Experiences in Developing a Decision Support Tool for Agricultural Decision-Makers—Australian CliMate

Abstract

Australian agriculture managers deal with climates that are characterised by high variability and unpredictability. A simple framework for decision-making is used to structure weather-related inquiries using recent and long-term climate data to better inform decisions based on current conditions and future expectations. This paper describes the rationale, design philosophy, and development journey of Australian CliMate (CliMate), a contemporary climate analysis tool built to consolidate and modernise the functionality of earlier computer-based decision support tools (DSTs). CliMate aimed to be simple, transparent, and user-driven, supporting tactical and strategic agricultural decisions. Ten core analyses were included from previous DSTs. With over 20,000 registered users and widespread adoption among farmers, consultants, and other professionals over a decade, CliMate demonstrates the enduring demand for accessible, mobile climate analysis tools. We reflect on lessons learned in the development process, advocating for minimalism, iteration with users, and integration of transparent data sources. This experience underscores the necessity for long-term support and evaluation to sustain the value of agricultural DSTs.

1. Introduction

Given the variable and unpredictable nature of seasons (weather) across Australia [1] and the impact of rainfall and other climate variables on management in agriculture [2], it is not surprising that discussion amongst farmers and their advisors is often around “the weather” and seasonal outlooks. These discussions generally begin with “How much rain did you get?” which often moves on to “What’s likely in the coming weeks or season?”. These discussions, whether social or part of strategic and tactical decision-making, are often biased by recent weather and shaped by varying expectations from forecasts.

This natural preoccupation with “the weather” led to the development of several computer-based decision support tools (DSTs), including Rainman [3,4], Howwet? [5], and CropMate [6]. With time, most of these DSTs became redundant due to lack of maintenance, emergence of new technologies, and loss of support for established technologies. Computing power moved from the desktop to being readily available on mobile devices, which offered promise [7]. While these older DSTs were widely applied and valued [3,6,8,9], they were not maintained and degraded with time, with most not being available beyond their active funding period (1–5 years).

Traditional responses to managing variable water supply for crops include fallowing between dryland crops to store soil water and irrigation where available, adjusting planting windows to avoid frost and heat stress, and modifying inputs to suit seasonal expectations [2]. Short-term (<14 day) and seasonal forecasts (3–6 months) are provided by public [10] and private providers with variable credibility in the rural community [11].

Anderson et al. [12] maintained that decision-making in agriculture would be improved if we asked questions such as “What chances, what choices, and what consequences?”. This paper describes the rationale behind the redevelopment of a set of climate-related decision support analyses designed to support Anderson et al.’s [12] proposition. The resulting product, Australian CliMate [13], hereafter referred to as CliMate, incorporates analyses from ten products developed over previous decades. We describe the development pathway and design philosophy and reflect on its use and lessons learnt during its development.

2. Design Philosophy

The initiative for CliMate’s development came from the recognition that previous DSTs developed for climate analyses, while valued, had become redundant due to developer “champions” no longer being available, ceased funding, and developments in technology (from desktop to mobile devices). CliMate was developed under the Managing Climate Variability Program (MCVP), a consortium of rural research and development corporations managed by the Australian Grains Research and Development Corporation. A draft concept was proposed in 2009 for a “generic climate analyser” to incorporate the best features of previous climate-related DSTs into one package by 2013. The initial design planned for a desktop application only, but observations of train commuters engrossed in their mobile phones highlighted the potential for mobile platforms to support agricultural decision-makers [7]. A second phase (2015–2018) increased the availability of CliMate to include Android and www versions with minor maintenance funding thereafter. The vision was that farmers and their advisors could consult a range of DSTs in their fields where they make decisions along with discussions with their advisors.

It was recognised at the planning stage that a simple inquiry-focused approach was more likely to be useful and educative (exploration of “what chances, what choices”) than a complex treatment of climate information, which was possible with available technology. The concept was that rather than provide answers, a DST should support iterative exploration of the decision space, allowing users to gain a better understanding of current conditions and future expectations based on local climate data.

A common belief in the “science” community, whose members are typically the developers of DSTs, is that adding detail or completeness to an analysis or a more realistic treatment of the real world (added complexity) would result in a “better” DST and better decisions. Ward [14] proposed that technology can follow several paths as it matures. Some developments add complexity to the belief that this will improve the reliability of the innovation, yet the outcome for usefulness or “goodness” can follow two pathways (Figure 1).

