Modelling future uptake of solar photo-voltaics and water heaters under different government incentives

https://doi.org/10.1016/j.techfore.2013.07.006Get rights and content

Highlights

  • Diffusion model to forecast uptake of solar PVs and water heaters

  • Applied to New South Wales housing stock (population 7.1 million) with geographical resolution of 250 households

  • Evaluation of financial incentives funded by government

Abstract

To accelerate the adoption of technologies to reduce energy consumption and greenhouse gas (GHG) emissions in the residential sector, government policy makers offer a range of fiscal instruments and incentives. Despite the high costs of these schemes, methods are lagging to systematically evaluate their likely effectiveness amongst a geographical landscape of heterogeneous consumers. To address this need, a model was developed for spatial adoption of technologies such as water heaters and solar photo-voltaic panels (PVs), across housing stock, given government policy incentives. By combining features of choice modelling, Multi-Criteria Analysis (MCA) and diffusion models, it provides a capability to analyse future adoption patterns of the competing technology options under a range of features for purchase timing and choice. The model was implemented across 2.7 million residential dwellings in the State of New South Wales (NSW) of Australia to estimate future stock of PV and water heater options at geographical units of 250 households. Validation against actual numbers of PV installations at each postcode showed the model was effective at identifying high versus low adoption locations. Application to a wide range of policy scenarios, ranging from feed-in tariffs to upfront rebates, showed substantial differences in their effectiveness to accelerate uptake, and the government expenditure required.

Introduction

Governments employ various policy instruments, including incentives, rebates, tax breaks and other fiscal instruments to accelerate adoption of technologies towards meeting national environmental performance targets. Incentives are usually part of the suite of government initiatives aimed at achieving energy consumption targets through improved supply or demand management. The Australian Government has committed to the long-term goal of reducing Australia's GHG emissions by 60% of 2000 levels by 2050, and has set a medium-term target of a reduction in greenhouse gas emissions by 2020 of between 5% and 15% less than the level they were in 2000 [1]. One of the biggest challenges for government and industry to reduce resource use and GHG emissions in the residential sector is the selection of the most cost-effective intervention schemes. Any regulatory measure, having legal authority and compliance requirements, can be expected to have widespread impact. However the effectiveness of any voluntary scheme, relying on consumer adoption, is very difficult to evaluate. Despite large budgets for government policy schemes, systematic evaluation of their effectiveness and the likely impact on long term diffusion across a landscape of heterogeneous households is rarely undertaken. New methods and tools are needed for this purpose.

Choice experiments have been an effective approach to identifying key features important to the adoption of technologies such as solar photo-voltaic panels (PVs) and water heaters, along with their sensitivity to policy incentives. This involves modelling consumer preferences for these technologies to derive the effect of key attributes such as installation cost, energy efficiency and out of pocket expenses after rebates. Islam and Meade [2] conducted a discrete choice experiment in Ontario, Canada to estimate key attributes of consumer adoption intention of PVs over time. The authors use these attributes in a discrete time survival mixture analysis to estimate future probabilities of adoption for different consumers. Yamaguchi et al. [3] used the consumer preferences to estimate the marketing effort component of a Bass diffusion model. They forecasted the uptake of PV and solar hot water in Japan under various financial subsidies. In Australia, discrete choice data gathered from stated preference experiments have been used to understand the drivers and interaction effects of uptake of water heaters. Bartels et al. [4], [5] surveyed 129 plumbers and 312 consumers in Sydney who purchased a water heater. The stated choice experiment allowed the respondent to indicate important features to the purchase of their gas or electric water heaters. When used in a regression analysis for consumer ratings, the most significant variables were system capacity, warranty, price, running cost, and plumbing certificates. The analysis also showed consumers put a A$135 value on saving A$5 per month. A recent study by Wasi and Carson [6] considered various multinomial logit models for choice probabilities amongst the water heater options. The embedded utility function included cost, rebate and running costs, with the parameters estimated using maximum likelihood. By using the model to analyse data from discrete choice experiments, they tested a scenario of New South Wales rebates (A$300 for gas and 50% reduction in upfront cost of solar and heat pump) that increased the probability of households (with no gas connection) choosing solar or heat pump water heaters by 38%. Whilst these methods help understand drivers of consumer behaviour in the presence of incentives, their application to forecast future uptake has been limited. Despite this, the literature is rich with diffusion and choice models that can be adapted to forecast the uptake of water heater and PV options, and we will review the suitability of such methods.

