A cost-effective and transferable methodology for rooftop PV potential assessment in developing countries

: The e ﬃ cient uptake of decentralized solar rooftop photovoltaics (PV) is in some cases hindered by ine ﬀ ective energy and political framework conditions. These may be based on inaccurate and uncertain potential assessments in the early development stage of the solar market. This paper develops a more accurate, cost-e ﬀ ective, and robust potential assessment for emerging and developing economies. Adjusting the module e ﬃ ciency corresponding to regional and household conditions improves the output accuracy. The rooftop PV market changes are simulated regarding di ﬀ erent input changes and policy designs, including changing the Feed-In Tari ﬀ (FIT), grid tari ﬀ , and technology development. In the case study, the market potential in Vietnam is estimated at 260–280 TWh / a and is clustered into six groups in priority order, in which Hanoi and Ho Chi Minh need the most policy focus. Changing the FIT from 8.83 to 9 Euro cent / kWh and using di ﬀ erent regional FITs can activate an additional 16% of the market and lead to a possible 28 million Euro beneﬁt. Increasing the grid tari ﬀ to 8.7 cents / kWh could activate the self-consumption model, and the self-su ﬃ cient market can be guaranteed in the case of CAPEX and OPEX being lower than 650 Euro / kWp. Future developments of the method should focus on combining this top-down method with detailed bottom-up approaches.


Introduction
Renewable energy technologies have gained momentum in cost competition with conventional thermal electricity generation and have become significant in the global electricity transition. They had a share of 14% in global electricity generation by 2018, with around 53% of all renewable energy investments in developing and emerging economies [1]. In the countries concerned, distributed solar PV systems not only play a prominent role in spreading energy access to households in remote areas [1] but also promise an effective alternative to serving the highly increased urban electricity demand [2][3][4].
Many sub-national governments have become leaders of solar development by setting even more ambitious targets than their national counterparts [1]. However, national and local governments in the countries are dependent on generic rooftop PV assessments. These assessments often have a rather low accuracy, which could direct investors into suboptimal locations and configurations [5,6]. As a result, the diffusion of PV power projects is spatially heterogeneous between the different regions in a country [7][8][9]. If the development goes in one direction only, this could have a far-reaching impact on both grid and market congestion [9][10][11][12]. To avoid this, policymakers urgently need a superior method

Methodology for Rooftop PV Potential Assessment
Figure 1 describes the method of PV potential assessment proposed in this paper. The outputs of each step are highlighted in blue, and the dark blue arrows show the direction of the mathematical flow. Input data are collected from (*) national/international statistical sources, (**) the assumptions based on previous studies, and/or (***) a collection from manufacturers' catalogues. The iterative process to create datasets are represented by orange arrows. There are in total four different loops numbered in Figure 1. To calculate a high-resolution PV potential, previous studies have used advanced and accurate technologies such as 3-D models to calculate geometry, insolation, and shading of buildings [5,[18][19][20][21]. Even though the methods are more precise, the main barrier to employing them at national and international scales is the associated challenge of an exponentially increasing number of uncertain parameters with an increase in the sample size or studied space. This leads to computationally intensive models and expensive data collection [25], especially for 3D models [18][19][20] or solar detective models [31]. Therefore, these methods are not currently economically viable for developing countries [24,27].
This paper provides a less costly and more accurate solution that is able to cover a large scale by estimating rooftop PV potential for each region corresponding to its climatic, geographic, and demographic conditions. The available roof for installing solar PV and the expected electricity production are calculated based on housing characteristics such as the type of roof, the azimuth of the building, and the roof. Statistical data and previous research results are used to determine housing architecture. Uncertainty in the data coming from the natural variability of the data generating, measuring, and sampling processes is then simulated by using an iterative process. The results of the simulation are the output distributions (Section 5.3). The details of the method are presented in Section 3.

Methodology for Rooftop PV Potential Assessment
Figure 1 describes the method of PV potential assessment proposed in this paper. The outputs of each step are highlighted in blue, and the dark blue arrows show the direction of the mathematical flow. Input data are collected from (*) national/international statistical sources, (**) the assumptions based on previous studies, and/or (***) a collection from manufacturers' catalogues. The iterative process to create datasets are represented by orange arrows. There are in total four different loops numbered in Figure 1. The method proposed in the paper consists of three steps. The first step is determining the geographical potential using ArcGis and geostatistical analysis (Section 3.1), which is proposed in previous studies [25]. The second step is identifying the technical potential (Section 3.2), in which the authors improve the assessment accuracy by determining the electricity output of the solar cell regarding the expected operating temperature of the PV module and the corresponding solar cell The raster maps [17,35] only provide the solar irradiation at 90° to the horizontal. However, the irradiation changes depending on the solar position and its relation to an oriented panel. To improve the accuracy of the geographical potential, this paper calculates relative solar irradiations corresponding to the northern, central, and southern horizontal surface in different azimuth and inclination angles.
The geographical potential is determined by using ArcGis to match available roof area, , and relative solar irradiation with data of the topographic, population distribution, and cadastral maps.

Technical Potential
A PV module will typically be rated at standard test conditions (STC), which indicates a condition of a cell temperature of 25 °C and an irradiance of 1000 W/m 2 with an air mass 1.5 (AM1.5) spectrum [36], ensuring a relatively independent comparison and output evaluation of different modules. However, when operating in the field, they typically operate at higher temperatures and at somewhat lower insolation conditions. To improve the accuracy of technical potential assessments, the paper determines the power output of the solar cell regarding the expected operating temperature of the PV module and the corresponding solar cell efficiency.
Energies 2020, 13, x FOR PEER REVIEW 6 of 46 over a calendar year, we would see an effect of both irradiance and temperature on the electricity output of a solar cell across the seasons. Another factor that affects the electricity output is the solar cell efficiency. Because the efficiency depends on the spectrum and intensity of the incident sunlight and the temperature of the solar cell [40], the paper calculates the corrected energy conversion efficiency ( ) of a solar cell corresponding to the operating temperature , which is reflected in the latitude and climate of the considered region. The corrected temperature factor and the corrected efficiency of the module cell are determined via Equation (4). A constant value (CT) can be extracted from the module datasheet (%/K). and are the temperature and efficiency at STC conditions.
Moreover, to be more precise in assessing technical potential, this paper identifies the least space distance between PV modules to minimize shading effects. To calculate the module row spacing between two lines of modules, the first step is to calculate the height difference, HD (m), from the back edge of the module to the surface [34]. Equation (5) is used for this purpose. β (degrees) is the tilt angle and (m) is the module width.
= sin( ) × The module row spacing ( ) is determined by using the Sun Elevation Angle. The row width, , which is the distance from the trailing edge of one row to the trailing edge of the subsequent row, is calculated as in Equation (6), in which α is the solar elevation angle (degrees) ( Figure 3) and κ is the azimuth correction angle (degrees) (Figure 4). The shadow between rows falls perpendicular to a south-facing array only when the sun is located at true south in the sky. At solar noon, the length of the shadow cast between rows would be equal to the minimum row spacing ( ). In this paper, is calculated in the worst case from 9 AM to 3 PM of the winter solstice, therefore an azimuth angle correction is required (Figure 4).

