*3.2. Meteorological Data and Land Surface Temperature Evaluation*

#### 3.2.1. Air Temperature and LST Trend and Relationship Analysis

Figure 6 shows the monthly maximum (Tmax), mean (Tmean), and minimum (Tmin) air temperature trends from 2014 to 2018. The values were taken from the diurnal data and were averaged per month to clearly show the monthly trend. This observation was discussed in [45] showing an upward trend in the values starting from March and continuing to April and May while values start to drop in October until around January and February. Such an observation is the same as what was presented by Estoque et al. [28] and Manalo et al. [65] in their framework showing the climate and seasons in the Philippines based on combined rainfall and temperature. Between March to May, the Philippines experiences a hot dry season which explains the high recorded air temperature.

**Figure 6.** Monthly maximum (Tmax), mean (Tmean), and minimum (Tmin) air temperature trends from 2014 to 2018.

Additionally in our paper [45], we found a significant linear correlation between air temperature (maximum, mean, and minimum) and land surface temperature (day and night) as analyzed from available daily data shown in Table 12. On the other hand, the relative humidity shows a weak correlation with the LST data although it is shown to be significant for LST\_Night.


**Table 12.** Corresponding interpretation of the quantitative values from the correlation analysis [45]. (\* not significant).

#### 3.2.2. Outdoor Thermal Comfort Assessment

Using the same meteorological data (Tmean, Relative Humidity, and Wind Speed) taken in Port Area, Manila City, from 2014 to 2018, the Physiological Equivalent Temperature (PET) thermal index was estimated through the RayMan model. The diurnal data were computed and then averaged per month and are shown in Figure 7. Additionally, the corresponding physiological stress levels for each of the values are indicated.

**Figure 7.** Monthly estimated Physiological Equivalent Temperature (PET) based on the RayMan model from 2014 to 2018.

As shown, moderate heat stress can be consistently felt in May and at some points in April and June. From July to December, light heat stress was observed, while the thermal comfort zone where there is no thermal stress only appeared in January and February. Understanding the thermal comfort in this area can also give us an idea on what is the expected outdoor thermal comfort in the other parts of Manila City. These results will be used as part of the assessment method in the latter part of the study.

#### *3.3. LULC Indicators and Evaluation Methods*

#### 3.3.1. Multivariate Cluster Analysis

From the space-time cube generated for spectral indices (NDVI, NDWI, and NDBI) used as land use and land cover indicators and top-of-atmosphere land surface temperature (TOA\_LST), the k-means clustering algorithm was used to identify the clusters within the dataset. Four groups were initialized to see a cluster for high LST (1 cluster), mid-LST (2 clusters), and low LST values (1 cluster). Standardized parameter values were plotted to clearly show the distribution of clusters, as the measurement units are not the same.

Figure 8 shows the boxplot of the result of the multivariate cluster analysis. The clustering results indicate that for the high LST cluster, values with low NDWI, moderate NDVI, and high NDBI values are clustered together. This is also expected since low NDWI correlates to low water content and high NDBI corresponds to urbanized regions. In contrast, mid-range NDVI values correspond to urbanized areas. For the low LST cluster, values are clustered with high NDWI values, low NDVI values, and low NDBI values. A high NDWI refers to a high-water content, a negative NDVI to water bodies, and a low NDBI to undeveloped regions. Consequently, two mid-LST clusters were produced because of varying parameter combinations. The first set of clusters for mid-LST (orange line) is seen to be a combination of negative NDBI, high NDVI, and a higher mid-value of NDWI which translates to lowly built-up, high vegetation with a fair amount of moisture content. On the other hand, the second set of mid-LST clusters (light blue line) is composed of NDBI, NDVI, and NDWI values close to zero which can be interpreted as areas with low to no built-up and low water content.

3.3.2. LULC Indicators and LST Correlation Analysis

The same dataset was used to see the correlation of these parameters (NDVI, NDWI, NDBI) with land surface temperature (TOA\_LST). GeoDa software was used to calculate the Pearson correlation and plot the results.

Figure 9 shows the relationship between LST and LULC indicators with their corresponding slope of linear fit and frequency distribution chart while all indicators are significant at *p* < 0.01. The results show that there is a direct relationship between LST and NDBI at a *r* = 0.361 which means that highly built-up areas have high recorded temperature values. This observation agrees with the multivariate analysis. An indirect relationship is, however, observed between LST and NDVI (*r* = −0.064) and LST and NDWI (*r* = −0.365). The low Pearson correlation value between LST and NDVI indicates that both water body values and vegetation are expected to have low temperatures while mid values correspond to being built-up. With LST and NDWI, areas with high water/moisture content are more likely to have lower surface temperatures compared to areas with low water/moisture content. Based on these results, it can be inferred that the correlation values suggest that NDWI is a better indicator than NDVI for land surface temperature, which is aligned with the findings of Alexander et al. [66]. In addition, results also suggest that NDBI is a good indicator for LST.

**Figure 9.** Relationship between LST and spectral indices with their corresponding slope of linear fit and frequency distribution chart. \*\* significant at *p* < 0.01.

#### *3.4. LST Spatiotemporal Pattern Analysis*

#### 3.4.1. Emerging Hotspot Analysis

Based on the generated Emerging Hotspot Analysis (ESHA) Map, a reclassified map was also produced to indicate areas to preserve, monitor, and intervene.

As shown in Figure 10, cold spot and hot spot areas were mapped using the trend categories and a corresponding new class.

**Figure 10.** Emerging Hotspot Analysis Map and the reclassified map with the corresponding new class.

#### 3.4.2. Local Outlier Analysis

Based on the generated Local Outlier Analysis (LOA) Map, a reclassified map was also produced to indicate areas to "preserve", "monitor", and "intervene". In Figure 11, the trend categories of clusters and outliers are shown on the left while the corresponding new class is also provided in the map on the right.

#### *3.5. Intra-Urban Island Map*

Using the generated maps presented in Sections 3.2.1 and 3.2.2, a suitability analysis model was used to combine the raster maps. The suitability analysis was carried out by giving numerical equivalents for the new classification maps for emerging hotspot analysis and local outlier analysis with a common suitability scale.

Figure 12 (left) shows the resulting suitability map with suitability values per pixel. Consequently, the equivalent Intra-Urban Heat Island (IUHI) Class of Action was produced as shown in Figure 12 (right).

**Figure 12.** Suitability Map and the reclassified suitability (IUHI) map with the corresponding new class.

In Figure 13, the final Intra-Urban Heat Island (IUHI) Map of Manila City (2013–2022) was created. To keep the map as intuitive as possible, the class of action as well as the administrative boundaries at the city, district, and barangay levels were provided. This

allows an easy understanding of the map while still showing the locations where areas need preservation, monitoring, and intervention.

**Figure 13.** Intra-Urban Heat Island Map of Manila City (2013–2022).
