See discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/310425563
Examining sustainable landscape function across the Republic of Moldova
Article in Habitat International · February 2018 DOI: 10.1016/j.habitatint.2016.11.002
CITATIONS READS 10 341
1 author:
Richard Ross Shaker Ryerson University
48 PUBLICATIONS 253 CITATIONS
SEE PROFILE
Some of the authors of this publication are also working on these related projects:
Sustainable Development & Environmental Visualization using Spatial Analysis View project
Sustainability indicators, Patterns of Sustainable Development, and Sustainable Landscape Change View project
All content following this page was uploaded by Richard Ross Shaker on 30 November 2019.
The user has requested enhancement of the downloaded file. Habitat International 72 (2018) 77e91
Contents lists available at ScienceDirect
Habitat International
journal homepage: www.elsevier.com/locate/habitatint
Examining sustainable landscape function across the Republic of Moldova
* Richard Ross Shaker a, b, a Department of Geography & Environmental Studies, Ryerson University, Toronto, Ontario, Canada b Graduate Programs in Environmental Applied Science & Management, Ryerson University, Toronto, Ontario, Canada article info abstract
Article history: Sustainability remains an undeniable, yet obscure, destination for humanity to reach. Although progress Received 8 August 2016 has been made, there remains no agreed upon method for spatial scientists, nor landscape and regional Received in revised form planners to use during sustainable development assessments. Furthermore, limited examples exist that 29 October 2016 investigate relationships between-landscape form (e.g. urban configuration) and population dynamics Accepted 9 November 2016 (e.g. number of settlements)- and a local measure of sustainable development. Using a recently pub- Available online 15 November 2016 lished local sustainable development index (LSDI) for Moldova, a regional spatial analysis was created to further elucidate strengths and weaknesses of index-based assessments of sustainable landscape func- Keywords: Indicator-based assessment tion. Using a one-to-many relationship, sixty-six landscapes were joined to 399 mean LSDI sample lo- ¼ Landscape science cations for the quantitative spatial assessment (n 399). A rarity of this study was that it employed the Regional development Eastern School of Geography's “landscape units” for Moldova during geospatial data aggregation and Spatial autoregressive modeling spatially enabled regression. Moran's I scatterplot and spatial correlogram were used to visualize spatial Sustainable development planning autocorrelation dynamics of LSDI. Three local conditional autoregressive (CAR) models were made, with Sustainable urbanization all explaining over 70% of LSDI variation. The two strongest positive predictors of LSDI were city pop- ulation density and road intersection density, while the two most consistent negative were settlement density and distance between urban land cover patches (ENN_AM). Findings suggest index-based landscape valuations could suffer from spurious inferential correlations when landscape-calculated sub-metrics (i.e., proportion agricultural land) are included within evaluation indices. This phenome- non complicates the interpretation of results during regional analyses, thus potentially hindering sus- tainable development planning and policy responses across spatial scales. © 2016 Elsevier Ltd. All rights reserved.
