Page 124 - JCAU-7-3

Journal of Chinese
Architecture and Urbanism Spatial analysis of urban garden space

Figure 9. Example of a principal component analysis image from Landsat
Figure 7. Sample satellite image used in classification 8 for extracting urban, rural, and vegetated environments
Source: Image by the authors. Source: Image by the authors.

Table 3. Area and percentage of land cover changes between
2000 and 2020

Year Class Area (sqkm) Percentage
2000 1 228/98686 39/64
2 292/11666 50/57
3 18/51549 3/21
4 14/35161 2/48
5 23/65527 4/09
2005 1 229/099265 39/66
2 286/892978 49/66
3 20/076934 3/48
4 13/495503 2/34
5 24/471896 4/24
6 3/671962 0/64
2010 1 77/177266 13/31
Figure 8. Normalized difference built-up index image for extracting 2 391/633878 67/54
human-made environments 3 61/689714 10/64
Source: Image by the authors 4 24/479491 4/22
5 23/626623 4/07
RF models, a paired t-test was conduced, with the null 6 1/261454 0/22
hypothesis stating that the average accuracy of the two 2015 1 67/09551 11/61
models is equal. The test results are as follows: 2 402/323197 69/64
• t-statistic: 9.24524508292494 3 53/340273 9/23
• p-value: 0.0007609242652860933
4 28/017788 4/85
Since the p-value (0.0007) is <0.05, we can conclude 5 20/545517 3/56
with 95% confidence that a significant difference exists 6 6/386251 1/11
between the accuracy of the two models. Moreover, the 2020 1 63/778973 11/04
t-statistic value of 9.24 indicates that the NNET model 2 362/704745 62/78
achieves significantly higher accuracy than the RF model. 3 94/112337 16/29
Based on these statistical results, the NNET model
demonstrates superior performance and was found to 4 28/724081 4/97
have higher accuracy than the RF model in predicting the 5 22/624144 3/92
target variable and classifying satellite images. Therefore, 6 5/764258 0/99

Volume 7 Issue 3 (2025) 8 https://doi.org/10.36922/jcau.6234