During the last two decades, the laboratory incubation method has been used to measure the net NO release from soil (Yang and Meixner, 1997; Otter et al., 1999; Kirkman et al., 2001; Feig et al., 2008; Yu et al., 2008; Ashuri, 2009; Gelfand et al., 2009; Bargsten et al., 2010; Behrendt et al., 2014; and Mamtimin et al., 2015). However, today's knowledge of soil biogenic NO emission rates from managed arid soils is still rather limited (cf. Behrendt et al., 2014; Mamtimin et al., 2015; Delon et al., 2015). The laboratory incubation method is usually a dynamic chamber system which consist of gas dilution, a thermostat valve, a thermostat cabinet, and analyzers (Behrendt et al., 2014). The dynamic chamber in this study was made of Plexiglas with a diameter and height of 9.2 and 13.6 cm (Behrendt et al., 2014), respectively. Five chambers were used for soil samples, while one was kept empty as a reference and the thickness of the sample in the dynamic chamber kept the same (approximately 5 cm). The wetted soil samples were placed into five chambers, while one was kept empty as reference chamber. Then, the net potential NO release rates were measured until the soil was completely dried out. NO fluxes measured in the laboratory by means of the incubation method have repeatedly been shown to be in good agreement with those measured in the field (Mayer et al., 2011; van Dijk et al., 2002; Ludwig et al., 2001). The net release of NO from soil is the result of microbial production and consumption processes which occur simultaneously (Conrad, 1996). In our study, we investigated net NO releases (JNO; in terms of mass of nitrogen per mass of dry soil and time), as well as potential NO fluxes (FNO; in terms of mass of atomic nitrogen per area and time) of cotton, grapes, and desert soils. The methodology for the laboratory soil measurements which is used in the frame of this study is described in detail in Behrendt et al. (2014) and Mamtimin et al. (2015). By application of the laboratory dynamic chamber method, the net potential NO flux FNO (ng m-2 s-1, in terms of mass of N) is defined by FNOθg,Tsoil=JNOθg,TsoilmsoilA, where JNO (in ng kg-1 s-1) is derived from the laboratory measurements and θg is the dimensionless gravimetric soil moisture. The soil temperature Tsoil (in ∘C), the total mass of the dry soil sample msoil (in kg), and the cross section of the dynamic chamber A (in m2) were directly measured.

Soil moisture and soil temperature had been identified as the most dominant influencing factors of the net NO release; therefore, JNO is usually parameterized by these two quantities (Yang and Meixner, 1997; Otter et al., 1999; Kirkman et al., 2001; van Dijk et al., 2002; Meixner and Yang, 2006; Yu et al., 2008, 2010; Feig et al., 2008; Ashuri, 2009; Gelfand et al., 2009; Bargsten et al., 2010). While the dependence of JNO on Tsoil is exponential, the dependence of JNO on θg has the form of an optimum curve. These dependencies are described by two explicit dimensionless functions: the optimum soil moisture curve g(θg) and the exponential soil temperature curve h(Tsoil), which are described in detail by Behrendt et al. (2014). With θg,0, the optimum gravimetric soil moisture content (i.e., where the maximum NO release has been observed), introduced and Tsoil,0= 25 ∘C as the reference soil temperature, the net potential NO flux, specific for the soils of each land use type i(i: grape fields, cotton fields, desert), is given by FNO,iθg,i,Tsoil,i=JNO,iθg,o,i,Tsoil,o,igiθg,ihiTsoil,i⋅(msoil/A).

The land-use-type-specific exponential soil temperature curve hTsoil,i and the land-use-type-specific optimum soil moisture curve gθg,i are defined by Eqs. (3) and (4): hTsoil,i=exp⁡ln⁡Q10,NO,i10Tsoil,i-Tsoil,0,gθg,i=θg,iθ0,iaexp⁡-aθg,iθ0,i-1.

Land-use-type-specific shape factors “a” of the corresponding optimum soil moisture curves were derived from laboratory drying-out incubation studies on the corresponding soil samples according to Behrendt et al. (2014). The land-use-type-specific quantity Q10,NO is the (logarithmic) slope of h(Tsoil), which was also derived from the laboratory measurements; with two different soil temperatures applied, Tsoil,0 and Tsoil,1 (Tsoil,1-Tsoil,0=10 K), the land-use-type-specific Q10,NO is defined by Q10,NO=ln⁡FNOθg,0,Tsoil,1-ln⁡FNOθg,0,Tsoil,0Tsoil,1-Tsoil,0.

We used Landsat images since Landsat provides land surface information in a fine-scale matrix (30 m×30 m). The Landsat TM/ETM+ sensors acquire land surface information and store it as a digital number (DN) which ranges between 0 and 255. Landsat images are widely applied for the estimation of biospheric applications, such as the normalized differenced vegetation index (NDVI), land surface temperature (LST), soil moisture index (SMI), and fluxes of matter between different ecosystems (Goward and Williams, 1997; Liang et al., 2002; Lu et al., 2002). To perform land use classification and to derive the corresponding land-use-specific surface soil temperature, we have developed a tool by using ARCGIS 10.x, namely the Geoscience General Tool Package (GGTP; see Supplement, Sects. S1.1–S1.7). Within the GGTP model, we also calculated the 2-D distributions of soil biogenic NO emissions (see Sect. S1.9).