1. Introduction borders through land cover change, and devastating entire eco- systems across the developing world through globalization (Rands Humanity's understanding on how to live sustainably is as open et al., 2010; Shaker, 2015a). More than 40% of Threatened or En- as ever. As we progress through the early stages of the Anthro- dangered species are at risk of extinction due to human propagated pocene, not one of the original eight Millennium Development invasive species (Grime, 2006). Lastly, marine resources continue to Goals (MDGs) were reached by 2015, nor has the rate of global be overharvested (Worm et al., 2006), their waters are becoming warming or sea level rise slowed (Dutton et al., 2015), nor has the more acidic due to pollution deposition from the continual burning eradication rate of our life-supporting ecosystems decreased of fossil fuels (Hoegh-Guldberg et al., 2007), and terrestrial garbage (Butchart et al., 2010). The world's richest countries have seen continues to collect in natural oceanic gyres (Jambeck et al., 2015). improvements in socioeconomic and material well-being These direct and indirect sustainability stressors are driven pri- (Weinzettel, Hertwich, Peters, Steen-Olsen, & Galli, 2013); albeit marily by population growth, which was recently projected to at the cost of metabolizing natural habitats within their own continue into the next century and surpass 12 billion globally (Gerland et al., 2014). Despite the current condition of humanity's life-supporting ecosystems, Griggs et al. (2013) reiterated that in- equalities between groups, high-levels of poverty, and human well- * & Corresponding author. Department of Geography Environmental Studies, being need to be addressed while restoring biophysical stability. Ryerson University, Toronto, Ontario, Canada. “ E-mail address: [email protected]. According to Hales and Prescott-Allen (2002), Making progress http://dx.doi.org/10.1016/j.habitatint.2016.11.002 0197-3975/© 2016 Elsevier Ltd. All rights reserved. 78 R.R. Shaker / Habitat International 72 (2018) 77e91 towards sustainability is like going to a destination we have never measures and various spatial analysis tools (e.g. FRAGSTATS) (i.e., visited before, equipped with a sense of geography and the prin- Shaker et al., 2010; Shaker & Ehlinger, 2014). Recently, for sus- ciples of navigation, but without a map or compass” (6). With over tainable development planning purposes, Shaker (2015a) recorded 300 working definitions of sustainability and sustainable devel- significant relationships between-landscape and population vari- opment (Dobson, 1996), and some definitions contradicting each ables- and development indices when investigating sustainable other (Goodland & Daly, 1996), some feel that achieving a sus- urbanization at the macroscale. That said, sustainable landscape tainable destination is more remote than ever (Jickling, 2000). In function research at regional and local scales using evaluation example, a paradox is present within the term “sustainable devel- indices of sustainability conditions remain virtually nonexistent. opment” e development has been used synonymously with While studies have employed evaluation indices to understand growth, and sustainable implies increase endlessly, which is not landscape pattern on process, few have specifically used a local possible on a planet with finite natural resources (Bartlett, 2006). sustainable development measure to assess sustainable landscape Despite its deficiencies, sustainable development remains an function across a country. Furthermore, although likely the most appropriate guide for creating a long-term, positive relationship appropriate areal unit for understanding sustainable development between humankind and life-supporting ecosystems; albeit, at the landscape scale, no studies have incorporated “landscape bombastic and inconsistent goals hamper humanity's ability to units” into explorations of landscape patterns on sustainable determine if this relationship has been or will be achieved (Mayer, development process. To address these issues and guide this study, Thurston, & Pawlowski, 2004). Slow evolution of sustainable the following two null hypotheses are tested: (1) no significant development has been linked to progress being mostly conceptual relationship exists between-landscape form and population dy- and methodological (Hezri & Dovers, 2006). For this study, the namics- and the local sustainable development index (LSDI) applied definitions for “sustainability” and “sustainable develop- created for Moldova; and (2) local spatial autoregressive modeling ment” are adopted from Shaker (2015b: 305) as: “‘sustainability’ does not corroborate global ordinary least squares (OLS) regression should be viewed as humanity's target goal of human-ecosystem methodology. This study also aims to deliver regional sustainability equilibrium (homeostasis), while ‘sustainable development’ is the planners and landscape scientists an applied example for system- holistic approach