Our aim was to follow the simplicity slope of Figure 1 with reduced complexity and improved functionality, even though we experienced peer pressure to add functionality.

Our aim in developing CliMate was to adopt a minimalist approach, with emphasis on user involvement during design and development [15] while avoiding some of the traps of previous developments [16,17].

Given CliMate was designed to bring a range of existing analyses from several organisations into a common framework; the interface design was based on an existing DST, Howoften? [18], a simple product distributed on disc that had been favourably received and met a key design criterion of simplicity. Subsequent interface designs were based on trial and error of alternative presentations with over ~50 iterations. The large number of iterations was facilitated by separating the need for computing skills from interface design through an enhanced backend to the system’s database, where much of the interface and options were easily accessible.

2.1. Conceptual Framework

A conceptual framework proposed for weather-based decisions consists of two components: understanding the status or current conditions of a system based on recent weather and future expectations based on probabilistic analysis of long-term data (so-called “climatology”) (Figure 2). CliMate facilitates exploration of local data to support tactical and strategic decisions with two categories of analysis included:

• Assessing the current condition of a system (rainfall to date, soil water, heat sum, and drought) using near real-time data from a local weather station, the “knowable” element shown in Figure 2.
• Probability of future weather events based on “climatology” derived from a probabilistic analysis of historic weather data (rainfall, temperature, and radiation) and derivatives (heat sum and soil moisture).

This framework is like the ADOPT approach used to evaluate information and adoption of new technology [19]. Kuehne et al.’s [19] framework also includes a third component, which considers a model’s sensitivity and associated discussive potential. CliMate‘s suitability for iterative exploration for each analysis implicitly meets this third element. In applying the framework (Figure 2), the decision point travels with time, with recent weather history often an important part of any decision. For example, rainfall in the previous weeks or months will be highly influential on soil water status (measurable or knowable). This balance between system status and future expectations can be particularly important when 30–80% of a crop’s water supply is derived from soil water at planting (starting condition) in the northern Australian grain regions [20].

CliMate’s interface was designed to encourage a “discussive” or explorative approach to current conditions and future expectations given its ability to iterate through adjustments to queries of the climate record. CliMate’s reliance on “climatology” rather than forecasts was a purposeful decision in the spirit of maintaining simplicity and transparency and based on user feedback. An application of a seasonal forecast signal was initially included (How likely?), but during construction, skills in forecasts were in a state of flux [21,22], and specialised forecast information was available elsewhere [23,24].

2.2. Specifications

Considering previous DSTs and feedback from prototype users, ten analyses were initially included (Table 1). Key functions of the proposed DST were to achieve the following:

• Develop a common interface between the previously valued but lapsed analyses.
• Support transparent and open-ended queries to accommodate users’ rules and models across a wide range of agricultural production systems (grazing, cropping, horticulture, and apiculture).
• Involve stakeholders early in development and revise prototypes based on feedback.
• Aim for a minimalist interface (input and output), applying a principle of “if in doubt, leave it out”, and use easily recognised graphical presentations such as “fire risk” charts, histograms, and line graphs.
• Provide output as text and graphics to accommodate different learning styles [25];
• Enable mobile device apps to be used “offline” by farmers when in the “field”. This required accessing and storing climate data sourced when last connected to the internet.
• Access continuous (patched) local weather data from SILO [26,27], originally sourced from the Australian Government’s Bureau of Meteorology.
• Include a “backend” to support iterative tuning of interfaces based on user feedback, shortening development cycles.
• Embed the ability to monitor use of each analysis, including spatial and industry applications.

Table 1. Summary of the origin, function, considered variables, and interface for ten analyses within Australian CliMate.