=
× cos × = tan( ) × cos × cos × (6) Figure 3. Schematic of two simulated photovoltaic rows at solar noon. Reproduced with permission from the authors of "Submerged and Floating Photovoltaic Systems", Copyright Elsevier Science Publishing Co Inc., 2018 [33].   The number of PV modules, , on a rooftop for each raster map unit corresponding to the available roof area for PV, , is obtained from Equation (7), in which is the length of the PV module (m).
The electricity output of the solar system, , is obtained from Equation (8), in which PR is performance ratio of the module, SI is the adjusted solar irradiation (kWh/m 2 ) and is the module area (m 2 )

Levelized Cost of Electricity (LCOE)
The Levelized Cost of Electricity (LCOE) is a measure of the average net present cost of electricity generation for a generating plant over its lifetime, which is assumed to be 20 years in this study [41]. The LCOE depends on costs (i.e., the initial capital investment, maintenance, and operational costs), local condition (load profile), and financial condition (i.e., the discount rate) The LCOE is calculated as in Equation (9) [42]  The method proposed in the paper consists of three steps. The first step is determining the geographical potential using ArcGis and geostatistical analysis (Section 3.1), which is proposed in previous studies [25]. The second step is identifying the technical potential (Section 3.2), in which the authors improve the assessment accuracy by determining the electricity output of the solar cell regarding the expected operating temperature of the PV module and the corresponding solar cell efficiency for the specific considered region. The last step is the economic and sensitivity analyses (Section 3.3). Section 4 presents data, assumptions and results for our case study, which is used to illustrate the methodology proposed.

Geographical Potential
In the case of unavailable high-resolution street maps, household data from government statistical offices can be used to calculate the available roof area for PV installation A roo f (m 2 ) for detached houses and apartment buildings. Equation (1) is suggested based on previous studies [25].
The total roof area, S roo f (m 2 ), is calculated by subtracting the exterior area, b 1 (percent of total roof area), such as garden, back yard, etc., and adding the interior area, b 2 (percent of total roof area), Energies 2020, 13, 2501 7 of 46 such as stairway, corridors, etc. for an apartment building, to the ground area of a household S F (m 2 ). For apartment buildings, n A F is the average number of apartments, n A , in a building on one floor, F. The available roof for installing rooftop PV, A roo f (m 2 ), is determined by multiplying the total roof area, S roo f (m 2 ), with the utilization factor, µ (%), which is a percentage of the roof area that can be used for PV installations. The factor µ excludes the utilized area of antennas, chimneys, HVAC systems, and unusable area such as shades from the total roof area, S roo f (m 2 ). It can be identified based on national statistical data of household architecture or based on previous studies [25]. ν is the average house/building roof slope (degrees). Figure 2 demonstrates the two angles related to the sun position, which are the solar altitude angle α s , and the solar azimuth angle γ s , the angles β and γ that define the PV module position. β is the tilt angle, and γ is the surface azimuth angle. Further, the angle θ between the normal to the PV surface and the incident beam radiation is indicated [33,34].
The solar irradiation at different azimuths, inclination angles, and horizontal positions regarding different locations is calculated as in Equation (2), in which SI dir and SI di f f are the direct and diffuse solar irradiations, respectively.
The geographical potential is determined by using ArcGis to match available roof area, A roo f , and relative solar irradiation with data of the topographic, population distribution, and cadastral maps.
The raster maps [17,35] only provide the solar irradiation at 90 • to the horizontal. However, the irradiation changes depending on the solar position and its relation to an oriented panel.
To improve the accuracy of the geographical potential, this paper calculates relative solar irradiations corresponding to the northern, central, and southern horizontal surface in different azimuth and inclination angles.
The geographical potential is determined by using ArcGis to match available roof area, A roo f , and relative solar irradiation with data of the topographic, population distribution, and cadastral maps.

Technical Potential
A PV module will typically be rated at standard test conditions (STC), which indicates a condition of a cell temperature of 25 • C and an irradiance of 1000 W/m 2 with an air mass 1.5 (AM1.5) spectrum [36], ensuring a relatively independent comparison and output evaluation of different modules. However, when operating in the field, they typically operate at higher temperatures and at somewhat lower insolation conditions. To improve the accuracy of technical potential assessments, the paper determines the power output of the solar cell regarding the expected operating temperature of the PV module and the corresponding solar cell efficiency.
First, the paper identifies the operating cell temperature T mod ( • C) based on nominal operating cell temperature (NOCT • C) Equation (3).
NOCT is defined as the temperature reached by open-circuited cells in a module under the conditions of the maximum irradiance, G NOCT = 800 W/m 2 , air temperature is 20 • C, wind velocity is 1 m/s, and an open rear surface mounting. NOCT can be calculated from the NOCT model [37] or taken from the information given by the module specification. The equation is a literature model [38] that gives a reasonable estimate of the operating temperature of the module as a function of the ambient temperature (T air • C) and the irradiance at the instant G mod (W/m 2 ) when the ambient temperature is T air ( • C). T air ( • C) can be obtained from Raster data or extracted manually from the climate map [39]. As can be seen in Equation (3), because the irradiance and temperature changes over a calendar year, we would see an effect of both irradiance and temperature on the electricity output of a solar cell across the seasons. Another factor that affects the electricity output is the solar cell efficiency. Because the efficiency depends on the spectrum and intensity of the incident sunlight and the temperature of the solar cell [40], the paper calculates the corrected energy conversion efficiency (η cor ) of a solar cell corresponding to the operating temperature T mod , which is reflected in the latitude and climate of the considered region. The corrected temperature factor η TC and the corrected efficiency of the module cell η cor are determined via Equation (4). A constant value (CT) can be extracted from the module datasheet (%/K). T STC and η STC are the temperature and efficiency at STC conditions.
Moreover, to be more precise in assessing technical potential, this paper identifies the least space distance between PV modules to minimize shading effects. To calculate the module row spacing between two lines of modules, the first step is to calculate the height difference, HD (m), from the back edge of the module to the surface [34]. Equation (5) is used for this purpose. β (degrees) is the tilt angle and W mod (m) is the module width.
The module row spacing (MRS mod ) is determined by using the Sun Elevation Angle. The row width, W row , which is the distance from the trailing edge of one row to the trailing edge of the subsequent row, is calculated as in Equation (6), in which α is the solar elevation angle (degrees) ( Figure 3) and κ is the azimuth correction angle (degrees) (Figure 4). The shadow between rows falls perpendicular to a south-facing array only when the sun is located at true south in the sky. At solar noon, the length of the shadow cast between rows would be equal to the minimum row spacing (MRS mod ). In this paper, MRS mod is calculated in the worst case from 9 AM to 3 PM of the winter solstice, therefore an azimuth angle correction is required (Figure 4).
The number of PV modules, n mod , on a rooftop for each raster map unit corresponding to the available roof area for PV, A roo f , is obtained from Equation (7), in which L mod is the length of the PV module (m).
The electricity output of the solar system, M el , is obtained from Equation (8), in which PR is performance ratio of the module, SI is the adjusted solar irradiation (kWh/m 2 ) and A mod is the module area (m 2 )

Levelized Cost of Electricity (LCOE)
The Levelized Cost of Electricity (LCOE) is a measure of the average net present cost of electricity generation for a generating plant over its lifetime, which is assumed to be 20 years in this study [41]. The LCOE depends on costs (i.e., the initial capital investment, maintenance, and operational costs), local condition (load profile), and financial condition (i.e., the discount rate). The LCOE is calculated as in Equation (9) [42] where I 0 (Euro) stands for the total initial investment, which is the product of module size (in kW) and specific investment i 0 as the investment cost per installed kW (€/kW); A t (Euro) is the yearly operating cost; i (%) is the discount rate; n s (year) is the life span; and M el (kWh) is the annual electricity production.

