Abstract
Detecting the influence of temperature on urban vegetation is useful for planning urban biodiversity conservation efforts, since temperature affects several ecosystem processes. In this study, the relationships between land surface temperature (LST) and vegetation phenology events (start of growing season, SOS; end of growing season, EOS; peak phenology) was examined in native savannah woodland and grass parcels of a hot climate town. For comparison, similar woodland and grass parcels on the town’s periphery, and a wetland, were used. The vegetation parcel LST values (°C) in one calendar year (2023) were obtained from Landsat-8 (L8) and Landsat-9 (L9) thermal imagery, whose combination yielded an 8-day image frequency. Phenology changes relative to seasonal air temperature and LST were determined using vegetation index (VI) values computed from accompanying 30 m resolution L8-L9 non-thermal bands: the Normalised Difference Vegetation Index (NDVI) and one improved VI, the Soil Adjusted Vegetation Index (SAVI). Higher imaging frequency, 250 m resolution NDVI and Enhanced Vegetation Index (EVI) MOD13Q1 layers supplemented the L8-L9 VIs. LST correlated highly with air temperature (p < 0.001). On nearly all L8-L9 image dates, the urban vegetation parcel’s mean LST was higher (p < 0.001) than that at its peri-urban equivalent. Improved VIs (SAVI, EVI) detected some phenology events to have occurred slightly earlier than detected by the NDVI. Associated with the higher LST, the SOS was earlier in the urban than in the peri-urban woodland. This association has scarcely been demonstrated in savannah vegetation, necessitating proactive efforts to reduce potential biodiversity effects.
Similar content being viewed by others
Avoid common mistakes on your manuscript.
Introduction
Air temperature, which is related to land surface temperature (LST), has been identified as an indicator of ecosystem processes, such as vegetation phenology (Caparros-Santiago et al. 2021). Due to the heat island phenomenon (Heisler and Brazel 2010), urban environments can have slightly different thermal temporal patterns compared to the surrounding unaltered landscapes (Alexander 2020; Portela et al. 2020). In addition to affecting plants, urban temperature can have effects on insects and other faunal species (Geppert et al. 2022). Given the need for biodiversity conservation in urban areas (Rastandeh et al. 2019; Simons et al. 2022), increased understanding of the effects of urban thermal regimes can contribute to nature conservation.
Although not established as the cause, changes in the LST have been shown to be related to the start of the growing season (SOS) and the end of the growing season (EOS) in temperate grasslands (Li et al. 2021) and forests (Park and Jeong 2023). However, its relationship with these phenological events in urban areas located in warmer climates is less well documented. At continental scales, NourEldeen et al. (2020) and Yan et al. (2020) showed that there are climate-related regional variations in long-term LST changes. Zou et al. (2020) showed that non-urban LST was correlated with atmospheric (air) temperature, reaching its peak in summer. Vegetation in leaf, particularly trees, often induces cooler LST and, therefore, its removal or senescence can result in high LST.
Thermal images from satellite-borne sensors provide a useful, repetitive source of LST data (Li et al. 2023). Due to their high spatial resolution (100 m, resampled to 30 m in the delivered product), Landsat-8 and -9 thermal images were some of the most widely used for LST retrieval at the time of this study. They were, in part, generated using ASTER (Advanced Spaceborne Thermal Emission and Reflection Radiometer) Global Emissivity Database (GED) data, which were found to have low error (Hulley et al. 2015). The thermal infrared sensor-2 (TIRS-2) on Landsat-9 has radiometric improvements over the Landsat-8 TIRS, its predecessor (Montanaro et al. 2022). These improvements suggest that TIRS-2 thermal images should be preferred to Landsat-8’s TIRS images. However, despite the TIRS-2 comparative advantages, the joint use of TIRS and TIRS-2 images has the benefit of providing more LST data of a given site, at a higher temporal frequency (8 days) than using imagery from only one of them (16 days).
Vegetation index (VI) values that employ the contrast between high near infrared (NIR) and low red energy reflectance (ρ) by turgid leaves, due to absorption by chlorophyll, can potentially indicate the onset of vegetation greening (SOS, when VI values to start increase consistently) and senescence (EOS, when VI values start to decrease consistently). The Normalized Difference Vegetation Index (NDVI; Rouse et al. 1974) has been widely used in studies of urban vegetation greenness or its detection relative to LST (Abdulmana et al. 2021; Palafox-Juárez et al. 2021; Worku et al. 2021). However, it is susceptible to background soil reflectance and can saturate at high green vegetation biomass (Huete et al. 2002; Rondeaux et al. 1996). The Soil Adjusted Vegetation Index (SAVI; Huete 1988) and its derivatives like the Modified SAVI (MSAVI; Qi et al. 1994) and Optimized SAVI (OSAVI; Rondeaux et al. 1996) overcome the background soil reflectance problem. In the SAVI, this is accomplished by including a canopy background adjustment factor, L (= 0.5) in the NDVI formula’s denominator. The Enhanced Vegetation Index (EVI) maximises the vegetation signal by using NIR, red and blue ρ in its formula, as well as L set to 1 and two coefficients (C1 = 6, C2 = 7.5) which correct for atmospheric aerosol scattering (Huete et al. 2002).
There are differences in the accuracies of different VIs at detecting vegetation phenology events. In some studies, the EVI has been found to be more accurate than the NDVI. Cui et al. (2023) compared the abilities of the NDVI, EVI, and a third VI (the near-infrared reflectance of vegetation) to detect deciduous broadleaf forest phenology and found that the EVI outperformed the NDVI at detecting the SOS, but that the NDVI was better at detecting the EOS. Zou et al. (2020) showed that LST was negatively correlated with vegetation greening as indicated by the OSAVI. Other studies have shown this negative relationship between LST and a VI, using the NDVI (Guha and Govil 2022; Jaber 2021) and EVI (Xu et al. 2020). The high frequency (daily) MODIS (MODerate resolution Imaging Spectroradiometer) images have proved to be a useful way of monitoring vegetation phenology in large areas through VI changes (Gao et al. 2023; Xu et al. 2020), but imagery with higher spatial resolution can be advantageous (Cui et al. 2023; Ma et al. 2022). Atmospheric correction of the multitemporal images is required if changes in vegetation phenology based on multi-date VI values are to be detected accurately (Zeng et al. 2020).
