**The E**ff**ects of Historical Housing Policies on Resident Exposure to Intra-Urban Heat: A Study of 108 US Urban Areas**

#### **Jeremy S. Ho**ff**man 1,2,\*, Vivek Shandas <sup>3</sup> and Nicholas Pendleton 1,2**


Received: 5 November 2019; Accepted: 3 January 2020; Published: 13 January 2020

**Abstract:** The increasing intensity, duration, and frequency of heat waves due to human-caused climate change puts historically underserved populations in a heightened state of precarity, as studies observe that vulnerable communities—especially those within urban areas in the United States—are disproportionately exposed to extreme heat. Lacking, however, are insights into fundamental questions about the role of historical housing policies in cauterizing current exposure to climate inequities like intra-urban heat. Here, we explore the relationship between "redlining", or the historical practice of refusing home loans or insurance to whole neighborhoods based on a racially motivated perception of safety for investment, with present-day summertime intra-urban land surface temperature anomalies. Through a spatial analysis of 108 urban areas in the United States, we ask two questions: (1) how do historically redlined neighborhoods relate to current patterns of intra-urban heat? and (2) do these patterns vary by US Census Bureau region? Our results reveal that 94% of studied areas display consistent city-scale patterns of elevated land surface temperatures in formerly redlined areas relative to their non-redlined neighbors by as much as 7 ◦C. Regionally, Southeast and Western cities display the greatest differences while Midwest cities display the least. Nationally, land surface temperatures in redlined areas are approximately 2.6 ◦C warmer than in non-redlined areas. While these trends are partly attributable to the relative preponderance of impervious land cover to tree canopy in these areas, which we also examine, other factors may also be driving these differences. This study reveals that historical housing policies may, in fact, be directly responsible for disproportionate exposure to current heat events.

**Keywords:** urban heat islands; environmental justice; climate change; redlining

#### **1. Introduction**

No other category of hazardous weather event in the United States has caused more fatalities over the last few decades than extreme heat [1]. In fact, extreme heat is the leading cause of summertime morbidity and has specific impacts on those communities with pre-existing health conditions (e.g., chronic obstructive pulmonary disease, asthma, cardiovascular disease, etc.), limited access to resources, and the elderly [2–4]. Excess heat limits the human body's ability to regulate its internal temperature, which can result in increased cases of heat cramps, heat exhaustion, and heatstroke and may exacerbate other nervous system, respiratory, cardiovascular, genitourinary, and diabetes-related conditions [5]. As heat extremes in urban areas become more common, longer in duration, and more intense across the US and globe [6,7] due to unmitigated human emissions of heat-trapping gases from fossil fuels [8] as well as urban expansion [9], the number of deaths and attendant illnesses are expected to increase around the US [10].

Urban landscapes amplify extreme heat due to the imbalance of low-slung built surfaces to natural, non-human manufactured landscapes [11,12]. This urban heat island effect can cause temperatures to vary as much as 10 ◦C within a single urban area [13], even without comparison to a "traditional" rural baseline for assessing UHI. Others, including Li et al. (2017), found that the density of total impervious surface area (ISA) is a major predictor for land surface temperatures, or the "surface urban heat island" studied here [14]; yet others describe the apparent cooling effects of urban green spaces. In general, greenspace, trees, or water bodies within a city have been correlated with cooler land surface temperatures (LST), and more greenspace or water is related to lower urban LST at the location of that greenspace [15–18]. Hamstead et al. (2016) studied the role of landscape composition on surface temperatures by dividing New York City into 22 classes at 3-m resolution and identifying the specific ranges of land surface temperature represented within each class [19]. The authors conclude that urban areas contain discernable "classes" of form—the integration of land use and land cover—and that those sets have "distinct temperature signatures".

Emerging research suggests that many of the hottest urban areas also tend to be inhabited by resource-limited residents and communities of color [20,21], underscoring the emerging lens of environmental justice as it relates to urban climate change and adaptation. In one study, Voelkel and others (2018) found that residents living in neighborhoods with higher racial diversity, extreme poverty, and lower levels of formal education were statistically more likely to be exposed to greater heat—the neighborhood heat effect [21]. Still other studies have found that those with the least access to resources, more advanced in age, and people with pre-existing face some of the greatest burden [22]. While the evidence about the distributional implications of heat waves mounts, we still do not have a clear uniting principle to explain consistent patterns between an emerging challenge like intra-urban heat and observable records of excess mortality and morbidity among underserved populations. If heat varies across urban environments, then why are communities of color and resource-limited communities living in the hottest areas? Could a plausible explanation be the presence of past urban planning programs and housing policies that have heightened disproportionate exposure to intra-urban heat in US cities?

The present study further examines the relationship between present-day spatial patterns of inequitable exposure to intra-urban heat and historical housing policies, which were applied to many US cities in the early 20th century. We specifically examine maps generated by the Home Owners' Loan Corporation's (HOLC) practice of "redlining" [23,24] in the 1930s. As part of a national program to lift the US out of a recession, HOLC refinanced mortgages at low interest rates to prevent foreclosures, and in the process created color-coded residential maps of 239 individual US cities with populations over 40,000. HOLC maps distinguished neighborhoods that were considered "best" and "hazardous" for real estate investments (largely based on racial makeup), the latter of which was outlined in red, leading to the term "redlining." These "Residential Security" maps reflect one of four categories ranging from "Best" (A, outlined in green), "Still Desirable" (B, outlined in blue), "Definitely Declining" (C, outlined in yellow), to "Hazardous" (D, outlined in red), relating directly to subsequent access to mortgage lending and at least partially to the racial makeup of that neighborhood.

Though redlining was banned in the US as part of the Fair Housing Act of 1968, a majority of those areas deemed "hazardous" (and subsequently "redlined") remain dominantly low-to-moderate income and communities of color, while those deemed "desirable" remain predominantly white with above-average incomes [24]. Those living in redlined areas experienced reduced credit access and subsequent disinvestment, leading to increased segregation and lower home ownership, value, and personal credit scores, even when compared to those similar-sized US cities that did not receive a HOLC map [23]. Increasingly evident is the legacy of these historic policies in racial disparities in health care, access to healthy food, incarceration, resources allotted for schools, and public infrastructure investment such as the privileging of the suburban highway system at the expense of the city's public transportation [25].

Similarly, as areas that received severely limited real estate investment over time, we might expect those areas to have fewer environmental amenities that help to clean and cool the air, including urban tree canopy [26]. Recent studies describe the increased likelihood that those who are poor and communities of color are more likely living in areas with fewer trees and poorer air quality [27,28]. At the same time, the extent to which these policies may have resulted in environmental disparity as a consequence of systematic disinvestment nationally largely remains an open question. We seek here to assess if evidence of disproportionate environmental stressors (specifically anomalous urban land surface temperatures) exists through the lens of these long-term housing policies, and if a national-scale signal varies by region in the US.

By assessing HOLC maps from aggregated urban areas in the United States (Figure 1) in relation to the relative anomaly of land surface temperature within and outside redlined areas, we ask two questions: (1) do historical policies of redlining help to explain current patterns of exposure to intra-urban heat in US cities? and (2) how do these patterns vary by geographic location of cities? Our intent is not to explain why precisely these patterns exist; instead, we seek to describe their relation through spatial analysis of historical redlining maps and present-day warm season intra-urban land surface temperature anomalies. By examining these patterns, we aim to assess how current patterns of intra-urban heat inequities may result from a combination of historical policies that may be further exacerbated by present-day planning practices that fail to center communities that have been historically underserved in adaptation and mitigation of these patterns.

