Indicator-based assessment of local and regional progress toward the Sustainable Development Goals (SDGs): An integrated approach from Romania

In order to measure progress in achieving the Sustainable Development Goals (SDGs) by 2030, 169 targets have been approved globally. Even though interest in implementing these goals is high, many states have not yet established a set of subnational indicators to measure the implementation of the SDGs and have not com-pleted their own assessment of progress in achieving these global goals. This study aims to measure the progress toward achieving the SDG at local and regional level in Romania by calculating the SDG Index. For the calculation of the SDG Index at subnational level, we propose an integrated approach based on 90 indicators, stored and processed in a PostgreSQL object-relational database. The results show the concentration of the highest performances of sustainable development in some specific geographical areas. The rural areas and the extended peripheral regions in the eastern and southern part of the country are the poorest performers.

Southern Europe were the worst performers, among them the Romanian capital Bucharest (54.4 SDG Index score) (Lafortune et al., 2019).
Off course, the list of studies working on the operationalization of SDGs is long (see also Allen, Reid, Thwaites, Glover, & Kestin, 2020;Bolcárová & Kolostá, 2015;Campagnolo et al., 2018;Dalampira & Nastis, 2019;Hametner & Kostetckaia, 2020;Nhemachena et al., 2018;Tanguay, Rajaonson, Lefebvre, & Lanoie, 2010;Weitz, Carlsen, Nilsson, & Skånberg, 2017;Wichaisri & Sopadang, 2018). We identified two major gaps in the relevant literature: a strong focus on the national scale and a spatial focus on cities. Accordingly, based also on our previous study (Nagy, Benedek, & Ivan, 2018) and some recent research (Lowery, Dagevos, Chuenpagdee, & Vodden, 2020), we argue strongly that rural areas may not been left outside the SDG analysis. They are generally peripheral and marginalized areas, and, in addition, in the specific case of Romania, they concentrate broadly half of the country total population (Benedek & Lembcke, 2017;Török & Benedek, 2018). For an effective implementation and assessment of the SDGs, we cannot ignore the evaluation of the rural areas, and therefore we argue for an integrated territorial approach of measuring the SDGs. This approach offers three novelties: the generation of a territorial database on local level, which includes both rural and urban areas; the large employment of earth observations methods in the measurement of the SDG indicators; and the development of a new data model for the measurement of the SDG Index.
More explicitly, in order to measure the progress of each commune, city, and county (subnational, NUTS 3-level statistical units) in achieving the SDGs, we proposed to calculate the SDG Index at local and county level. The SDG Index is a composite index that sums up individual scores calculated for each of the 17 SDGs. In calculating the individual scores, we developed a set of monitoring indicators that were subsequently aggregated at the level of each SDG, thus obtaining 17 specific composite indices for measuring the SDGs. These indices allow the establishment of the level and stage reached by each local government and county for each specific objective of sustainable development. The overall SDG composite index resulting from the sum of specific composite indices assesses the general performance of each commune, city, and county in Romania in achieving all 17 SDG.
We outline that the present study is an attempt to quantify the performance of each LAU (Local Administrative Units) on the SDGs, contributing in this way to calls for more progress in the operationalization of SDGs and in the evaluation of indicators' relevance (Hák, Janousková, & Moldan, 2016;Halisçelik & Soytas, 2018;Holden, Linnerud, & Banister, 2017). It will provide a policy-relevant assessment tool for the LAUs in establishing their position within each SDG, which in turn will help them to set up empirically sound and politically relevant Local Development Strategies. Moreover, understanding the differences in sustainable development across multiple scales and resources will enhance the ability of central authorities to balance sustainable development between national and various subnational levels.

| DATA AND METHODS
The calculation of the SDG index at local (LAU, 3181 local administrative units: communes and cities) and county (NUTS-3, 41 counties and the Municipality of Bucharest) level in Romania involved several work steps.

| Data collection
As mentioned in the Introduction, one important original contribution of the article is the mix of data resources employed in the generation of 90 indicators and in the consequent measurement of the SDGs, and the outstanding role of Earth Observation methods, which are employed, to our knowledge, for the first time at this scale in SDG measurement oriented studies.  years 1990, 2000, 2006, 2012, and 2018 and include the following main classes: (a) artificial surfaces, (b) agricultural areas, (c) forests and semi natural areas, (d) wetlands, and (e) water bodies. From the CLC data, the information related to 2012 and 2018 was used; these being derived from the satellite images IRS P6 LISS III and RapidEye (CLC, 2012) in (EEA, 2020a), respectively, Sentinel-2 and Landsat-8 (CLC2018) (EEA, 2020a). "HRL imperviousness" data are available for the years 2006, 2009, 2012, and 2015 and are derived from the IRS-P6/Resourcesat-2 LISS-III, SPOT 5, and Landsat 8 satellite images (EEA HRL, 2020b). The information extracted from these databases was processed using GIS tools and subsequently aggregated at the level of Romanian municipalities, T A B L E 1 Summary of indicators used to construct the SDG Index.