In this study, variations in LST at native vegetation sites in and near a hot climate, semi-arid town are examined relative to meteorological station air temperature data, and then related to the vegetation’s phenology. The aim was to establish whether urban temperatures influence the dates of vegetation phenology events. The objectives were: (1) to establish the strength of the relationship between land surface temperature and seasonal air temperature changes in the urban environment, and (2) to determine the relationship between land surface temperature and the greenness cycle of native vegetation in the urban environment. The greenness cycles at the urban native vegetation parcels were established using VI images, relative to similar vegetation parcels outside the urban environment. The ecological significance of the relationship between urban LST and vegetation greenness is then evaluated.
Material and methods
For a small town in a developing world setting, the phenology of urban and peri-urban vegetation parcels was compared using trends in VI values from coarse resolution (30 m – 250 m) open access imagery. LST values were also derived from open access imagery and related to the phenological changes. Rainfall and air temperature data from meteorological station records aided the interpretation of the relationship between LST and phenological changes.
Study area
The study was conducted in Mahikeng (Fig. 1a), the capital city of the North West Province of South Africa. The Mahikeng Central Business District (CBD) is centred at approximately 25°38′40"E, 25°52′0"S (Fig. 1b). Vegetation parcels near and just outside the Mahikeng CBD (detailed in ‘Study sites and experimental design’) were used as study sites. Although there were suggestions that South Africa’s season demarcations needed to be re-defined (van der Walt and Fitchett 2020), at the time of this study the country’s seasons (which are the opposite of northern hemisphere seasons) were generally recognised as: summer from December to February, autumn from March to May, winter from June to August, and spring from September to November (Klopper et al. 1998).
Mahikeng has a semi-arid climate, receiving 500–600 mm of rainfall annually, mainly during the rainy season from October/November in one year to March/April in the following year (Mucina and Rutherford 2006). It is in the transition zone between the savannah and grassland biomes, with estimated mean annual potential evaporation of 2 825 mm (Mucina and Rutherford 2006). The savannah vegetation typically consists of scattered trees interspersed with grass patches; grass phenology being driven mainly by rainfall (i.e., water as opposed to temperature), with tree phenology being driven by temperature and length of day (Boke-Olén et al. 2016; Currier and Sala 2022). Different savannah tree species can have different phenological cycle dates in the same location (Sekhwela and Yates 2007). In the study area, days lengthen with an increase in the solar zenith angle. The savannah trees are predominantly deciduous and the grasses perennial, entering senescence soon after the end of the rainy season and re-sprouting soon after consistent rainfall. Wildfires, which are mainly human-induced, are common in the area. They occur as soon as the grass is dry enough to burn, mainly during the ‘fire season’ from April to October each year. Fire in grass parcels usually induces early re-sprouting of the grass during the dry season, which can result in the detection of a false SOS. However, it damages tree leaves, to the extent that they may not re-emerge if the fire occurs soon after the spring leaves have established. This can result in the detection of a false EOS and no peak phenology.
Study sites and experimental design
The analysis was performed for all seasons in one calendar year, from January to December 2023. To detect whether LST and the dates of phenological events (SOS, EOS, peak vigour) differed between within-town (i.e., urban) vegetation and that at the periphery of the urban area (i.e., ‘peri-urban’ vegetation), one woodland (centred at 25°37′52"E, 25°50′57"S; Fig. 1b), and one grass (centred at 25°37′48"E, 25°50′53"S) parcel in Mahikeng were selected. The woodland parcel (9.1 ha) consisted of remnants of the original savannah woodland on a piece of land where no urban construction had yet occurred. The grass parcel (2.3 ha) was adjacent to the woodland (Fig. 2a, Fig. 2b), and both were located near the town’s industrial area (the Industrial Site; Fig. 1b), at 1 281 m (m) above sea level (a.s.l.). For comparison with these respective urban vegetation parcels, one woodland (4.7 ha, at 1 292 m a.s.l.) and one grass parcel (2.1 ha, at 1 290 m a.s.l.) located just outside Mahikeng (centred, respectively, at: 25°40′26"E, 25°51′37"S and 25°40′11"E, 25°52′5"S; Fig. 1b) were selected. These peri-urban parcels were in the grazer- and browser-excluded section of a conserved area (Mahikeng Nature Reserve), to the southeast of the Mahikeng central business district (CBD). They had the same soil type (petric calcisol) and underlying geology (andesite) as their equivalent urban vegetation parcels and had similar (flat) topography. Both the urban and peri-urban grass parcels had scattered woody shrubs. An additional, 6.1 ha wetland vegetation parcel (centred at 25°39′9"E, 25°52′14"S; Fig. 1b) was included in the analysis. In this water-limited environment, the wetland, whose vegetation was hydrophyte grasses and reeds (e.g., papyruses; Fig. 2c), was expected to exhibit less influence from rainfall on vegetation phenology. This permitted the isolation of the effect of changes in temperature.
After field verification, the extents of the vegetation parcels were selected by carefully examining Landsat images of the area that had been acquired during the 2023 calendar year (see ‘Satellite image data’) to identify locations that had experienced no fire events. Very few locations in the Mahikeng urban area were spared from wildfires in that year, which limited the spatial extent of the vegetation parcels that were studied. Therefore, the boundaries of the vegetation parcels were delineated by manually digitising them on a high (10 m) spatial resolution Sentinel-2 MSI image using ArcMap 10.8.2®, to exclude sections that had evidence of prior fire (dark fire scars, Fig. 1b) in the 2023 fire season. Parts of the urban woodland parcel, for example, experienced fire events and were, therefore, excluded from the analysis. The dominant trees in the two woodland parcels were of the Vachellia (previous taxonomy: Acacia) genus. There were diverse but similar mixtures of climax and sub-climax grasses in the studied grass and woodland parcels, whose most common genera included Aristida, Brachiaria, Eragrostis, Heteropogon, and Hyparrhenia.