**Figure 1.** Map of 108 US cities with HOLC Residential Security maps included in this study. These areas may include several smaller-area HOLC maps that have been aggregated into a larger urban area (Supplementary Materials I).

#### **2. Materials and Methods**

We use the University of Richmond's Digital Scholarship Lab's "Mapping Inequality" database (Figure 2a, Richmond, VA, USA, [29]) to download each available city's HOLC map shapefile individually (*n* = 239). To make analysis of Landsat-derived LST maps less computationally complex, we then condense the 239 unique HOLC maps into a database of 108 US cities or urban areas that overlap within Landsat 8 imagery tiles, and excluding any cities that were not mapped with at least one of all four HOLC security rating categories (*n* = 4). In some cases, HOLC map shapefile boundaries needed to remove overlapping security rating boundaries, boundary crossings over bodies of water, and to merge overlapping maps drawn in the same generalizable urban area and/or because they were drawn during different years (Supplementary Materials I).

We assess patterns of intra-urban land surface temperatures in the 108 HOLC areas using readily-accessible Landsat 8 satellite-derived northern hemisphere summertime (June–August) land surface temperatures (LSTs) following accepted United States Geological Survey calculation protocol (30 × 30 m resolution, TIRS Band 10, Normal Difference Vegetation Index [NDVI] emissivity corrected LST, Figure 2b [20,30,31]. This LST method relies on transforming raw Landsat 8 TIRS Band 10 data into top-of-atmosphere spectral radiance and then into at-sensor brightness temperatures. LST is then calculated by correcting the at-sensor brightness temperatures by surface emissivity calculated from the NDVI (derived from Bands 4 and 5 [30]). LST maps were only generated from imagery that satisfied a threshold for less than 10 percent scene cloud coverage and had to have been collected in the northern hemisphere summertime between 2014 and 2017. While these LST descriptions of intra-urban heat are coarse in spatial resolution and not the most representative of human-level, experiential air temperatures which are better resolved by dense networks of air temperature and humidity monitors [13,32], LST maps such as these have been widely applied to questions of large-scale patterns related to urban land use and heat-related public health outcomes for individual US cities [20,33].

We then use Zonal Statistics in ESRI's ArcGIS Spatial Analyst toolbox to estimate the mean of the derived Landsat 8 LSTs within each individual HOLC security rating polygon within a given urban area (e.g., Figure 2b,c). We then estimate each individual HOLC security rating polygon's land surface temperature anomaly from the area-wide mean LST from all HOLC security rating polygons (referred to as δLST, Equation (1)).

$$\text{LSTM}\_{\text{area, polyym}} = \overline{\text{LSTM}\_{\text{area, polyym}}} - \overline{\text{LSTM}\_{\text{area, all polyym}}} \tag{1}$$

This δLST estimate gives us the ability to show relatively how much warmer or cooler a particular HOLC security rating polygon is from the entire set of HOLC security rating polygons for a given urban area, and then compare these anomalies between cities in a quantitative manner.

We also estimate average percent developed impervious surface land cover [34] and tree canopy cover [35] within each HOLC polygon (Figure 2e,g) in each urban area as derived from the National Land Cover Database (NLCD) 2011 [36]. NLCD tree canopy percent is a 30 m raster dataset covering the coterminous United States, providing continuous percent tree canopy estimates derived from multi-spectral Landsat imagery for each 30 m pixel. NLCD imperviousness reports the percentage of urban developed surfaces that is impervious over every 30 m pixel in the coterminous United States and beyond. These estimates of underlying land use and overlying tree canopy may not sum to 100 percent, as tree canopy can exist over all land use types within a HOLC polygon and not all land use is necessarily impervious.

To compare δLST variations within and among HOLC security ratings between cities, we then average the estimated δLST by HOLC security rating category within each city. This binning by HOLC category yields how δLST varies between HOLC security ratings within each city. We then binned the δLSTs for each city at the national scale (*n* = 108) and by US Census Bureau regions: Northeast (*n* = 26), South (*n* = 29), Midwest (*n* = 41), and Western (*n* = 12). To estimate the significance of mean temperature differences between the HOLC security ratings by region and nationally, we apply a post-hoc ANOVA multiple comparisons test known as Tukey's Honest Significant Differences (HSD) Test. Tukey's HSD test estimates differences among group sample means for statistical significance. This pairwise post-hoc ANOVA test determines the statistical significance of differences between the mean of all pairs of group means using a studentized range distribution.

**Figure 2.** Demonstration of HOLC Security Grade δLST and land cover analysis for Richmond, VA (grey outline). (**a**) HOLC Polygons for Richmond, VA [29] (see Introduction text for explanations of HOLC security grade color designations), (**b**) LST map for Richmond, VA derived from Landsat 8 TIRS Band 10 imagery collected on 2 July 2016, and (**c**) Resulting δLSTs in HOLC polygons calculated as the anomaly of an individual HOLC polygon to the city-wide HOLC polygon average LST (see Equation (1)), (**d**) box–whisker plot of the δLSTs presented in (**c**) binned by HOLC security rating (see Introduction text for explanations of designations), (**e**) percent tree canopy from NLCD 2011 [36] averaged into HOLC polygons, (**f**) box–whisker plot of the δLSTs presented in (**e**) binned by HOLC security rating, (**g**) percent developed impervious surface from NLCD 2011 [36] averaged into HOLC polygons, (**h**) box–whisker plot of the average imperviousness presented in (**g**) binned by HOLC security rating.

#### **3. Results**

Our LST maps were generated from Landsat 8 acquisitions that satisfied a < 10 percent scene cloud coverage threshold collected from 3 June 2014 to 25 August 2017 (Supplementary Materials Table S1). These mostly sunny days provide the best conditions for Landsat 8 to reliably capture a strong LST pattern in urban areas [32]. Approximately 40 percent of the Landsat 8 imagery was collected during the 2016 northern hemisphere summer, while ~10 percent of the imagery was collected during the 2014 northern hemisphere summer. Regression tests reveal an insignificant relationship between the day that the imagery was collected and the resulting δLST patterns (Supplementary Materials Table S1).

Our analysis reveals three major trends that help to address our research questions. First, LST differences across the cities follow a non-uniform distribution of differences, suggesting that historical redlining policies are reflected in present-day intra-urban heat differentially (Supplementary Materials Table S1). Notable, intra-city δLST differences between areas given "D" and "A" HOLC security ratings range between +7.1 ◦C (Portland, OR) to −1.5 ◦C (Joliet, IL, USA), with ~94% of urban areas included in this study showing warmer present-day LSTs in their "D"-rated areas relative to their "A"-rated areas (Supplementary Materials Table S1). While Portland (OR) and Denver (CO) had the greatest "D" to "A" security rating differences within a city, the warmest δLST temperatures in formerly redlined areas relative to the city-wide average LST were identified in Chattanooga (TN, 3.3 ◦C) and Baltimore (MD, 3.2 ◦C). These cities were in contrast to formerly redlined areas that displayed, on average, cooler surface temperatures than their non-redlined counterparts (e.g., Joliet, IL, USA and Lima, OH, USA), a consistent pattern in several cities across the Midwest (Supplementary Materials Table S1). Patterns of relatively pronounced or muted δLST are underscored by attendant patterns of land use type and cover within the same HOLC security rating polygons, whereby the urban areas with the highest D-A difference and largest δLST in D-rated polygons show considerable HOLC rating-specific trends in average tree canopy and developed impervious surface percentages as compared to the Midwestern cities that exhibit cooler-than-average δLST patterns in their D-rated areas (Supplementary Materials Figure S1). The coolest δLST temperatures in areas assigned "A" HOLC security ratings relative to the city-wide average LST were identified in Birmingham (AL, −4.7 ◦C) and Roanoke (VA, −4.5 ◦C).