Sensitivity Analysis
The proposed model is computationally intensive and highly parameterized with model input uncertainty, which can reduce the accuracy of the model outputs. To provide a robust assessment at medium resolution, this paper suggests a solution to estimate the impact and significance of uncertain parameters in the model by using variance-based sensitivity analysis. A variance-based method is a probabilistic approach, which quantifies the input and output uncertainties as probability distributions and decomposes the output variance into parts attributable to input variables and combinations of variables.
The sensitivity of the output to an input variable is therefore measured by the amount of variance in the output caused by that input.
Considering our model in Equation (10): As Y = f (X) for X = {X 1 , X 2 , . . . , X k }, a measure of the sensitivity of the ith variable X i is given as{\displaystyle \operatorname {Var} \left(E_{{\textbf {X}}_{\sim i}}\left(Y\mid X_{i}\right)\right)} Var(E X ∼i (Y X i )) , where "Var" and "E" denote the variance and expected value operators, respectively, and X ∼i denotes the set of all input variables except X i . This expression essentially measures the contribution of X i alone to the uncertainty (variance) in Y (averaged over variations in other variables) and is known as the first-order sensitivity index or main effect index, S i , which is determined as in Equation (11). Importantly, it does not measure the uncertainty caused by interactions with other variables.
This method allows exploration of the input space, accounting for linear and nonlinear responses. As demonstrated in Figure 1, four loops, represented by orange arrows, repeatedly run to generate the final outputs until all the possibilities of the inputs are covered. The first input set is built based on the available data and assumptions from previous studies.
Subsequently, this paper fits a linear and/or nonlinear regression to the model response and the standardized regression coefficients are used to measure the level of impact of each uncertain parameter on the output. The relationship between the uncertain parameters and the output is presented as in Equation (12), for example.
This paper uses hierarchal clustering in the manner of the Pareto principle to generate the ranking of the uncertainty factors x i according to their relative contribution to the output variability. The influencers are categorized into different groups, including "strong influencer", "moderate influencer", and "negligible influencer". The goals of the analysis are to provide different references for policymakers and stakeholders to choose an appropriate interaction with the current rooftop PV market status. One goal focuses on quantifying the uncertainty in the output of the model, thus it is used as a control tool for modelers and policymakers in order to ensure the accuracy of the output. The other goal focuses on apportioning output uncertainty to the different sources of uncertainty inputs and will be used for designing policy and for supporting the decision-making of investors in the rooftop PV market.

Data and Assumptions
Vietnam is taken as a case study for demonstrating our method and findings. It is one of the 65 emerging countries in the world with typical features and struggles to develop solar PV [2][3][4]. The Vietnamese population is 94 million people, with annual GDP growth of around 7%, the total electricity demand of 227,421 GWh, total supply capacity of 48.57 GW [43], and power demand surging at an average rate of 12% per year since 2010 and expected to continue growing by 8%/year until 2030 (Vietnam energy outlook 2018). Residential demand for electricity accounts for almost a third of the total electricity demand [43], with the average retail tariff of 9.3 Euro cents/kWh (Appendix A Table A1). However, during the period 2016-2030, the original estimation of the additional capacity is estimated to be decreased by 16% (~15.2 GW) because of the delay of some natural gas projects. This would be likely to increase the power shortage in the south, especially from 2022 onwards (EVN June 2019).
Under these circumstances, solar and wind power are expected to play significant roles as alternative resources and turn Vietnam into a hot spot for energy investors. Renewable energy supply, including small hydro, wind, solar, and biomass power plants, currently covers 7.16% of total national supply [43]. However, given a local overload, transmission losses, and uneven infrastructure between different regions, the Vietnamese government has been provoking significant investment in power infrastructure with net metering. As a result, by June 2019, the country has shown a surge of centralized solar capacity connected to the national grid of 4.46 GW. This accounts for 8.28% of the Vietnamese electricity supply mix, which is far more than the 1000 MW target set for 2020 (EVN report, June 2019), while PV rooftop projects only grow at a very modest level. Moreover, because of lacking integrating policy, management, and compensation mechanisms to support rooftop PV development in the whole country, it has developed rapidly solely in Ninh Thuan. This region accounts for 90% of the total national solar capacity, which has caused significant grid congestion and reduced the spread of the development.
To solve the issue of the lacking integrating policy and mechanisms to support rooftop PV development, this paper focuses on assessing rooftop PV potentials in 63 provinces in Vietnam and proposes different suggestions for the government to unleash their PV potentials.
This paper uses a previous study of house design in Vietnam [32], in which the average number of apartments on one floor of the building, n A F , is 5, the exterior area, b 1 , is 0.2, and the interior area, b 2 , is 0.1 to calculate the available roof area for installing rooftop PV. The roof of the apartment buildings is considered flat, so ν for them is 0. For single-detached houses, ν is presented in Table 1. A sensitivity analysis for these assumptions is conducted in this paper. The utilization factor of the flat roof is estimated to be 50% based on previous studies [5,9,14,44], which concluded that 35% to 65% of residential flat roofs are available for PV because 5% of the roof is covered by HVAC equipment or other building components, which would shade over 35% of the entire rooftop, and trees and other construction would shelter from 0% to 30% of the rest. The utilization factor of a slanted roof is assumed to be 58%, which considers that a north-facing roof also has the potential for installing PV systems [25].
Based on Equation (2), the relative solar irradiation for different roof types in three different parts of Vietnam is calculated. The results are presented in Table 1.
In this paper, two main models of PV modules, i.e., monocrystalline and polycrystalline, are considered. The characteristics of these two models, extracted from the catalogue, are given in Table 2. The performance ratio is considered as 0.85 [25]. To assess the effect of uncertainty parameters on the outputs, this paper alters the assumptions and the values of input variables, which have been referenced from other studies but for which sensitivity analyses have not been performed in previous studies. The used values of these inputs and the considered variation range are presented in Table 4. (1) to control the accuracy of the output; and (2) to measure the output variability regarding possible input changes in order to support policy decision making.