Satellite image data
Landsat
To obtain LST data at the respective vegetation parcels that were studied (see ‘Study sites and experimental design’), as well as VI values at high spatial resolution, Level-2 Science Product (L2SP) Landsat-8 Operational Land Imager (OLI)_TIRS and Landsat-9 OLI_TIRS-2 images of the study area (path 172, row 078) were downloaded from the United States Geological Survey (USGS) site EarthExplorer®. The L2SP images had atmospheric and geometric corrections and were in the Universal Transverse Mercator (UTM) projection (zone 35 South), and WGS 84 (World Geodetic System 1984) datum. There were 45 (8 days apart) image dates for the study area in 2023 (Table 1), consisting of 23 Landsat-8 and 22 Landsat-9 acquisitions.
MODIS
To detect the SOS, EOS, and peak vigour dates more accurately at the respective vegetation parcels that were studied (see Sect. 2.2), the high temporal frequency MOD13Q1 version 6.1 (v6.1) vegetation index images (Didan 2021) were downloaded from the LPDAAC (Land Processes Distributed Active Archive Center). These Terra MODIS VI data were generated every 16 days at a spatial resolution of 250 m. Included in the MOD13Q1 v6.1 product were NDVI and EVI images. Images of the calendar year period from January to December 2023 were downloaded. The algorithm that generated the VI data chose the best available pixel value from all the acquisitions from the 16-day period (Didan 2021).
Weather station data
For use in interpreting the changes in VI values and LST from the image data, rainfall and temperature data for the period from 1 January to 31 December 2023 were obtained from the South African Weather Service (SAWS). The ideal would have been rainfall and temperature data at each of the vegetation parcels. However, there were no weather stations at the sites. Therefore, the data recorded at the main weather station at Mahikeng International Airport, approximately 10 km northwest of the Mahikeng CBD (Fig. 1b), were used. The data consisted of daily total rainfall, daily minimum air temperature (usually attained between midnight and dawn on a given calendar date), and daily maximum air temperature (usually attained between noon and sunset on a given calendar date).
Image processing
Pre-processing of Landsat data
Out of the 45 image dates (Table 1), only 28 (14 Landst-8, 14 Landsat-9) images had no cloud cover over the study sites. On the 24 February Landsat-8 image, a cloud obscured the wetland parcel, but the other study sites were not covered by clouds. This image was used for the analysis despite the omission of the wetland parcel, to add to the 28 cloud-free images. Therefore, only 29 Landsat image dates were used for further analysis, due to cloud cover problems (mainly during the January-April and November–December rainy season periods).
LST values were retrieved using ERDAS Imagine 2022®, from the TIRS/TIRS-2 band 10 (B10) that was supplied in each of the 29 images that were downloaded. This was accomplished by multiplying the unsigned 16-bit integer digital number (DN) values by the B10 surface temperature (ST) multiplication parameter (0.00341802), and then adding the addition parameter (149.0) supplied in the metadata. This process yielded temperature values in Kelvins (K), which were converted to °C by subtracting 273.15 from each temperature. To retrieve the reflectance (ρ) values from the unsigned 16-bit integer values, the pixel DN values of the nonthermal bands were multiplied by the surface reflectance (SR) scale factor (0.0000275), and then the offset (+ -0.2) value was added to each value. These values were supplied in the metadata. VI values were then computed from the ρ images using Eq. (1) – Eq. (3), with the following band specifications: ρBlue as OLI band 2 (B2), ρRed as OLI B4, and ρNIR as OLI B5. The goal of using VI values was merely to derive the dates of vegetation phenology events, as opposed to determining the optimal VI. Having been developed and optimized for MODIS bands (Huete et al. 2002), the EVI did not perform well with the OLI bands and was, therefore, not employed further. Therefore, the NDVI and one VI with an improvement over the NDVI, the SAVI, were used.
Using the respective shapefiles of their boundary polygons, the vegetation parcels were then extracted from each LST and VI image using image sub-setting procedures. Summary LST and VI statistics from the extracted pixels were then derived for use in graphing and statistical test analyses (see Sect. 2.5).
Pre-processing of MODIS data
NDVI and EVI values were retrieved from the MOD13Q1 v6.1 VI images using ERDAS Imagine 2022®, by multiplying the signed 16-bit integer DN values by their scale factor (0.0001) which was supplied in the metadata. For each 16-day VI image, the respective vegetation parcels were then extracted by image sub-setting procedures using their digitised boundary polygon shapefiles. Summary VI statistics from the extracted pixels were then derived for use in graphing and statistical test analyses (see ‘Statistical analysis’).
Statistical analysis
Analysis of variance
From the five-study site experimental design, the statistical significance of differences in the vegetation parcel LST values per Landsat image date was assessed using one-way analysis of variance (ANOVA), using all pixels per respective vegetation parcel. The resulting variance ratio (F) statistic from the ANOVA indicated the extent to which the variance between groups (the five vegetation parcels) was higher than that within the groups. The statistical significance of differences in mean LST between the paired vegetation parcels in the experimental design (i.e., urban grass versus peri-urban grass, urban woodland versus peri-urban woodland) was assessed using the t-test (two tail, assuming unequal variances).
Correlation analysis
The relationship between air temperature (from the weather station) and LST was examined using Pearson’s product moment correlation coefficient (r). For each vegetation parcel on a given cloud-free Landsat image date (Table 1), r values were computed for the following relationships, using all pixels: (1) maximum temperature on the day before the Landsat image acquisition date (i.e., pre-image date maximum temperature) versus mean LST, (2) the minimum (morning) temperature on the Landsat image acquisition date (i.e., image date minimum temperature) versus mean LST, and (3) the average (mean) of daily maximum temperatures from the eight days before the image date (i.e., pre-image date 8-day maximum temperature mean) versus mean LST.