Analysis Origin 1	Function and Application (Variables 2)	Interface
How’s the season? Rainman [3,4] Qld Govt.	Current season relative to long term. Adjust expectations and inputs. (1, 2, 4, 5)
How often? Howoften? [18] Qld Govt.	Probability of weather events. Risk assessment key operations, e.g., planting, rain, heat/frost stress, and grazing days. (1, 2, 4)
How Wet/Nitrate? Howwet? [5] Qld Govt.	Soil water and nitrate accumulation in fallows. Yield expectations, nitrogen inputs, and crop choice (1, 2, 6, 7)
Potential yield? WUE [28] PYCalc [29], DAWA	Expected crop yield. Adjust inputs and marketing. (1, 8, WUE)
Drought? SCOPIC [30,31] Qld Govt.	Drought status, rainfall deficit, and decile method. Stocking rate alert and financial relief (1, 9)
How hot/cold? CropMate [6] NSW Govt.	Coincident probability of min. and max. temperature. Risk assessment for new crops and new managers (1, 2)
How’s the past? Standard statistics. MCVP	Historic weather. Seasonal overview. Land purchase (1, 2, 3, 4)
What trend? Graphical analysis MCVP	Long-term trend graphics. Assess trends and variability (1, 2, 4, 10)
How likely? SCOPIC [30] BoM, Qld Govt.	Season forecast and skill (ENSO). Assess forecast skill to complement “climatology” assessment (11)
How’s El Nino Direct lookup BoM, Qld Govt., MCVP	SOI and ENSO status. Seasonal forecast (11)

Notes: ¹ Origin: Qld Govt—Queensland Government; NSW Govt—New South Wales Government; MCVP—Managing Climate Variability Program; DAWA—Department of Agriculture, Western Australia; BoM—Bureau of Meteorology, Australian Government. ² Variables: 1—rainfall; 2—temperature; 3—evaporation; 4—radiation. Derived variables: 5—heat sum; 6—soil water; 7—soil nitrate; 8—crop yield; 9—drought index; 10—incidences of weather events. Lookup variables: 11—SOI, ENSO, and IOD; WUE—water use efficiency.

Table 1. Summary of the origin, function, considered variables, and interface for ten analyses within Australian CliMate.

Analysis Origin 1	Function and Application (Variables 2)	Interface
How’s the season? Rainman [3,4] Qld Govt.	Current season relative to long term. Adjust expectations and inputs. (1, 2, 4, 5)
How often? Howoften? [18] Qld Govt.	Probability of weather events. Risk assessment key operations, e.g., planting, rain, heat/frost stress, and grazing days. (1, 2, 4)
How Wet/Nitrate? Howwet? [5] Qld Govt.	Soil water and nitrate accumulation in fallows. Yield expectations, nitrogen inputs, and crop choice (1, 2, 6, 7)
Potential yield? WUE [28] PYCalc [29], DAWA	Expected crop yield. Adjust inputs and marketing. (1, 8, WUE)
Drought? SCOPIC [30,31] Qld Govt.	Drought status, rainfall deficit, and decile method. Stocking rate alert and financial relief (1, 9)
How hot/cold? CropMate [6] NSW Govt.	Coincident probability of min. and max. temperature. Risk assessment for new crops and new managers (1, 2)
How’s the past? Standard statistics. MCVP	Historic weather. Seasonal overview. Land purchase (1, 2, 3, 4)
What trend? Graphical analysis MCVP	Long-term trend graphics. Assess trends and variability (1, 2, 4, 10)
How likely? SCOPIC [30] BoM, Qld Govt.	Season forecast and skill (ENSO). Assess forecast skill to complement “climatology” assessment (11)
How’s El Nino Direct lookup BoM, Qld Govt., MCVP	SOI and ENSO status. Seasonal forecast (11)

Notes: ¹ Origin: Qld Govt—Queensland Government; NSW Govt—New South Wales Government; MCVP—Managing Climate Variability Program; DAWA—Department of Agriculture, Western Australia; BoM—Bureau of Meteorology, Australian Government. ² Variables: 1—rainfall; 2—temperature; 3—evaporation; 4—radiation. Derived variables: 5—heat sum; 6—soil water; 7—soil nitrate; 8—crop yield; 9—drought index; 10—incidences of weather events. Lookup variables: 11—SOI, ENSO, and IOD; WUE—water use efficiency.

As an example (Figure 3), the Howoften? analysis demonstrates some interface features common across CliMate with inputs including variable selection (rainfall, consecutive maximum/minimum temperature, and solar radiation); greater than or less than; a target metric value; the number of days over which a test is applied; a location (selected from a list or map); the start and end date of a period being explored; and the decades of interest. Default values are provided as a starting point for new users, while input values from previous analyses are retained. Results are presented as a fire chart with the percentage of years the query was true, as text, and as a histogram of “hits and misses”. Users can iteratively change inputs and view changes immediately, supporting exploration of their local climate.