Spatial Potential Distribution of Rooftop PV
This paper breaks down the estimated theoretical potential of solar with a total of around 120 thousand TWh per year (TWh/a) from the solar irradiation raster map [17,34,49] to geographical and technical potentials of rooftop PV with 1407.2 km 2 available rooftop and 278 TWh/a, respectively. The technical potential is calculated with the results of roof-mounted solar cells in the whole country being 228.3 and 215.8 GWp or 278.1 and 262.7 TWh/a with monocrystalline and polycrystalline solar cells, respectively (detailed results for each region are given in Table A3).
Comparing different regions in Vietnam, 58% of the total national rooftop PV potential is concentrated in three territories, including the two river deltas and the southeast with an almost equally divided potential of around 20%, equivalent to an average installable capacity of 43.9 GWp. The remaining potential is divided among the coastal areas and the midlands at approximately 10% or 22 GWp, and the mountains and highlands have the lowest potential with about 5% or 11 GWp in each territory.
Taking economic factors into account, the cost per kWh of rooftop PV varies greatly depending on the region and the prevailing economic conditions, in particular regarding different capital costs and discount rates. This paper determines the LCOE (Euro/kWh) of rooftop PV for each province in order to assess the economic attractiveness of the rooftop PV assumed at the current discount rate of 8%.
The distributed PV potentials, including the specific production potential per total area (GWh/km 2 ) and the LCOE (Euro/kWh) in Vietnam, are shown in Figures 5 and 6. The potential density is significantly high in the two river delta regions, which are Red River Delta (RRD) and Mekong River Delta (MRD) in Vietnam, especially in the two largest cities of the country, Hanoi (in the RRD) and Ho Chi Minh (in the MRD), with the most enormous potential of electricity production of about 12 and 5 GWh/km 2 ( Figure 6) or around 24.8 and 15.2 TWh/a or 18 and 15.1 GWp, respectively. These cities are the largest metropolitans and most populous cities with an estimated population of 7.7 million and 8.4 million, corresponding to 8.6% and 9.3% of the national population, respectively, and they also have the largest household area in Vietnam (Government statistical reports, 2018). Hanoi has 407 km 2 of housing land, of which around 93.9 km 2 rooftop available to install solar cells, while Ho Chi Minh City has about 282 km 2 housing land but can provide an estimated 110 km 2 roof for installing solar. Moreover, since Ho Chi Minh City is in the southern part of Vietnam, it could potentially generate in total 24.8 or 23.4 TWh/a with monocrystalline or polycrystalline solar cells, respectively, which is nearly 60% more than the electricity production potential from solar in Hanoi. This gives Ho Chi Minh City the most significant potential for solar power throughout the country, while Hanoi is ranked second. The LCOE ranges from 7.5 to 9.2 Euro cents/kWh at a discount rate of 8% across 63 provinces in Vietnam ( Figure 5). In the regions with a vigorous radiation intensity such as Central Highland, Mekong River Delta, South Central Coast, and Southeast (see Figure 5 for regions), rooftop PV is more economical than in other areas with the LCOE being around 0.075 Euro/kWh. Among the eleven cities, which account for 50% of the total specific potential (GWh/km 2 ) in the whole country, there are only four cities, namely Ho Chi Minh City, Can Tho, Tien Giang, and Vinh Long, with the average LCOE being less than 0.076 Euro/kWh ( Figure 6). Despite having a sizeable available roof area, Hanoi and the rest do not have a convincing economic potential due to their low radiation intensity, which leads to a relatively high LCOE of around 0.085-0.089 Euro/kWh. The other quarter of potentials consists of 12 provinces that are relatively promising for investment with high potentials of approximately 2-3 GWh/km 2 and a low LCOE of about 0.075-0.080 Euro/kWh, except for Vinh Phuc and Ninh Binh. However, Ninh Thuan and Binh Thuan, which are the best places for investing in rooftop PV with the lowest LCOE in Vietnam at around 0.073 Euro/kWh, have relatively small solar resources at about only 0.6 GWh/km 2 . Figure 7 illustrates the cost-potential curves for monocrystalline solar deployment in Vietnam at different discount rates ranging from 5% to 11%. With varying rates of discount, the cost curve for solar PV in Vietnam shifts vertically. The total electricity potential in the case of considering The LCOE ranges from 7.5 to 9.2 Euro cents/kWh at a discount rate of 8% across 63 provinces in Vietnam ( Figure 5). In the regions with a vigorous radiation intensity such as Central Highland, Mekong River Delta, South Central Coast, and Southeast (see Figure 5 for regions), rooftop PV is more economical than in other areas with the LCOE being around 0.075 Euro/kWh. Among the eleven cities, which account for 50% of the total specific potential (GWh/km 2 ) in the whole country, there are only four cities, namely Ho Chi Minh City, Can Tho, Tien Giang, and Vinh Long, with the average LCOE being less than 0.076 Euro/kWh ( Figure 6). Despite having a sizeable available roof area, Hanoi and the rest do not have a convincing economic potential due to their low radiation intensity, which leads to a relatively high LCOE of around 0.085-0.089 Euro/kWh. The other quarter of potentials consists of 12 provinces that are relatively promising for investment with high potentials of approximately 2-3 GWh/km 2 and a low LCOE of about 0.075-0.080 Euro/kWh, except for Vinh Phuc and Ninh Binh.
However, Ninh Thuan and Binh Thuan, which are the best places for investing in rooftop PV with the lowest LCOE in Vietnam at around 0.073 Euro/kWh, have relatively small solar resources at about only 0.6 GWh/km 2 . Figure 7 illustrates the cost-potential curves for monocrystalline solar deployment in Vietnam at different discount rates ranging from 5% to 11%. With varying rates of discount, the cost curve for solar PV in Vietnam shifts vertically. The total electricity potential in the case of considering monocrystalline and polycrystalline cells adds up to around 278 and 262 TWh/a, respectively. The total electricity demand in 2018 in Vietnam is 192. 36 TWh, which implies that the total solar rooftop could cover the current demand of the country. However, Vietnam has only exploited about 0.1% of its potential [43]. The step in the curve (see Figure 7) results from the LCOE jumping from high irradiation in the southern part to low radiation to the northern part of Vietnam. It reveals the significant difference in solar irradiation along with the country, which causes a drastic increase in LCOE from south to north. Referring the LCOE results to FIT [51] reveals a favorable condition for the rooftop solar development because the FIT can cover the LCOE cost. Details of the relation between LCOE and FIT are discussed in Section 5.2.
Finally, we analyze the results of the present paper in the context of additional studies relating to rooftop PV in Vietnam. As compared to other studies in Vietnam, this method poses its advantages of providing a reasonably accurate result at low-cost (Table 4), even though it deals with insufficient data and meager budget. When normalizing for the area covered to 80% of the cities' area, this paper determines the potentials for rooftop PV in HCM and DN with a relative error of 4% and 6% compared with the field study, respectively.  The step in the curve (see Figure 7) results from the LCOE jumping from high irradiation in the southern part to low radiation to the northern part of Vietnam. It reveals the significant difference in solar irradiation along with the country, which causes a drastic increase in LCOE from south to north. Referring the LCOE results to FIT [51] reveals a favorable condition for the rooftop solar development because the FIT can cover the LCOE cost. Details of the relation between LCOE and FIT are discussed in Section 5.2.
Finally, we analyze the results of the present paper in the context of additional studies relating to rooftop PV in Vietnam. As compared to other studies in Vietnam, this method poses its advantages of providing a reasonably accurate result at low-cost (Table 3), even though it deals with insufficient data and meager budget. When normalizing for the area covered to 80% of the cities' area, this paper determines the potentials for rooftop PV in HCM and DN with a relative error of 4% and 6% compared with the field study, respectively.