Trend inflection point analysis of phenological change
Graphical plots of VI (MOD13Q1, OLI) and LST values, respectively, against image dates were used to detect phenological trends and the SOS, EOS, and peak vigour dates. Along the respective date progressions, the successive inter-date (date 1, date 2) change in VI value (ΔVI = VIdate 2 – VIdate 1), i.e., the first derivative, was then used for identifying the turning (inflection) point in the plotted phenology graphs. The EOS was identified as the autumn (March–May) 16-day (MOD13Q1) or 8-day (Landsat) period when the VI values started to decrease consistently (i.e., consistent negative ΔVI values), and the SOS was identified as the late winter/early spring (August–September) 16-day or 8-day period when the VI values started to increase consistently (i.e., consistent positive ΔVI values). Variants of this derivative technique, whose advantage is simplicity, have been used in related studies (Arjasakusuma et al. 2021; Jeganathan et al. 2014; Zhang et al. 2022). Peak vigour was identified as the period when the VI values were at their highest. In the plots that used MOD13Q1 VI (NDVI, EVI) data, the mean maximum air temperature values from the 16-day period leading up to the image date, and the total rainfall for the same 16-day period, were included. In the plots of VI (NDVI, SAVI) values from the Landsat images, the vegetation parcel mean LST on image date, the mean maximum air temperature values from the 8-day period leading up to the image date, and the total rainfall for the same 8-day period, were included.
Results
LST and its relationship with air temperature
Air temperature had a statistically significant positive correlation with LST (Table 2). For each of the studied vegetation parcels, the pre-image date maximum temperature had the strongest positive correlation with the image date mean LST (r > 0.900, p < 0.001), followed by the pre-image date 8-day maximum temperature mean (r > 0.900, p < 0.001), and the image date minimum temperature (r > 0.800, p < 0.001). The relationship that was derived was that LST varied with the air temperature in the study area. Therefore, variations in LST can be taken to be indicative of air temperature variations. However, the maximum air temperature was not equal to the LST on a given image date. The Landsat image derived LST was at the time of image acquisition in the morning (Table 1), whereas the maximum air temperature was reached in the afternoon.
For the five-study site experimental design, ANOVA revealed that the differences in LST values were statistically significant, as indicated by the predominantly very high F-statistic values in Table 3, all of which were very highly significant (p < 0.001). On a given Landsat image date, the woodland parcel’s mean LST was slightly lower than the grass parcel mean LST in the same location context (i.e., urban woodland versus urban grass, peri-urban woodland versus peri-urban grass), and the wetland was cooler than all other vegetation parcels. Table 3 shows that the urban vegetation parcel’s mean LST was statistically significantly higher than that of its equivalent peri-urban parcel (i.e., urban woodland versus peri-urban woodland, urban grass versus peri-urban grass). The anomaly was on 31 May when the differences were not statistically significant.
Detection of phenology event dates by the vegetation indices
MOD13Q1 vegetation indices
Figure 3 shows the trends in the NDVI and EVI at the five studied vegetation parcels from all 2023 MOD13Q1 images of the study area. The EVI values were lower than the NDVI values. For any pixel, the EVI value was approximately 60% of its NDVI equivalent. The detection of phenology events also differed slightly between the EVI and the NDVI. The NDVI seemed to indicate the SOS and peak vigour approximately one 16-day period later than the EVI (Fig. 3). Given the problems with the sensitivity of the NDVI (susceptibility to background soil reflectance, saturation at high biomass), the phenology change dates that were detected by the EVI were considered more accurate. They were, therefore, used for the comparison with the phenology change detections by the OLI-derived vegetation indices.
Vegetation indices from Landsat OLI images
Figure 4 shows the trends in the NDVI and SAVI values for the five studied vegetation parcels from all the cloud-free Landsat OLI_TIRS and OLI_TIRS-2 images of the study area in 2023. The NDVI values were lower than the SAVI values. For any pixel, the NDVI value was approximately 67% of its SAVI equivalent. The two VIs had similar detections of the dates of phenological events (Table 4). However, the magnitudes of the SAVI values were considered more reliable indicators of vegetation vigour, given the improved sensitivity of the SAVI over the NDVI.
Influence of LST on vegetation phenology events
Figure 4 shows that there was covariance between the image date mean LST and the mean air temperature, particularly during the hotter pre-rainy season weather of spring and early summer (September to December). During that period, the LST on the dry vegetation parcels (grass, woodland) was much higher than the air temperature, reaching more than 51 °C when the maximum air temperature was only 38.7 °C (on 4 December). During the hotter weather, the urban vegetation parcels had higher LST than that at equivalent peri-urban vegetation parcels (i.e., urban grass versus per-urban grass, urban woodland versus peri-urban woodland; Table 3). The differences in LST were statistically significant (Table 3) and seemed to result in slight differences in the dates of phenological events (Table 4).
The spring leaf seemed to establish earlier (i.e., earlier SOS) in the urban woodland (OLI: 19 August, MOD13Q1: 13 August—28 August) than in the peri-urban woodland (OLI: 4 September, MOD13Q1: 29 August—13 September), which had lower LST. The spring leaf also seemed to establish faster in the urban woodland, as indicated by its steeper SAVI curve compared to that of the peri-urban woodland (Fig. 4c). The EVI curves in Fig. 3c, which are likely to be more accurate due to the higher MODIS imaging frequency, also show that the urban woodland curve was steeper than the peri-urban woodland curve from the 30 September—15 October to the 17 November—2 December 16-day periods, after which the two curves became equal (at peak leaf). They indicate that the EOS was slightly earlier in the urban (22 March—6 April) than in the peri-urban (7 April—22 April) woodland.