and temporal processes that lead humanity to its atically assessing, describing, and monitoring sustainable land- end goal of sustainability.” scape function across space. Policy and decision makers have encouraged researchers to improve existing models and develop new techniques for man- 2. A landscape unit approach aging coupled human-natural systems associated with local and regional sustainable development planning (Grosskurth, 2007). In The landscape scale, “geographical landscape,” or specifically response, the planning community sees a need for sustainable the “landscape unit,” may be the best management scale for development initiatives that goes beyond lip-service and puts assessing and monitoring sustainable landscape function across a concepts into action. “Along with the questions ‘should we?’ or ‘can region. A landscape unit is an areal unit that is created from a we implement sustainable development?’ more the question of collection of in situ (disaggregated) spatial data, and is methodo- ‘how can we apply this concept?’ dominates the literature” (Chifos, logically rooted to Russian soil science (Shaw & Oldfield, 2007). The 2007). Despite uncertainty about operationalization, the field of first landscape units were published within Lev Semenovich Berg’s planning acknowledges that sustainable development is an influ- (1947) seminal work Geographical Zones of the Soviet Union.Itwas ential concept and should shape future methodology and practice in this research that Berg spelled out the pioneering definitions of (Godschalk, 2004; Jepson, 2004). That said, there remains no ‘ideal’ geographical landscape and the foundational principles of the instrument for attaining sustainability on neither regional nor local Russian landscape unit. According to Berg (1947), “A geographical planning scales (Keiner, 2006). As suggested by Wu (2008), land- landscape is that combination or grouping of objects and phe- scape ecology appears to be the most relevant solution-driven and nomena in which the particularities of relief, climate, water, soil, place-based discipline for moving humanity towards sustainability vegetation, fauna, and to a certain degree human activity, is across geographical scales. Landscape ecology is the study of: i) blended into a single harmonious whole.” Other physiographic land spatial relationship among landscape elements and/or ecosystems; management systems have been created throughout the world, but ii) the flow of energy, minerals, nutrients, and species (including the inclusion of human aspects makes the Russian-influenced Homo sapiens) among the elements; and iii) the ecological dy- physiographic planning system unique. Using this Eastern School namics of the landscape mosaic through time (Forman, 1995). of Geography's conception of landscape science, landscape units Despite numerous urban planning, environmental management, were created from the culmination of over ten years of field surveys conservation, and restoration projects completed, Naveh (2007) in the Republic of Moldova (Proka, 1978, pp. 69e72, 1983). The stated that landscape ecology has had limited impact on sustain- Moldavian multi-hierarchical land management system was orga- able landscape management. nized into four nested spatial scales: two zones, five regions, 74 Evaluation of landscapes can be accomplished through use of landscape units, and 120 elementary landscape features (Fig. 1). existing sustainable development indices, which allows for Increasingly, sustainability planners, scientists, and policy- assessing connections between landscape patterns and develop- makers have focused on understanding coupled human- ment processes (Mander & Uuemaa, 2010). At the global level, the environmental systems and there remains a need for interna- need for indicators was expressed in Chapter 40.4 of Agenda 21: tional integration between the various landscape traditions. “indicators of sustainable development need to be developed to Neither in Russia nor the West have scientists succeeded in speci- provide solid bases for decision making at all levels and to fying an agreed and unproblematic understanding of landscape, or contribute to a self-regulatory sustainability of integrated envi- more broadly promoted a common geographical conception of ronment and development systems” (UN, 1992). At local and human-environmental relationships (Shaw & Oldfield, 2007). The regional scales, indicator-based assessment of landscape function Moldavian multi-hierarchical land management system was provides a fundamental tool for evaluating relationships during designed to provide a useful basis for decision-making about in- sustainable landscape planning (Leitao~ & Ahern, 2002; Mander & tegrated human and natural systems. For multidisciplinary projects Uuemaa, 2010). Regional planning studies of coupled human- with applied geographical and ecological aims (i.e., sustainable natural systems have been further elucidated using landscape development), the employment of landscape units has been R.R. Shaker / Habitat International 72 (2018) 77e91 79
