This study was conducted on an alluvial plain in the southern Gurbantunggut Desert region, China. With the involvement of human activity, this area has become a typical desert–oasis ecotone. The climate of the region is arid temperate. Soils in this area are poorly weathered, typically with high pH and salt content, low moisture availability and low organic matter content (Table 1); consequently microbial biomass and soil biota abundance are also very low compared to other ecosystems.

To determine the potential abiotic contribution to total soil CO2 flux, eight typical sites were selected based on apparent differences in land use types from oasis to desert: cotton field, hops field, halophyte garden, reservoir edge, native saline desert, alkaline soil, dune crest and interdune lowland in Gurbantunggut Desert. These formed gradients in biological activity, pH and electrical conductivity (EC, a proxy for soil salt content). Soil properties, including soil C content (organic and inorganic C), pH, EC, soil moisture and living root biomass are listed in Table 1.

Based on the above description, a total of eight types of soil samples were selected to represent biological activity, pH and EC gradient in the field. Undisturbed soil was obtained by stainless steel tube (height 25 cm, inner diameter 20 cm and outer diameter 21 cm). Specific sampling processes were as follow: first, stainless steel tubes were pounded vertically into the soil by a hammer until the upper edge was about 4 cm from the soil surface, which represented the parameter “offset” in the subsequent CO2 flux measurements. The soils around the tubes were then dug out, and stainless steel circular plates (3 mm thick), with the diameter slightly greater than the tubes (approximately 20.5 cm), were carefully inserted into the soil along the bottom edge of the tubes. After that, the soil columns were lifted out and the bottom plates were carefully sealed with waterproof fabric to prevent any kind of material exchange (e.g. water or gas). To reduce damage to soil cores in the process of pounding the tubes into the soil, an approximately 15∘ slope was designed on the outer edge of the bottom end of the stainless steel tube. For each sample site, a total of six undisturbed soil columns were obtained (three for sterilization treatment and another three in natural condition).

A series of sterilization experiments were conducted to partition the potential abiotic contribution to total soil CO2 flux. The fluxes over sterilized and natural soil were considered abiotic (Rabiotic) and total flux (Rtotal), respectively, with the difference between the two representing the biotic flux (Rbiotic). For sterilized soils, the tops of the stainless steel tubes were sealed by layers of filter and brown paper to minimize water infiltrating into the soil column. Sterilization was achieved in a medical autoclave for 24 h at 120 ∘C. After sterilization treatment, the tubes were placed in an ultraviolet (UV) radiation sterilized room to allow soil cores to equilibrate with surrounding conditions. The non-heated soil remained at ambient field temperature. The tubes were then moved out and reburied in the field with the soil surface inside the tube at an equivalent height to the surrounding soil, so that the tube “wings” were flush with the soil. Doing so allowed the soil temperature to maintain natural fluctuations. It should be noted that all tubes were reburied in the site of native saline desert, which was the nearest site to the laboratory. The aboveground parts of plants, for the heat treatment soils, were removed before the soil column was sealed; for the non-heated soils, to maintain root activity in the flux measurement, the plant aboveground parts were removed immediately before measurement started. Furthermore, all measurements were conducted on clear days within 1 month (July) to preserve similar amplitudes and peak times in temperature fluctuation.

CO2 flux was measured with an Automated Soil CO2 Flux System (LI-8150, Lincoln, Nebraska, USA), equipped with six long-term monitoring chambers (LI-8100-104, Lincoln, Nebraska, USA). Fluxes were recorded at 30 min intervals for 2 days for each set of soil samples. We denote CO2 flux from soil to atmosphere with positive values; thus, negative values indicate CO2 moving from atmosphere into soil.

To evaluate the contribution of abiotic flux to the total soil CO2 flux, ratios of Rabiotic to Rtotal were calculated for the eight sites when Rabiotic was positive: Ratio=RabioticRtotal|Rabiotic>0

Cumulative CO2 exchange of Rbiotic and Rabiotic were calculated by numerical integration of Rbiotic or Rabiotic during a particular period (e.g. the period of Rabiotic>0 or Rabiotic<0) as follows: CumulativeCO2exchange ofRabiotic=∑Rabiotic>0orRabiotic<0Rabiotic×44×1800/1000

Soil temperature (Tsoil) was measured at 5 cm below the soil surface in a soil profile close to the chambers, using a thermocouple connected to the LI-8150, and recorded when each flux measurement was taken.