Sensitivity Analyses
Similar to many other developing countries, Vietnam has deficient and inadequate data conditions even to estimate rooftop PV potential at medium spatial resolution. It is worth mentioning that conducting such an assessment for countries with similar unfavorable data condition requires significant manual processing of inputs and maps, for example matching geographic and population distributions and building densities with 11,050 clusters of the topographic map in the case of Vietnam.
This section assesses the effect of each mentioned input or assumption on technical and economic potentials and identifies an acceptable confidence interval of every particular variable required to achieve specific confidence in the result. The results are demonstrated in Figures 8 and 9.

Sensitivity Analyses
Similar to many other developing countries, Vietnam has deficient and inadequate data conditions even to estimate rooftop PV potential at medium spatial resolution. It is worth mentioning that conducting such an assessment for countries with similar unfavorable data condition requires significant manual processing of inputs and maps, for example matching geographic and population distributions and building densities with 11,050 clusters of the topographic map in the case of Vietnam.
This section assesses the effect of each mentioned input or assumption on technical and economic potentials and identifies an acceptable confidence interval of every particular variable required to achieve specific confidence in the result. The results are demonstrated in Figures 8 and 9. The results show that, among the uncertain parameters that are used to calculate the geographical and technical potentials, the utilization factors for flat and slanted roofs and the efficiency or performance ratio of a considered PV system are the most critical parameters influencing the accuracy of the technical potential output (Figure 8). Considering the economic potential, all the uncertain geographical inputs, such as the average number of apartments on one floor in a building, area added to the ground floor area, and utilization factors for different roofs, are irrelevant to the LCOE output. However, technical inputs, especially the efficiency of the PV module, show significant effects on the LCOEs. Lastly, the economic data, such as costs and interest, have the most significant impact on the LCOE output, especially the capital cost ( Figure 9). To verify the correlation of the variation of these variables with the output, we use regression techniques, including curve fitting and linear fitting, to specify the model that provides the best explanation of the relationships between the uncertain parameters and outputs. The considered outputs are the technical potential (y1) and the LCOE (y2). The uncertain parameters are indicated below: x1 Area added to a ground floor area x6 Azimuth of building x2 Utilization factor for flat roof x7 Capital cost x3 Utilization factor for slanted roof x8 O&M cost x4 Average number of apartments on one floor x9 Discount rate x5 Performance ratio Figure 10 shows the strong influence of the uncertain parameters on the outputs in descending order. Using the Pareto principle [53,54] indicating that roughly 80% of the effects come from 20% of the causes, this paper divides them into three groups. The first group is called "strong influencer", which consists of x3, x5, x5, and x7 that are responsible for more than 80% of the impact on the technical and economic outputs, respectively. The second one is the "moderate influencer", which includes x2 for technical output and x8 for economic output, and accounts for the next 15% of the impact on the outputs. The other factors belong to the "negligible influencer" group. By selecting impacts based on their strength, we can measure the change of the outputs based on the different possible alterations of the uncertain parameters, as in Equations (13) and (14). The results show that, among the uncertain parameters that are used to calculate the geographical and technical potentials, the utilization factors for flat and slanted roofs and the efficiency or performance ratio of a considered PV system are the most critical parameters influencing the accuracy of the technical potential output (Figure 8). Considering the economic potential, all the uncertain geographical inputs, such as the average number of apartments on one floor in a building, area added to the ground floor area, and utilization factors for different roofs, are irrelevant to the LCOE output. However, technical inputs, especially the efficiency of the PV module, show significant effects on the LCOEs. Lastly, the economic data, such as costs and interest, have the most significant impact on the LCOE output, especially the capital cost (Figure 9).
To verify the correlation of the variation of these variables with the output, we use regression techniques, including curve fitting and linear fitting, to specify the model that provides the best explanation of the relationships between the uncertain parameters and outputs. The considered outputs are the technical potential (y 1 ) and the LCOE (y 2 ). The uncertain parameters are indicated below: x1 Area added to a ground floor area x6 Azimuth of building x2 Utilization factor for flat roof x7 Capital cost x3 Utilization factor for slanted roof x8 O&M cost x4 Average number of apartments on one floor x9 Discount rate x5 Performance ratio Figure 10 shows the strong influence of the uncertain parameters on the outputs in descending order. Using the Pareto principle [53,54] indicating that roughly 80% of the effects come from 20% of the causes, this paper divides them into three groups. The first group is called "strong influencer", which consists of x 3 , x 5 , x 5 , and x 7 that are responsible for more than 80% of the impact on the technical and economic outputs, respectively. The second one is the "moderate influencer", which includes x 2 for technical output and x 8 for economic output, and accounts for the next 15% of the impact on the outputs. The other factors belong to the "negligible influencer" group. By selecting impacts based on their strength, we can measure the change of the outputs based on the different possible alterations of the uncertain parameters, as in Equations (13) and (14). The results show how dependent the output is on a particular input value. The performance ratio -x5 has the most significant impact on both outputs of the model, which are the electricity potential -y1, and the LCOE -y2. While the electricity potential is also changed considerably when the utilization factors change, the LCOE is not be affected by these two variables. It alters when the costs and discount rates vary.
These results can be used to assist policymakers for decision making by predicting the outcome of a decision in case they intend to change inputs using policy incentive/subsidy. It helps in assessing the riskiness of a strategy and in making informed and appropriate decisions. A detailed discussion of and examples for this argument are presented in Section 5.

Deriving Policy Implications from the Potential Assessment Results
Based on the results from the sensitivity analysis, we identified that factors including the performance ratio of the PV system, capital and O&M costs, and the discount rate are the variables with the most significant impact on the results of LCOE. To design a practical and transferable tool for policymakers in planning rooftop PV and a reference for stakeholders, this paper explores changes in each area when these variables change. First, we identify target groups based on their PV potential and demographic characteristics (Section 5.1). Then, we build the market distribution curves based on the techno-economic results and then observe the possibilities of shifting and bending distribution curves under different possible impact changes (Sections 5.2 and 5.3). This paper compares different desirable policies to help policymakers select effective strategies to promote their desire for the rooftop PV market.

Defining Target Groups
From the results in Section 4.2.1, national and regional governments can choose the most economic provinces directly from the potential ranking in Figure 6. However, the main drivers for solar development in developing countries are urbanization and GDP [2,55]. This means the chance of higher-income people investing in rooftop solar PV is higher in urban than in rural areas. Figure 11 shows the available rooftop PV distribution corresponding to the level of urbanization (left) and the GDP per capita in 2018 (right) distributions in 63 regions in Vietnam. The distributions highlighted in red, which consist of 14 provinces, indicate the minimal potential area, with less than 5 km 2 of roof area, 15% of urbanization, and 1000 Euro/a income. The green distributions illustrate the outstanding potential area, including nine provinces with more than 35 km 2 of the available roof, 60% of urbanization, and 2000 Euro/a income. Details of potential assessment results and demographic indicators can be found in Appendices 2 and 3. The results show how dependent the output is on a particular input value. The performance ratio -x 5 has the most significant impact on both outputs of the model, which are the electricity potential -y 1, and the LCOE -y 2 . While the electricity potential is also changed considerably when the utilization factors change, the LCOE is not be affected by these two variables. It alters when the costs and discount rates vary.
These results can be used to assist policymakers for decision making by predicting the outcome of a decision in case they intend to change inputs using policy incentive/subsidy. It helps in assessing the riskiness of a strategy and in making informed and appropriate decisions. A detailed discussion of and examples for this argument are presented in Section 5.