Fig. 1. Original multi-hierarchical land management structure systematically dividing the Republic of Moldova through four nested scales: 120 elementary landscape features (A); 74 landscape units, five regions, and two zones (B) (Proka, 1983). considered an ideal management scale (Zonneveld, 1989). Although entirely contiguous. Moldova lies in the northeast Balkan region, the Russian-based geographical landscape is a justifiable aggrega- occupies 33,846 km2 of land with over 4.3 million people (2010 tion scale for assessing sustainable landscape function, very few census), and is an ideal nation for conducting a study on sustainable development studies exist utilizing them. For various purposes, landscape function for several reasons. Besides both the afore- Proka’s (1983) four scales of landscape aggregation were digitized mentioned history of landscape science, and the recently published within a geographic information system (GIS) by the Institute of LSDI for Moldova, substantial economic (i.e., governmental cor- Geography and Ecology, Academy of Sciences of Moldova. Based on ruption), environmental (i.e., water pollution), and social (i.e., hu- the aforementioned literary support, the geographic landscape man trafficking) challenges remain. Moldova is a very young nation, scale (74 landscape units) was utilized in the forthcoming spatial only declaring its liberation from the Union of Soviet Socialist Re- analysis of sustainable landscape function across Moldova. publics (USSR) on 27 August 1991; however its independence was not officially recognized until 2 March 1992 when Moldova became a United Nations member Moldova remains one of the least affluent 3. Material and methods countries in Europe, undergoing 65 percent decline in income since its independence. 3.1. Study area Anthropogenic forces have altered all locations in the Republic of Moldova. This modification of natural land cover has resulted in a This study of sustainable landscape function incorporated 66 of great variety of landscape patterns (Fig. 3), presenting an oppor- the possible 74 landscape units within the Republic of Moldova tunity to investigate sustainable landscape function across the (Fig. 2). The 66 geographical landscapes were selected based on 399 country. The Moldavian landscape mosaic is primarily a result of global positioning system (GPS) sample locations from the 2005 millennia of agrarian practices, and a little more than a century of Demographic and Health Survey (MDHS, 2006) used to create the industrialization. Its predominant land cover remains over- local sustainable development index (LSDI) for Moldova (see whelmingly agricultural, while farming is the dominant land use Shaker & Sirodoev, 2016). The 2005 Demographic and Health Sur- activity. Correspondingly, fertile chernozem origin soils cover over vey did not occur within the autonomous region of Transnistria, 75% of the Republic. Annual rainfall varies only slightly throughout which exempts four landscape units. Therefore, four landscape the country, reporting an average annual rainfall of 555 mm units did not have LSDI point representation; however the (1969e1990). Geology is relatively consistent throughout the remaining 66 Moldavian geographical landscapes were almost 80 R.R. Shaker / Habitat International 72 (2018) 77e91
Fig. 2. Study area map illustrating the selected 66 of 74 landscape units in Moldova (47 240N, 28 220E), and 399 demographic and health survey cluster locations used to create Moldova's local sustainable development index (LSDI). nation, with the majority of exposed rock features having sedi- there via the Danube River. Large-scale silviculture operations are mentary consistency. It should also be noted, there is low but minimal but some small-scale illegal deforestation exists. Although prevalent seismic activity here. The Republic is occupied by gentle urban settlements can be traced back to the 15th century in steppe from the east side of the Carpathian Mountains great arc and Chis¸inau, almost 58% of the country's present population lives in a has a maximum elevation under 430 m. The lands of Moldova are rural setting. Chis¸inau, the capital of Moldova, is found in the south mostly undulating hills, marked with dendritic river basin, ravines, central part of the country and is the largest city. As of the local and gullies, which drain southeast to the Black Sea or indirectly 2014 census, Chis¸inau's city population was just under 500,000 R.R. Shaker / Habitat International 72 (2018) 77e91 81