At the completion of each group of flux measurements, approximately 200 g of soil was collected from each soil core to a depth of 10 cm. All the samples were divided into two parts: one part was sealed in aluminum specimen boxes to estimate soil moisture content by conventional balance-weighing and oven-drying method; the other part was sealed in a hermetic bag, taken to the laboratory and used to determine pH, EC and soil C contents (i.e. organic and inorganic C). All the soil samples used for chemical analyses were air-dried and sieved (<1 mm) in advance. Soil pH and EC were determined on a 1:5 soil:deionized water suspension, using PP-20 Professional Meter (Sartorius, Germany) and a portable conductivity meter (Hach, USA), respectively. Soil total C and inorganic C were measured using a total organic C/total nitrogen analyzer (multi C/N 3100, Analytik Jena, Germany), and the difference between the two values was taken to represent the organic C content.

In addition, for each intact soil core, living roots were sieved out (100-mesh sieve) and weighed to estimate the living root biomass.

One way analysis of variance (ANOVA) was used to test for differences in mean soil properties among the eight landscapes. Stepwise multiple regressions were used to identify predominant factors for total soil CO2 flux and its biotic and abiotic components. All data analysis was performed with SPSS 16.0 and Origin 8.0 software.

Soil organic C content differed significantly among the eight sites (F=102.5, P<0.001), with a maximum of 15.85 (±0.38) gkgsoil-1 in the hops field and a minimum of 0.57 (±0.07) gkgsoil-1 in dune crest (Table 1). The average concentration of soil organic C content was 5.98 gkgsoil-1. The eight sites differed in inorganic C contents (F=92.54, P<0.001), with an average of 5.08 gkgsoil-1, which is comparable with the average content of organic C. However, there was no significant correlation between organic and inorganic C contents (P=0.83). For example, soil organic C contents were significantly higher than inorganic C in cotton field, halophyte garden, hops field and interdune lowland but inorganic C dominated in dune crest, reservoir edge, alkaline soil and native saline desert.

Soil properties varied significantly among the eight sites (Table 1). Gravimetric soil moisture content was highest in the cotton field and lowest in the alkaline soil (F=79.24, P<0.001), with coefficient of variation of 89.36 %. Soil pH was high for all eight sites, with a minimum of 8.00 (±0.10), indicating that the soils were all strongly alkaline. EC, a proxy for soil salt concentration, was in the range of 0.09 (±0.01) to 14.23 (±0.87) dSm-1. In addition, the living root biomass, regarded as the most active contributor to CO2 flux (Hanson et al., 2000), was low for all eight sites, with average of 18.78 gm-2, but with significant differences among sites. The maximum of living root biomass was 47.75 (±4.46) gm-2 in the cotton field; whereas there was little root biomass in alkaline soil, with 0.32 (±0.14) gm-2. Thus, the eight sites showed significant differences in soil properties and root biomass, along with a wide range of soil organic and inorganic C contents, and provided a natural gradient to differentiate the contributions of Rbiotic and Rabiotic.

Rtotal exhibited similar diurnal patterns across the eight sites, with positive values in the day and single peaks during 13:00–16:00, but with significant differences in flux rates. For example, the maxima of Rtotal were 3.72, 2.03, 2.12, 1.28, 1.13, 0.53, 0.42 and 0.45 µmolm-2s-1 in cotton field, hops field, halophyte garden, interdune lowland, reservoir edge, native saline desert, dune crest and alkaline soil, respectively. In particular, Rtotal was negative during the night in sites of native saline desert, dune crest and alkaline soil, in which Rbiotic was relatively low in the range of 0.001–0.364 µmolm-2s-1 (Fig. 1f–h). By comparing CO2 fluxes from soils after sterilization treatment with those in natural condition, Rtotal was partitioned into Rbiotic and Rabiotic (Fig. 1). There were significant differences in Rbiotic among the eight sites, either in flux rate or diurnal pattern. Maximum CO2 flux rates decreased following the sequence of hops field, halophyte garden, hops field, interdune lowland, reservoir edge, native saline desert, dune crest and alkaline soil, with the average flux rate being 0.579 µmolm-2s-1. The diurnal patterns of Rbiotic were all similar to those of Rtotal (Pearson's correlation coefficients r values were in the range of 0.939–0.996, P<0.001). However, for the native saline desert, dune crest and alkaline soil sites, Rtotal had a significant relationship with Rabiotic, with r=0.949, 0.965 and 0.993 (P<0.001), respectively. These variations implied that there were different dominant factors (i.e. abiotic or biotic processes) in Rtotal. For Rabiotic, diel variations were all of alternating positive and negative CO2 fluxes over a day, and hourly flux rate fluctuated in the range of -0.67 to 0.538 µmolm-2s-1 across the eight sites. For each site, the daily sum of hourly flux rate approximated zero (Fig. 1).