To classify these field-specific vegetation types, preprocessing of the Landsat imagery and image interpretation has to be performed. To effectively record the image data, histogram equalization and edge detector techniques (Jensen, 2005) were applied to image preprocessing. By using suitable edge-detecting functions, Landsat images were enhanced, and boundaries between different classes were strongly highlighted for effective subsequent image interpretation. Due to the significant requirements for the different areas of interests (AOIs) of image classification, in situ GPS data (160 points) from the field campaigns (2010) as well as previous field investigations (2006, 2008) were used. Spectral signatures of in situ data were collected by digitizing and were consequently used to classify all pixels. For that, image spectral analysis and the most used supervised classification method of maximum likelihood by Jensen (2005) was performed. This procedure ensured that (a) spectral characteristics of all AOIs were used and (b) every pixel (both within and outside AOIs) was assessed and associated with the land cover which it has the highest likelihood of being a member.

Knowledge of the land surface emissivity (ε) is indispensable for retrieving the land surface temperature (Ts) from remotely sensed data. As the land cover varies greatly from place to place, land surface emissivity widely varies from one location to another. Therefore, the land surface temperature Ts is defined as Ts=1ε0,25TB. The land surface emissivity (ε) can be obtained from red and near-infrared (NIR) spectral reflectance, while at-sensor brightness temperature (TB) can be calculated from the satellite image's thermal band in high-gain mode (band 6.2); more details are given in the Supplement (Sects. S1.2–S1.6).

Bottom-up calculations in this study aim for regional, seasonal (monthly), and land-use-specific soil biogenic NO emissions via laboratory-derived net potential NO fluxes, which depend on soil temperature, gravimetric soil moisture (Sect. 2.3.1), and fertilizer factors (i.e., amplification of NO fluxes due to the application of fertilizer). However, the necessary timescale for these data must be much shorter than a seasonal (monthly) timescale: (a) there is a strong nonlinear response of the net potential NO flux to soil temperature and gravimetric soil moisture (Eqs. 3, 4), as well as to fertilizer applications, and (b) there are large diel variations in the topsoil temperature (e.g., 10–60 ∘C in July; Tohsun Oasis desert soil) and large diel to weekly variations in soil moisture and soil nutrient content caused by the applied irrigation and fertilization schedules. Consequently, seasonal (monthly) land-use-type-specific soil NO emissions have to be averaged from data calculated on much shorter timescales (< 1 h).

Due to the very low soil temperature before and after the growing season (see Fig. 1b), most of the soil surface is expected to be frozen, and biogenic emissions from any soils of the Tohsun Oasis can be neglected. Thus, mean monthly soil NO emissions were only calculated for the period between April to September. In situ measurements of soil temperature (and soil moisture; 2.5 cm depth, 5 min resolution) at the Tohsun Oasis have only been performed for the land use type “desert”, and only for the period 1 July to 30 September 2010 (see Sect. 2.2). However, soil temperature (and soil moisture; 2.5 cm depth, 5 min resolution) have also been measured at two other Taklimakan oases during the same period, namely at Kuche (41.5360∘ N, 82.8546∘ E; “cotton soils”) and Minfeng (37.0534∘ N; 82.0760∘ E; “jujube soils”). For this study, data from Kuche were adopted for the land use type “cotton soils” of the Tohsun Oasis, while data from Minfeng were chosen to substitute grape soils data of the Tohsun Oasis. The latter is justified by the fact that both grapes and jujube belong to the bushy agricultural landscape type; they have similar height, distance in rows, water requirements, and percent ground cover by their canopy. For each of the three land use type data sets, each original measured data point (5 min, July–September, 2010) of a particular day has been normalized by the mean value observed at 10:45 LT (Landsat satellite overflight, Supplement Fig. S7). Then, a mean diel variation data set (30 min) was created by averaging all respective data between 1 July and 30 September constituting the representative diel variation in the (normalized) soil temperature of the respective land use type valid for the entire growing period (Supplement Sect. S3.1; Fig. S8). As described in Sect. S3.1 of the Supplement, land-use-type-specific surface temperatures (Ts; Sect. 2.3.2) derived from Landsat observations on 25 April, 28 July, 13 and 21 August, and 6 and 22 September 2010 (Sect. 2.2) have been temporally upscaled for the entire growing season (1 April–30 September 2010). Finally, the diel variation in soil temperatures of the land use types “desert”, “cotton soils”, and “grape soils” for every day of the growing season in 2010 was obtained by multiplying the normalized soil temperatures.

The interpolation of the satellite-derived data was done by a third-order polynomial. To estimate the corresponding uncertainties, we compared results using polynomials of degree 2 and 3 for the fitted curve. We found that R2 in the second-degree model was 89 %, while in the third-degree model it is 98 %.

