*4.2. Deformation Time Series of the SFM–Def Region from 2007 to 2020*

#### 4.2.1. Long-Term Deformation in the Spatial Dimension

The long-term (2007–2010 and 2015–2020), multidimensional (along with up–down and east–west directions), and high-spatial-resolution displacements are obtained from the four frames of the ALOS-1/PALSAR data and the ascending/descending tracks from the Sentinel-1 data, using advanced IPTA and MSBAS technologies (Figure 4). The ground deformation is mainly distributed in a plain area south of the Flaming Mountains fault line (Figure 4d). There are large areas of farmland in this region (Figure 3a), and the irrigation relies heavily on groundwater. The ground deformation is mainly vertical, with small horizontal movement (Figure 4e,f), which is typical for displacements caused by groundwater extraction [60–62]. The red and magenta dotted lines in Figure 4 delineate the settlement funnel centers in 2007–2010 and 2015–2020, respectively. The area and magnitude of the subsidence in the northwest of the SFM–def region gradually decrease, but the subsidence area in the southeast gradually expands and becomes connected. The center of the funnel shifts from the northwest to the southeast, and form a giant funnel with a larger subsidence rate and area in the southeast region. See Section 5 for detailed analysis and discussion.

#### 4.2.2. Long-Term Deformation in the Time Dimension

The long-term deformation rate can reflect the spatial distribution and evolution characteristics of ground deformation. We select two profiles (AA and BB ) and two points (P1 and P2) in the SFM–def region (Figure 4) to investigate the variation characteristics of deformation in the time domain. The long-term time-series displacements at the corresponding position in the two monitoring periods, i.e., 2007–2010 and 2015–2020, are shown in Figures 5 and 6. The time-series cumulative deformation at AA and P1, BB and P2 can represent the deformation characteristics of the central region of the subsidence funnels during the two monitoring periods.

**Figure 5.** The long-term cumulative deformation at the profiles (**a**,**c**) A-A and (**b**,**d**) B-B in Figure 4.

**Figure 6.** The time-series cumulative deformation at P1 and P2 in Figure 4. The blue dots represent the InSAR observations. The magenta lines are the linear fitting results of the corresponding InSAR observations.

The long-term deformation at AA and P1 shows that the subsidence of the area with the most significant subsidence in the first monitoring period tends to be stable, and slows down in the second period. In the first period, the subsidence rate of the section northwest of AA is higher than that of the southeast section. However, in the second period, this phenomenon is reversed. The subsidence center moves from northwest to southeast, which is consistent with the spatial evolution of the global subsidence funnel. In the first period, the subsidence rate of BB is small, and presents two separate funnels. In the latter period, the two funnels merge into a giant funnel. The subsidence area and rate increases significantly.

Both ALOS-1/PALSAR and Sentinel-1 data can reflect the overall change characteristics of the subsidence in time and space well (Figures 4 and 5). However, compared with

the ALOS-1/PALSAR data, which have a revisit period of ≥46 days, the Sentinel-1 data can capture more detailed changes to the deformation signals with obvious periodicities in the time dimension due to its higher temporal resolution (≥12 days) (Figure 6). The subsidence mainly occurs in summer. The ground tends to be stable or slightly uplifted in winter. See Section 5 for detailed analysis and discussion.

#### 4.2.3. Reliability Assessment

As can be seen from Figure 4, the deformation results of the SFM–def region from the ALOS-1 data of different frames have good consistency. The deformation obtained by the Sentinel-1 data of ascending and descending tracks also has good consistency in spatial distribution and magnitude. This indicates that the TS–InSAR results have a good consistency. To quantitatively assess the reliability of the TS–InSAR results, we compare the average subsidence rates extracted from the overlapped areas of two adjacent InSAR frames acquired at the same period, e.g., ALOS-1/PALSAR datasets from AT496F840 and AT496F850, and Sentinel-1 datasets from AT143F136 and DT121F449 (Figure 2). Due to the different observation geometry of each monitoring point in different frames, we convert the LOS deformation to the vertical direction for comparison. The correlation between the results at AT496F840 and AT496F850, and Sentinel-1 results at AT143F136 and DT121F449, are 0.98 and 0.99, respectively. The root-mean-square errors (RMSEs) between them are 0.02 and 0.01 mm/year, respectively. These results show good consistency, and the differences at most points are smaller than three times the RMSE (between the red dotted lines in Figure 7).