The phenology change dates indicated by the MOD13Q1 EVI images at the wetland site were expected to be more accurate indicators of the effect of temperature (as opposed to rainfall) on vegetation phenology. Based on them, the SOS for temperature-dependent vegetation was the period 13 August—28 August, peak phenology was during the period 18 February—5 March, and the EOS was in the period 22 March—6 April. However, the SOS dates that were detected by the MOD13Q1 VI images of the urban and peri-urban grass parcels were the same as those at the wetland site. This suggested that there was some measure of influence by LST on the grass SOS. However, the influence was not as pronounced as it was for the woodland parcels. Based on the OLI-derived VI values, the SOS seemed to occur later in the peri-urban (12 September) than in the higher LST urban (19 August) grass parcels.
Discussion
Factors influencing the detected vegetation phenology and their significance
Rainfall is the primary factor that influences the phenology of herbaceous vegetation in the study area. The small amount of rain in September 2023 initiated flushes of fresh grass shoots. The resulting rise in grass vigour was detected as increases in VI values at the grass parcels (Fig. 3b, Fig. 4b). Peak grass vigour, which also marked the EOS, occurred in March towards the end of the rainy season. This was indicated as the period with the highest VI values. Most of the studies that have assessed savannah vegetation phenology have used the NDVI (Boke-Olén et al. 2016; Higgins et al. 2011; Zeidler et al. 2012). Therefore, NDVI values from this study will be used in the comparison with these studies.
From herbaceous vegetation in pristine savannahs to the north of the study area, Zeidler et al. (2012) similarly detected the strongest VI (NDVI) value as being from a mid-March image (out of a Worldview and Landsat TM/ETM + image data set), with an average NDVI of 0.5 (maximum of 0.75). In this study, the maximum MOD13Q1 NDVI value in the peri-urban (undisturbed) grass parcel was 0.670 (mean = 0.656), from the 6 March—21 March period. These peak NDVI values were attained after the 2–17 February period, which had the highest total rainfall (121 mm). Given these comparable NDVI values, the peri-urban grass parcel that was used in this study was sufficiently natural (undisturbed) for comparison with the disturbed urban grass patch. For the same amount of rainfall, the urban grass patch yielded lower VI values, which can be attributed to human disturbance. Its maximum MOD13Q1 NDVI value during the 6 March—21 March period was 0.543 (mean = 0.541). The cloud cover-limited Landsat OLI image data set used in this study yielded a mean NDVI of 0.293 for the peri-urban grass patch and 0.262 for the urban grass patch, both from the 12 March image.
Rainfall influenced the phenology of the trees in the woodland patches to a much less extent. In the savannahs of the Kruger National Park, to the east of the study area, Higgins et al. (2011) observed that tree leaf flush occurred before the onset of the rains, whereas grass leaf flush followed the rain. Therefore, in the woodland patches, temperature seemed to have a greater influence. This was why the woodland VI parcels had steeper VI curves after the 13 August—28 August SOS (which was before the first rains in September) when temperatures and day length started to recover from their winter lows. The existence of scattered shrubs in the grass patches (e.g., Fig. 2b) is likely to be the reason why the grass patches had the same SOS as the woodland patches. The temperature and increased day length-induced emergence of shrub leaves was detected by the increasing VI values, as the SOS. The wetland patch phenology was controlled by a mixture of rainfall and temperature. Figure 5 shows that it maintained a core section with high VI and low LST values across the seasons, unlike the other vegetation parcels.
The higher LST in the urban grass and woodland is likely to be due to the urban heat island phenomenon, as has been indicated in many cities globally (Neil and Wu 2006). It seems to be related to the urban woodland’s earlier SOS and the steeper urban woodland VI curves during spring, which indicate a faster rate of leafing (Fig. 4c, Table 4), in comparison to peri-urban woodland. The higher LST indicates a microclimate with high air temperature, which could have stimulated earlier spring leafing in the urban woodland. Although their work was conducted in a colder climate zone, Zakrzewska et al. (2023) showed that differences in air temperature can indicate tree health. Therefore, physiologically, air temperature affects trees, indicating that differences in air temperature can affect tree phenology. For a city in a hot (tropical) climate, Kabano et al. (2021) showed that a high LST resulted in a shorter growing season length. Their results are consistent with the earlier EOS (22 March—6 April) in the urban woodland parcel in this study compared to that in the peri-urban woodland (7 April—22 April; Table 4) parcel.
Relationship between land surface temperature and vegetation phenology
As indicated in Sect. 4.1, high urban LST induced earlier spring leafing in the studied savannah woodland trees, compared to the same species in cooler peri-urban locations. This phenomenon has been observed in non-savannah vegetation in related studies. From an analysis of multiple Chinese cities, Jia et al. (2021) showed that high urban temperatures induced early SOS (by just under 10 days) and delayed EOS. Ding et al. (2020) also detected the earlier spring and later EOS in urban areas than in rural areas. In urban areas of New York State, USA, Li et al. (2022) observed early vegetation greening by about a day.
Related studies in colder climates, therefore, have shown that urban heat can result in a longer growing season compared to that in rural areas. The slightly shorter length of growing season that was detected in this study, and by Kabano et al. (2021) in a hot climate city, might be due to the moisture (seasonal rainfall) dependence of vegetation phenology in the hotter climate, compared to temperature dependence in colder climates. In the hotter climates, high urban temperatures perhaps result in higher evaporation than in rural locations, leading to moisture stress and earlier senescence. Ibrahim et al. (2021) showed that the SOS (as opposed to EOS) was the key determinant of the growing season length in the woody savannahs of their tropical (hot) study area. Therefore, increased urbanisation in tropical savannah areas is likely to affect growing seasons in the future, due to the early tree leafing effect of higher urban LST.
Comparative reliability of the tested vegetation indices
For the same pixel, the NDVI values were approximately 40% higher than the EVI values from the MOD13Q1 layers and 33% lower than the SAVI values derived from the Landsat OLI images. This is consistent with values derived by related studies, like Huete et al. (2002). The differences are likely to be due to the susceptibility of the NDVI to background soil reflectance in sparsely vegetated areas and its lower sensitivity to differences at high vegetation density (during the peak vigour period in the study area). Consequently, in studies of savannah vegetation, some authors have preferred not to use NDVI values, in favour of an improved VI. Kabano et al. (2021) used the EVI layer from the MOD13Q1 data set, instead of the NDVI. Hyvärinen et al. (2019) preferred to report SAVI values in place of NDVI values, and Folega et al. (2019) similarly opted to employ the SAVI. For predicting savannah herbaceous biomass, the SAVI has been found to be slightly more accurate than the NDVI (Munyati 2022; Svinurai et al. 2018).