Total soil CO2 fluxes were all linearly related to temperature (both for Tsoil and ΔT/Δt), but with intriguing differences in explanatory degree of the variation in total soil CO2 flux among the eight landscapes (Table 2). In the cotton field, halophyte garden and hops field, where biotic flux dominated the total soil CO2 flux (Fig. 1), Tsoil accounted for more than 60 % of the total soil CO2 flux variance, while ΔT/Δt alone accounted for <40 % of this variance. In contrast, for reservoir edge, native saline desert, alkaline soil and dune crest, ΔT/Δt explained more variance of total soil CO2 flux than did Tsoil (Table 2). Based on the above partitioning results, temperature dependence of biotic and abiotic components of the total soil CO2 flux was separately analyzed (Table 2). For Rbiotic, natural variation of Tsoil accounted for most of daily Rbiotic variation for most sites except alkaline soil and dune crest, where Rbiotic was extremely low and with irregular variation (Fig. 1). Thus, the diel temperature cycle was the predominant physical control over Rbiotic. For Rabiotic, variation in Tsoil was significantly related to diel Rabiotic variations, but explained far less of the variation in Rabiotic than did ΔT/Δt (Table 2). The ΔT/Δt accounted for an average of approximately 71 % of the Rabiotic variation across the eight sites. Moreover, the negative values of Rabiotic coincided with naturally decreasing soil temperature (when ΔT/Δt<0), and positive values with increasing soil temperature (when ΔT/Δt>0) (Fig. 2).

To evaluate the contribution of Rabiotic to Rtotal, ratios of Rabiotic to Rtotal were calculated for the eight sites during the periods of Rabiotic>0. The instantaneous ratios were in the range of 0.007–0.995, with an average of 0.33 in all eight sites (Fig. 3). The maximum of Rabiotic/Rtotal ratio was in alkaline soil while the minimum was in the cotton field. When the ratios for each site were grouped, the average Rabiotic/Rtotal ratio followed a trend of decreasing logarithmically as the cumulative CO2 release of Rbiotic rose during the period of Rabiotic>0 (Fig. 4). Thus, the contribution of Rabiotic to Rtotal was obviously negatively related to increasing Rbiotic. It is noteworthy that the reason that we used cumulative CO2 release of Rbiotic as the target variable was that Rabiotic was balanced in a day – CO2 drawn into soil in the night was released during the day (Fig. 1). From this point of view, under the influence of Rabiotic, apparent Rbiotic was overestimated during the period of Rabiotic>0 (Fig. 5a). The overestimated ratio for Rbiotic was within the range of 1.07–7.72, with an average of approximately 2. For example, the real value of cumulative CO2 release from Rbiotic was 72.13 mgm-2d-1, which was up to 340.64 mgm-2d-1 lower than the apparent flux in alkaline soil. Conversely, for the period of Rabiotic<0, apparent Rbiotic was obviously underestimated due to the abiotic part of the total soil CO2 flux, and even to the extent of altering CO2 transport direction (Fig. 5b) – despite Rbiotic being always positive according to the conventional wisdom concerning soil respiration. For example, the real cumulative CO2 exchange through Rbiotic was 203.17, 287.61, 89.38 and 63.72 mgm-2d-1 in dune crest, native saline desert and interdune lowland, respectively; whereas, corresponding apparent Rbiotic, offset by negative Rabiotic, all became negative (indicating absorption) with -118.00, -29.82, -445.42 and -329.45 mgm-2d-1.

Once Rbiotic and Rabiotic were partitioned, the predominant factors for Rtotal and the corresponding biotic and abiotic components were analyzed (Table 3). Root biomass was significantly correlated with Rbiotic, and explained 91 % of Rbiotic across the eight sites. Soil moisture was significantly correlated with Rabiotic, but explained less of the variance in abiotic flux than soil pH did. Thus, Rabiotic was determined by soil pH and soil moisture. While Rabiotic had an approximately zero sum over a diel cycle, daily cumulative CO2 exchange from Rtotal equaled that for Rbiotic. As a result, variation in Rtotal was also significantly related to living root biomass (model R2=0.91, P<0.001).