Deriving Policy Implications from the Potential Assessment Results
Based on the results from the sensitivity analysis, we identified that factors including the performance ratio of the PV system, capital and O&M costs, and the discount rate are the variables with the most significant impact on the results of LCOE. To design a practical and transferable tool for policymakers in planning rooftop PV and a reference for stakeholders, this paper explores changes in each area when these variables change. First, we identify target groups based on their PV potential and demographic characteristics (Section 5.1). Then, we build the market distribution curves based on the techno-economic results and then observe the possibilities of shifting and bending distribution curves under different possible impact changes (Sections 5.2 and 5.3). This paper compares different desirable policies to help policymakers select effective strategies to promote their desire for the rooftop PV market.

Defining Target Groups
From the results in Section 4.2.1, national and regional governments can choose the most economic provinces directly from the potential ranking in Figure 6. However, the main drivers for solar development in developing countries are urbanization and GDP [2,55]. This means the chance of higher-income people investing in rooftop solar PV is higher in urban than in rural areas. Figure 11 shows the available rooftop PV distribution corresponding to the level of urbanization (left) and the GDP per capita in 2018 (right) distributions in 63 regions in Vietnam. The distributions highlighted in red, which consist of 14 provinces, indicate the minimal potential area, with less than 5 km 2 of roof area, 15% of urbanization, and 1000 Euro/a income. The green distributions illustrate the outstanding potential area, including nine provinces with more than 35 km 2 of the available roof, 60% of urbanization, and 2000 Euro/a income. Details of potential assessment results and demographic indicators can be found in Tables A2 and A3. Because most of the provinces have a relatively low (15-60%) urban rate and humble economic (1000-2000 Euro/a income) development, the success of the first adopters plays a crucial role in the new market due to their potential impacts on followers. To improve rooftop solar development, the role of local and government authorities is equally essential. However, in practice, sub-national or local governments are constrained by limited resources, weak institutional capacity, inadequate mechanisms, and limited availability of information [3,7,26,29]. These circumstances, as well as the complication of land ownership and attractiveness, constitute a narrow space for the policymakers to encourage rooftop PV development at the local level without significant support from the government. This means that there is an obvious need for selecting target groups that is unified from national to local levels.  Because most of the provinces have a relatively low (15-60%) urban rate and humble economic (1000-2000 Euro/a income) development, the success of the first adopters plays a crucial role in the new market due to their potential impacts on followers. To improve rooftop solar development, the role of local and government authorities is equally essential. However, in practice, sub-national or local governments are constrained by limited resources, weak institutional capacity, inadequate mechanisms, and limited availability of information [3,7,26,29]. These circumstances, as well as the complication of land ownership and attractiveness, constitute a narrow space for the policymakers to encourage rooftop PV development at the local level without significant support from the government. This means that there is an obvious need for selecting target groups that is unified from national to local levels.
Based on the market characteristics, this paper suggests determining target groups based on k-means clustering of the development dynamics of 63 provinces in Vietnam with the criteria, i.e., their techno-economic potentials including the available roof (km 2 ), the production potential (TWh/a), LCOE (Euro cent/kWh), the level of urbanization (%) [47], and GDP (Euro/a) [47] (See Table A2 for detailed information). The results are shown in Table 5, Figure 12, and Figure A1. Because there are more than two dimensions (variables), Table 5 shows the principal component analysis and plots the data points according to the first two principal components (Dim1 and Dim2) that explain the majority of the variance and shows an illustration of the clusters. Figure A1 shows the clusters corresponding to their geographic position in the country. Ranked in order of priority for rooftop PV policies, Hanoi and Ho Chi Minh are classified as Priority 1, which requires the most policy focus due to their most favorable condition for rooftop PV development, followed by Ba Ria Vung Tau, Binh Duong, Can Tho, and Da Nang. Detailed clustering and priority ranking results for each region can be found in Table A4. The results do not only recommend policy priority but also support national policymakers to redirect and balance the PV development between different regions. For example, the government Ranked in order of priority for rooftop PV policies, Hanoi and Ho Chi Minh are classified as Priority 1, which requires the most policy focus due to their most favorable condition for rooftop PV development, followed by Ba Ria Vung Tau, Binh Duong, Can Tho, and Da Nang. Detailed clustering and priority ranking results for each region can be found in Table A4.
The results do not only recommend policy priority but also support national policymakers to redirect and balance the PV development between different regions. For example, the government can provide a favorable policy to the higher LCOE but potentially fast-uptake regions, e.g., cities and provinces of Priorities 1 and 2.

Assessing the Market Attractiveness and FIT
Rather than just analyzing the sensitivity of the output analysis to changes in the variable assumptions, as above, this section also looks at the probability distribution of the outcome in order to make decisions or take actions accordingly. Scenario analysis is designed to analyze the change of output regarding the different sets of inputs and then based on the various possible outcomes. In this paper, we use competitive prices to compare the potential benefit of installing rooftop PV between different scenarios. The competitive prices are the differences between the FIT and LCOE and between LCOE and the grid tariff (T G ) (Equation (15)). Table 6 summarizes the different possibilities of different metering models concerning LCOEs and grid tariffs. The business model indicates that residential customers can sell generated electricity on the grid to the system operator. The self-consumption model indicates that consumption of PV electricity takes place directly at the house/building-either immediately or delayed through the use of storage systems. Table 6. Competitive prices and corresponding metering models in the rooftop PV market ("x", beneficial; "-", not beneficial).

Competitive Price Metering Model
Delta 1 Delta 2 Business Self-Consumption The results support policymakers in making decisions and testing the robustness of an effect. By understanding the relationships between input and output variables, policymakers can encourage rooftop PV market development in certain areas by keeping the risk of investors at a comfortable level. With each specific decision, policymakers can measure the effect coverage, as well as the cost and benefit of their choice, and can compare and therefore decide which action should be taken.
Vietnam has an attractive solar market. However, it is undergoing a period of intense transformation. If the FIT 2.0 was approved, it would create a momentum pushing the entire market moving up and to the right of the coordinate axes. This means the expected profit per kWh would increase by 0.12 cents in all localities (orange line). By increasing 0.07 cents/kWh of the average FIT and specified regional price, FIT 2.0 would increase the possible activated market from 76% to 92% and create an additional exploitable 56 TWh potential, equivalent to a possible of 28 million Euro of profits. More than 30 provinces could show their aggressive competition of more than 1 Euro cent profit per kWh.
While the FIT 2.0 promises a bright future for rooftop PV, only some areas may qualify for the FIT 2.1 if the Prime Minister agrees with this approach. More than 60% of the market would be pushed into a difficult situation (red line), causing damage of 900 million Euro in the whole market (Table A5). The possibility of suppressing development or even market recession due to this FIT 2.1 needs to be borne in mind [57].