Fig. 3. Map displaying 2004 land cover for the Republic of Moldova. with about 5400 people/km2.Balt ¸ i is the third largest city by 3.2. Moldova's local sustainable development index population after Tiraspol and Chis¸inau, but remains the second most important city logistically and is located north central. As of Evaluation indices are increasingly acknowledged as useful tools the 2014 census, Balt ¸ i's city population was just over 140,000. for development planning and policy-making because they provide Agricultural practices, urbanization, historical and illegal forest numerical support for entities seeking guidance towards their practices, landsliding, and floodplain alterations are the main individualized sustainability goals (Shaker & Zubalsky, 2015). It has contributors modifying landscape configuration in the Republic of been over two decades since the Earth Summit, where 178 nations Moldova. adopted Agenda 21 and indicators to help implement action plans; 82 R.R. Shaker / Habitat International 72 (2018) 77e91 however there remains no agreement on how best to design or use FRAGSTATS (ver. 4.2), which quantifies the spatial arrangement of these metrics for sustainable development planning. Societies have land cover and land use (Leitao,~ Miller, Ahern, & McGarigal, 2006; engaged in policy-building that supports Agenda 21 goals and its McGarigal, , Cushman, & Ene., 2012). Source land cover data came addendums, yet few local initiatives display evidence of success- from hand digitized orthophotographs (circa 2004) by experts at fully combining all three spheres of sustainability (economic wel- the Academy of Sciences of Moldova. These land cover data were fare, social equity, environmental quality). In response, Shaker and originally classified using the United Nations (UN) Food and Agri- Sirodoev (2016) created the first local sustainable development culture Organization (FAO) system. However, to improve overall index (LSDI) for assessing geographical patterns of development classification accuracy, the FAO divisions were reorganized into across Moldova. The forthcoming index-based assessment of sus- Anderson Level I land cover categories (Anderson, Hardy, Roach, & tainable landscape function utilized mean Moldavian LSDI to Witmer, 1976). The reclassified land cover data remained at 30 m evaluate conditions of landscape form and population dynamics resolution, and an 8-neighbor rule was used for metric delineation. across the selected 66 landscape units. Eight Anderson Level I land cover classes were applicable for the The landscape evaluation index, mean LSDI, was created using Republic of Moldova (Fig. 3). Five distinct land cover compositions household and property composition data from a 2005 de- were assessed for the 66 landscapes yielding: agricultural land mographic and health survey (MDHS, 2006). Using data from (61%), forest land (15%), urban or built-up (10%), wetland (1%), and 11,066 households, the multi-metric LSDI for Moldova used a 15 water (1%) (percentages obtained from this analysis). Since no sub-metric optimum, equal weighting, 1 to 5 ordinal scale stan- minimum set of landscape ecology metrics exists for capturing the dardization, and additive construction for holistically capturing the majority of landscape structure (Wagner & Fortin, 2005), 60 urban development needs of sustainability (Shaker & Sirodoev, 2016). The class and six landscape diversity measures were calculated then LSDI included the following ten categories: education [1 indicator]; statistically reduced into a highly independent subset. Principle goods production [2 indicators]; household abundance [1 indica- components analysis (PCA) and Pearson's correlation coefficients tor]; income [1 indicator]; psychology [1 indicator]; public health (r > 0.75) extracted 16 urban class configuration and two landscape [2 indicators]; structural composition [2 indicators]; technology [1 diversity metrics for inferential analyses (Table 1). indicator]; transportation [2 indicators]; and utility services [2 in- dicators]. LSDI ranged between 36 (very bad) and 64 (very good), 3.4. Data analysis with increasing values corresponding to improved conditions of sustainable development (n ¼ 11,066). When averaging household A regional spatial analysis was created to elucidate strengths LSDI scores to their corresponding Thiessen polygons index values and weaknesses of index-based assessments of sustainable land- compressed and ranged 45.23 (bad) to 58.12 (good) (n ¼ 399). A scape function by testing relationships between mean LSDI- and map illustrating mean LSDI (Shaker and Sirodoev (2016) super- landscape form and population dynamics variables-across 66 imposed with Moldavian geographical landscapes is provided in Moldavian landscapes. Although there is no ubiquitous rule for Fig. 4. “landscape scale,” Forman (1995) stated that a ‘landscape’ is “a kilometers-wide mosaic over which local ecosystems recur.” For 3.3. Population dynamics and landscape form data this study, the area descriptive statistics for the selected 66 Mol- davian landscape units were: 100 km2 (min), 466 km2 (mean), and Using the 66 selected landscape units, landscape form and 1293 km2 (max). Using a one-to-many relationship, the sixty-six population dynamics data were aggregated and organized into the landscapes were spatially joined to the 399 mean LSDI GPS loca- following six categories: landscape unit data, population variables, tions creating landscape replicates and a total