Regional aggregation of the city-specific trends reveals that average δLST differences between HOLC security rating categories exhibit a pattern of incremental warming relative to worsening HOLC security rating (Figure 3b–e). However, the magnitude of the δLST differences varies considerably by region, with the Midwest (*n* = 41) showing more muted δLST differences than the Southeast (*n* = 29) and West (*n* = 12), respectively (Figure 3b–e). Honest Significant Difference tests on urban areas at the regional scale reveal that the greatest δLST differences exist between "A" and "D" HOLC security rating areas across US regions, with "D"-rated areas progressively warmer than each subsequent rating in the present day. These amplified differences in the West and Southeast, as well as the relatively muted response in the Midwest (Figure 3b–e), are attended by similar differences in underlying percent land use cover (Figure 4b–e), and especially apparent in the available tree canopy (Figure 5b–e) for the areas assigned "A" HOLC security ratings.

A third trend that is consistent in a national-scale aggregation of δLSTs in these cities is the finding that "D"-rated areas are now on average 2.6 ◦C warmer than "A"-rated areas (Figure 3a). Each HOLC security rating category warms systematically relative to the more favorable neighbor security rating category (Figure 3a). Honest Significant Difference tests reveal that areas given "D" HOLC security ratings are significantly warmer than all of the other HOLC security rating categories at the national scale, in progressively larger magnitudes. These LST differences are underscored by similar, but opposing, national-scale patterns in underlying land use and tree canopy within the same redlined cities (Figures 4a and 5a), showing that areas assigned a "hazardous" HOLC security rating in US cities exhibit quantitatively less coverage by tree canopy and more coverage by impervious surfaces in the present decade [35,36].

**Figure 3.** (**a**) National-scale Land Surface Temperature Anomalies by HOLC security rating (Green, "Best," A; Blue, "Still Desirable," B; Yellow, "Definitely Declining," C; Red, "Hazardous," D) (**b**) same as (**a**)**,** but for the Midwest region; (**c**) same as (**b**), but for Northeast region; (**d**) same as (**b**), but for West region; (**e**) same as (**b**), but for South region.

**Figure 4.** (**a**) National-scale averages of underlying percent developed impervious surface [36] by HOLC security rating (Green, "Best," A; Blue, "Still Desirable," B; Yellow, "Definitely Declining," C; Red, "Hazardous," D), (**b**) same as (**a**), but for the Midwest region; (**c**) same as (**b**), but for Northeast region; (**d**) same as (**b**), but for West region; (**e**) same as (**b**), but for South region.

**Figure 5.** (**a**) National-scale averages of percent tree canopy [35,36] by HOLC security rating (Green, "Best," A; Blue, "Still Desirable," B; Yellow, "Definitely Declining," C; Red, "Hazardous," D), (**b**) same as (**a**), but for the Midwest region; (**c**) same as (**b**), but for Northeast region; (**d**) same as (**b**), but for West region; (**e**) same as (**b**), but for South region.

#### **4. Discussion and Conclusions**

We sought to understand the extent to which historic policies of redlining help to explain current patterns of intra-urban heat and the extent to which these patterns were consistent across US cities. Questions about the increasing economic inequality in US society motivated our inquiry and suggest several patterns related to historical federal housing policies and which communities experience the hottest areas of a city in the present day. Most notably, the consistency of greater temperature in formerly redlined areas across the vast majority (94%) of the cities included in this study indicates that current maps of intra-urban heat echo the legacy of past planning policies. While earlier studies document the lack of present-day services for and lower income of communities living in formerly redlined areas, this analysis presents an argument for understanding how global climate change will further exacerbate existing, historically-codified inequities in the US. We highlight three important

dimensions of our findings—built environment, policies, and current inequities—as they relate to implications of these results.

First, our findings corroborate earlier studies that describe consistent patterns between the lack of tree canopy and historically underserved urban areas, at the national and regional scales (Figures 4 and 5). The prevalence of impervious surfaces as opposed to tree canopy points to the fact that green spaces have been consistently more abundant in wealthier and majority White-identifying neighborhoods [26]. At the same time, intra-urban heat is not only affected by tree cover, since the use of different materials within varying urban typologies also amplifies temperatures [19,37]. Two features of the urban landscape—roadways and large building complexes—are well known to transform solar radiation into heat. These landscape features absorb the energy-filled short-wave radiation coming from the sun, and re-emit long-wave radiation during the diurnal heating-cooling process. As a result, large roadways and building complexes gain heat during the day and, as the evening cools ambient temperatures, the retained heat is released back into the neighborhoods, which is captured by overhead satellite sensors. These evening temperatures are precisely the factors that can exacerbate excess mortality and morbidity [38].

An earlier body of evidence from the regional studies and economics literature makes the connection between federal programs that provided incentives for major roadway and building construction projects and the fact that many of these occurred in the lowest income neighborhoods of cities [39–41]. In fact, the 1950s were an important decade for the creation of major roadways across the US, and many redlined neighborhoods were transformed and divided by road and highway infrastructure projects [42]. These changes came at a time when intra-urban heat was not recognized as a major public health hazard, and yet, given the well-known heat-absorbing capacity of asphalt and concrete [43], the selection of these materials may underlie the differences revealed in these results.

Similarly, throughout the mid-1900s large building complexes, including housing complexes, industries, and university campuses, often subsidized by the federal government, were also placed in redlined areas, largely due to the inexpensive land, and current population of largely lower income and communities of color [44]. From the 1940s through the 1970s, large buildings were made of high-density materials, such as cinder block and brick, which retain heat, and maintain high temperatures through the night [45,46]. Many of these buildings still stand, and the LST maps investigated here partially describe the thermal signature of these buildings. Areas that were in non-redlined areas, often built of other materials but also dispersed across a more natural, maintained landscape, which allows for greater circulation of air [47,48] are hence the cooler neighborhoods registered by satellites.

Second, differences in implementing policies and landscapes may help to explain the variation of temperatures across different regions of the US. The cities of Portland (OR), Denver (CO), and Minneapolis (MN), for example, notably reflect the largest differences between the formerly redlined areas and their non-redlined counterparts (Supplementary Materials Table S1). We can speculate that the redlined areas of all three cities are currently located in areas with extensive physical infrastructure, including housing complexes, railway terminals, industrial or manufacturing sites, and/or adjacent to major business centers. The presence of these current day land uses may suggest a relationship between formerly inexpensive land and large-scale development. These results, when combined with more pernicious modern-day policies that support development of high-asphalt and low tree canopy areas such as massive shopping complexes that contain large surface parking lots, are further strengthened. In Portland (OR), for example, decades of development code allowed for multifamily complexes to cover 100% of the lot area with no provisions for greening. Only recently, and due to extensive support from local researchers and community organizations did the city evaluate earlier asphalt-driving policy to require 85% lot coverage and green spaces [49,50]. Such reversals of policy are the forms of planning that can help to reverse decades of amplifying temperatures in areas that have historically been underserved. Denver and Minneapolis are also making strides, though without further understanding about the historic and present-day drivers that generate these asphalt-rich and tree canopy-poor land uses on intra-urban heat, and local communities, progress will be slow.