Evaluation of the MSR^® 165 data logger measurements (see Sect. 2.2) at the site of the Tohsun County meteorological station (bare desert soil) has demonstrated a temporally quite constant, very low gravimetric soil moisture content of 0.0028, which has been adapted for the entire growing season (April–September 2010) land use type “desert” of the Tohsun Oasis. Temporal variation in gravimetric soil moisture content of the two other land use types, cotton soils and grape soils, depends strongly on irrigation amounts and schedules. Unfortunately, there is no official information on the Tohsun Oasis's irrigation available, neither on amounts nor schedules, and, consequently, none on the gravimetric soil moisture contents during the drying-out phases between individual irrigation events. Therefore, as for data assimilation of soil temperatures (see above), gravimetric soil moisture content data observed at Taklimakan's oases Minfeng (cotton soils) and Kuche (jujube soils) have been adapted for the Tohsun Oasis's land use types “grape soils” and “cotton soils”, respectively. The irrigation farming consists of temporally constant irrigation schedules, starting on 1 April 2010 and repeated every 2 weeks. From the observations at Minfeng and Kuche, “drying-out” shape functions could be defined, which were scaled by multiples of the soil-specific physical quantity “field capacity” (FC), the amount of water content held in the soil after excess (irrigation) water has drained away. The quantity FC is typically used in irrigation scheduling, calculation of plant available water, and calculation of water depth to be applied by irrigation (Diallo and Mariko, 2013). FC data of soil samples from Minfeng and Kuche oases have been determined by laboratory measurements (see Sect. S1.8, Supplement). Finally, for the Tohsun Oasis's land use types, cotton soils and grape soils, the following gravimetric soil moisture contents have been chosen (for details, see Sect. S3.2, Supplement): (a) 0.086 (grape soils) and 0.167 (cotton soils) before the first irrigation event (corresponding to mean gravimetric soil moisture contents at conditions of 0.25 × FC and 0.5 × FC, respectively) and (b) 0.514 (grape soils) and 0.586 (cotton soils) after each irrigation (corresponding to mean gravimetric soil moisture contents at conditions of 1.5 × FC and 1.75 × FC, respectively). Since for each irrigation event (every 14 days) the starting and end conditions in terms of gravimetric soil moisture content were known, as well as the functional relationship for the temporal drying-out of cotton soils and grape soils between the irrigation events, temporally high-resolution data (30 min) could be calculated. The result, the seasonal variation (1 April 2010 (DOY 90)–30 September 2010 (DOY 273)) of the gravimetric soil moisture content (at 30 min resolution) for the land use types “cotton fields” and “grape fields” of the Tohsun Oasis is illustrated in Fig. S9 (Supplement).

To maintain the productivity of the grape and cotton fields of the Tohsun Oasis, the soils regularly receive considerable N-containing fertilizer amounts (see Sect. 2.1). The impact of the fertilizer application on the NO fluxes from arable soils of Taklimakan oases was recently investigated by Fechner (2014) using the laboratory dynamic chamber system described by Behrendt et al. (2014). Depending on the applied fertilizer amount (FA), NO fluxes from all soil samples increased considerably and immediately after fertilizer application, which was accomplished by dissolving the mineral fertilizer in the same amount of water that was used to wet the soil samples (just to mirror the fertilization procedures of the Taklimakan oases). The impact of fertilization is twofold: (a) it strongly amplifies the net release (net potential flux) of NO over the entire range of gravimetric soil moisture and (b) there is a moderate amplification of the net NO release (net NO potential flux) with soil temperature. Furthermore, the impact of fertilization also has a temporal dimension: it has been shown that the fertilizer effect on the emission of N-containing gasses is temporally declining after each fertilization event (Chen et al., 2011). Therefore, the fertilizer effect may be considered (a) by the multiplicative, FA-dependent, time-dependent, dimensionless “fertilizer factor” (FF) applied to the standard net release JNO(θg,0, Tsoil,0) in Eq. (2) and (b) by the FA-dependent, time-dependent, dimensionless “Q10 factor” (Q10F), which is also multiplicatively applied to Q10 (the logarithmic slope of the exponential soil temperature curve h(Tsoil)). According to Fechner (2014), initial fertilizer effects are FF = 142, 179, 205, and 226 and Q10F= 1.21, 1.24, 1.25, and 1.26 for fertilization amounts FA = 100, 200, 300, and 400 kg (N) ha-1, respectively. From recommendations of the Tohsun Oasis's local agricultural administration (personal communication, 2010) and our interviews with farmers there in 2010, data for applied fertilizer amounts as well as for fertilizing schedules could be inferred. With the assumption of an exponential decay of the fertilizer effects, corresponding functional relationships for the temporal behaviors of FF and Q10F have been developed (see Sect. S3.3, Supplement), and temporally high-resolution data (30 min) of FF and Q10F were calculated. The result – i.e., the seasonal variation (1 April 2010 (DOY 90)–30 September 2010 (DOY 273)) in the (dimensionless) factors FF and Q10F for the land use types “cotton fields” and “grape fields” of the Tohsun Oasis – is illustrated in Fig. S10 (Supplement).

Estimates of soil biogenic NO fluxes can be calculated from net potential NO fluxes (see Sect. 2.3.1) according to Eqs. (1)–(4), since those NO fluxes that are derived from laboratory incubations have repeatedly been shown to be in good agreement with those measured in the field (Mayer et al., 2011; van Dijk et al., 2002; Ludwig et al., 2001). Consequently, soil biogenic NO fluxes for the Tohsun Oasis were calculated every 30 min from land-use-specific net potential NO fluxes (see Sect. 2.3.1) using land-use-specific data of soil temperature (see Sect. 2.3.3), gravimetric soil moisture contents (see Sect. 2.3.3), and fertilizer factors (see Sect. 2.3.3). Bottom-up soil biogenic NO emissions for the entire region of the Tohsun Oasis were calculated monthly by (a) averaging temporally high-resolution data (30 min) of the three land-use-specific soil biogenic net NO fluxes for each month of the growing period (April–September 2010), (b) weighting by the areal contribution of each of the three land use types (1117, 80, and 150 km2 for desert, grape, and cotton soils; see Sect. 2.3.2), and (c) summing up for each month of 2010.