sample size of 399 infrastructure variables, land cover composition, urban class (n ¼ 399). To meet the requirement of Gaussian frequencies for configuration, and landscape diversity (Table 1). Landscape unit parametric tests, independent variables were transformed using data included: landscape unit area, landscape unit perimeter, standard methods where needed. Using the statistical software JMP perimeter-area (P/A) ratio, and mean degree slope angle. The Na- (ver. 11) (SAS, 2013), Shapiro-Wilk test of normality determined if tional Aeronautics and Space Administration (NASA) and the Na- transformation was required, and which mathematical function tional Geospatial-Intelligence Agency (NGA) provided the source was most appropriate for reaching Gaussian distribution. Using a digital elevation model (DEM) data used in this analysis. The raster four-step method, the following empirical method was created to elevation data were collected in 2000e2001 via SRTM instrument test this study's two guiding hypotheses for evaluating and un- at 60 m resolution and after projecting can be used in raster format derstanding sustainable landscape function. at 90 m resolution. Population variables calculated were: number of First, an exploratory spatial data analysis (ESDA) was conducted settlements, total settlement population, number of cities (popu- to assess the level of spatial autocorrelation of the dependent and lation > 50,000), and total city (population > 50,000) population. 35 independent variables of this study. The fact that nearer things All population variables were normalized by landscape area to are more similar than further ones is now commonly understood as create density measures, which decrease sampling bias and errors the First Law of Geography (Tobler, 1970). This spatial univariate associated with modifiable area unit problem (MAUP). Infrastruc- phenomenon is commonplace to socio-ecological data associated ture variables included: total road distance, total number of road with sustainability research, and can be a benefit and hindrance intersections, total railroad distance. Measures of transportation when trying to understand spatial patterns of development. A infrastructure were also normalized by landscape area, with total positive aspect of this nonstationarity (spatially autocorrelation) is number of road intersections also being corrected by total road that it can provide statistically significant meaning to geographical distance within a landscape. The predictors of landscape form and patterns of sustainable development (Shaker, 2015a). However, population dynamics were created using various tools within ESRI's since the presence of spatial autocorrelation violates the assump- (2014) ArcMap 10.2. These geospatial data were circa 2004, and tion of randomness, traditional parametric tests (i.e., ordinary least sourced from the National Bureau of Statistics of Moldova, Ministry squares; OLS regression) are no longer appropriate for under- of Transpiration and Road Infrastructure of Moldova, and the standing inferential relationships (Dormann et al., 2007; Lichstein, Academy of Sciences of Moldova. Simons, Shriner, & Franzreb, 2002). Lennon (2000) called attention Measures of Land cover composition, urban class configuration, to problems associated with spatial autocorrelation and argued and landscape diversity were calculated using the freeware virtually all inferential studies over space should be redone. For this R.R. Shaker / Habitat International 72 (2018) 77e91 83
Fig. 4. The spatial distributions of Moldova's local sustainable development index (LSDI). Index values are from 11,066 household scores averaged to 399 Thiessen polygons, with higher values representing improved development. (Map modified from Shaker & Sirodoev, 2016). study, the ESDA technique, Global Moran's I-test (Moran, 1950)was menu, row standardization was chosen because there was a po- applied to evaluate spatial nonstationarity levels of all variables tential for bias due to sampling design and aggregation scheme. using ESRI's ArcGIS 10.2 Spatial Analyst Tools (ESRI, 2014). The Lastly, using the publically available freeware Spatial Analysis in Global Moran's I distance threshold of 35 km was established using Macroecology (SAM) (ver. 4) (Rangel, Diniz-Filho, & Bini, 2010), the Incremental Spatial Autocorrelation tool within ArcGIS, with Moran's I scatterplot and spatial correlogram were created to conceptualization of spatial relationships was Euclidean inverse illustrate spatial autocorrelation dynamics of mean LSDI. distance weighting (IDW). As suggested by ESRI's ArcGIS help Second, a two-tailed Pearson's Product-Moment Correlation test 84 R.R. Shaker / Habitat International 72 (2018) 77e91
Table 1 Global spatial autocorrelations for all study variables using Global Moran's I-test; Pearson product-moment correlation coefficients (two-tailed) between Moldova's local sustainable development index (LSDI) and all independent measures of landscape form and population dynamics (n ¼ 399).