In addition, the coupling of landscape and historic designs of urban development in these cities may also play a role in helping to explain differences across the country. Portland, like Minneapolis, are in landscapes where tree canopy is relatively easy to sustain. Unlike arid and drought-prone areas, where planting trees can require extensive maintenance, the warm, sunny summers and wet/snowy winters of Portland, Minneapolis and other cities of the Northwest and Midwest, provide ideal conditions for expanding an urban forest, which can, in turn, reduce surface temperatures of a neighborhood. Tree planting efforts often took place as part of urban development projects in the early and mid-1900s, and were used as a way to mark special designations [42]. Similarly, metropolitan areas that conform to the concentric zone model (for example, places like Chicago, Los Angeles, and Philadelphia) tend to be larger and more densely populated metros, often with a higher degree of both affluence and inequality, a larger African American population, and a greater share of population in the suburbs. In the remaining metropolitan areas, there is greater integration between the affluent and the poor [44]. In these places, such as Seattle (WA), Charleston (WV), and Birmingham (AL), the rich are concentrated in the urban core, where redlining and tree planting efforts coincide.

Finally, indicators of and/or higher intra-urban LSTs have been shown to correlate with higher summertime energy use [51,52], and excess mortality and morbidity [20,53,54]. The fact that residents living in formerly redlined areas may face higher financial burdens due to higher energy and more frequent health bills further exacerbates the long-term and historical inequities of present and future climate change. As the results from earlier studies have documented income inequality between formerly redlined areas another other parts of US cities, we recognize that hotter areas will amplify these current inequities. Such historic income inequality leads to income segregation because higher incomes, which are further supported by past and current housing policy, allow certain households to sort themselves according to their preferences—and control local political processes that continue exclusion [55]. Other explanatory factors of these patterns, though too many for the current study and setting the stage for future studies, include disinvestment in urban areas, suburban investment and land use patterns, and the practices generally of government and the underwriting industry [39,56].

To our knowledge, this is the first study to link a historical federal housing policy to the creation (or at least the exacerbation) of a climate stressor and potential variability in resident exposure to it. While redlining most likely did not create the microenvironments that mediate LSTs relative to the rest of the urban environment, our findings suggest a strong and significant likelihood of the cauterization of current day exposure to the hottest parts of a city. While patterns of who experiences the most exposure to intra-urban heat may change as a result of (green) gentrification, which many formerly redlined neighborhoods are undergoing (e.g., wealthier communities can afford to green and change the physical landscape, and, over time, cool the hottest areas of a city), we observe consistent patterns that can be inferred as in part due to the creation of HOLC maps [23]. Future studies will need to describe the mechanisms by with planning practices—past and present—are likely to amplify the effects of climate change on historically underserved communities and communities of color.

While a growing body of evidence describes the intra-urban variation of temperatures due to characteristics of the built environment, few have asked why we observe a pattern of historically-marginalized communities living in the hottest areas. Here we have presented results from an analysis of 108 US cities that aimed to examine the role of historic "redlining" policies in mediating exposure to intra-urban heat. We found that in nearly all cases, those neighborhoods located in formerly redlined areas—that remain predominantly lower income and communities of color—are at present hotter than their non-redlined counterparts. Although the extent of differences in temperatures varies by region, the preponderance of evidence establishes that those experiencing the greatest exposure to present and potentially future extreme heat are living in neighborhoods with the least social and ecosystem services historically.

As more and more communities race to develop plans to react to and adapt to worsening extreme heat and its attendant effects on human health [57], a research agenda focused on developing place-specific, heat-mitigating urban designs and interventions [58–61] will be critical toward not only alleviating heat disparity but ensuring that the urban forms and policies that gave rise to these inequities in our past (like redlining) are recognized and altogether avoided. Furthermore, crafting climate equity-centered policies that recognize decades of disproportionate exposure to environmental stressors can help any new discoveries in urban design get implemented with focus and rapidity.

**Supplementary Materials:** The following are available online at http://www.mdpi.com/2225-1154/8/1/12/s1, Figure S1: Comparison of urban areas of relatively large or small differences in LSTs between HOLC grades, Table S1: Urban area-specific results from our LST analysis.

**Author Contributions:** J.S.H. conceived the project and coordinated the analysis, advised interpretation, and contributed to the manuscript, created figures, and coordinated the responses to reviewers. V.S. provided major contributions to the manuscript including context, interpretation, editorialization, and literature review and references. N.P. performed HOLC map and satellite imagery download and spatial analyses. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research involved no external funding.

**Acknowledgments:** J.S.H. thanks the NOAA Office of Education Environmental Literacy Program, the Virginia Academy of Science, Groundwork RVA, Virginia Commonwealth University SustainLab, University of Richmond Spatial Analysis and Digital Scholarship Labs, and the City of Richmond Sustainability Office. J.S.H. and V.S. acknowledge support from the NOAA Climate Program Office, and U.S. Forest Service's National Urban and Community Forestry Challenge Grants Program (No. 17-DG-11132544-014). N.P. acknowledges the work study program at the Virginia Commonwealth University. The authors thank four anonymous reviewers for their thoughtful and thorough consideration.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

### **Retrospective Analysis of Summer Temperature Anomalies with the Use of Precipitation and Evapotranspiration Rates**

#### **Andri Pyrgou 1, Mattheos Santamouris 2,\*, Iro Livada <sup>3</sup> and Constantinos Cartalis <sup>3</sup>**


Received: 5 July 2019; Accepted: 28 August 2019; Published: 30 August 2019

**Abstract:** Drought and extreme temperatures forecasting is important for water management and the prevention of health risks, especially in a period of observed climatic change. A large precipitation deficit together with increased evapotranspiration rates in the preceding days contribute to exceptionally high temperature anomalies in the summer above the average local maximum temperature for each month. Using a retrospective approach, this study investigated droughts and extreme temperatures in the greater area of Nicosia, Cyprus and suggests a different approach in determining the lag period of summer temperature anomalies and precipitation. In addition, dry conditions defined with the use of the Standardized Precipitation-Evapotranspiration Index (SPEI) were associated with positive temperature anomalies at a percentage up to 33.7%. The compound effect of precipitation levels and evapotranspiration rates of the preceding days for the period 1988–2017 to summer temperature anomalies was demonstrated with significantly statistical R squared values up to 0.57. Furthermore, the cooling effect of precipitation was higher and prolonged longer in rural and suburban than urban areas, a fact that is directly related to the evaporation potential of the area in concern. Our work demonstrates the compound effect of precipitation levels and evapotranspiration rates of the preceding days to summer temperature anomalies.

**Keywords:** Mediterranean; semi-arid; drought; standardized precipitation evapotranspiration index (SPEI); climate warming; soil moisture

#### **1. Introduction**

Weather regimes drive climate change and influence temperature variation [1] and may persist from a few days to a few weeks. Weather regimes in Cyprus depend on mid-latitude flow dynamics, yet they are regulated by several external factors, such as dry soils [2,3] and sea-surface temperature anomalies [4,5] that subsequently affect the development and the duration of heat waves. The feasibility of prediction of extreme temperatures in the summer using numerical models largely rests on the variability of soil moisture, sea surface temperature, and heat fluxes [6]. Variations of surface temperature after a precipitation event in the summer suggest that, due to the wet ground, more energy is likely to go into evaporation at the expense of sensible heating [7,8]. Precipitation is also associated with clouds blocking the sun and provides less energy by further reducing the temperature [7,9].

