Recently, processes of urbanization have become a major global driver of land use/land cover change [1
]. At present, about 3%–5% of global land area has been converted into urban and developed land-use (hereafter referred to as urban) [2
], 13%–17% of which were intensively developed [4
]. With the current expansion rate, the global urban areas by 2030 would nearly triple those of the year 2000 [1
]. A major finding of urban ecological research in the past decade is that urban ecosystems could account for a significant portion of terrestrial C storage at both local and regional scales [5
]. Rapid urban expansion accompanied by intensive human disturbances could significantly alter ecosystem C pools [12
]. Compared to other ecosystems, urban ecosystems are characterized by highly dynamic landscape structures and strong spatial heterogeneity in C storage. Some urban land types, like lawns or urban forests, were found to have higher C storages/sequestration capacities than their rural counterparts [5
]. In contrast, the installation of impervious surfaces could result in significant C loss. Even in a single urban park, the soil organic C densities (SOCD hereafter) of different land cover types could differ by 10 times [14
]. Therefore, an accurate assessment of the urbanization effect on C cycle requires detailed information regarding the types and locations of land-cover changes, as well as knowledge about the C densities in different pre-urban and urban land-cover types. Spatially explicit land-cover changes and C sinks/sources datasets could also help decision makers improve urban land planning and ecosystem management for the purpose of C management [15
However, many previous studies tended to consider urban areas as a homogeneous land cover when assessing the urbanization effects on ecosystem C [4
]. Furthermore, significant uncertainties still exist in the C density of urban land-cover types. Field observations of the SOCD in impervious surface areas (SOCDISA
), which could account for more than 63% of the urban area [21
], are particularly scarce due to the difficulties in sampling under pavement or buildings [22
]. Previous studies that relied on untested assumptions of SOCDISA
], 1 [25
], or 3.3 kg·C·m−2
]) were prone to underestimation of urban C storages and overestimation of urbanization-induced C losses [11
]. Moreover, all of the few [11
] field studies on SOCDISA
focused in humid areas, while SOCDISA
observations in dryland that covers more than 30% of the global land surface are still unavailable [28
In this study, we measured the SOCDISA and C densities of other land-cover types in the largest dryland city of China—Urumqi city, and developed 30 m resolution multi-temporal land-cover datasets with Landsat TM images to track the land-cover changes between 1990 and 2010. Our objectives were to (1) access the amount and distribution of the ecosystem C pools of the dryland city; (2) reveal the spatial-temporal change of urban land cover from 1990 to 2010 and its impacts on the urban C dynamics; and (3) identify the major C sources/sinks and provide a scientific basis for effective C management of the city.
Land cover change in Urumqi from 1990 to 2010.
Since urban ecosystem C storage is related to pre-urban land-cover types as well as the current land cover, knowledge related to land-cover change patterns is important for our understanding of urbanization effects on the C cycle and locating of C sinks/sources in urbanized areas [59
]. The land-cover change in Urumqi from 1990 to 2010 has been characterized by dramatic ISA expansion at the cost of cropland and remnant desert/bare soil (Table 4
). Although the urban greenspace increased slightly, it, like all other pervious land-cover types, experienced dramatic fragmentation (Figure 5
). As small patches of pervious surfaces were being separated by the ever-expanding impervious surfaces, most native plants and animals that lived in these “urban green-islands” would be isolated from each other. Meanwhile, the increased contagion index, the shrinking pervious patch sizes, and the more complex patch shapes (Figure 5
) indicated intensified interactions between the ecosystems in urban and the paved areas (buildings, roads, etc.
) where humans live and work. In other words, the environments (climate, atmosphere, ground water, etc.
) of the urban ecosystems became more susceptible to human disturbances [60
]. The intensified human disturbances together with the habitat loss and isolation threatened the stability of the urban ecosystem functions (e.g., C sequestration) as well as the ecosystem structure (e.g., native species abundances) in Urumqi [59
]. Because the city is located in an arid climate, and because the ecosystem is relatively fragile compared to other eco-zones [62
], rapid urbanization in this area may have more serious environmental/ecological consequences than in other areas.
The effect of urbanization on ecosystem C storage.
Our study showed that urbanization from 1990 to 2010 resulted in about 1.23 kg·C·m−2
loss in Urumqi. The urbanization effects in Urumqi were weaker than those found in the southern US (−2.6 kg·C·m−2
], United Kingdom and the Northern Ireland (−8 ± 4 kg·C·m−2
]. This is because the dryland ecosystem had extremely low VEGC storage (0.04 kg·C·m−2
), and thus suffered less VEGC loss (compared to other regions) when the biomass was removed due to ISA expansion (Table 4
indicates about 63% of the newly developed ISA was converted from remnant deserts). More importantly, we found that the SOCD of the impervious surfaces (5.36 ± 0.51 kg·C·m−2
) was only slightly lower than that of the remnant deserts (5.55 ± 0.60 kg·C·m−2
), meaning the SOC loss due to remnant deserts to ISA conversion was also small. Previous studies that assumed the ISA conversion would remove 100% [25
] or 50% [63
] of the SOC might overestimate the negative effect of urbanization on ecosystem C balance [8
]. Furthermore, we found that about 15% of the disappeared remnant deserts have been converted to urban greenspaces, which had much higher VEGC (1.69 kg·C·m−2 vs.