The soil temperature is usually the dominating factor of the uncertainties of NO fluxes because of the exponential dependence of the NO flux on soil temperature (typically expressed as the increase in the NO flux for a 10 ∘C increase in soil temperature known as a Q10; see Sect. 2.3.1). The mean Q10 value is 2 for both cotton and grape soils. The total uncertainties of bottom-up NO emissions were determined from (a) the errors of the concentration measurements of NO, which was conservatively assumed to be 5 % (Behrendt et al., 2014), and (b) errors of the simulated soil temperatures, which were assumed to be 1.5–2.5 ∘C in both estimates (satellite-derived (see Fig. S4) and estimated land surface temperatures (see Fig. S7)).

The studies by Su et al. (2011) and Oswald et al. (2013) pointed out that the HONO and NO emissions have to feature similar dependencies on soil properties, and there exists a co-emission of HONO with NO from the soil; therefore, it could be expected that the HONO and NO emission processes follow similar parameters (Naegele and Conrad, 1990; Skopp et al., 1990, Gödde and Conrad, 2000; Oswald et al., 2013).

The recent findings of Oswald et al. (2013), namely that a large amount of gaseous nitrous acid (HONO) is concurrently emitted along with NO, have been considered for the monthly estimation of biogenic emissions of the Tohsun Oasis. The authors investigated a total of 17 soil samples, which were taken from a variety of natural and agriculturally managed soils around the globe. Fortunately, five of these samples (no. S13–S17 in Oswald et al., 2013) were cotton, jujube, and desert soils of the Taklimakan region; they have the same arid climate condition and irrigation regime as well as fertilizer application as those used for this study. The strong dependency of HONO emissions on soil water content and soil properties was determined by laboratory experiments with dynamic chamber systems (Oswald et al., 2013). Since the HONO emissions were not directly measured within the present study, the land-use-specific constant scale factor – namely the ratios of HONO to NO releases for the optimum conditions to estimate the HONO emissions ((FN, opt(HONO) to FNopt(NO)) – were adapted from the study of Oswald et al. (2013). This is justified by (a) the strong assumption that the soil from our study emits the same magnitude of HONO as the soil by Oswald et al. (2013) because they have the same soil properties and experienced the same irrigation regimes and fertilizer applications, and (b) the soil release of HONO and NO by Oswald et al. (2013) has also been measured in the Max Planck Institute for Chemistry by a next-laboratory version of the dynamic chamber system which was used to measure the soil release of NO in the present study. Consequently, corresponding ratios for the optimum conditions of HONO and NO emissions (i.e., FN,opt(HONO) to FN,opt (NO)) have been used to calculate monthly HONO emissions from the Tohsun Oasis from the measured NO emissions. Monthly land-use-type-specific contributions of HONO emissions for the Tohsun Oasis during the growing period 2010 were estimated by multiplying the corresponding NO emissions by the mean, land-use-type-specific factors (0.8 for grape and desert soils, and 1.6 for cotton soils). These were calculated from individual values provided by Oswald et al. (2013), which are derived from the ratio for the optimum condition FN,opt(HONO) to FN,opt(NO). To quantify the uncertainties of calculated HONO emissions, the mean relative errors for concentration measurements of 12 % for HONO (Sörgel et al., 2011) and 5 % for NO (Behrendt et al., 2014) were adopted. Based on these estimates, we performed a conservative error propagation calculation of the additional HONO emissions (under optimum conditions), which resulted in a relative error of 10 %.

In the present study, the total bottom-up biogenic emission estimates (NO and HONO) were expressed in terms of mass of atomic nitrogen – partly because the reported mass of emitted nitrogen compounds becomes independent of the partitioning ratio between NO and NO2. Additionally, the anthropogenic and top-down emissions estimates are primarily derived for NO2. In this study, the mean primary NO2 emissions were not considered, and all NOx emissions are being reported as the mass of nitrogen. For that purpose soil biogenic emissions of NO and HONO (given as mass) were multiplied by the respective ratios of the corresponding molecular weights (MN/MNO, MN/MHONO).

For the comparison of biogenic bottom-up emissions (Fbiogenic) with top-down emissions (Fsatellite) only the data around the time of the satellite overpass around 13:00 LT were considered (bottom-up results for 12:45 LT (12:30–13:00) and 13:15 LT (13:00–13:30)). To estimate the effect of the diurnal cycle of the emissions, we calculated the ratio of the bottom-up emission at the satellite overpass (at 13:00 LT) and the daily average emissions.

OMI satellite data of tropospheric NO2 were chosen because of their smaller pixel size (13 km × 24 km in nadir) compared to data from SCIAMACHY (60 km × 30 km) or GOME (320 km × 40 km). OMI is operated in a Sun-synchronous orbit such that measurements are taken at about 13:00 LT. OMI data also provide daily global coverage (Boersma et al., 2007). For our study, we used the OMI monthly mean tropospheric NO2 data, which are calculated on the basis of the results of daily OMI overflights (around noon) and are provided by TEMIS (Tropospheric Emission Monitoring Internet Service, European Space Agency (ESA); http://www.temis.nl/). The OMI monthly mean tropospheric NO2 data product was averaged on a 0.125∘ × 0.125∘ grid and has been retrieved for cloud fractions < 20 %.