Hirschi et al. [10] divided the European domain into two sectors based on the soil moisture variations: southeast Europe with transitional soil-moisture-limited evapotranspiration regime and central European characterized by a wet soil-moisture regime (energy-limited evapotranspiration regime) [10]. A strong relationship between soil-moisture deficit and summer hot extremes in southeast Europe was noted. Droughts and heatwaves have been shown to intensify and propagate via land–atmosphere feedbacks [3]. Fischer et al. [2] argued that a large precipitation deficit together with early vegetation green-up and strong positive radiative anomalies in the months preceding the extreme summer event contributed to an early and rapid loss of soil moisture [2], resulting in low latent cooling and increased temperatures. Soil moisture deficits induce higher temperatures of about 5–6 ◦C over the initially drier region [11]. Several studies have suggested that the variations of summer climate are regulated by the soil moisture-atmosphere interactions [12–14], because soil moisture acts as a storage component for precipitation and affects plant transpiration and photosynthesis with subsequent impacts on water, energy, and biogeochemical cycles [15]. Drivers of evapotranspiration vary with climate regimes, particularly in the transitional Mediterranean climate where soil moisture is limited. Regions may switch between energy-limited and soil moisture-limited evapotranspiration regimes through the year due to land cover [15]. McHugh et al. [16] studied soil moisture in semi-arid regions and showed that atmospheric moisture may significantly contribute to variations in soil water content. The study additionally showed that maximum respiration rates could arise in the early morning [16] when soils are warm enough to stimulate microbial activity and carbon cycling, and they still contain moisture trapped through water vapor adsorption [17]. In semi-arid climates, such as Cyprus, depletion of soil moisture occurs in the early summer (May–June), but other sources of soil moisture may be fog deposition, dew formation, and water vapor adsorption [17,18].

Liu et al. [19] articulated that soil moisture memory is approximately 2–3 months in mid-latitudes and that dry initial soil moisture anomalies lead to a decrease of precipitation and an increase of surface temperature in the subsequent months, resulting in an increase of droughts and hot and cold extremes [19]. Several drought indices have been adopted that investigate droughts using precipitation data or estimation of evaporative losses, which seriously alter the natural water availability [20]. In the case of limited precipitation, moisture stays only in the upper layers, whereas in abundance of rainfall, moisture reaches the lower layers and recharges the bedrock fractures. Increased atmospheric evaporative demand due to warming, solar radiation, humidity, and wind speed lead to further drying of the areas where precipitation reduces, resulting in droughts [20] as the drying of the surface is enhanced with water scarcity. Eliades et al. [21] studied the transpiration of *Pinus bruita* trees in the mountainous area of Cyprus for the years 2015 to 2017 and evidenced that high levels of rain and soil moisture in the preceding fall months can recharge the bedrock fractures, leading to higher transpiration in the early summer [21]. However, this mechanism also depends on leaf area and rooting depth. Enhancement of air moisture in the early summer may also be dependent on transpiration and the vegetation type. Extremely high temperatures and extended drought also affect the physiological processes in plants by regulating the stomatal openings, increasing the rate of photorespiration in leaves and irreversibly damaging leaves, leading to plant death [22].

Temperature anomalies are mostly affected by external climatic conditions, such as precipitation frequency, amount of precipitation, and synoptic weather conditions. The adaptation strategies should therefore aim to modify the vulnerability component by changing the adaptive capacity of a region to withstand extremely high or low temperatures. Vulnerability may change based on human capacity, social and cultural habits, governance of a region, and physical and biological parameters [23]. However, social vulnerability differs for heatwaves and drought for people who live in poorly constructed homes, older people, and those who work in hot conditions. Management options may accelerate adaptation to climatic variability because the response of each area to environmental conditions at any moment in time depends on the current state of the system and not on its past history of exposure to events.

In this study, the relationship between ambient air temperature anomalies in Cyprus and the preceding deficit in precipitation from the previous months was investigated via a retrospective approach and a solid statistical methodology for the period 1988–2017 (inclusive). This study used the cross-correlation analysis to determine the lag period of summer temperature anomalies and precipitation. The role of land albedo with soil moisture is important, thus we compared the lag period of three different areas under the same climatic conditions with contrasting land cover. Even though the land albedo was not quantified, the different characteristics of the urban and the rural layouts were obvious through the satellite images and the noteworthy results of the analysis. Moreover, this study examined the effect of summer precipitation and related relative risk factors for higher temperatures under drought conditions in each area; the analysis was comparatively applied in urban, suburban, or rural areas in order to identify how the built environment affects urban temperatures. Drought was defined with the use of the Standardized Precipitation-Evapotranspiration Index (SPEI) multi-scalar drought index that represents both the supply and the demand sides of the surface moisture balances by investigating the evapotranspiration rate of the preceding months for three nearby stations with different land-use in a semi-arid Mediterranean country. Results demonstrate the feasibility of the development of an operational early warning system and adaptation measures in southern Europe considering the vulnerability of the area to droughts.

#### **2. Study Area and Datasets**

Cyprus (Figure 1) is an island in the eastern basin of the Mediterranean Sea with an area of 9251 km2. Cyprus has a hot summer Mediterranean climate and a hot semi-arid climate (in the northeastern part of island) according to Köppen climate classification signs Csa (Mediterranean hot summer climates) and BSh (Hot semi-arid climates) [24], with warm to hot dry summers and wet winters. The hot, dry summer lasting from May to September is affected by the low barometric centered in Southwest Asia, which contributes to the persistence of high temperatures and low precipitation levels.

Three meteorological stations were investigated: an urban station (35.17◦ N, 33.36◦ E) in the city center, a suburban station (35.15◦ N, 33.40◦ E), and a rural station (35.05◦ N, 33.54◦ E) at a distance of 21.3 km from the urban station (Figure 1). The urban, the suburban, and the rural stations are located at altitudes 160, 162, and 175 m above mean sea level, respectively (Figure 2). The maximum height of buildings is 24 m (six floors) at the urban area, 17 m (four floors) at the suburban area, and 8.3 m (two floors) at the rural area [25].

**Figure 1.** Map with urban, suburban, and rural meteorological stations in Nicosia.

**Figure 2.** Geophysical map showing the landscape surrounding the three investigated areas (urban, suburban, rural).

The daily ambient air temperature (mean, maximum, and minimum) as well as the daily accumulated precipitation were obtained from the Meteorological Service of Cyprus for the period 1988–2017 (inclusive) [26]. Only the months April to September were chosen from the continuous dataset for further investigation. No outliers or missing data existed in the final dataset, ensuring normality and homogeneity of variance throughout the series. The mean ambient air temperatures for the months May to September were 27.6 ◦C, 27.1 ◦C, and 26.7 ◦C for the urban, the suburban, and the rural areas, respectively.

#### **3. Methodology**

#### *3.1. Ambient Air Temperatures and Total Precipitation in the Urban, Suburban and Rural Areas*

For the investigated years (1988–2017), a linear trend analysis was used to estimate the statistical significance of the slope (b) of trend lines and reveal specific patterns of the local climate of the monthly values of temperatures and precipitation for months April to September for the three stations. The t-test analysis was used to allow for comparisons with other studies that investigate increasing and decreasing trends of temperature, precipitation, and climatic abnormalities [27–30]. According to the t-test analysis (Table 1) for the regression lines, the maximum air temperatures showed a steady profile throughout the years (values less than 2.048 for α = 0.05 and 28d.f), but the minimum and the mean temperatures showed a statistically increasing trend (values over than 2.048 for a = 0.05 and 28d.f).



The following table (Table 2) presents mean monthly maximum, minimum, and mean air temperatures and the total monthly precipitation for the three investigated areas (urban, suburban, and rural) for months May to September. The highest average monthly temperatures developed in

July, followed by August for all areas. Precipitation was the lowest in August with values close to zero. Moreover, the histograms (Figure 3a–i) show the distribution of these reference values.