The initial target audience for CliMate was farmers, with advisors and consultants across all agricultural industries (cropping, grazing, horticulture, and apiculture) considered secondary. This audience later extended to education, banking, and insurance assessors. To accommodate this broad user base, analyses were designed to be generic yet customisable for diverse agricultural contexts. The app was packaged for iOS and Android native apps and any device that supported a web browser.

2.3. Software and Data Sources

CliMate consists of four separate applications: a website for analysis and administration, a web API for mobile device communication, and separate iOS and Android applications. When development began in 2012, a mature cross-platform development technology was not available to use a single set of source code across all platforms. Therefore, separate technologies were used for each platform, including Microsoft asp.Net for the web application and web API (ASP.NET MVC (Net Framework 4.8) and native technologies for the mobile applications (Objective-C: iOS 15.6 Framework and Java: Android SDK - now obsolete).

To simplify development and maintenance, a common structure and naming convention was applied across all technologies. Despite having three sets of code for each analysis, core syntax was almost identical across each coding language. Data are stored on an SQL Server database, and a custom-built synchronisation algorithm ensures that user data are synchronised across multiple devices via a web API. This allows offline operation of mobile applications with user data and settings updated once an internet connection is detected. The app is administered from a common web service that supports maintenance, adjustment of default values, generation of user analytics for the three platforms, and links to support material.

Long-term climate data are curated by the Australian Government’s Bureau of Meteorology [10], while a continuous data stream of the main climate variables is hosted by the Queensland Government’s SILO [26], with a national coverage of ~8000 locations available from 1889 to present. SILO’s data stream is freely available under the Creative Commons Attribution 4.0 International (CC BY 4.0) licence. SILO is the only available, curated, long-term, and continuous daily weather data record available in Australia and, while many of SILO’s sites have interpolated and modelled data based on the nearest available data, presents the best available data source [25]. CliMate could be extended to any source of continuous daily weather data. Selection of soil types for the Howwet? analysis [5] used a matrix based on soil depth (shallow, average, and deep) and texture (heavy clay, light clay, clay loam, sandy loam, and sand) for each Australian state (15 generic soil types).

2.4. Ethical Considerations

Users were asked to register an email address (username) and indicate their role (farmer, advisor, academic, or other) and production industry group (grain, cotton, horticulture, sugar, or animal production). Registration enabled users to retain previous analysis settings. We specified that this information would not be forwarded to any other party.

3. The Analyses

Ten analyses were originally included, of which nine accessed data directly from the SILO database [26], while How’s El Nino? sourced data from a link to the Bureau of Meteorology. A summary of the ten initial analyses is shown in Table 1 along with each analysis’s origin, function, and application. Further detail is available in a Supplementary Material (Australian CliMate factsheet 2025).

How’s the Season? tracks the current season’s rainfall, temperature, radiation, and heat sums in relation to previous years. Applications include adjusting agricultural inputs as a season progresses, forward marketing based on yield expectations, and providing an objective basis for discussions around seasonal prospects for multiple paddocks and clients. Farmers and insurers are the most common users of this analysis: farmers a guide to crop prospects and marketing, and insurers support actuarial assessments.

Howoften? calculates a probability of future events based on climatology only (frost, heat stress, planting rain, and extreme events) to assess risks of key agronomic and pasture management actions (e.g., crop establishment, heat sum to achieve maturity, break of season for pasture production, and wet weather during key operations) within a season. A common application is assessing the risks of receiving a planting opportunity for crops and pastures, especially when future planting opportunities are regarded as high risk and yield penalties for later plantings are high.

How Wet?/Nitrate? estimates soil water accumulation and nitrate mineralisation during a non-crop (fallow) period. Tactical applications include yield expectations based on soil water at planting and nitrogen fertiliser requirements based on nitrate N accumulation relative to other years. This analysis is used by advisors to support field sampling and as a basis for discussions with clients, while farmers use estimates of soil water at planting to set yield expectations.

Potential Yield? provides a progressive yield expectation based on soil water at planting, rainfall-to-date, and expected rainfall to maturity using a simple water use efficiency model (WUE) [28]. The WUE’s simplicity is attractive to farmers and consultants in tactical and strategic decision-making by adjusting inputs during the season and exploring long-term yield expectations. Potential Yield? is particularly applicable to Western and South Australia, where it has been applied since the 1980s [28,29].