) and SOC (8.08 kg·C·m−2 vs.
) than the native deserts (Table 5
). Management activities such as irrigation and fertilization enable urban greenspace to sequestrate more C than natural ecosystems in dryland [6
]. The carbon sinks from the urban greenspaces could partially offset the C loss during urbanization. However, it should be noted that any gains in C by urban greenspace management were at the cost of increasing water, fertilizer, and fossil fuel usages.
Locating the C sinks and sources in Urumqi.
In the face of rapid global urbanization in the 21st century, the potential for C management in urban and developed areas has drawn attention from both ecologists and decision makers [64
]. Effective C management requires detailed information about the locations and magnitudes of the C sinks/sources. Our results showed large patches of intensive C sources in the northern and northwestern outskirts of Urumqi, where many croplands were converted to ISA as the built-up areas expanding northward. Although the northeastern expansion of the ISA seemed to be more intensive than the northwestern expansion (Figure 4
d), fewer strong C sources were found in the northeast, because most ISA in this area were converted from remnant deserts that had low C density (Figure 4
c, Table 4
). Furthermore, the newly developed built-up areas in the northeast had relatively high urban greenspace coverage (Figure 4
b), which acted as C sinks (Figure 6
g). Meanwhile, small patches of C sources emerged across the built-up areas, mainly due to the urban infill that converted the greenspaces to ISA (Figure 4
d). The conversions of two large greenspaces in the mid-eastern and southeastern downtown areas to ISA from 1990 to 2010 (Figure 4
c,d) especially created two strong C sources (Figure 6
Based on these results, we recommend the municipal government to (1) limit northwestern urban expansion and direct the expansion to the desert areas in the northeast, trying to avoid conversion of cropland to ISA; (2) protect urban greenspaces and recover the large parks in the downtown area. Although the local government did a good job in maintaining the overall greenspace coverage in Urumqi, we found that the landscape pattern of the urban greenspaces has become highly fragmented during 1990–2010. The environmental services (e.g., climate regulation) and ecological functions (e.g., wildlife habitats) provided by the few large urban parks in downtown could not be fully compensated by many small patches of greenspaces that scattered across the metropolitan area [65
When using the post-classification method, errors in the classification maps could influence the accuracy of land-cover change detection, and affect the carbon assessment results in this study. According to the overall classification accuracies in the 1990 (90%) and 2010 (92%), the change detection error was 17%, or 64 km2
in the study area (Table 1
). Our results showed the land-cover change area from 1990 to 2010 was 191 km2
). Therefore, the change detection error might have influenced up to one-third of the actual changed area. We used the direct estimation adjustment to reduce the impacts of land-cover misclassification on our estimation of the carbon stock. Comparing the pre-adjustment and the post-adjustment results, we found the urban carbon stock could be overestimated by 2.7% in 1990 and be underestimated by 0.3% in 2010, due to the land-cover misclassifications. The relatively low impacts from the land-cover misclassifications on the carbon stock estimations were due to the similar carbon densities between the ISA (5.36 kg·C·m−2
) and remnant desert (5.59 kg·C·m−2
) lands (Table 5
), which were the major sources of the overall classification error (Table 1
). When applied in other regions with greater differences in the carbon densities among the converted land-cover types, the change detection errors from post-classification method may severely affect the carbon dynamic estimations. In such cases, some other multi-temporal classification change detection (i.e.
, pre-classification methodology) may be more suitable, especially if there is no need to quantify the detailed “from-to” change trajectory information.
Another major source of uncertainty is related to the estimated carbon density values for the urban ecosystems [66
]. Ideally, a randomized sampling with large number of sampling plots is required for a confident field investigation on carbon density. In reality, however, our sampling of ISA was limited to the 11 excavation sites during an urban construction project. Although the location of these sampling sites generally represented the major ISA types, land-use types, and soil types of the study area, the approach was not a true random sampling approach. Other SOCISA
studies faced similar problems, due to the inaccessibility of impervious-covered soil. Moreover, the SOCD of the cropland and the remnant deserts in the Urumqi metropolitan area were estimated based on the soil data from a national soil survey that focused on rural area. The underlying assumption was that urbanization-induced environmental changes might not significantly alter the SOCD of the croplands and remnant deserts that were mainly distributed in the city outskirts. Although this was a popular assumption in many previous urban carbon studies [7
], its validity in the dryland city is yet to be tested. In general, more observation data from the SOC and VEGC of urban ecosystems were needed to improve statistical confidence and reduce uncertainties.