**Figure 3.** Histograms of summer daily measurements of Tmax (**a**–**c**), Tmin (**d**–**f**), and Tmean (**g**–**i**) for urban, suburban, and rural stations.

Figure 3 shows the percentage distribution of daily mean, maximum, and minimum temperatures for the three investigated areas. According to the percentage values of Figure 3, the maximum daily values of months May until September were observed at the urban and the suburban stations with values of 36–38 ◦C appearing more frequently (highest percentage), whereas at the rural station, values of 34–36 ◦C appeared more frequently (Figure 3). The minimum daily temperatures appeared slightly increased at the urban station with values between 22–24 ◦C, whereas at the other two stations, they were lower and fairly equal (Table 2) with values 20–22 ◦C (Figure 3). The mean daily temperatures of May until September ranged mainly between 28–30 ◦C for all stations, but a closer investigation showed a steady decrease of 0.5 ◦C during the thirty investigated years.

The mean monthly total precipitation was usually lower at the urban station during the months May, June, and September, whereas for the months July and August, due to the extremely low precipitation levels (Table 2), a significant variation between the three stations could not be corroborated.


**Table 2.** Average monthly maximum (Tmax), minimum (Tmin), and mean (Tmean) air temperatures (and daily absolute maximum and minimum ambient air temperatures) and total precipitation using data from years 1988–2017 for urban, suburban, and rural stations.

#### *3.2. Temperature Anomalies*

The term temperature anomaly (Tanomaly) means a deviation from a long-term average, with positive/negative Tanomaly values indicating that the observed temperature was warmer/cooler than the reference value. Reference values were computed on local scales over a defined time period, establishing a baseline from which the anomalies were calculated. This resulted in normalization of the data in order for them to be compared and combined to a more accurate temperature pattern with respect to normal climatic values of a specific region. The average maximum temperatures of each month (Tmax of Table 2) were considered as the baseline values from which anomalies were calculated and were used for the calculation of temperature anomalies.

#### *3.3. Standardized Precipitation and Evapotranspiration Index (SPEI)*

Droughts are identified by their effect at different levels, such as duration, intensity, magnitude, spatial extent, and onset, but there is not a physical variable to quantify them. Over the years, several drought indices have been developed with the most wide usage of the Palmer Drought Severity Index (PDSI) [31,32] and the Standardized Precipitation Index (SPI) [33,34]. PDSI is based on a simplified water balance equation that incorporates prior precipitation, moisture supply, runoff, and evaporation demand at the surface level [32], whereas SPI is based on precipitation anomalies and has the advantage of analyzing different temporal scales [33].

In this study, we utilized the Standardized Precipitation and Evapotranspiration Index (SPEI), which is a commonly used index that combines the sensitivity of PDSI with changes in evaporation demand and the multi-temporal nature of the SPI [35]. Several studies showed that SPEI more accurately captures the impacts of droughts on hydrological, agricultural, and ecological variables compared to SPI or PDSI. The SPEI allows comparison of drought severity through time and space since it can be calculated over a wide range of climates and is statistically robust with clear and comprehensible calculation procedure [35–37].

The following table (Table 3) shows the categorization of the area according to SPEI values. The SPEI allows the comparison of drought severity through time and intensity and can identify the onset and the end of drought episodes. For the calculation of SPEI, the preceding month's precipitation is required for the water balance equation. SPEI was calculated on a daily basis in order to relate drought episodes to soil water content and river discharge in headwater areas. Larger time scales are used to monitor drought conditions in different hydrological subsystems, such as reservoir and groundwater storages [38].


**Table 3.** Categorization according to the Standardized Precipitation-Evapotranspiration Index (SPEI) values.

The SPEI index was calculated based on precipitation and potential evapotranspiration (PE), which was evaluated according to the SPEI package [36] in RStudio by implementing the Hargreaves equation and the log-logistic distribution of the water surplus or deficit. The Hargreaves equation [39] was preferred over other equations of potential evapotranspiration (Penman or Thornthwaite) due to its simplicity and accuracy, as it gives an estimate of the potential evapotranspiration based mainly on temperature adjusted for the sunshine hours per day and is given by:

$$PE = 0.0023 \cdot (Tmean + 17.8) \cdot (Tmax - Tmin)^{0.5} \cdot R\_a \tag{1}$$

where Tmean, Tmax, and Tmin are mean, maximum, and minimum daily temperatures (Celsius), respectively, and Ra is the extra-terrestrial radiation (MJm<sup>−</sup>2day−1), which is calculated as:

$$R\_{\alpha} = \frac{1440}{\pi} \cdot 0.082 \cdot (1 + 0.033 \cdot \cos\left(\frac{2\pi \cdot \text{Julian day}}{\text{Number of days in year (366 in leap year)}}\right) \tag{2}$$

A simple measure of the water surplus or deficit for each analyzed day (Di) is then calculated as the difference between the precipitation (PR) and the PE of each day.

$$D\_i = PR\_i - PE\_i \tag{3}$$

Vicente-Serrano et al. [35] further explored this water surplus or deficit at different time scales, adjusted it to a log-logistic probability distribution [F(D)], and proposed the climatic drought index SPEI [35].

According to Vicente-Serrano et al. [35], the standardized values of the log-logistic probability distribution [F(D)] and the soil water balance (W) values could be used for the SPEI calculation by following the classical approximation of Abramowitz and Stegun [40] and resulted in the following equation:

$$SPEI = \mathcal{W} - \frac{\mathbb{C}\_0 + \mathbb{C}\_1 \mathcal{W} + \mathbb{C}\_2 \mathcal{W}^2}{1 + d\_1 \mathcal{W} + d\_2 \mathcal{W}^2 + d\_3 \mathcal{W}^3} \tag{4}$$

where the constants are *C*<sup>0</sup> = 2.515517, *C*<sup>1</sup> = 0.802853, *C*<sup>2</sup> = 0.010328, *d*<sup>1</sup> = 1.432788, *d*<sup>2</sup> = 0.189269, and *d*<sup>3</sup> = 0.001308. The average value of the SPEI is 0, and the standard deviation is 1. For this study, daily SPEI index was evaluated using the data of the investigated time period (years 1988–2017) for the three stations.

#### *3.4. Retrospective Approach with Cross Correlation*

Soil moisture is increased with precipitation, and this consequently modifies the total energy used by latent heat flux. Therefore, more energy is available for sensible heating, resulting in the increase of ambient air temperature [15]. The effects of precipitation may prolong for a number of days and may vary according to the investigated area. The lag period of the precipitation effect on lowering daily temperatures was found using the cross correlation function (CCF analysis), which computes the correlation between two variables, x and y. If ⊗ denotes correlation, then the cross-correlation function is defined as [41]:

$$\mathbf{R\_{xy}(t) = x(t) \otimes y(t) = \int\_{-\infty}^{\infty} x(h) \, y(t+h) \, dh} \tag{5}$$

where y(t) are the precipitation levels shifted to the left by h-lag time, and x(t) is the temperature anomalies deviating from the average maximum temperatures of each month (Tmax of Table 2). The lag period used in our study was h = 0, 1, 2, ... , 30 days, thus we also had to employ daily precipitation levels in the month of April.

The cross-correlation analysis was also followed for correlating the temperature anomalies [x(t) component] with the SPEI [y(t) component]. The same lag period h = 0, 1, 2, ... , 30 days was used. The variances of the cross-correlation coefficient under the null hypothesis of zero correlation for both cross-correlation analyses were approximately 0.0002.