The Role of Grid Tariffs and PV Module Costs in Self-Sufficiency Developments
According to Vietnam EVN reports, the country's retail electricity price is projected to gradually increase by around 6.12% per year so that the electricity supply units, including EVNs, can offset costs. This paper analyzes the grid tariff changes reflecting on the self-consumption rooftop PV market by ranging it from 7.2 Euro cent/kWh in 2018 to a projected tariff of 8.7 Euro cent/kWh in 2025 0.12 cents in all localities (orange line). By increasing 0.07 cents/kWh of the average FIT and specified regional price, FIT 2.0 would increase the possible activated market from 76% to 92% and create an additional exploitable 56 TWh potential, equivalent to a possible of 28 million Euro of profits. More than 30 provinces could show their aggressive competition of more than 1 Euro cent profit per kWh.
While the FIT 2.0 promises a bright future for rooftop PV, only some areas may qualify for the FIT 2.1 if the Prime Minister agrees with this approach. More than 60% of the market would be pushed into a difficult situation (red line), causing damage of 900 million Euro in the whole market (Table A5). The possibility of suppressing development or even market recession due to this FIT 2.1 needs to be borne in mind [57].

The Role of Grid Tariffs and PV Module Costs in Self-Sufficiency Developments
According to Vietnam EVN reports, the country's retail electricity price is projected to gradually increase by around 6.12% per year so that the electricity supply units, including EVNs, can offset costs. This paper analyzes the grid tariff changes reflecting on the self-consumption rooftop PV market by ranging it from 7.2 Euro cent/kWh in 2018 to a projected tariff of 8.7 Euro cent/kWh in 2025 ( Figure 14). The left chart displays the current market with the grid tariff of 7.2 Euro cent/kWh with different discount rates of 4% and 12%. The right one presents the market when the grid tariff increases to 8.7 Euro cent/kWh. Increasing the grid tariff undeniably creates a favorable condition for the self-consumption business by increasing the expected benefit to −0.5 and 1.2 Euro cent/kWh at a 12% and 4% discount rate, respectively. Detail results for each region can be found in Table A6. As the grid tariff increases, the chance for self-consumption will also increase. However, it should be noted that, even though the policy seems to be effective in encouraging rooftop PV development, it faces many mixed reactions from the public. While researchers and banks complain that electricity price policy in Vietnam is inappropriate for investment because of its relative low grid tariff compared with the grid tariff in neighbor countries [58,59], the increasing grid tariff has led to an increase in the price of finished products, making it particularly tricky for domestic manufacturing enterprises and public consumers. In this paper, we observe the effect of the reduction in specific investment for rooftop PV in Vietnam by adjusting the capital expenditure (CAPEX) of the small rooftop PV systems (1-15 kWp), which will also affect O&M costs (OPEX). The current total costs (CAPEX + OPEX) range from 635 to 1038 Euros/kWp in Vietnam. At a discount rate of 8% and under the FIT 1.0 or FIT 2.0, the business project can expect an average benefit ranging from −1.7 to 3.3 Euro cent/kWh and the selfconsumption project can expect from −3.5 to 1.5 Euro cent return per kWh (Figure 15). If the total cost is lower than 650 Euros/kWp, it is feasible to invest in either of the metering models. However, the projects, whose total cost is higher than 836 Euros/kWp, are not recommended in Vietnam.
Rooftop PV technology is becoming more efficient and cost-competitive compared to other traditional energies. A dramatic decrease in the average wholesale price for crystalline modules is In this paper, we observe the effect of the reduction in specific investment for rooftop PV in Vietnam by adjusting the capital expenditure (CAPEX) of the small rooftop PV systems (1-15 kWp), which will also affect O&M costs (OPEX). The current total costs (CAPEX + OPEX) range from 635 to 1038 Euros/kWp in Vietnam. At a discount rate of 8% and under the FIT 1.0 or FIT 2.0, the business project can expect an average benefit ranging from −1.7 to 3.3 Euro cent/kWh and the self-consumption project can expect from −3.5 to 1.5 Euro cent return per kWh (Figure 15). If the total cost is lower than 650 Euros/kWp, it is feasible to invest in either of the metering models. However, the projects, whose total cost is higher than 836 Euros/kWp, are not recommended in Vietnam.
is lower than 650 Euros/kWp, it is feasible to invest in either of the metering models. However, the projects, whose total cost is higher than 836 Euros/kWp, are not recommended in Vietnam.
Rooftop PV technology is becoming more efficient and cost-competitive compared to other traditional energies. A dramatic decrease in the average wholesale price for crystalline modules is reported in China with a reduction from about 530 to 400 Euros/kWp in less than two years from 2016 to the first quarter of 2018 [60]. Moreover, while in 2005 solar modules constituted almost 75% of the system costs, this share amounts to around 50% today, even for rooftop PV systems [60]. In light of the reduced prices for crystalline modules in the global market [60,61], if the total cost for a rooftop PV system is lower than 650 Euros/kWp, it will provide a significant impetus to self-sufficient developments for the rooftop PV market in Vietnam.  Rooftop PV technology is becoming more efficient and cost-competitive compared to other traditional energies. A dramatic decrease in the average wholesale price for crystalline modules is reported in China with a reduction from about 530 to 400 Euros/kWp in less than two years from 2016 to the first quarter of 2018 [60]. Moreover, while in 2005 solar modules constituted almost 75% of the system costs, this share amounts to around 50% today, even for rooftop PV systems [60]. In light of the reduced prices for crystalline modules in the global market [60,61], if the total cost for a rooftop PV system is lower than 650 Euros/kWp, it will provide a significant impetus to self-sufficient developments for the rooftop PV market in Vietnam.