It should be noted that the tropospheric VCDs derived from satellite measurements can be systematically affected by the a priori assumptions needed for the calculation of the air mass factor (AMF; reflecting the satellite sensitivity for tropospheric NO2), such as cloud fraction, pressure at cloud level, surface albedo, and trace gas profile (Boersma et al., 2007, 2011). Fortunately, the region of Tohsun County provides relatively favorable conditions for accurate calculation of tropospheric AMFs, since the desert region is typically only slightly affected by clouds and aerosols; also, lightning activity, due to the predominant absence of convective rains, is rather weak, if not negligible. The AMF for OMI DOMINO v2.0 has been calculated using TM4 model NO2 profiles with a longitude–latitude resolution of 3∘ × 2∘, a surface reflectance climatology (0.5∘ × 0.5∘), and cloud information (Boersma et al., 2011). To test whether the assumed surface albedo is appropriate for the Tohsun Oasis, we examined the OMI surface reflectance climatology for the study area and compared it with the high-resolution MODIS surface reflectance product (0.05 × 0.05∘, MOD43B). We used the MODIS albedo products from channel 3 (at 459–479 nm) because this channel is close to the spectral window (405–465 nm) used for the OMI retrieval. According to Kleipool et al. (2008), the surface albedo difference between 470 nm (center of MODIS channel) and 440 nm (OMI surface reflectance climatology) is very small for land surfaces.

To account for any background NO2 from upwind (desert) areas, representative VCDNO2 observed over the Taklimakan Desert (400 km away from the oasis, at 85.5129∘ E, 40.0669∘ N) were subtracted from the VCDNO2 observed over the Tohsun area.

To check the representativeness of satellite-derived NO2 on the regional scale, we investigated the interannual variability in tropospheric VCDNO2 from 2006 to 2010. This is because a mean value of longer periods leads to significantly reduced uncertainties, and it can better demonstrate the representativeness of satellite-derived NO2. Evidence of a clear seasonal pattern in the long-term observations can reflect the true seasonal effect on tropospheric VCDNO2, which is of substantial importance with respect to the intended comparison of “bottom-up” vs. “top-down”. Seasonal patterns (i.e., December–February, March–May, June–August, September–November) of satellite-derived NO2 column densities were evaluated using long-term cloud-free (cloud fraction < 20 %) measurements from 2006 to 2010.

Satellite observations reflect contributions from different emission sources. In principle, the retrieval of NO2 emissions over a city or oases is similar. However, usually the emissions from cities are higher, and the corresponding enhancements can be better identified compared to the background. In order to establish the relationship between biogenic NO2 and satellite-derived NO2, it is necessary to understand their spatial patterns with respect to the locations and shapes of the potential sources. Thus, four different areas (see Fig. 3) were selected to represent (1) typical agricultural areas (study area) as the biogenic source, (2) mixed land use areas (agricultural and small urban) as the biogenic and anthropogenic sources, (3) large urban areas as the anthropogenic source, and (4) desert area as the background source.

For the top-down inventory, the given molecular weight of NO2, Avogadro's number, a lifetime of NOx, and pixel area (m2) of NO2 were considered. Top-down NOx emissions from the Tohsun Oasis were calculated from all VCDNO2 – whose pixels cover the area of the Tohsun Oasis and where each represents the particular monthly mean of 2010 – as follows: (a) from each individual VCDNO2 (in NO2 molecules cm-2) the representative “desert” VCDNO2 value was subtracted to account for any background NO2 from upwind (desert) areas; (b) all of these VCDNO2 were multiplied by their respective pixel area and summed up to obtain the mean number of (vertically integrated) NO2 molecules for each month of 2010 over the Tohsun Oasis (for the area indicated in Fig. 3); (c) this value was converted to mass of NO2 per unit area (m2) and divided by the characteristic lifetime of NOx, τNOx (with respect to its reaction with the hydroxyl radical OH); and (d) in order to compare the top-down and bottom-up emission estimates, the monthly mean top-down emissions and their daily averages were expressed in terms of mass of atomic nitrogen.

The corresponding result represents the monthly mean total NOx emission (i.e., from anthropogenic and biogenic sources) from the area of the Tohsun Oasis at the time of the OMI satellite overflight (and a few hours before). Here, it should be noted that meteorological information does not have to be taken into account explicitly because, from the satellite observations (Fig. 9), we can conclude that most of the NOx (about more than 90 %) is destroyed before it is blown outside the considered area by wind. The characteristic lifetime of NOx can be expressed as a short lifetime. It has been investigated by Beirle et al. (2011), for example, and can be estimated at about τNO2=4 h for low and mid-latitudes, particularly for the dryland cities of Isfahan (32.6∘ N) and Riyadh (24.6∘ N).

Results of the laboratory-derived net potential NO fluxes (FNO) from soils of the desert and the irrigated and fertilized cotton and grape fields are shown in Fig. 4. The highest mean net potential NO flux at 25 ∘C occurred from the cotton soils, with a maximum net potential NO flux of 57.5 ng m-2 s-1 followed by grape soils (18.9 ng m-2 s-1) and desert soils (4.1 ng m-2 s-1). Common to all net potential NO fluxes from the major land use types of the Tohsun Oasis is the exponential response to soil temperature, due to strong enzymatic activities associated with microbial processes of soil NO production. During the growing period (April to September), soil temperatures of 45 ∘C plus are easily reached in the soils of the Tohsun Oasis, particularly for the desert soils (see Supplement, Fig. S1). At Tsoil=50 ∘C, the maximum net potential NO fluxes of 384 (cotton soils), 115 (grape soils), and 46 ng m-2 s-1 (desert soils) are reached at an optimum gravimetric soil moisture content (θg,opt) of 0.021, 0.146, and 0.0158, respectively. These θg,opt values, particularly those of the cotton and desert soils, are much lower than those of natural and agricultural soils of temperate latitudes (> 0.25; cf. Oswald et al., 2013; Behrendt et al., 2014).

