**3. Results**

#### *3.1. Accuracy Assessment of Land Cover Maps*

Table 4 presents the accuracy assessment for the land cover types in each study year. The overall accuracies of all the classification results were more than 0.93, which means that our classification results were consistent with those obtained from the validation points.

**Table 4.** Summary of land cover classification accuracies in 1990, 2000 and 2015.


Note: Pro denotes producer accuracy; Use denotes user accuracy; the value after the symbol "±" represents the margin of error at confidence level 95%.

#### *3.2. Temporal and Spatial Changes of Land Cover Types*

Figure 4 illustrates the spatio-temporal distribution of each land cover type in the study area from 1990 to 2015. The comparisons of the percentage of each land cover type are depicted in Figure 5. Table 5 shows the ALCA and ALCR of each land cover type. The results indicate that the natural wetlands (i.e., swamp/marsh and natural open water) of the Wusuli River Basin experienced a gradual decrease from 13.79% of the total area (26,892.99 km2) in 1990 to 10.91% of the total area (21,267.23 km2) in 2015, with an ALCA of −225.03 km<sup>2</sup>/y and an ALCR of −0.96%/y. Swamp/marsh decreased by 2.96% during the period 1990–2015, while the area of natural open water increased by 0.08%. From 1990 to 2000, swamp/marsh decreased dramatically, with a change rate of −42.81 km<sup>2</sup>/y. During 2000–2015, the reduction rate of the swamp/marsh area slowed over time, with an average rate of loss of 90.45 km<sup>2</sup>/y. Despite an annual reduction area of 85.87 km2, woodland remained the dominant landscape type during the period 1990–2015, with the vast majority distributed in the Russian part of the catchment. Cropland expanded markedly with an average rate of gain of 236.01 km<sup>2</sup>/y, especially in 1990–2000 with an ALCR of 417.91 km<sup>2</sup>/y. Specifically, the distribution range of paddy field expanded from sporadic patches in 1990 to large-scale continuous areas in the Chinese section of the basin in 2015, and the area quadrupled. In contrast, dry farmland decreased overall from 15.33% of the total area in 1990 to 12.06% of the total area in 2015, mainly because of a rapid rate of decrease from 2000 to 2015 (ALCR was −545.27 km<sup>2</sup>/y). Human-made wetland, grassland, built-up land, and barren land had a slight areal increase during the period 1990–2015 with an ALCA of 4.58 km<sup>2</sup>/y, 18.61 km<sup>2</sup>/y, 13.72 km<sup>2</sup>/y, and 38.03 km<sup>2</sup>/y, respectively.

In terms of the different countries, in the Russian section of the Wusuli River Basin, woodland still accounted for more than three-fourths of the total landscape area from 1990 to 2015, despite an annual decline of 55.09 km2. For croplands, both dry farmland and paddy field decreased by a small proportion. However, swamp/marsh and natural open water increased with an ALCA of 11.68 km<sup>2</sup>/y and 4.91 km<sup>2</sup>/y, respectively.

**Figure 4.** Land cover maps of the Wusuli River Basin in 1990, 2000 and 2015.

**Figure 5.** Area percentage of each land cover type for the Wusuli River Basin in 1990, 2000 and 2015.



74

In the Chinese section, croplands were the largest land cover type from 1990 to 2015. Paddy field and dry farmland showed the opposite areal change trends. During 1990–2015, paddy fields increased more than five-fold in area, while dry farmlands decreased by 9.65%. The area of natural wetland almost halved, and their areal proportion reduced from 20.05% of the total area of the Chinese section to 10.04%, with an ALCA of 241.62 km<sup>2</sup>/y and an ALCR of −2.21%/y. The areal reduction of swamp/marsh accounted for the overwhelming majority of change in natural wetland with a loss rate of 243.08 km<sup>2</sup>/y, whereas no significant change occurred in the area of natural open water.

#### *3.3. Conversion between Natural Wetland and Other Land Cover Types*

Figure 6 and Table 6 illustrate the conversions between natural wetland and other land cover types in terms of spatial distribution and area. Most of the natural wetland conversion occurred in the Chinese section of the Basin, while only a small proportion took place in the Russian section. Across the entire Wusuli River Basin, most of the natural wetland recession occurred from 1990 to 2000, attributed to the conversion of a large area to dry farmland and paddy field, which accounted for 78.51% and 15.16% of the natural wetland reduction in this stage, respectively. During the period 2000–2015, the percentage of conversion to dry farmland and paddy field was 43.78% and 50.87%, respectively. Between 1990 and 2000, 125.88 km<sup>2</sup> of natural wetlands were converted into human-made wetlands which reduced the area of natural wetlands. In terms of the transformation of other land cover types into natural wetlands, in both stages (1990–2000 and 2000–2015), the proportion of dry farmland converted into natural wetlands was the highest, accounting for 78.76% and 54.94%, respectively. Paddy field also contributed to the increase of natural wetlands, accounting for 14.22% and 16.36% of natural wetland area recovery in the stages 1990–2000 and 2000–2015, respectively. There was little reciprocal conversion among natural wetlands and woodland, grassland, built-up land or barren land during the period 1990–2015. These results sugges<sup>t</sup> that the change in natural wetlands area can be attributed mainly to cropland reclamation and natural restoration from cropland.

For the Russian portion of the basin, the reclamation of natural wetlands covered a smaller area than their expansion during the two periods. Especially in the stage of 1990–2000, a total of 457.20 km<sup>2</sup> of natural wetlands were restored from cropland. Nevertheless, the reduced area of natural wetlands in the Chinese part of the basin was much larger than that of natural wetland restoration, suggesting a serious areal loss process.


**Table 6.** Conversion comparison between natural wetlands and other land cover types in the study area.

**Figure 6.** Dynamic conversion comparisons between natural wetlands and other land cover types. (**a**) Sankey diagram for comparison of total land cover dynamics from 1990 to 2015; (**b**) Spatial distribution of conversion between natural wetlands and other land cover types in two stages, 1990–2000 and 2000–2015.

#### *3.4. Fragmentation and Improvement of Natural Wetland and Trend of Climate Change*

Table 7 presents a comparison of the landscape metrics for natural wetlands in the Chinese and Russian sections of the Wusuli River Basin in 1990, 2000 and 2015. The landscape pattern of natural wetlands changed significantly in the Chinese part of the basin, while there was no significant change from 1990 to 2015 in the Russian part. During 1990–2015, despite a small increase in average patch area (MPS), the NP, LPI and AWMSI of natural wetlands decreased significantly in the Chinese portion, indicating that natural wetlands had undergone a loss and fragmentation process. In the Russian section, the NP, MPS, LPI and AWMSI of natural wetlands increased slightly, supporting the

improvement which the natural wetlands had experienced in the Russian portion of the basin. There were more obvious changes in the IJI of natural wetlands in the Chinese part than in the Russian part of the basin, suggesting that the existence of natural wetland was subjected to more outside interference in the Chinese portion of the Wusuli River Basin.


**Table 7.** Landscape metrics comparison of the natural wetlands for the China portion and Russia portion of the Wusuli River Basin in1990, 2000 and 2015.

Based on the Mann–Kendall test, there were no significant changes, at the 5% significant level, in the trend for the annual average temperature and annual precipitation in the Wusuli River Basin during the study period.
