the Yucatan in these years (Figure 5). Variations in annual dust P inputs are an order of magnitude smaller than the variation in the total annual P input from the field estimate, suggesting that dust is not the most important determinant of interannual variability in P inputs. Mean annual dust P input between 2000 and 2010 varied twofold in the MODIS and threefold in the MATCH estimate, which is similar to observations of variations in annual dust deposition in other studies [Mahowald and Luo, 2003; Prospero, 1996]. Greater variations on decadal scales are possible, and some modeling studies report that there have been particularly “dusty” periods when the Sahel region experienced severe drought and dust concentrations in Saharan air masses went up several times [Prospero and Lamb, 2003]. Hence dust may be more important in determining long-term average variations in P input, but local sources are more important in determining short-term interannual variation in total atmospheric P deposition.

[22] Seasonal variation in atmospheric P inputs reveals the influence of different sources on atmospheric P inputs in the Yucatan. Biomass burning occurs every year in April and May, primarily for the clearing of land for agriculture. This pattern is evident in MODIS fire detection products that show a large peak in the number of fires every year in the Yucatan in April (J. Rogan, personal communication, 2010). The peak in P inputs in April in the field estimate, the peak in fine AOD in April and May, and the higher proportion of organic P in P inputs measured during this period are all characteristic of P inputs from biomass burning [Redfield, 2002; Mahowald et al., 2008]. Saharan dust is transported to the Caribbean by the Trade winds annually from June through August, which is reflected in the satellite and model based dust P estimates that show peaks during those months. The field estimate has relatively high inputs during this period, which is the beginning of the wet season, when wet deposition leads to higher removal of aerosols from the atmosphere, and thus greater P inputs to the biosphere. The field estimate has another peak in P input in September, before values decline in October and November toward the end of the wet season. September is past the period of dust transport, and the high P input during this time is probably due to high wet deposition rates that result from heavier rainfall from tropical storms during September. A lag in rainfall often occurs during June and July in the Yucatan, as occurred in 2007, when the field estimate was substantially lower than the same months in 2010, due to very low rainfall during those months. The influence of rainfall on P inputs is a characteristic of bulk deposition estimation of inputs [Redfield, 2002].

5 Conclusions

[23] In this study, we estimated the contribution of long-distance dust transport to total atmospheric P inputs in the Yucatan peninsula between 2000 and 2010 using a combination of field sampling, MODIS remote sensing of atmospheric aerosols and MATCH atmospheric transport modeling. Atmospheric P input measured in atmospheric bulk deposition is 265 ± 80 g P ha-1 yr-1, and P input from dust is 46 ± 12 g P ha-1 yr-1. We estimate that dust is approximately 25% of the annual inorganic P from atmospheric deposition, based on the MODIS estimate of dust transport. The MATCH model underestimates dust transport to the Yucatan, and consequently also underestimates P inputs from dust by 75%, with an average annual input value of 12 g P ha-1 yr-1. The remote sensing method may be useful in improving estimates of P inputs from dust in remote regions, where models tend to underestimate long-range dust transport. There is a well-defined seasonal pattern in atmospheric P inputs in the Yucatan, with dust inputs occurring from June through August, and local sources of atmospheric P input such as biomass burning and biogenic aerosols dominating P inputs from March through May and September through November. Dust is not the largest source of atmospheric P input to the Yucatan peninsula because of its relatively small contribution to total input and the dominance of local sources, especially biomass burning.

[24] Even though dust is not the largest atmospheric P input, it may still be a critical source of P for tropical dry forests in the region in the long term. Atmospheric P inputs and the contribution from dust are up to 2 orders of magnitude smaller than P fluxes like litterfall, mineralization of organic matter and P uptake by biota in the tropical dry forest, which are on the order of several kg P ha-1 yr-1 (Lawrence et al., 2007). However, when compared to leaching losses from the ecosystem, estimated at 70 ± 110 g P ha-1 yr-1 [Lawrence et al., 2007], atmospheric P inputs of 265 g P ha-1 yr-1 are large enough to offset declining P supply in the long term. Dust inputs play a particularly important role, because these are external inputs to the terrestrial P cycle in this ecosystem, unlike local inputs, which are effectively being recycled within the ecosystem. Over the long term, dust transport is essential for the maintenance of soil fertility and productivity in these ecosystems. The seasonality of dust is also important. Dust transport occurs between June and August, which is the beginning of the wet season, when deciduous plants begin growing leaves, and ecosystem P demand is greatest. Dust provides new inputs of a limiting nutrient at a critical growth stage for tropical dry forests. We have previously shown that forest canopies act like traps for dust and aerosols, enhancing dry deposition particularly, and may increase atmospheric P inputs 5 to 10 times compared to background atmospheric bulk deposition [Das et al., 2011; Heartsill-Scalley et al., 2007]. This would amplify the importance of atmospheric inputs, including dust as a source of P for the ecosystem.

[25] Previous studies have found that dust is not an important input in the P cycle for terrestrial ecosystems, especially at great distances from dust source regions. Mahowald et al. [2008] found that regions like the Yucatan have high losses of P from biomass burning, making it a net source region for atmospheric P. We agree that the clearing of forest areas results in a much larger loss of P than annual dust transport could replace, but for intact tropical dry forests such as in the southern Yucatan, dust transport is still an important P input that helps maintain ecosystem productivity over the long term. This study shows that atmospheric P inputs to tropical forests, especially in regions that receive long-distance dust transport, need to be better quantified, and that a combined approach of field sampling, models and remote sensing can provide more robust estimates.


[26] The authors wish to thank N. Mahowald for providing the MATCH model simulations and guidance for using the model outputs for this manuscript. We also thank one anonymous reviewer and the editor for helpful comments on an earlier version of this manuscript. Partial funding for this work was provide by NOAA Climate Program Office grant NA11OAR4310157.