Significance and limitations of the study
Earlier spring leafing in urban savannah woodlands, in association with the urban heat island effect, is a scarcely demonstrated phenomenon in the literature. This study has succeeded in showing it. Furthermore, the utility of remotely sensed image VIs for assessing vegetation phenology is demonstrated. However, cloud cover during the height of the summer rainy season in the study area is a limitation. Related studies in similar locations have reported this limitation (Zeidler et al. 2012). In this study, most of the cloud-free images were acquired in the May–October (dry season) period when the grasses were largely in senescence. During the January to April period, five of the seven cloud-free images were Landsat-8 OLI_TIRS images. Despite their lower radiometric resolution, they provided an indication of the LST, which was better than not using them. They have been found to be useful for this purpose in related studies (Carrillo-Niquete et al. 2022).
The early spring leafing of urban trees in the study area can have ecological effects. One effect is that spring flowering trees tend to flower earlier in urban areas than in their surroundings due to the heat island effect (Jochner and Menzel 2015; Neil and Wu 2006). Early flowering and elevated temperatures can then affect tree insect pollinators, such as bees (Fisogni et al. 2020; Hamblin et al. 2017) and butterflies. In some butterfly species, Kaiser et al. (2016) showed that the effects can include increased larval survival rates and the modification of the physical size of males in urban areas. Such effects can have impacts along the food chain, and further studies are needed to verify them in the predominantly Vachellia (previous taxonomy: Acacia) tree woodlands of the study area. They would then have effects on biodiversity. Mahikeng has high urban biodiversity, particularly birdlife, due to its location in a relatively undeveloped part of South Africa.
Measures to reduce the temperature contrast between urban areas and their surroundings are required to lessen the possible ecological effects. They include increasing the number of urban green spaces, which have a colling effect; the use of material that reflects, rather than absorbs, solar energy from pavements and buildings; and planning the distribution of buildings so that there is reduced build-up of heat in one location (Gago et al. 2013). The benefits would include promoting and maintaining urban biodiversity.
A longer-term study, as opposed to one year only, would have enhanced the derivations from the study. A multiple year analysis at the study sites would only be possible if the sites were protected from urban development pressures and arson fires, for purposes of the experiment. The land at the experimental sites is owned by the town’s municipality and a conservation agency, respectively, who manage it according to their goals. As a result, vegetation destruction by arson fires is uncontrollable, so the vegetation leaf content in the experimental sites would perhaps differ from year to year. Additionally, images of higher spatial resolution would have yielded more accurate results. This is because the phenology of individual tree crown would have been separated as opposed to the aggregated phenological changes and LST that were detected in the employed 30 m – 250 m pixels. Cloud cover (see Table 1) limited the use of the Landsat images and, therefore, the length of the analysis period at their higher spatial resolution.
Conclusion
Land surface temperature has a strong positive relationship with air temperature, even in the urban environment of hot climates. Therefore, its seasonal fluctuations can be inferred from meteorological station air temperature records. During hot, dry weather in the study area, the dry land LST is higher than the air temperature. There is an association between LST and the earlier spring leaf flush in urban areas, where LST is possibly also associated with a shorter growing season. In comparison to temperate locations, this association has rarely been shown in the deciduous savannah woodlands of the tropics where there have been fewer studies. It is likely to impact urban biodiversity. Therefore, the municipal authorities in these areas need to proactively adopt strategies for reducing the thermal regime contrasts between urban areas and their surrounding rural lands.
Data availability
The image data used in this study can be obtained from https://earthexplorer.usgs.gov/.
References
Abdulmana S, Lim A, Wongsai S, Wongsai N (2021) Land surface temperature and vegetation cover changes and their relationships in Taiwan from 2000 to 2020. Remote Sens Appl Soc Environ 24:100636
Alexander C (2020) Normalised difference spectral indices and urban land cover as indicators of land surface temperature (LST). Int J Appl Earth Obs Geoinf 86:102013
Arjasakusuma S, Kusuma SS, Saringatin S, Rafif R (2021) Assessing land surface phenology of the savanna ecosystem in Southeast Asia using Moderate Resolution Imaging Spectroradiometer (MODIS) leaf area index from 2002 to 2020. Appl Geomat 13:515–525
Boke-Olén N, Lehsten V, Ardö J, Beringer J, Eklundh L, Holst T, Veenendaal E, Tagesson T (2016) Estimating and analyzing savannah phenology with a lagged time series model. PLoS ONE 11(4):e0154615
Caparros-Santiago JA, Rodriguez-Galiano V, Dash J (2021) Land surface phenology as indicator of global terrestrial ecosystem dynamics: A systematic review. ISPRS J Photogramm Remote Sens 171:330–347
Carrillo-Niquete GA, Andrade JL, Valdez-Lazalde JR, Reyes-García C, Hernández-Stefanoni JL (2022) Characterizing spatial and temporal deforestation and its effects on surface urban heat islands in a tropical city using Landsat time series. Landsc Urban Plan 217:104280
Cui K, Yang J, Dong J, Zhao G, Cui Y (2023) Comparing different spatial resolutions and indices for retrieving land surface phenology for deciduous broadleaf forests. Remote Sens 15:2266
Currier CM, Sala OE (2022) Precipitation versus temperature as phenology controls in drylands. Ecol 103(11):e3793
Didan K (2021) MODIS/Terra vegetation indices 16-day L3 global 250m SIN grid V061 . NASA EOSDIS Land Processes Distributed Active Archive Center. https://doi.org/10.5067/MODIS/MOD13Q1.061. Accessed 4 January 2024.