#### **4. Results**

#### *4.1. Temperature Anomalies and Lag Period*

In Section 3.1, the average maximum temperatures of each month were found (Tmax of Table 2) and were later used for the calculation of the temperature anomalies for the months May to September. These anomalies were divided into positive (above the average) or negative (below the average) values with the majority of them varying between −2 ◦C and 2 ◦C (as shown in Figure 4). Specifically, for months May to September and years 1988–2017 (inclusive), around 52.6%, 53.6%, and 56.7% of the temperature anomalies in the urban, the suburban, and the rural stations, respectively, varied between −2 ◦C and 2 ◦C. The most positive anomalies were found at the suburban station, and the most negative anomalies were found at the rural station. Moreover, about 2.6–3% of the temperature extremes exceeded the average monthly maximum temperature by 6 ◦C at all stations.

**Figure 4.** Air temperature anomalies above the average maximum temperatures of each month (Tmax of Table 2) for urban, suburban, and rural stations.

Figure 5 illustrates the results of the cross correlation analysis (Section 3.4) for the temperature anomalies happening after a precipitation event for the three investigated areas. The blue dashed lines in Figure 5 represent the significance limit at α = 0.05 of Rxy(t) (Equation (5)) in order to determine the statistical significance of a null-hypothesis. The variance of the cross-correlation coefficient under the null hypothesis of zero correlation for this study was approximately 0.0002, thus the approximate critical values (at the 5% level) were ±0.029 (to three decimal places). On a rainy day, day 0, an immediate drop of temperature appeared, which prolonged for six, seven, and nine days at the urban, the suburban, and the rural stations, respectively. The rural station was influenced by precipitation, resulting in a delayed increase of temperature after a rainfall event, which explained the higher percentage of negative anomalies as well as the relatively steady temperature profile without any extremities. Negative temperature anomalies prolonged for up to nine days at the rural station, signifying the importance of soil moisture for preventing extremely high temperatures in the summer. On the contrary, the urban station seemed the most susceptible to extremely high temperatures above the average maximum temperatures of each month (Tmax of Table 2), with the most days over 4 ◦C above the average maximum temperatures of each month. In urban areas, the urban environment resulted in high water runoff through the concrete structures and rapid evaporation of the overlay water, leading to a small decrease of temperature that only lasted for five to six days. He et al. [42] also indicated stronger impacts on diurnal temperature range extremes from short-term rather than long-term precipitation deficits and that low soil moisture due to precipitation deficits increase air temperatures through higher sensible heat flux [42].

**Figure 5.** Cross correlation (y) for daily maximum temperature anomalies and precipitation (left column) or SPEI (right column) in the urban station, the suburban station, and the rural station. Dotted blue horizontal lines show 95% significance limits.

The cross-correlation analysis of SPEI with temperature anomalies revealed the stronger relationship and the importance of this index. The correlation coefficient increased from −0.2 for precipitation to −0.5 for SPEI, a percentage increase of 150%. Temperature anomalies and SPEI had a negative correlation and evolved concurrently, i.e., when one parameter increased, the other decreased, and vice versa. In the case of SPEI with regards to temperature anomalies, the lag period was significantly longer: 15, 11, and 16 days at the urban, the suburban, and the rural stations, respectively.

Figure 6 shows the time-series variation of daily temperature anomalies with respect to the daily accumulated precipitation. Negative temperature anomalies were clearly observed for a rainy day, as well as for the days following a precipitation event, suggesting local climatic variations strongly controlled by the evapotranspiration of small soil moisture after the precipitation event. From Figure 6, it was also noted that there was no significant trend towards increasing or decreasing temperature anomalies in Cyprus within the last 30 years during the summer period. In contrast, temperatures showed some changes over time, with average and minimum temperatures increasing, and this was accompanied by a significant decrease in the daily temperature range (DTR) [9].

**Figure 6.** Time series of precipitation levels (blue bars) and temperature anomalies (black line) for months May to September for years 1988–2017.

Precipitation is directly linked to regional evapotranspiration, but this is trivial in cases of intense precipitation or areas under extreme drought. In the investigated areas of this study, moderate drought conditions dominated, thus it was expected that, after rainfall, the higher soil moisture would increase the evapotranspiration. Positive soil moisture anomaly led to a negative temperature anomaly mediated through a positive anomaly of evapotranspiration. Small soil moisture indicated a small evapotranspiration rate, which, according to Seneviratne et al. [15], is stronger in transitional zones between dry and wet climates. In the case of precipitation, the soil moisture may increase, leading to an evapotranspiration rate increase and a consequent decrease of temperature and negative temperature anomalies. Typically, low evapotranspiration rate is linked to lower energy used by latent heat flux and an increase in sensible heat flux and thus an increase in positive temperature anomalies. Therefore, even a small increase of the evapotranspiration rate after a precipitation event suggested higher energy used by latent heat rather than sensible heat flux, leading to fewer positive temperature anomalies compared to days not following a precipitation event.

#### *4.2. Analysis of SPEI*

The SPEI was calculated for years 1988–2017 for the months of May to September based on the temperatures and the precipitation of the preceding days. The regression lines of Table 4 show the existence of a negative relationship between the parameters SPEI and Tanomalies. They could not be used for the estimation of the daily variation of Tanomalies, as there was large variation around the mean values and the respective standard deviations of the two parameters (SPEI and Tanomalies). A statistically significant increasing trend for the time-series of the mean values of mean and minimum ambient air temperatures (Table 1) was previously proven. However, the regression lines of Table 4 reveal the higher decreasing trend of SPEI during the thirty investigated years at the suburban and the rural station, which was probably attributed to by external factors (change of land cover, meteorological conditions, etc.). More frequent temperature anomalies were observed in August for the urban and the suburban stations and in July for the rural station.


**Table 4.** Monthly mean and standard deviation (sd) values of SPEI and temperature anomalies (Tanomaly) for urban, suburban, and rural station.

According to Figure 7, the urban, the suburban, and the rural stations were mainly characterized by a normal climate (SPEI between −1 to 1) with 73.5%, 73.9%, and 74.2% SPEI values, respectively, for the five months. The highest percentages (77.6 to 86.9%) of normal climatic conditions (SPEI between −1 and 1) were observed in May, June, and September.

**Figure 7.** SPEI values for urban, suburban, and rural stations for the months May to September of years 1988–2017.

As depicted in Figure 7, July and August were the major drought months in the study area with SPEI below −1, contributing to about 35% of the total SPEI values. It is worth noting that, during July and August, no days were observed with wet conditions (SPEI over 1), whereas in May, a small occurrence of wet conditions (SPEI over 1) associated with negative Tanomalies was observed at a percentage of 13.4–14.5% (Figure 7 and Table 5).



Dry conditions with SPEI lower than −1 were associated with positive temperature anomalies (Tanomalies > 0 ◦C) at percentages from 10.7 to 31.7%. Dry conditions were associated with negative temperature anomalies (Tanomalies < 0 ◦C) at percentages from 1.4 to 15.4% (Table 5). This frequency was increased in July and August, confirming the overall drought in the area during these two summer months.

In summary, there were no large discrepancies in the monthly SPEI values between the three areas, but more severe and extreme dry conditions (SPEI less than −1.5) occurred at the rural area in July and August.