Critical Discussion
A cost-effective and transferable methodology for rooftop PV potential assessment in developing countries has been developed. The accuracy of the proposed method has been improved. However, challenges related to the data input still exist. The paper uses digital maps, which are prepared by using satellite imagery, combined with a manual spatial data collection method to conduct geospatial analysis. For example, the final map of Vietnam requires matching geographic and population distributions and building densities within 11,050 clusters of the topographic map in the case of Vietnam. The uncertainty issue arises when different maps often exhibit different accuracy levels depending on their application and spatial extent. The final map incorporates the mistakes of all component maps. The geo-referencing errors, the tilt angle of the sensor, the spatial resolution of the digital data, etc. affect the accuracy of the final map. The performance of the raster map of Vietnam in monthly means compared to measurements at 11 ground stations results in a mean bias error of −0.05 kWh m −2 day −1 (which represents −1.2% in relative mean bias error) and a root mean square error of 0.32 kWh m −2 day −1 (8.3% in relative root mean square error) [49]. The data collecting methods and the limitations/errors associated with those methods disturb the spatial accuracy. For example, different classification and interpolation methods were used in preparing the population distribution map for data presented with different errors of around 2-5% in the final map [47]. This paper conducts a sensitivity analysis to assess the effect of these uncertainties on the final results.
This paper considers the temperature and solar radiation, which have significant effects on the performance of photovoltaic (PV) systems; PV cell temperature, which is related to the ambient temperature; and the solar radiation incident on the PV surface, which depends on the slope and Energies 2020, 13, 2501 23 of 46 azimuth of the PV panels. Another factor that affects the solar radiation incident on the PV surface and hence influences its performance is the ground reflectance (albedo). This paper does not consider this factor due to the renowned complicated vertical distribution of cloud and aerosol layers of Vietnam [62]. Moreover, based on the literature, the albedo effect seems to be relatively small in Vietnam. Vietnam has a diffuse broadband solar irradiance (310-2100 nm) with the normalized solar wavelength of 532 mm [62]. The corresponding albedo spectrum and the effective albedo value, thus, range from around 5% to 15% and 0.1 to 0.35, depending on the rooftop materials such as brick, red shingle, ceramic tiles, and green paint, respectively [63]. The effective albedo can lead to errors of less than 1.6% and 6.7% for the module tilt angles from horizontal of 40 • and 90 • [64]. Further studies, especially field studies on a small scale, can use this effective albedo value to optimize PV system design and take advantage of the reflected albedo resource.
This study improves the accuracy of technical potential output by calculating the corrected temperature based on the NOCT model. As mentioned in Section 3, NOCT is obtained under predetermined environmental conditions, while STC is the test condition used to rate the performance of a PV panel. STC is the idealized condition, which can overpredict PV performance [65,66]. Since the NOCT is based on realistic data, it is simple and easy to apply. Manufacturers usually include NOCT in the module characteristic data. However, NOCT can overpredict the PV cell temperature by around 10% [67], and can also vary around 2-5 • C, depending on the module materials [68]. Thus, it can underpredict the PV performance [36]. Another option for calculating the expected PV system yield for specific provinces or even buildings is the free online tool PVGIS, provided by the European Commission [69]. The tool uses the real measured values of solar radiation at specific location and provides an overview information about PV system performance. However, while PVGIS is certainly more accurate for a specific building, it would not be feasible to perform a national analysis with this tool.
When calculating the cost for decentralized PV generation technologies, this paper uses LCOE, which is the most widely used metric for a cost comparison between different generation technologies. However, since the renewable energy share in total electricity supply increases, it poses a difficulty for LCOE to make comparisons between different categories of supply, e.g., residual load, load following, and variable load. Another concern regarding LCOE is that it is becoming more difficult to justify applying the same weighted average cost of capital to technologies with very different climate policy risks. To deal with these drawbacks, the new measure, called the Value-Adjusted Levelized Cost of Electricity (VALCOE), is now recommended by the IEA [70], which includes three adjustments to LCOE: energy, value, and flexibility. Although the VALCOE may be a better measure of cost for grid managers, the IEA needs to provide the data for the added factors from their World Energy Model (WEM) regional hourly dispatch models. Once these data are available and the flexibility value, outside of the WEM model, in the VALCOE equation, are well defined, future studies can replace LCOE by this new measurement.

Conclusions and Policy Implications
Rooftop PV is driving the decentralization of electricity production, which is a cornerstone of the sustainable development concept, whereby households become more independent of the electricity grid due to a thorough integration of renewable energy sources. However, it is still often seen as a novelty, and legally, local authorities have difficulties enforcing this specific technology in their detailed development plans partly because of the lack of expertise and capital. Therefore, it is relevant to address decentralized PV potentials as early as possible when the state and local governments start planning for the type of energy source. Since different regions have different climatic, geographic, and demographic profiles, this paper provides an affordable, robust and high-resolution potential assessment for rooftop PV and suggests target groups alongside policy solutions with clear national and local goals, based on the findings. The proposed method considers construction and design constraints, obstacles on the rooftop, and the proper orientation on the roof, including shading effects and the slum areas (informal settlements). The paper considers Vietnam as an example for other emerging countries to apply the method. The results can provide an overview plan as well as an integrated strategy for developing rooftop PV between different administrative levels.
The physical potential in the raster maps, which encompasses the maximum amount of solar energy received in a particular area, is broken down into the geographic potential of in total 1407 km 2 available rooftop by gradually excluding the zones reserved for other uses and restricting the locations where solar energy can be gathered. The technical potential is calculated with the results of roof-mounted solar cells in the whole country being about 228 and 216 GWp or about 278 and 263 TWh/a with monocrystalline and polycrystalline solar cells, respectively. The LCOE is also calculated, ranging from about 7.6 to 9.2 Euro cents/kWh at a discount rate of 8% in 63 provinces in Vietnam. However, since the techno-economic potential of PV is associated with a certain degree of uncertainty, we measure the error regarding the uncertainty of data assumptions and inputs and the inherently unknowable reaction of the market. Overcoming the unfavorable data conditions, this paper provides a practical and useful tool for policy-making processes. We suggest dividing the market by target group and priority level based on their PV potentials, demographics, and economic status. Hanoi and Ho Chi Minh City should be classified as the highest priority cities for policy focus based on their high potential and robust internal economic and demographic conditions. This paper also provides systematic guidance to help policymakers design political strategies to support market development. For example, only projects with total costs (CAPEX + OPEX) and discount rates lower than 850 Euros/kWp and 8% respectively, can be beneficial in the current market. With total costs of around 650-850 Euros/kWp, the only attractive investment option is the business model, which is selling generated PV electricity directly on the state grid, while total costs lower than 650 Euros/kWp make a choice between the business and the self-consumption models possible for their rooftop PV projects. For self-consumption, there is no clear strategy, but the situation improves due to increasing grid tariffs. Consumers must, however, wait until around 2025 when their grid tariff should rise to at least about 8.7 cents/kWh and self-consumption will become attractive, but this date could be sooner if costs for solar technology continue to decrease. By simulating the rooftop PV market in different projected FITs, we recommend that the government consider their tariff strategies carefully. In the case of changing the average FIT from 8.83 to 9 Euro cents/kWh and establishing regional FITs, they can activate an additional 16% of the total market, equivalent to around 56 TWh and approximately 28 million Euro of profit. However, if they were to reduce the FIT to 7.79 Euro cents/kWh, it would deactivate about 60% of the rooftop PV market and would reduce total market benefits of approximately 900 million Euro (Table A5).
This paper presents a methodology to carry out resource assessments for rooftop PV in developing countries, as applied to Vietnam. The method is a top-down method that can be employed where detailed, accurate data on buildings, their outlines, and associated infrastructure are lacking. It represents an improvement beyond state of the art and has an absolute error of about 5% compared to detailed bottom-up city models, which is very low for a method at this spatial scale and resolution. Uncertainties relating to the input data are significant but are considered in the sensitivity analysis, which should be borne in mind when interpreting the results. Especially the accuracy of the assumptions relating to the building location, size, orientation, and shading impacts are country-specific and require further research to be employed elsewhere. One promising avenue in this context is the combination of open mapping data with satellite images to automatically identify suitable geometries for rooftop PV [31]. By further validating the top-down results from this study with bottom-up results from 3D models (e.g., Effigis Geo-Solutions 2018), the method could be improved and applied to other contexts. Finally, this and related studies have tended to focus on the supply side for rooftop PV, but there is a need for future research to analyze the demand side, especially public opinion and willingness to pay for PV systems.