Journal of Geographical Sciences

Investigating nonlinear factors influencing multi-scale urban land surface temperature using machine learning models

SHI Yue, FAN Qiang, SUN Shuang, SONG Xiaonan, ZHANG Bing

Journal of Geographical Sciences, 2025, 35(9): 1998-2014. DOI: 10.1007/s11442-025-2400-x

Index abbreviation	Full name of the landscape index	Landscape significance
PD	Patch density	This index reflects the overall heterogeneity and fragmentation of the landscape (Zheng et al., 2023).
LPI	Largest patch index	This index indicates the proportion of dominant patch types in the landscape and reflects the level of human activity disturbance (Zhang et al., 2021).
LSI	Landscape shape index	This index reflects the complexity of landscape shape (Ye et al., 2021).
CONTAG	Contagion index	This index indicates the degree of connectivity among different patches in the landscape (Wang et al., 2022).
SHDI	Shannon’s diversity index	This index reflects the abundance of land-use types in the landscape system (Ye and Qiu, 2021).
SHEI	Shannon’s evenness index	This index measures the evenness of land-use types across the landscape system (Ye and Qiu, 2021).
AI	Aggregation index	This index indicates the degree of aggregation or dispersion of land-use types within the landscape system (Zhang et al., 2022).

Table 2 Landscape index and its significance

Other figure/table from this article

Figure 1 Location of Shenyang city, Liaoning province, Northeast China (Note: With the map approval number GS(2024)0650, the base map remains unmodified.)
Table 1 Classification rules for different temperature zones
Table 3 Factors influencing LST
Figure 2 Spatial variation map of influencing factors in Shenyang for 2020 (Note: Owing to space limitations, only the spatial distribution of the influencing factors for 2020 is shown.)
Figure 3 Mean standard deviation classification map of LST in Shenyang for 2005, 2010, 2015 and 2020
Table 4 Area and proportion of temperature classifications
Figure 4 Pearson’s analysis results at different scales for 2005, 2010, 2015 and 2020 (*p < 0.05, ** p < 0.01)
Figure 5 Training and test set results
Figure 6 Significance ranking of factors (The size of the symbols represents the strength of the influence, while the colour gradient visually illustrates the impact of each factor on LST.)