**Figure 7.** Comparison between the results obtained by (**a**) ALOS-1/PALSAR AT496F840 and AT496F850 data, and (**b**) the Sentinel-1 AT143F136 and DT121F449 data. The red dotted lines denote the value three times the root-mean-square error.

#### **5. Discussion**

#### *5.1. Anthropogenic Factors of Ground Deformation in the Turpan–Hami Basin*

We obtained variable-scale deformation products in the Turpan–Hami basin using the proposed WAVS–InSAR method. The distribution of most detected deformation funnels (Section 4.1) is highly consistent with human activity, such as agriculture cultivation and mineral mining. The agricultural area in the SFM–def region has a funnel cluster with the largest deformation area and magnitude in the Turpan–Hami basin. We obtained the long-term and multidimensional ground deformation in the SFM–def region in Section 4.2. The subsidence center of the first period (2007–2010) shifted from the northwest to the southeast in the second period (2015–2020).

We collect optical images of the SFM–def region in 2007 and 2018 (Figure 8a,b), corresponding to the two monitoring periods. The green lines mark the locations of the greenhouses that appeared in the latter period. As the optical images show, the majority of farmland in the SFM–def region in 2007 was open-air farmland. However, in 2018, there was a large area of greenhouses, especially in the farmland far from the Flaming Mountains

fault line. Many open-air farmlands in 2007 had been changed to greenhouses (Figure 8). In traditional open-air farmland, the crops are mainly grain and cotton, which are planted in spring, managed in summer, and harvested in autumn. However, in greenhouse farmland, the expected proportion of fruit and vegetable cultivation is more than 70% [63]. After 2009, many greenhouses were built in Turpan, especially in the agricultural areas far from the southern margin of the Flaming Mountains fault line (Figure 8b). Advanced agricultural planting technologies have brought huge economic benefits to Turpan, but also increased the environmental burden, especially the demand for water [53]. Irrigation water in the SFM–def region is mainly groundwater. Hence, ground subsidence caused by groundwater overexploitation is more significant in the greenhouse areas of the SFM–def region, resulting in aquifers carrying net deficit and the subsidence center shifting to the southeast (Figure 8a,b).

**Figure 8.** (**a**,**b**) Optical images of the SFM–def region in 2007 and 2018. Green lines delineate greenhouse planting areas. (**c**,**d**) Zoom-ins of the blue rectangular area in (**a**,**b**) in 2007 and 2018. Background image: Google Maps satellite image.

Karezes are an important water supply in arid agricultural areas, known as "the fountains of life". In China, karezes are mainly distributed in the Turpan–Hami basin (Section 3.1). A karez is composed of vertical shafts, culverts, water outlets, open channels, and waterlogging dams, with length ranging from several to dozens of kilometers [51]. The number and distribution of karezes can reflect the changes to the ecological environment in the Turpan–Hami basin. It is important to evaluate the health of the aquifer. We compared two high-resolution (0.44 m) optical images covering the blue rectangular region of Figure 8a,b in July 2003 and May 2013 (Figure 9), where ground subsidence funnels in the

second period (Figure 9a,b) were developed. In July 2003, lots of small mounds—the shaft part of a karez—are linearly distributed in this area (Figure 9c). However, a lot of small mounds have disappeared in the optical image taken in May 2013, indicating the karezes in the area were severely damaged (Figure 9d). Some areas that karezes passed through were turned into farmland (Figure 9e,f). The water supply of karezes was destroyed. The water supply in this area will depend mainly on the extraction of groundwater using electromechanical wells. The karezes may have ceased to function, and dried up.

**Figure 9.** Deformation rate in the blue rectangular area of Figure 8 from (**a**) ALOS-1/PALSAR, (**b**) ascending track Sentinel-1 data. (**c**,**d**) Optical images of this area in July 2003 and May 2013, respectively. (**e**,**f**) Zoom-ins of the yellow rectangular area in (**c**,**d**). Background image: Google Maps satellite image.