Global Moran's I z-score P-value Pearson's r P-value
Dependent variable Local sustainable development index (LSDI) 0.634 33.430* <0.001 Landscape unit data Landscape unit area 0.582 30.776* <0.001 0.23 <0.001 Landscape unit perimeter 0.519 27.651* <0.001 0.06 0.270 Landscape perimeter - area (P/A) ratio 0.338 17.930* <0.001 ¡0.38 <0.001 Mean degree slope angle 0.704 30.514* <0.001 0.15 0.003 Population variables Number of settlements/area 0.574 30.370* <0.001 ¡0.25 <0.001 Settlement population/area 0.809 42.625* <0.001 0.76 <0.001 Number of cities (>50,000)/area 0.645 33.973* <0.001 0.69 <0.001 City (>50,000) population/area 0.798 42.079* <0.001 0.76 <0.001 Infrastructure variables Road distance/area 0.514 27.210* <0.001 0.28 <0.001 Road intersections/area 0.568 30.051* <0.001 0.40 <0.001 Road intersections/road distance 0.384 20.373* <0.001 0.30 <0.001 Rail distance/area 0.500 26.422* <0.001 0.55 <0.001 Land cover composition Agriculture, percent 0.617 32.642* <0.001 ¡0.20 <0.001 Forest, percent 0.506 26.973* <0.001 0.41 0.413 Urban, percent 0.797 41.976* <0.001 0.71 <0.001 Wetland, percent 0.344 19.354* <0.001 ¡0.26 <0.001 Water, percent 0.609 32.282* <0.001 0.02 0.720 Urban class configuration Mean of patch area distribution (AREA_MN) 0.703 37.073* <0.001 0.55 <0.001 Class area (CA) 0.713 37.557* <0.001 0.72 <0.001 Area-weighted mean core area index (CAI_AM) 0.581 30.665* <0.001 0.59 <0.001 Cohesion index (COHESION) 0.653 34.424* <0.001 0.66 <0.001 Area-weighted mean patch contiguity index (CONTIG_AM) 0.581 30.674* <0.001 0.59 <0.001 Mean of core area distribution (CORE_MN) 0.707 37.255* <0.001 0.57 <0.001 Area-weighted mean of Euclidean nearest neighbor distance (ENN_AM) 0.659 34.794* <0.001 ¡0.69 <0.001 Area-weighted mean of fractal area dimension (FRAC_AM) 0.761 40.110* <0.001 0.67 <0.001 Interspersion and juxtaposition index (IJI) 0.542 23.501* <0.001 ¡0.12 0.016 Landscape shape index (LSI) 0.555 24.033* <0.001 0.35 <0.001 Perimeter area fractal dimension (PAFRAC) 0.545 23.699* <0.001 0.16 0.001 Area-weighted mean of perimeter area ratio (PARA_AM) 0.674 29.159* <0.001 ¡0.59 <0.001 Patch density (PD) 0.560 24.257* <0.001 0.27 <0.001 Area-weighted mean of shape index (SHAPE_AM) 0.841 36.291* <0.001 0.71 <0.001 Range of shape index (SHAPE_RA) 0.713 30.815* <0.001 0.65 <0.001 Total core area (TCA) 0.786 33.924* <0.001 0.72 <0.001 Landscape diversity Patch richness density (PRD) 0.380 20.286* <0.001 ¡0.25 <0.001 Shannon's diversity index (SHDI) 0.572 30.260* <0.001 0.09 0.061
Notes: All landscape ecology metrics computed based on raster data with 30 m cells, and using queen contiguity (8-neighbor rule). See Leitao~ et al. (2006) and McGarigal et al. (2012) for landscape ecology metric details and equations. Landscape ecology technical notes: search area for isolation/proximity metrics was 200 m; edge depth for core area calculations was 100 m. Spatial clustering was determined using a threshold distance of 35 km; symbol designation: * denotes <1% change random pattern. A Pearson's correlation coefficient in bold depicts a statistically significant relationship above the 95% confidence level.