Drought? tracks drought status using the Rainfall Percentile Method [30]. A “drought” condition is flagged when rainfall for a specified “residence time” is below an arbitrary 10% value. Residence time is dependent on location and enterprise. For example, a short-duration horticulture crop may have a residence time of months, while extensive grazing may have a residence time of several years. “Drought” is a difficult condition to define, let alone anticipate, making this analysis challenging to apply.

How Hot/Cold? estimates a simultaneous probability of hot and cold temperature extremes and is suited to analysis of temperature conditions when considering a new location and crop, pasture, and cultivar types. In many parts of Australia, production is often constrained by heat or cold stress at the beginning or end of growing seasons.

How’s the Past? provides an overview of historic rainfall, temperature, radiation, and evaporation data. Applications include strategic analysis for land purchase and reviewing recent seasons.

What Trend? visualises long-term annual time series for rainfall, temperature, radiation, and incidences of extreme values for specified monthly windows. Applications include exploring trends in climate variables with the objective of providing users with a local view of climate variability and potential changes in selected decades since 1900.

How Likely? generated probabilistic seasonal rainfall and temperature forecasts, looking forward 3 to 6 months, along with an assessment of past forecast skill. The forecast was based on correlations between seasonal weather and sea surface temperature and El Niño Southern Oscillation index values. This analysis was included due to the preoccupation of farmers with seasonal forecasts, and it is acknowledged that more informed sources were concurrently available elsewhere.

How’s El Nino? accessed the status of the El Nino Southern Oscillation index and the Bureau of Meteorology (BoM) seasonal interpretation. This analysis was based on direct access from the Bureau of Meteorology.

Both How likely? and How’s El Nino? were discontinued in 2019 due to either unpublished methodology or loss of data streams.

4. Adoption

User characteristics were voluntarily collected during registration (profession and industry involvement), while location, sessions (>30 s), and analysis use were continuously collected from 2017. Administrator access allowed the development team to assess use patterns with minor adjustments implemented based on user feedback.

With >20,000 registered users since its release in 2013, CliMate has been applied across most of Australia (Figure 4). Figure 5 plots registrations and “user sessions” since 2017, while Figure 6 summaries users by agricultural industry and professional group. After CliMate’s main release across all platforms in 2017, registrations ramped up for 18 months even though there was no concerted publicity campaign. “Marketing” was by word of mouth, rural media, and industry meetings rather than a concerted campaign, a common feature of many agricultural DSTs.

Grain and animal production users accounted for 70% of users, while farmers were the most common user group by profession (Figure 6). The top 100 users have accessed CliMate for ~800 sessions (median), dominated by consultants and insurers. With approximately 85,500 agricultural enterprises in Australia [32] and ~6000 agricultural consultants [33], we estimate ~50% of agricultural consultants have used CliMate since its release.

Of the 10 analyses available, the most used analysis was “How’s the season?” (Figure 7), followed by the probabilistic How often? and the longer view How’s the past? Most analyses were used >5000 times annually. Farmers used How’s the season? the most, grain farmers used Yield potential? the most, and academics used What trend? the most.

5. Discussion

Ideally, any DST should be designed with a target audience in mind [34]. CliMate’s development cycle was somewhat organic, with a mandate to update several previously valued climate analysis tools based on expert advice from a multi-commodity funding source. Initial plans were to re-enact these analyses into a dedicated software package for desktop computers, but the age of mobile devices had arrived; thus, the plan for an iOS app was born. Development of a www-based tool that could be used on the originally planned desktop followed as a contracted deliverable. This dual development had to fit within a budget and timeline, as CliMate was built in a commercial consulting environment.

An initial challenge was to decide which analyses should be included, as several of the target DSTs had multiple functions and complexities with no prior evaluations of which analyses within existing products were most valued or indication of what interface (input and output) options were preferred. This lack of evaluation of related products is common in agricultural decision support developments and is part of the motivation for this paper. The development team did not consider evaluation frameworks previously proposed [35,36,37].