To emphasize this, significant net potential NO fluxes from desert soil only occurs if θg is higher than 0.01 and less than 0.03. Since the observed θg value of the Tohsun desert soils is 0.0028 (see Sect. 2.3.3), no significant biogenic NO emissions are expected from there. As stated in Sect. 2.3.3, the chosen irrigation schedules of cotton and grape fields resulted in certain, predefined ranges of gravimetric soil moisture content (depending on the field capacity (FC) of the respective soils). Immediately after each irrigation event, θg values of 0.514 (grape soils) and 0.586 (cotton soils) occur; according to Fig. 4, there will be no significant biogenic NO emissions at this stage. However, after 14 days (just before the next irrigation event), θg values as low as 0.086 (grape soils) and 0.167 (cotton soils) are found; consequently, regarding the results in Fig. 4, maximum biogenic NO emissions from grape soils have to be expected for the end of the 14-day irrigation period (θg,opt=0.146), while at that time, biogenic NO emissions from cotton soils will be (even at Tsoil=50 ∘C) less than 4.6 ng m-2 s-1 since then the θg value for cotton soils (0.167) is far above the corresponding optimum value (θg,opt=0.021). This means that despite the overwhelming potential of cotton soils to dominate the biogenic NO emissions of the Tohsun Oasis (see Fig. 4), their net actual contribution is almost negligible over the full 14-day irrigation cycle (see below). It should be noted that for other oases – with other irrigation cycles – the contribution to cotton soils might be much larger.

The land cover classification of the Tohsun Oasis has also been derived from Landsat images. Based on these results, the dominant land cover types were cotton and grapes, building/residential areas, other vegetation, and desert (see Fig. 5).

Biogenic net NO fluxes (in ng m-2 s-1, in terms of NO) from the three land use types of the Tohsun Oasis have been calculated from laboratory-derived, land-use-specific net potential NO fluxes (see above) by Eqs. (2)–(5) from assimilated data sets of soil temperature, gravimetric soil moisture content, and the enhancement of biogenic net NO fluxes caused by regular fertilizer application (see Sect. 2.3.3).

Three-dimensional graphs of the seasonal variation (1 April 2010 (DOY 90)–30 September 2010 (DOY 273)) in the mean diel soil temperature (at 30 min resolution) are presented in Fig. 6 for desert, grape, and cotton soils of the Tohsun Oasis. As expected, soil temperature in desert soils exhibits the largest variation on both diel and seasonal scales, while – due to the shading by the considerably dense canopy of cotton plants – the corresponding variations in the soil temperature in cotton fields are smallest. The corresponding seasonal variations in the gravimetric soil moisture content, as well as the (dimensionless) fertilizer factors FF and Q10F (at 30 min resolution) for the Tohsun Oasis's cotton and grape soils, are shown in Fig. S10 (Supplement), clearly demonstrating regular temporal patterns due to the chosen irrigation and fertilization schedules.

The seasonal variation in the diel variation in the biogenic net NO flux from soils of land use types “grape soils” and “cotton soils” are shown in Fig. 7. Results of the biogenic net NO fluxes from desert soils are not shown. Despite very high values of soil temperature, the net NO fluxes from the Tohsun Oasis's desert soils are ≪0.46 ng m-2 s-1 due to the extremely low gravimetric soil water content assumed for Tohsun's desert soils (θg,desert= 0.0028). The highest NO emissions were observed from grape soils (3220 ng m-2 s-1), while the highest NO emissions from cotton soils only reached 0.6 ng m-2 s-1. Considering the entire growing period, biogenic net NO fluxes from grape soils dominate and even exceed those from cotton soils by a factor of 1–2 × 103 on average. In both soils, the seasonal and diel variations in the net NO fluxes clearly mirror those of the corresponding soil temperature. On the seasonal scale, there is striking evidence for the impact of the chosen irrigation schedules particularly on the net NO fluxes from cotton soils (through the assumed regular 2-weekly variations in the corresponding gravimetric soil moisture content). Immediately after (flooding) irrigation, soils might be waterlogged to prevent any diffusion of gasses out of the soil, but even after the evaporation of the surface waters, the chosen “drying-out” curves (see Supplement Fig. S9) prescribe θg values of 0.514 (grape soils) and 0.586 (cotton soils), which are well above the corresponding values in field capacity conditions. There, soil diffusivity and soil microbial processes are severely suppressed (cf. Skopp et al., 1990). This is clearly reflected in the results of laboratory incubations (see Fig. 4), where the corresponding net potential NO fluxes are close to zero for θg>0.4. Near the end of the chosen 14-day irrigation period, θg values reach 0.086 and 0.167 for grape soils and cotton soils, respectively. As a result, soil diffusivity is no longer limited in the two soils, but soil microbial processes in cotton soils are still far from their maximum activity level (θg,opt=0.021), while soil microbial processes in grape soils have even passed their maximum activity level (θg,opt=0.146). This is why biogenic net NO fluxes from grape soils of the Tohsun Oasis dominate those from cotton soils. Furthermore, the impact on net NO emission by fertilizer application on 1 April (DOY 91), 27 May (DOY 147), 10 June (DOY 161), 22 July (DOY 203), and 5 August (DOY 217; Fig. S10, Supplement) amplifies the already higher net NO fluxes from grape fields. As a consequence, NO emissions from cotton soils can be neglected if compared with emissions from grape soils, even though the net potential emission of cotton soils is more than 3 times higher than for grape soils. It should, however, be emphasized that a small change to the low optimum soil moistures of cotton soils may cause higher NO emissions than initially estimated.