(r) was used to assess relative statistical strengths between mean into a multi-model selection framework (Burnham & Anderson, LSDI and 35 landscape form and population dynamics explanatory 2002; Diniz-Filho, Rangel, & Bini, 2008) for predicting mean LSDI. characteristics (Table 1). Pearson's correlation coefficients range Using the SAM freeware, all conceivable regressions were created from 1 to 1, with values closer to 1 indicating stronger bivariate (i.e., 32,767), and the Akaike Information Criterion (AIC) weight (wi) association. Pearson's Product-Moment Correlation test is one of of each model was contrasted to find the best regressions among all the most common non-spatial parametric tests for understanding possible options (Terribile, Diniz-Filho, Rodriguez, & Rangel, 2009). bivariate inferential relationships, and a P-value accompanies the The top three models were chosen and compared based on their coefficient value signifying its statistical significance. Thirty-one corrected (AICc) and coefficient of determination (R2). AICc is landscape form and population dynamics predictors had statisti- considered a preferred measure of model fit and a better approxi- cal significant bivariate correlations with Moldavian mean LSDI mation of reality (see Akaike, 1978, pp. 9e14; Fotheringham, (P < 0.05, Table 1). The 31 explanatory variables were then reduced Brundson, & Charlton, 2004). Fotheringham et al. (2004) has into an independent dataset using a bivariate Pearson's correlation stated that a ‘serious’ discrepancy between two models is when matrix (r > 0.75). When redundancy occurred between two vari- AICc values differ by at least three. Directionality of covariates, and ables, the predictor with best natural frequency and strongest their rank ‘effects’ on LSDI, was determined using the partial associate with mean LSDI was retained. The JMP software was used regression standardized (Beta) coefficients for the three multiple during this step of the analysis. regressions. To assess potential errors associated with multi- Third, ordinary least squares (OLS) multiple regressions were collinearity, the Variance Inflation Factor (VIF) was evaluated for the created to accomplish the goals of this study. 15 predictor variables three OLS multiple regression models. High values of VIF indicate a remained from the Pearson's correlation matrix and were entered potential for inflated covariance between regression coefficients, R.R. Shaker / Habitat International 72 (2018) 77e91 85 which creates difficulty separating and ranking predictors. Lastly, the response variable- and 31 landscape form and population dy- using the JMP software, the Shapiro-Wilk test of normality was namics explanatory variables were found using Pearson's Product- used to assess independence of model residuals and assure Moment Correlation test (r)(P < 0.05, Table 1). Correlation co- randomly distributed errors. efficients are commonly classified into very positive (>0.75), posi- Fourth, to reduce modeling errors from spatial autocorrelation tive (0.75e0.50), neutral (0.50 to 0.50), negative ( 0.50 to 0.75), in regression analysis, a local conditional autoregressive (CAR) or very negative (<