The decision framework proposed in Figure 2 and a template similar to Figure 3 were adopted from a previous single application, Howoften? [18], which set the scene for the inclusion of seven analyses, later extended to the ten analyses described here. How’s the past?, Drought?, and What Trend? were added in version 2. It was arbitrarily judged that these 10 analyses were logistically manageable within the allocated budget and timeline and kept the interface manageable for users. When a working draft was available for distribution in March 2013, a soft launch using ~50 known contacts was used to gain feedback on functionality and usability.

Typical user feedback was to add functionality rather than change the interface per sec. Watching users use CliMate was instructive in leading us to reframe the wording of inputs. A section “How do we know this?” was included at the end of each graphical output to provide further explanation of the basis for presented results based on user feedback. The high number of iterations of interface design facilitated by a non-computer specialist “backend” was essential to allow for tuning of what was initially a poorly specified software product and a contributing factor in CliMate’s relative success.

An observation is that DSTs are becoming less expensive and faster to build, but DST development remains a significant investment. For example, Rainman [4], with ~6000 units distributed over 10 years, had a real investment of >AUD 2.5 million (2015 AUD) compared to CliMate’s 20,000 registered users on three platforms over 7 years, costing AUD 750,000. A common feature of all DSTs in Australia to date is the paucity of evaluation of use and impacts.

Despite CliMate’s popularity, funding for upgrades and maintenance remained challenging. CliMate has been operating reliably since 2014, but several technical issues have led to the loss of some functionality and the availability of two analyses. These include removal of the Android app, as it was difficult to adequately accommodate operating system changes without significant additional support, and third-party data access issues leading to periods of unreliable access to climate data and indices. These issues were impossible to anticipate but are inevitable in DST development.

Lessons Learnt

The following points are offered to other developers based on our experience:

• DST development requires a partnership between the intended audience, a program manager, a software engineer, and a connection to people with empathy and knowledge of the audience and technical issues.
• If software development were to be contracted, specifications would need to be very tight. In this case, an organic development cycle allowed for synergy between software design, interface design, and user feedback. CliMate’s specification at project initiation would have been difficult, as the development team did not understand user preferences without a prototype, and the designers did not fully understand software and technology capabilities and limitations.
• There is a strong tendency for a DST to be comprehensive yet flexible—potentially leading to added complexity and reduced useability [16]. “Keep it simple, stupid” (KISS) is a hard but essential lesson in building a useful DST.
• Multiple platforms (iOS, Android, and www) increased costs and time to develop. An option might be to develop a rapid www-based prototype, interact with a sample of prospective users, and carry out an initial evaluation.
• DSTs may have simple interfaces, but technology requires maintenance and ongoing support (servers, etc.), which needs to be budgeted over the expected life of a DST.
• Technical problems in software development are inevitable as technologies and third-party data sources evolve, requiring continual support for the expected life of each DST. It seems that 10 years is an over-optimistic life expectancy for a DST.
• DSTs require promotion, evaluation, and maintenance after release. Investors underappreciate the importance of building on an initial investment, especially if early indications of acceptance and use are available.
• Evaluation is important to justify an initial investment and guide future development. CliMate was independently evaluated in 2018 [38], but there was no scope to act on findings.

6. Conclusions

Australian CliMate Version I was released in 2013 as an iOS application. A www app and version 2 for iOS and Android were released between June 2017 and February 2018. CliMate has provided a significant proportion of Australian primary producers and their advisers with ready access to long-term and near real-time weather data and analyses, which are accessible and easy to interpret [38]. CliMate provides a common format for 8 analyses, accessing near real-time weather data that are readily accessible to decision-makers in the field, where many discussions and decisions occur, while a flexible and open-question focus facilitates user incorporation of local rules of thumb and models.

User statistics indicate that CliMate has facilitated exploration of long-term weather data over a 12-year period, providing objective assessments of current conditions and probabilistic estimates of future weather events based on past climatology. Each analysis is cast as a question for a selected variable (temperature, rainfall, and radiation) and period of interest, encouraging users to frame their query using their own models of risk.

CliMate’s use appears reasonably stable with 500–1000 sessions and 50 new users each month after 12 years. Images of CliMate screens appear regularly in farmer discussion groups, as it is accepted as a simple tool for many weather-related decisions in Australian agriculture [33]. The positive acceptance of DST across a wide range of production systems and users dispels previous experiences [16,17,35,36] and offers support for the collaborative development of simple computer and mobile device decision support products.