Ding H, Xu L, Elmore AJ, Shi Y (2020) Vegetation phenology influenced by rapid urbanization of the Yangtze Delta region. Remote Sens 12:1783
Fisogni A, Hautekèete N, Piquot Y, Brun M, Vanappelghem C, Michez D, Massol F (2020) Urbanization drives an early spring for plants but not for pollinators. Oikos 129:1681–1691
Folega F, Atakpama W, Wala K, Mukete B, Shozo S, Akira O, Zhao X-H, Akpagana K (2019) Land use patterns and tree species diversity in the Volta Geological Unit. Togo J Mt Sci 16(8):1869–1882
Gago EJ, Roldan J, Pacheco-Torres R, Ordóñez J (2013) The city and urban heat islands: A review of strategies to mitigate adverse effects. Renew Sustain Energy Rev 25:749–758
Gao X, McGregor IR, Gray JM, Friedl MA, Moon M (2023) Observations of satellite land surface phenology indicate that maximum leaf greenness is more associated with global vegetation productivity than growing season length. Glob Biogeochem Cycles 37:e2022GB007462
Geppert C, Cappellari A, Corcos D, Caruso V, Cerretti P, Mei M, Marini L (2022) Temperature and not landscape composition shapes wild bee communities in an urban environment. Insect Conserv Divers 16:65–76
Guha S, Govil H (2022) Seasonal impact on the relationship between land surface temperature and normalized difference vegetation index in an urban landscape. Geocarto Int 37(8):2252–2272
Hamblin AL, Youngsteadt E, López-Uribe MM, Frank SD (2017) Physiological thermal limits predict differential responses of bees to urban heat-island effects. Biol Lett 13:20170125
Heisler GM, Bazel AJ (2010) The urban physical environment: Temperature and urban heat islands. In: Aitkenhead-Peterson J, Volder A (eds.), urban ecosystem ecology, agronomy monograph 55, American Society of Agronomy, Crop Science Society of America, Soil Science Society of America, Madison (USA), pp. 29–56. https://doi.org/10.2134/agronmonogr55.c2.
Higgins SI, Delgado-Cartay MD, February EC, Combrink HJ (2011) Is there a temporal niche separation in the leaf phenology of savanna trees and grasses? J Biogeogr 38:2165–2175
Huete AR (1988) A soil-adjusted vegetation index (SAVI). Remote Sens Environ 25(3):295–309
Huete A, Didan K, Miura T, Rodriguez EP, Gao X, Ferreira LG (2002) Overview of the radiometric and biophysical performance of the MODIS vegetation indices. Remote Sens Environ 83(1):195–213
Hulley GC, Hook SJ, Abbott E, Malakar N, Islam T, Abrams M (2015) The ASTER Global Emissivity Dataset (ASTER GED): Mapping earth’s emissivity at 100 meter spatial scale. Geophys Res Lett 42(19):7966–7976
Hyvärinen O, Hoffman MT, Reynolds C (2019) Vegetation dynamics in the face of a major land-use change: a 30-year case study from semi-arid South Africa. Afr J Range Forage Sci 36(3):141–150
Ibrahim S, Kaduk J, Tansey K, Balzter H, Lawal UM (2021) Detecting phenological changes in plant functional types over West African savannah dominated landscape. Int J Remote Sens 42(2):567–594
Jaber SM (2021) On the relationship between normalized difference vegetation index and land surface temperature: MODIS-based analysis in a semi-arid to arid environment. Geocarto Int 36(10):1117–1135
Jeganathan C, Dash J, Atkinson PM (2014) Remotely sensed trends in the phenology of northern high latitude terrestrial vegetation, controlling for land cover change and vegetation type. Remote Sens Environ 143:154–170
Jia W, Zhao S, Zhang X, Liu S, Henebry GM, Liu L (2021) Urbanization imprint on land surface phenology: The urban–rural gradient analysis for Chinese cities. Glob Change Biol 27:2895–2904
Jochner S, Menzel A (2015) Urban phenological studies - Past, present, future. Environ Pollut 203:250–261
Kabano P, Lindley S, Harris A (2021) Evidence of urban heat island impacts on the vegetation growing season length in a tropical city. Landsc Urban Plan 206:103989
Kaiser A, Merckx T, Van Dyck H (2016) The Urban Heat Island and its spatial scale dependent impact on survival and development in butterflies of different thermal sensitivity. Ecol Evol 6(12):4129–4140
Klopper E, Landman WA, van Heerden J (1998) The predictability of seasonal maximum temperature in South Africa. Int J Climatol 18:741–758
Li X, Guo W, Li S, Zhang J, Ni X (2021) The different impacts of the daytime and nighttime land surface temperatures on the alpine grassland phenology. Ecosphere 12(6):e03578
Li L, Li X, Asrar G, Zhou Y, Chen M, Zeng Y, Li X, Li F, Luo M, Sapkota A, Hao D (2022) Detection and attribution of long-term and fine-scale changes in spring phenology over urban areas: A case study in New York State. Int J Appl Earth Obs Geoinf 110:102815
Li Z-L, Wu H, Duan S-B, Zhao W, Ren H, Liu X, Leng P, Tang R, Ye X, Zhu J, Sun Y, Si M, Liu M, Li J, Zhang X, Shang G, Tang B-H, Yan G, Zhou C (2023) Satellite remote sensing of global land surface temperature: Definition, methods, products, and applications. Rev Geophys 61:e2022RG000777
Ma X, Zhu X, Xie Q, Jin J, Zhou Y, Luo Y, Liu Y, Tian J, Zhao Y (2022) Monitoring nature’s calendar from space: Emerging topics in land surface phenology and associated opportunities for science applications. Glob Change Biol 28:7186–7204
Montanaro M, McCorkel J, Tveekrem J, Stauder J, Mentzell E, Lunsford A, Hair J, Reuter D (2022) Landsat 9 Thermal Infrared Sensor 2 (TIRS-2) stray light mitigation and assessment. IEEE Trans Geosci Remote Sens 60:5002408
Mucina L, Rutherford MC (2006) Vegetation map of South Africa, Lesotho and Swaziland: An illustrated guide, strelitzia 19. South African National Biodiversity Institute, Pretoria.