Given the areal contributions of the three land use types of the Tohsun Oasis (1117, 80, and 150 km2 for desert, grape, and cotton soils), the corresponding monthly bottom-up biogenic NO and HONO emissions have been calculated by (a) summing up 30 min data of related biogenic net NO fluxes (presented in the preceding section) for each month of the growing period (April–September) and (b) multiplying the results by the area of appropriate land use types. Based on the results of Oswald et al. (2013), monthly biogenic emissions of nitrous acid (HONO) have been calculated by land-use-type-specific factors applied to the monthly means of biogenic NO emissions from grape and cotton soils (see Sect. 2.3.4). For comparison with anthropogenic emissions, respective numbers of monthly soil biogenic NO and HONO emissions have been multiplied by the respective ratios of corresponding molecular weights (MN/MNO, MN/MHONO) and are shown in terms of mass of N Fig. 8 (NO: green bars; NO+HONO: pink bars). Summing up the monthly data of NO emissions for the entire oasis, corresponding totals (calculation of grapes+cotton+desert) of April, May, June, July, August, and September are 13, 23, 52, 68, 66 and 3 t, respectively (see Fig. 8). Corresponding monthly bottom-up data for HONO are 10 (April), 19 (May), 42 (June), 55 (July), 53 (August) and 2.4 t (September), respectively. It is worth noting that the bottom-up biogenic soil emissions at the monthly scale seem to be under the primary control of the soil temperature; indeed, the seasonal variation in the soil biogenic emissions mirrors greatly that of the soil temperature (see Figs. 6 and S8). Particularly during July, when soil temperatures are highest, both monthly biogenic soil NO emissions (68 t) and soil HONO emissions (55 t) are also highest. It should be stressed, however, that the estimated HONO emissions are an upper bound to the HONO source since we used the ratio for the optimum condition ((FN,opt(HONO) to FN,opt (NO)). The lower bounds can be determined when the HONO emissions source is omitted (see Fig. 8).

Annual anthropogenic NOx emissions (in terms of N) of the Tohsun Oasis have been determined from published information on annual consumption of different fossil fuel types and related NOx emission factors, which are widely used in China (see Sect. 2.4). The results of the Tohsun Oasis's NOx emissions per fuel type are shown in Fig. S11 (see Supplement). Like other parts of China, coal combustion is the largest contributor to NOx emissions and accounts for approximately 58 % (1512 t) of the NOx emission inventory. As the second largest contributor, coke combustion contributed about 10 %. Monthly anthropogenic emissions (given in terms of mass of N) have been calculated from Tohsun County's annual NOx emission data, which were downscaled by appropriate mean monthly percentages (see Sect. 2.4); corresponding results are also shown in Fig. 8 (black bars). As expected, the overwhelming part of anthropogenic NOx (180–276 t N month-1) was emitted in the winter, when air temperatures (well below 0 ∘C; see Fig. 2) necessitate substantial domestic heating (predominantly by coal), while in July (approximately +35 ∘C) anthropogenic NOx emissions shrank to approximately 1/10 of the winter value (45 t N month-1), which might mainly be attributed to traffic emissions.

On an annual basis (not shown), anthropogenic sources contributed 81.4 %, and biogenic sources 18.6 % (10.3 % NO, 8.3 % HONO), to the Tohsun Oasis's total emissions (2174 t, in terms of N) of reactive nitrogen in 2010. However, if only the growing period (April–September) is considered, where 40 % (865 t) of the Tohsun Oasis's total annual emissions occur, the anthropogenic emissions contribute about 53.2 % and biogenic sources about 46.8 % (26.0 % NO, 20.8 % HONO). As already mentioned in Sect. 3.1.3, soil biogenic emissions at monthly scales seemed to be primarily controlled by the soil temperature. Consequently, during June, July, and August, total biogenic emissions (NO and HONO) exceeded anthropogenic emissions by factors of 1.8, 2.8, and 1.5, respectively (see Fig. 8). Therefore, biogenic soil emissions of NO and HONO during the growing period are much more important contributors to the regional NOx budget than previously thought.

