Modeling photovoltaic diffusion: an analysis of geospatial datasets

Abstract

This study combines address-level residential photovoltaic (PV) adoption trends in California with several types of geospatial information-population demographics, housing characteristics, foreclosure rates, solar irradiance, vehicle ownership preferences, and others-to identify which subsets of geospatial information are the best predictors of historical PV adoption. Number of rooms, heating source and house age were key variables that had not been previously explored in the literature, but are consistent with the expected profile of a PV adopter. The strong relationship provided by foreclosure indicators and mortgage status have less of an intuitive connection to PV adoption, but may be highly correlated with characteristics inherent in PV adopters. Next, we explore how these predictive factors and model performance varies between different Investor Owned Utility (IOU) regions in California, and at different spatial scales. Results suggest that models trained with small subsets of geospatial information (five to eight variables) may provide similar explanatory power as models using hundreds of geospatial variables. Further, the predictive performance of models generally decreases at higher resolution, i.e., below ZIP code level since several geospatial variables with coarse native resolution become less useful for representing high resolution variations in PV adoption trends. However, for California we find that model performance improves if parameters are trained at the regional IOU level rather than the state-wide level. We also find that models trained within one IOU region are generally representative for other IOU regions in CA, suggesting that a model trained with data from one state may be applicable in another state.

Regional distribution of photovoltaic deployment in the UK and its determinants: A spatial econometric approach

“…They confirm the importance of spatial neighbouring effects as well as built environment and policy variables, supporting the findings ofBollinger and Gillingham (2012), ofMüller and Rode (2013) and ofRode and Weber (2012). Another study byDavidson et al (2014) highlights the importance of home age, heating source, number of rooms, mortgage status and household education as key variables affecting PV diffusion in California, USA.…”

“…While both Davidson et al (2014) and Jager (2006) report positive influence of university and postgraduate education on PV uptake, our analysis reveals the effect of vocational and technical qualifications which are below university degree (captured by QL2). The findings presented here indicate that there is a statistically significant negative impact of population density on PV deployment which is in line with the existing literature, implying that residents located in less densely populated areas, characterised by a higher share of single and double family homes, are more likely to install a PV system (Müller and Rode, 2013).…”

“…Müller and Rode (2013) take into account the effect of density and income together and argue that consumers located in less densely populated areas, characterised by a higher share of single and double family homes, are more likely to be early adopters. While Davidson et al (2014) report statistically significant influence of higher education, Jager (2006) and Keirstead (2007) interview some early adopters and find that they are better educated and are from middle-age groups. Solar irradiation is another factor directly affecting PV electricity generation (Šúri et al, 2007) and thus the economics of PV installation.…”

Planning for a Distributed Disruption: Innovative Practices for Incorporating Distributed Solar into Utility Planning

“…Davidson et al (2014) use several types of geospatial informationincluding population demographics and housing characteristics-in a stepwise regression model to identify which subsets of geospatial information best predict historical PV adoption at the zip code level. Discrete choice models, as used in NWPCC's customer-adoption model described earlier, are also popular for modeling technology diffusion (Higgins et al 2014, Jun and Kim 2011, Lobel and Perakis 2011, Kim et al 2005 owing to their ability to model competition between several options; this class of models also has a well-defined methodology for soliciting customer preferences.…”

“…It then selects customers with the highest probability to adopt until the C -39 amount of DPV adopted matches the aggregate DPV adoption forecast for the utility. PG&E's approach is guided in part by similar research from NREL (Davidson et al 2014).…”