Munyati C (2022) Detecting the distribution of grass aboveground biomass on a rangeland using Sentinel-2 MSI vegetation indices. Adv Space Res 69:1130–1145
Neil K, Wu J (2006) Effects of urbanization on plant flowering phenology: A review. Urban Ecosyst 9:243–257
NourEldeen N, Mao K, Yuan Z, Shen X, Xu T, Qin Z (2020) Analysis of the spatiotemporal change in land surface temperature for a long-term sequence in Africa (2003–2017). Remote Sens 12:488
Palafox-Juárez EB, López-Martínez JO, Hernández-Stefanoni JL, Hernández-Nuñez H (2021) Impact of urban land-cover changes on the spatial-temporal land surface temperature in a tropical city of Mexico. ISPRS Int J Geo-Inf 10:76
Park C-E, Jeong S (2023) Land surface temperature sensitivity to changes in vegetation phenology over northern deciduous forests. J Geophys Res Biogeosci 128:e2023JG007498
Portela CI, Massi KG, Rodrigues T, Alcântara E (2020) Impact of urban and industrial features on land surface temperature: Evidences from satellite thermal indices. Sustain Cities Soc 56:102100
Qi J, Chehbouni A, Huete AR, Kerr YH, Sorooshian S (1994) A modified soil adjusted vegetation index. Remote Sens Environ 48(2):119–126
Rastandeh A, Pedersen ZM, Brown D, Vale R (2019) Analysis of landform and land cover: Potentials for urban biodiversity conservation against rising temperatures. Urban Policy Res 37:338–349
Rondeaux G, Steven M, Baret F (1996) Optimization of soil adjusted vegetation indices. Remote Sens Environ 55(2):95–107
Rouse JW, Haas RH, Schell JA, Deering DW (1974) Monitoring vegetation systems in the Great Plains with ERTS. In: Freden SC, Mercanti EP, Becker MA (eds) Third ERTS symposium. NASA, Washington, DC, pp 309–317
Sekhwela MBM, Yates DJ (2007) A phenological study of dominant acacia tree species in areas with different rainfall regimes in the Kalahari of Botswana. Journal Arid Environ 70:1–17
Simons AL, Caldwell S, Fu M, Gallegos J, Gatheru M, Riccardelli L, Truongi N, Viera V (2022) Constructing ecological indices for urban environments using species distribution models. Urban Ecosyst 25:1745–1764
Svinurai W, Hassen A, Tesfamariam E, Ramoelo A (2018) Performance of ratio-based, soil-adjusted and atmospherically corrected multispectral vegetation indices in predicting herbaceous aboveground biomass in a Colophospermum mopane tree-shrub savanna. Grass Forage Sci 73:727–739
Van der Walt AJ, Fitchett JM (2020) Statistical classification of South African seasonal divisions on the basis of daily temperature data. South Afri J Sci 116(9/10):7614
Worku G, Teferi E, Bantider A (2021) Assessing the effects of vegetation change on urban land surface temperature using remote sensing data: The case of Addis Ababa city. Ethiopia Remote Sens Appl Soc Environ 22:100520
Xu X, Zhou G, Du H, Mao F, Xu L, Li X, Liu L (2020) Combined MODIS land surface temperature and greenness data for modeling vegetation phenology, physiology, and gross primary production in terrestrial ecosystems. Sci Total Environ 726:137948
Yan Y, Mao K, Shi J, Piao S, Shen X, Dozier J, Liu Y, Ren H, Bao Q (2020) Driving forces of land surface temperature anomalous changes in North America in 2002–2018. Sci Rep 10:6931
Zakrzewska A, Kopeć D, Ochtyra A, Potůćková M (2023) Can canopy temperature acquired from an airborne level be a tree health indicator in an urban environment? Urban for Urban Green 79:127807
Zeidler J, Wegmann M, Dech S (2012) Spatio-temporal robustness of fractional cover upscaling: a case study in semi-arid savannahs of Namibia and Western Zambia. Proc SPIE 85380S:198–207
Zeng L, Wardlow BD, Xiang D, Hu S, Li D (2020) A review of vegetation phenological metrics extraction using time-series, multispectral satellite data. Remote Sens Environ 237:111511
Zhang X, Xiao X, Qiu S, Xu X, Wang X, Chang Q, Wu J, Li B (2022) Quantifying latitudinal variation in land surface phenology of Spartina alterniflora saltmarshes across coastal wetlands in China by Landsat 7/8 and Sentinel-2 images. Remote Sens Environ 269:112810
Zou F, Lib H, Hua Q (2020) Responses of vegetation greening and land surface temperature variations to global warming on the Qinghai-Tibetan Plateau, 2001–2016. Ecol Ind 119:106867
Acknowledgements
The North West Parks Board authorities are thanked for granting access to the Mafikeng Nature Reserve (MNR), where some of the vegetation sites used in the study were located. The South African Weather Service (SAWS) is thanked for providing rainfall and temperature data. Regrettably, the trees in the urban woodland study site were cleared for urban development in January 2024, soon after this study.
Funding
Open access funding provided by North-West University. This work was funded by the North-West University (NWU).
Author information
Authors and Affiliations
Contributions
CM designed the study, conducted field work, processed data, computer-processed the images required for the study, wrote the manuscript.
Corresponding author
Ethics declarations
Conflict of interest
The author declares no competing interests.
Additional information
Communicated by: H. Babaie
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
About this article
Cite this article
Munyati, C. Relating Urban Land Surface Temperature to Vegetation Leafing using Thermal Imagery and Vegetation Indices. Earth Sci Inform (2024). https://doi.org/10.1007/s12145-024-01443-y
Received:
Accepted:
Published:
DOI: https://doi.org/10.1007/s12145-024-01443-y