In addition, we collected land cover data of the SFM–def region in 2000, 2010, and 2020 (Figure 10) (data from global Land Cover Data Product and Service website of National Basic Geographic Information Center of China (http://www.globallandcover.com/, accessed on 10 July 2022)), and the corresponding area of land cover type in each period (Table 3). The agricultural area has continuously expanded in the past two decades. Artificial areas have expanded rapidly in the past decade, more than 10 times the rate of the previous decade. Water and wetland areas have decreased in the last decade. In 2000 and 2010, the lake area and the surrounding wetland area of Aydingkol Lake was stable, indicating that surface runoff and groundwater are still effective for supply of the lake. These water sources can also partially alleviate the overexploitation of groundwater for agricultural use. However, in 2020, land cover data showed that the waters and wetland of Aydingkol Lake had almost disappeared. Farmland area is in the inner ring of Aydingkol Lake (Section 3.1). The excessive use of surface and underground water in farmland areas has seriously reduced the water supply of the lake, resulting in the shrinkage of water and wetland, which will seriously endanger the ecological environment. The transformation of local agriculture and the economy has upset the ecological balance in the SFM–def region and the balance of aquifers. Conflicts between the development of the local agricultural economy and ecological environment should arouse the attention of local governments.

**Figure 10.** Spatio-temporal evolution of the land covers in the SFM–def region in 2000, 2010, and 2020. The red and magenta dotted lines delineate the central area of the subsidence funnels during the periods 2007–2010 and 2015–2020, respectively.

**Table 3.** The area of different land covers in the SFM–def region in 2000, 2010, and 2020, obtained from Globeland 30.


Unit: km2. a: Percentage of numerical growth in 2010 compared with 2000. b: Percentage of numerical growth in 2020 over 2010. c: Percentage of numerical growth in 2020 compared with 2000.

### *5.2. Geological Explanation of Ground Deformation in the Turpan–Hami Basin*

There are many farmlands in both the Turpan and Hami depressions. Facility agriculture planting areas are also developed in other agricultural areas, e.g., the oasis areas in Hami and the western part of Turpan. However, why is there a large area of ground subsidence funnels in only the agricultural areas of the SFM–def region?

We plotted the deformation results and the corresponding optical images and faults of the oasis areas in the Turpan depression and the Hami depression (Figure 11). Rainfall is scarce in the Turpan–Hami basin. Irrigation water in the oasis agricultural areas depends on rainfall and meltwater from the surrounding mountains (Section 3.1). The Flaming Mountains fault line lies east–west in the Turpan depression, blocking water flowing from the Tianshan mountain to the south. The other areas, e.g., the northern part of the Flaming Mountains fault line and Hami, can directly obtain abundant mountain water. The surplus water in the Hami oasis can even form a river to supply the downstream area in the southwest (Figure 11d). However, the SFM–def region is short of surface water and groundwater, and the only river channel has almost dried up. As the distance from the southern margin of the fault increases, the water supply gradually decreases. The limited surface water cannot meet the continuously increasing demand for irrigation water, resulting in the continuous overexploitation of aquifers, causing the development of many subsidence funnels in this area.

**Figure 11.** Comparison of ground deformation and optical images in (**a**,**b**) Turpan and (**c**,**d**) Hami oases. Background image: Google Maps satellite image.

The climate in the Turpan–Hami basin is dry and sunny, so evaporation is serious, especially in late spring and summer, when 75% of the year's evaporation occurs. Summer is also the main period of crop growth, which demands more water for irrigation. The agriculture in the SFM–def region relies heavily on groundwater exploitation, which directly leads to the short-term sharp loss of aquifers, and accelerates surface subsidence. From late autumn to early spring, groundwater exploitation intensity in farmland decreases. The aquifers are replenished by surface runoff and groundwater reflux. This explains why the subsidence of the funnels in farmland accelerates in summer and autumn, and slows down or turns to slight uplift in winter and spring (Figure 6).