Discrete choice modelling has been extensively used for technology diffusion [7], [8], due to the ability to consider the sensitivity of multiple product options available (e.g. different types of water heaters) to consumers. There have been applications for the diffusion of electric vehicle options [9], [10] and energy efficient appliances [11], [12]. To optimise the timing and size of a rebate to maximise long uptake of PVs, Lobel and Perakis [13] and Benthem et al. [14] combined choice models with mathematical programming. Lund [15] used a technology diffusion model to understand the total subsidies required for different electricity sources to reach market share targets by 2050. Linear utility functions have been extensively incorporated into a choice model to represent the different values of the product features (e.g. price, rebate, annual costs) to the consumer [16], [17], [18], [19]. A major challenge of utility functions in choice models is setting and calibrating the parameters, so that actual choice probabilities are reflected. In the absence of past data on uptake of PVs, large scale consumer surveys are conducted. Choice or diffusion models with high geographical sensitivity have had very limited attention in the literature to date, let alone in the case of water heaters and PVs. Bhat and Guo [20] and Sener et al. [21] develop spatially explicit choice models that accommodate socio-demographic differences and zone dependencies. In the case of forecasting uptake of different electric vehicle options in Victoria, Higgins et al. [10] implemented a model across over 9000 geographical units of residential housing stock (about 250 households per unit), which were further partitioned by demographics and building features. Each location is represented by an Australian Bureau of Statistics (ABS) Census Collection District (CCD). This current paper considers similar demographic and geographical granularity. Analysis at such high geographical granularity can also help electricity distributors plan future capacity across the network.

In this paper a diffusion-choice model is implemented and validated to forecast uptake of PV and water heater technologies across NSW considering different government policy schemes. To construct a suitable diffusion-choice model, the version developed by Higgins et al. [10] for electric vehicles, was redeveloped to include several novel features. When considering a long forecasting horizon, there is the need to incorporate product replacement along with first time purchasers into a choice or diffusion model [17], [22], [7]. Initially, consumers are mostly first time purchasers of PVs, though they can replace or upgrade to a larger system. For water heaters, consumers are faced with the decision of when to replace the existing system as well as what alternative water heater option to replace it with. A multi-product choice model by Jun and Kim [7] considers replacement timing to be dependent on a survivability (or failure) probability function. It assumes consumers will replace a PV or water heater upon failure of the existing one. A novelty in our paper is the probability of when a product is replaced includes a utility of other attributes associated with replacement. When implemented, it accommodates the socio-demographic differences of consumers who are early versus late replacers of their existing water heaters, as well as the effects of incentives to reduce the time to replacement. Incentives are for replacing electric water heaters with more energy efficient options. An additional variable is included to accommodate the reduced risk or cost of replacing the existing technology with the same type. For example, a consumer with a natural gas water heater will have less perceived risk and cost of installation by replacing it with another gas heater. With PVs, a consumer can upgrade his/her system with either adding more panels or upgrading to a larger inverter at a lower cost than a complete replacement.

A further novelty is the improved ability to accommodate the evolution of drivers of adoption over the forecasting horizon, through the incorporation of Multi-Criteria Analysis (MCA) features. Drivers such as upfront cost, running cost and demographic suitability need to be modelled in relation to their ideal points to ensure an accurate contribution to the utility in the choice model. The approach in this paper uses the ‘Displaced Ideal’ feature of compromise programming [23] that obtains a normalised weighted distance from ideal points. It is a preferable approach to the standard normalisation function used in many MCA methods, as it is independent of changing best and worst values of scores amongst water heating and PV options over time. For example, the effect of water heater cost on a consumer decision to purchase needs to be measured against an ideal purchase price across the forecasting horizon rather than the best price amongst the competing options at a specific time only.

Our paper is organised as follows: the next section outlines the model in a generic form, followed by its adaptation to the PV and water heater case study in Section 3. The model is validated against actual numbers of PV installations by postcode for all of NSW in Section 4, demonstrating the model's geographical sensitivity to adoption. In Section 5, several policy incentives are tested and compared using the model to identify an option that could accelerate adoption at least cost to government.