Figure 9 shows the seasonal means for spring, summer, autumn, and winter (from 2006 to 2010) of the tropospheric NO2 VCD around the Tohsun Oasis. Comparing the mean distribution (2006–2010) of the different seasons, it is clearly visible that enhanced values are found over the area of the Tohsun Oasis (compared to the surrounding desert) for all seasons. Even stronger enhancements of VCDNO2 values are also found in other regions, which are always dominated by anthropogenic emissions, especially the growing city of Urumqi (cf. Mamtimin and Meixner, 2011). To investigate these seasonal and spatial variations in more detail, time series of monthly means are calculated for four selected areas (see Fig. 3): an area exclusively used for agriculture (A: study area), an area of mixed land use (B: agricultural and urban), a large urban area (C: Urumqi), and a desert area (D). The seasonal variation in mean tropospheric VCDNO2 of areas A–D is shown in Fig. 10. These data can be distinctively classified: highest VCDNO2 values (3–15 × 1015 NO2 molecules cm-2) over the large urban area of Urumqi (C), 1–6.5 × 1015 NO2 molecules cm-2 over the Tohsun Oasis (A) and the Tohsun and Turpan area (B), and less than 1 ×1015 NO2 molecules cm-2 over the chosen desert area (D). From these seasonal cycles, clear indications of the relative contributions from fossil fuel combustion and soil emissions can be drawn: over Urumqi (C), clear maxima are found during winter, indicating the dominating contribution from fossil fuel combustion, while there are evidently clear minima during the summer months. In contrast, over the desert (D), where only low VCDNO2 values prevail, small but distinctive maxima are found during summer, which might be caused by soil emissions. Here, we need to mention that the desert VCDNO2 values are higher than the NO emission of the GGTP model results (see Supplement Sect. S1.9, Fig. S6). This is indeed an interesting finding which can be attributed to the fact that, in the desert region, a small but still substantial diurnal variation in soil moisture is present. However, in our GGTP model, we used a constant value for the soil moisture (0.0028); as a result, the diurnal cycle of soil moisture was not considered in the model calculation. It might indeed indicate that the desert soil emissions are underestimated. Alternatively, it might also be partly caused by inflow from NOx-reach air from the surrounding regions.

Over the Tohsun Oasis (A) and Tohsun and Turpan area (B), a clear maximum, although 2–3 fold lower than that over (C), is also found in every winter and is certainly related to fossil fuel combustion during the heating period. Unique for all seasonal cycles are secondary maxima found over (A) and (B) during the summer months, which amount to approximately half of the winter maxima and indicate that, during summer, soil emissions substantially contribute to total NOx emissions.

As described in Sect. 2.5.2, top-down monthly total NOx emissions (particularly from anthropogenic and biogenic sources) of the Tohsun Oasis were calculated from satellite-derived, background-corrected VCDNO2 over the Tohsun Oasis and the characteristic lifetime of NOx, τNOx (with respect to its reaction with the hydroxyl radical OH). In addition, a correction for the partitioning between NO and NO2 in the atmosphere has to be made. For that purpose, we multiply the NO2 values by the typical ratio between NOx and NO2 of about 1.3 (see, e.g., Seinfeld and Pandis, 2012). The results are shown in Fig. 11. Since a prescribed and constant value for τNOx (4 h) has been applied, the seasonal course of monthly total NOx emissions is directly proportional to that of VCDNO2 over the area of the Tohsun Oasis ((A) in Fig. 10). In 2010, the highest total NOx emissions were observed in January and December (103–93 ng m-2 s-2, in terms of N), and the lowest (30–33 ng m-2 s-2) in May, September, and October (see Fig. 11). Between July and August, there is the already mentioned secondary maximum of total NOx emissions (53 ng m-2 s-2), approximately half of the value in January and December, and approximately twofold greater than the values of May, September, and October.

a.

Satellite retrieval: the air mass factor (AMF) has to be calculated for the conversion of the retrieved slant column densities into the vertical column densities. Uncertainties in the AMF can lead to uncertainties in the derived NO2 VCD of about 50 %. Here, particular assumptions on the trace gas profile and the surface albedo have to be made. For NO2, vertical profiles from a global model (TM4; see Sect. 2.5.1) with a rather coarse spatial resolution are used. The derived NO2 VCDs probably underestimate the true NO2 VCDs because the true NO2 profile peaks more closely to the ground than the profile used in the satellite retrieval. It should be noted that the uncertainties in the spectral retrieval are much smaller because many measurements are averaged. As a consequence, the true AMF is probably smaller than the one used in the OMI retrieval, and thus the VCDNO2 from the operational product probably underestimates the true tropospheric VCDNO2. In contrast, the surface albedo used in the operational product is very similar to those from other data sets (MODIS and MERIS). Here, it is interesting to note that the spatial patterns of the surface albedo and the tropospheric VCDNO2 show no similarities.

a.

Soil biogenic NO: the soil temperature at a 30 min resolution is calculated by using satellite-derived surface temperature and in situ measurements. The interpolation of the satellite-derived data was done by a third-order polynomial. From sensitivity analysis for second- and third-order models (see Sect. 2.3.3), we found that the uncertainties of the derived temperatures are up to about 6 ∘C. As a consequence, the interpolation of the data just by the third-order polynomial fit yielded the most consistent results. Substantial deviations of soil temperatures (from the interpolated values) between day 115 and day 225 might be unlikely since surface temperatures in (hyper)arid regions (even with sparse vegetation covering) are predominantly controlled by insolation only as long there is no impact of (convective) precipitation (indeed, Turpan's meteorological station did not report any precipitation events during the entire growing period). Since the dependence of the NO flux on soil temperature is exponential (see Sect. 2.3.4), by taking into consideration errors of soil temperature (1.5–2.5 ∘C) at Q10=2, the calculated relative errors of NO account for 20 %, so that a total uncertainty of 25 % was assumed for the NO emission estimates.