| Database creation and data processing
Due to a large volume of data collected at the local and county level in Romania, in order to process these data and calculate indicators, it was necessary to create a database and find a tool which enabled us to quickly reprocess the data in case of input changes and corrections.
The data were collected in multiple formats: ESRI Shapefile, CSV (comma-separated values), and JSON (JavaScript Object Notation), and in some situations, they were organized in sets for different years.
In this case, a PostgreSQL object-relational database was created with the PostGIS extension, which allowed us to organize the data into schemas and tables and also to separate input data from results.
To create the results from the input data, we used the dbt tool (data build tool) which enabled us to transform, aggregate, and join datasets. With this tool, 568 models were created to process the data and 282 tests for validation and verification of the input data and results. Each model represents an .sql file containing a select statement for selecting, aggregating, and processing data which can be joined from multiple source tables ( Figure 1). Tests are assertions that were made about the models in order to verify the correctness of the data and eliminate errors. These tests were running macroscripts to identify the null or zero columns and row counts to identify missing data. The models and the database thus created allowed us to quickly update the information and recalculate the SDG Index, respectively, to eliminate the errors that would occur in the manual processing of data.
Using the dbt tool and the SQL programming language, an individual model was created to calculate each indicator and workflow: 1. summation of values by administrative units: in this step, the data that were available on several categories or those available at locality level were summed, thus obtaining a value for each LAU and county; 2. joining data tables to LAUs and county limits; 3. calculation of indicators based on the data resulting from steps 1 and 2. An average of three-four models were used to calculate each indicator, such an example can be seen in Figure 1. This approach also allowed us to visualize the data processing flow and the connections between the data.
Based on the information extracted from the CLC dataset, the "forest area," the "change in forest area," the "agricultural area," and the "water bodies area" indicators were calculated related to each LAU and county in Romania. Using the information extracted from the "HRL imperviousness" dataset for the years 2006 and 2015, the "growth rate of built-up area" and the "Land Use Efficiency" (LUE) were calculated. LUE indicator was used in order to measure the F I G U R E 1 Workflow for calculating the forest area indicator [Colour figure can be viewed at wileyonlinelibrary.com] change rate of the built-up area per capita based on the following formula (Corbane, Politis, Siragusa, Kemper, & Pesaresi, 2017): where: LUE, land use efficiency indicator; BU t , built-up area at the initial year (t); Pop t , total population within the built-up area at the initial year (t); BU t+n , built-up area at the final year (t+n); Pop t+n , total population within the built-up area at the final year (t+n).
In calculating climate indices, we use the observed daily climatological spatial data (maximum air temperature, minimum air temperature, and relative humidity) interpolated on a grid at a spatial resolution of 0.1 (approximately 10 km). The data come from the ROCADA project and are available on the PANGEA portal at doi.
The comparison was made with data from the regional model THI combined temperature and humidity to measure the degree of discomfort an individual feels in hot weather. The original formula has been changed because it uses temperature expressed in degrees Fahrenheit. In our study, we used the following formula: where: T, air temperature ( C), H, relative humidity (%).
The National Meteorological Administration uses, in operational activity during winter, the cooling indicator. Legal measures to combat the cold are taken if the temperature drops below −20 C or if the wind chill drops below −35 C. In our analysis, we only used the minimum temperatures because wind data are not available in the ROCADA project.
4. normalization of indicators: the values were normalized to become easily comparable on a scale of 1 to 10, using the min-maxx ð Þ and max-min (xˇ) normalization method: where x is the value of raw data; min (x) and max (x) determine the lower and upper bounds for worst and best performance,x and x is the normalized value after the rescaling process.
For most indicators, the min-maxx normalization method was applied, where 0 indicates the worst performance and 10 the highest performance. In the case of indicators such as "Youth not in employment, education or training (NEET)," "Unemployment rate," "Traffic deaths rate" the max-min (x) normalization method was applied, where 10 indicates the worst performance and 0 the highest perfor- Kurtosis = where X is the mean, N is the number of data points, and s is the standard deviation.
The Shapiro-Wilk test has the following equation: where a i is the vector of expected values of normal ordered statistics, and x i is the ordered sample value. The Shapiro-Francia test has the same equation, using b instead of a, representing the square of the Pearson correlation coefficient resulted from random sample values and the expected normal order statistics (Royston, 1993;Shapiro & Wilk, 1965;Shapiro & Francia, 1975

| Calculation of the SDG Index
The calculation of the SDG Index followed the methodology The SDG Index offers the possibility to compare scores at the level of communes, cities, and counties in Romania.

| Data limitations
The study presents some limitations. First, even though we use the most recent available data for each indicator, the old data from the last census (2011)  23.6% in 2017 as compared to the EU average of 16.9%. What is more, the absolute poverty rate was four times higher in rural than in urban areas (Eurostat, 2020a (EM-DAT, 2020;Török, 2017). We must also take into account that projections related to climate change indicate an increase in the frequency and intensity of these phenomena (Micu, Dumitrescu, Cheval, & Birsan, 2015). All these have roots in the excessive industrialization, deforestation, and large-scale agriculture, complemented by the spatial expansion of the built environment, mainly in large urban centers and their metropolitan areas. Therefore, improving resilience and the adaptive capacity of countries to fight extreme weather patterns represents one of the major targets of the UN under the 13 SDG. Over the last years, tackling with climate change effects has represented an important issue also in Romania.
The National Strategy for Climate Change in Romania was approved in July 2013 by the GO 529/2013, focusing on the reduction of vulnerability in specific sectors like agriculture, energy, water resources, transport, industry, construction, urban planning insurance, biodiversity, human health, tourism, forestry, infrastructure, and recreational activities (MECC, 2013). In order to analyze the degree to which SDG 13 goals are achieved and to explore the reasons why some developed areas perform well below the national average, we have included indicators regarding the share of forest areas, growth rates of the built-environment, public expenses related to environmental protection, as well as indices with a great impact on climate change.
According to our results, most of the counties-including localities in mountainous areas (i.e., Caraş-Severin, Hunedoara, Vâlcea, Bistrița-N as aud, Neamț) have achieved a score (above 7) almost two times higher than that of developed counties (around 4). The situation gives reason for concern even in the capital and the surrounding Ilfov

| CONCLUSIONS
The Agenda 2030 has reinvigorated scientific and policy-led interest in indicator-based assessment of progress toward SDGs. Our study proposes a novel approach, called an integrated territorial approach,