To quantitatively describe the SPEI, we calculated the percentage of positive and negative temperature anomalies for SPEI lower than −1 (drought conditions) and higher than 1 (wet conditions). The results are shown in Table 5. In May, the percentage of negative temperature anomalies with SPEI > 1 was greater than the percentage of positive anomalies combined with either SPEI < −1 or SPEI > 1, indicating a greater proportion of low temperatures occurred under wet conditions. For the months June to September, the percentage of positive temperature anomalies with SPEI < −1 was greater than the percentage of negative anomalies. Zero positive temperature anomalies were found for SPEI > 1 for months June to September, which indicated that all higher air temperatures occurred during dry conditions. No wet climatic conditions appeared during the summer, mainly due to the lack of precipitation. Positive temperature anomalies reached a peak in August under dry conditions, with occurrences of 32.0%, 33.7%, and 27.4% at the urban, the suburban, and the rural stations, respectively. Most of the temperature anomalies occurred for SPEI values between −1 and 1, with greater values in June and September. Comparison of the percentage values between the three stations revealed that most positive temperature anomalies occurred in the urban and the suburban areas, and most negative temperature anomalies occurred in the rural area.

#### *4.3. Concurrent Drought and Hot Days*

In the next stage, we investigated the occurrence of positive temperature anomalies above the average maximum temperatures of each month (Tmax of Table 2) with respect to the SPEI in order to assess whether they appeared more frequently under dry conditions. The regression analysis showed the monthly relation between the two variables—SPEI and temperature anomalies (T anomalies). The results according to Figure 8 and Tables 6 and 7 showed:


$$T\_{\text{anomaly}} = A + B \cdot SPL \tag{6}$$

**Figure 8.** Regression lines for SPEI values and temperature anomalies (degC) for months May to September of years 1988–2017 for (**a**) urban station, (**b**) suburban station, and (**c**) rural station.


**Table 6.** Linear regression analysis of temperature anomalies with respect to SPEI for the three stations and for the months May to September of the years 1988–2017, showing the regression equation (Tanomaly = A + B·(SPEI)), the t-test of the statistical significance of A and B, and the adjusted R<sup>2</sup> correlation coefficients.

**Table 7.** Paired t-test for the statistical significance between Ai.Aj and Bi.Bj (for t values greater than 1.96, the differences were statistically significant for α = 0.05).


According to Table 6, for all cases, the coefficients A and B were found to be statistically significant (bold values) with |tA| > t0.05 = 1.96 and |tB| > t0.05 = 1.96.


#### **5. Discussion and Conclusions**

Through linear and cross-correlation statistical analysis, this study examined the compound effect of precipitation levels and evapotranspiration rates of the preceding days to summer temperature anomalies for years 1988–2017. The observations of the time-series figure (Figure 6) and the cross-correlation results showed that the cooling effect of precipitation was higher and lasted more in rural and suburban areas compared to urban areas, a fact directly related to the evaporation potential of the area concerned. We showed that precipitation was the dominant driving force of positive temperature anomalies and that varying evapotranspiration rates contributed to the development of moderate to severe drought in the investigated areas.

Particularly, the investigation of temperature anomalies showed a higher correlation for the concurrent month's precipitation compared with precipitation in the preceding months, suggesting that moisture was depleted faster. This showed that there was a lag effect of soil moisture memory of six, six, and nine days in the urban, the suburban, and the rural areas, respectively. In warmer areas (urban and suburban areas), the larger evaporative demand from the atmosphere exacerbated the existing drought conditions and its impacts. Also, the higher urban and suburban temperatures (Table 2) compared to the rural area could significantly reduce the natural storage of water. With view of the precipitation events, the negative temperature anomalies suggested local climatic variations strongly controlled by the evapotranspiration of small soil moisture after the precipitation event. The SPEI was later used that employed both precipitation and evapotranspiration rates to characterize dry or wet conditions. The cross-correlation analysis of SPEI with temperature anomalies revealed the stronger relationship with negative correlation coefficient of −0.5 and highlighted the importance of this index. In the case of SPEI with regards to temperature anomalies, the lag periods according to the cross-correlation analysis were significantly longer: 15, 11, and 16 days at the urban, the suburban, and the rural stations, respectively. The higher surface albedo of the urban infrastructure may have led to additional warming. This does not necessarily translate to drier conditions and longer droughts, but it creates challenges for better water reservoir management.

According to this study, the SPEI has a high correlation with temperature anomalies and may be considered as a key tool for the identification of abnormal weather conditions and extremely high temperatures. Moreover, it confirmed that rainfall events combined with evapotranspiration, which could be effectively represented by SPEI index variation, may be the main regulators of soil moisture rather than the amount of monthly rainfall [43,44]. In the results section, the temperature anomalies were inversely correlated with precipitation anomalies, and the SPEI index and the linear regression coefficients were found. High temperatures during the summer months may be understood by the investigation of the soil moisture to understand the impact of soil storage memory on ambient air temperatures. Further analysis could focus on the division of temperature anomalies based on the amount of rainfall as well as the intervals between rainfall events. We should consider the effects of not only precipitation but also evapotranspiration in future studies to better understand the length of extreme weather conditions.

Further analysis focused on the statistical investigation of the linear regression lines of the SPEI with temperature anomalies for the three stations and for each month. The results of the paired t-test for the statistical significance showed that the coefficients A of Table 7 were considered statistically equal between them for all pairs, indicating that the three investigated areas were nearby. The B coefficients suggested that external factors (land cover, meteorological conditions, etc.) differently affected the three stations during the thirty investigated years. This study focused on the analysis of the effect of precipitation during the summer period on temperatures and particularly the deviation of temperature from the mean monthly value. The spatial investigation revealed a similar climatic profile in all three investigated areas but showed a noteworthy different lag effect of precipitation. Particularly, precipitation in rural areas led to a longer decrease of temperature compared to the urban and the suburban areas because the wet ground favored the increased evapotranspiration and the

decrease of sensible heat flux. Later, the investigation of SPEI further supported the above statement, because SPEI was strongly negatively correlated with positive temperature anomalies.

Future work should focus on the effect of the intervals between precipitation events in urban, suburban, and rural areas. In this study, the semi-arid climate in Cyprus and the infrequent precipitation allowed a more comprehensive understanding of the lag effect of precipitation during the dry period (summer) in areas with different land cover. The lag period may vary seasonally; therefore, further investigation during the winter is necessary. The investigation of the transitional phase of dry and wet climates in Cyprus will likely confirm the strong soil-moisture climate coupling, which is the strong dependency of evapotranspiration on soil moisture during the dry periods and the little impact of soil moisture on evapotranspiration during the wet periods.

**Author Contributions:** M.S. conceived the research topic. A.P. obtained the datasets, created the figures and analyzed the results. I.L. designed the methodology and did the statistical analysis of the data. All authors (A.P., M.S., I.L. and C.C.) contributed in the discussion of the results and reviewed the manuscript.

**Funding:** This research received no external funding.

**Acknowledgments:** The authors are grateful to the Ministry of Agriculture, Rural Development and Environment (MADRE) of the Republic of Cyprus for the Department of Meteorology historical meteorological data. Special thanks to Marinos Eliades for the creation of Figure 1 in ArcGis software version 10.3 (www.ESRI.com).

**Conflicts of Interest:** The authors declare no competing interests.

#### **References**


© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

MDPI St. Alban-Anlage 66 4052 Basel Switzerland Tel. +41 61 683 77 34 Fax +41 61 302 89 18 www.mdpi.com

*Climate* Editorial Office E-mail: climate@mdpi.com www.mdpi.com/journal/climate

MDPI St. Alban-Anlage 66 4052 Basel Switzerland

Tel: +41 61 683 77 34 Fax: +41 61 302 89 18