Section snippets

Model for forecasting uptake

In the generic model described in this section, adopters are households, where each residential dwelling contains a single household. In the case of water heaters, a household will have a single installation. For PVs, a household can have up to one installation, though the size can be increased over time. The main output from the model is:Adot=stockoftechnologyoptionoforcategorydattimetwhere:

    d

    household category within the typology. It is broken down into housing type by location by household

Adaptation to NSW residential housing

NSW is Australia's most populated state with 7.2 million people (2010) and a residential housing stock of 2.73 million (2006). Data was gathered from multiple sources to set up the model represented by Eqs. (1), (2), (3), (4), (5), (6), (7), (8). The first step was to construct a typology to represent all possible categories D. To do this, data representing the 2006 census from the ABS was available at a granular spatial scale, called census collection district (CCD). Each CCD represents about

Validation

A spatial validation of the model was conducted by comparing forecasted uptake of PVs versus actual number of installations by postcode across NSW. This validation was more difficult for water heaters since data on the number of installations was not available by postcode in earlier years (e.g. 2007). Data on PV installations were obtained from the NSW Office of Environment and Heritage for 2010. The model was applied starting from 2006 to 2018, with a calibration (Table 3) only for the

Policy analysis

Australian federal and state government policy, in the form of financial rebates has been a major strategy to accelerate the uptake of PVs and phasing out of electric water heaters in NSW and across Australia.

Electric water heating accounted for 11% of total energy use in Australia in 2007 [1], and 55% of dwellings still use electric systems [29]. Government subsidies are considered an effective means to accelerating adoption of more efficient water heating. Menanteau [30] outlines different

Conclusions and further research

An integrated diffusion-choice model was developed and applied to all of NSW to forecast the adoption of PVs and water heaters under various policy incentives. The model was implemented and validated at high geographical granularity using spatial units of approximately 250 households. For the greater Sydney region, the number of PV installations by postcode had a correlation coefficient of 0.8 between the model and actual. For the rest of NSW, the correlation coefficient was 0.58. These

Acknowledgements

This project was co-funded by the NSW Government — Office of Environment and Heritage. The authors thank Dr Charles Xu from this department for contributions in defining the case studies and providing access to data sets. The authors also thank the journal reviewers for their suggestions to improve the paper.

Dr Andrew Higgins is a Principal Research Scientist at CSIRO, and specialises in mathematical forecasting of technology at different locations and demographics, and under various incentives. He has previously developed methods to optimise supply chain problems found in freight, agriculture and natural resource management. He is currently the vice-president of the Australian Society for Operations Research. Andrew received his PhD from Queensland University of Technology in 1996.

References (37)

  • D.K. Sidiras et al.

    The effect of payback time on solar hot water systems diffusion: the case of Greece

    Energy Conserv. Manag.

    (2005)
  • K.C. Chang et al.

    Subsidy programs on diffusion of solar water heaters: Taiwan's experience

    Energy Policy

    (2011)
  • Z. Ren et al.

    A local-community-level, physically-based model of end-use energy consumption by Australian housing stock

    Energy Policy

    (2012)
  • Australian Government

    Carbon Pollution Reduction Scheme: Australia's Low Pollution Future

    (2008)
  • R. Bartels et al.

    Consumers and experts: an econometric analysis of the demand for water heaters

    Empir. Econ.

    (2006)
  • N. Wasi et al.

    The influence of rebate programs on the demand for water heaters: the case for New South Wales

  • Z. Lin et al.

    Who will more likely buy PHEV: a detailed market segmentation analysis

  • M. Jaccard et al.

    Estimating home energy decision parameters for a hybrid energy-economy policy model

    Environ. Model. Assess.

    (2006)
  • Cited by (0)

    Dr Andrew Higgins is a Principal Research Scientist at CSIRO, and specialises in mathematical forecasting of technology at different locations and demographics, and under various incentives. He has previously developed methods to optimise supply chain problems found in freight, agriculture and natural resource management. He is currently the vice-president of the Australian Society for Operations Research. Andrew received his PhD from Queensland University of Technology in 1996.

    Dr Greg Foliente is a Senior Science Leader at CSIRO, leading R&D initiatives and projects related to systems-based approaches to climate mitigation and adaptation, sustainability of the built environment, and transitions to urban sustainability. He received two Master's degrees and PhD at Virginia Polytechnic Institute and State University in the USA. He was the recipient of the ASCE James Croes Medal, the CSIRO Newton Turner Career Award, the SWST George Marra Award of Excellence in the USA, the STA Fellowship in Japan, and the DAAD Visiting Professorship and Wilhelm Klauditz Fellowship at Fraunhofer in Germany, amongst others. He has a diverse scientific publications record (including three books) and serves on the Editorial Board of four international journals.

    Cheryl McNamara currently works as a software developer for CSIRO Ecosystem Sciences. She also has expertise in database management, Geographical Information Systems and statistical analysis. She has preformed the role of project manager on numerous social science surveys and has provided support to the research activities and consultancy work of numerous and varied projects over many years.

    1

    Tel.: + 61 3 9252 6118.

    2

    Tel.: + 61 3 9252 6038.

    View full text