Utility-Scale Solar Design Optimization for Yield and CAPEX

Most of the value in a utility-scale solar park is fixed before the first pile is driven. The DC capacity behind the grid connection, the row pitch, the layout on the terrain, the foundations and the module bill of materials set both the CAPEX and the energy the plant can ever produce. Utility-scale solar design optimization means testing those choices against each other in simulation, with real site data, while changing them still costs only engineering hours. My position is that this is where profitability is won: design-stage decisions can raise yield and lower CAPEX at the same time, which strengthens the investment decision and the pace of portfolio growth more than anything done later.

Why utility-scale solar design optimization comes first

An investor sees a solar park through a few numbers: CAPEX, OPEX, the discount rate and the energy yield assessment. IEA PVPS Task 13 calls the yield assessment a prerequisite for long-term investment decisions, because together with the cost data it produces the LCOE, IRR and NPV that decide whether a project is funded ^[1]. Apart from the discount rate, all of these are shaped mainly at design stage.

The arithmetic is simple. LCOE is discounted lifetime cost divided by discounted lifetime energy, so 1% more energy at the same cost lowers LCOE by about 1% (1/1.01 = 0.990), and 1% less cost at the same energy does about the same. A change that does both, such as a layout that needs less earthwork and also shades less, compounds. At portfolio level, projects that need less capital per MWh let the same equity support more megawatts, and a design method proven on one site can be reused on the next.

Having worked across development, EPC management and asset management, with PVsyst, Helios 3D and RatedPower as everyday tools, my view is that this window closes quickly. Once the EPC contract is signed, every change becomes a variation order; once the plant is energized, O&M can protect the energy the design made possible but rarely add to it.

The position has limits. A simulation is only as good as its inputs: in an IEA PVPS benchmark where different modelers assessed the same site in Bolzano, initial yield assessments averaged 1,278 kWh/kWp with a standard deviation of 9.7%, and the differences came mostly from assumptions on irradiance data, degradation, soiling and shading ^[1]. Options must therefore be compared inside one consistent model, and the gains reach the investor only if construction delivers the design. Quality assurance and risk registers are a separate discipline and are not covered here.

Sizing DC against the grid connection: the Pnom ratio decision

In PVsyst, the Pnom ratio is the array’s nominal STC power divided by the inverter’s nominal AC output, the same quantity as the DC/AC or inverter loading ratio. Oversizing the array makes sense because it rarely reaches its STC rating in the field. When the array could deliver more than the inverter converts, a modern inverter shifts its operating point on the I-V curve and limits output, so the only loss is the potential energy above the limit ^[2]. In exchange, the plant produces more in the mornings, the evenings and on cloudy days.

PVsyst’s default criteria warn when simulated overload loss exceeds 0.2% and block the simulation above 3%, which leads to Pnom ratios of about 1.25 to 1.3 for well-oriented systems ^[2]. The documentation adds that an economic optimization, weighing clipped energy against the cost of more inverter capacity, often ends with inverters well below manufacturer recommendations, and that grid injection limits push designers toward highly oversized arrays ^[2]. The market agrees. Berkeley Lab reports a capacity-weighted average DC/AC ratio of 1.32 for US projects built in 2024 (1.41 for fixed-tilt, 1.32 for tracking), driven by module prices falling faster than inverter prices, and the highest DC/AC quartile had on average 1 percentage point higher capacity factors than the lowest ^[3].

The optimum is still site- and equipment-specific. A 2025 study in Energies simulated 28 inverters in 27 Brazilian cities at ratios from 0.8 to 2.0 and found LCOE optima of 1.1 to 1.3 for modern medium- and high-power inverters, but up to 1.8 for low-power units ^[4]. It also showed that first-year clipping overstates the lifetime penalty: at a ratio of 1.6, clipping was about 3.9% in year one and fell to 0.15% later in life as module degradation reduced array power ^[4]. The authors warn that rule-of-thumb ratios can leave a system’s potential untapped ^[4].

In Hungary, grid capacity has at times been the binding constraint. In 2022 the transmission and distribution operators announced zero available connection capacity for new renewable plants at all substations, a freeze later extended to a full year ^[5]. When the AC connection is the scarce asset, the DC behind it decides how many MWh each connected megawatt sells, and the optimum lies where the marginal cost of extra modules, structures, cabling and land meets the marginal value of the extra energy.

Two details are easy to miss. IEA PVPS notes that inverters may be sized on the module’s front-side nameplate only, although bifacial and albedo gains add energy that can then be lost to clipping ^[6]. And the maximum array current per inverter should not be exceeded, because it leads to premature inverter aging and yield loss ^[7]. My recommendation is to treat the Pnom decision as a scenario analysis: four or five ratios, each simulated hourly with lifetime degradation and priced through to LCOE and NPV with real module and inverter quotes.

Row spacing and topography: buying energy with land

Row pitch, expressed as ground coverage ratio (GCR), trades energy against land. Tighter rows put more megawatts on each hectare but shade each other for longer; wider rows need more land, fencing and cable. The IFC developer guide sets the goal of layout design as minimizing cost while achieving maximum revenue, by limiting inter-row shading and shortening cable runs ^[7].

The traditional shortcut spaces rows to avoid shading at a fixed hour on the winter solstice. A 2024 Applied Energy study instead optimized tilt and spacing of fixed monofacial and bifacial arrays to minimize LCOE per unit of land area. The optimized designs had 0.75% to 3.06% lower LCOE than the solstice rule and produced 22% to 51% more electricity per unit of land ^[8]. For trackers, IEA PVPS notes that with low module prices and high land costs the design goal is usually energy per unit of land rather than per kWp, and that backtracking avoids row-to-row shading only by holding modules at less than ideal angles ^[6]. The optimum GCR therefore depends on land price, cable cost and site shape, which is exactly what a simulation sweep tests.

Topography changes the shading geometry, so one pitch applied across undulating ground over-shades some rows and wastes land elsewhere. IEA PVPS notes that the established slope-aware backtracking model applies to a single uniform slope; for complex topography, one brute-force method raises the GCR parameter in the backtracking algorithm until shading falls to an acceptable level ^[6]. Most tracker suppliers surveyed by IEA PVPS allow a maximum north-south grade of 15% (8.5°) ^[6]. Beyond the structure’s limits, the options are grading, a different layout or a different mounting system, each with a cost that belongs in the comparison. The fixed-tilt versus tracker decision, where IEA PVPS cites typical tracker energy gains of 15% to 20% ^[6], is covered in a separate article on fixed tilt vs single-axis trackers.

In practice, I start from a digital terrain model, split the site into slope and aspect zones, and test zone-specific row distances against earthwork volumes before fixing the layout.

Surveys that right-size foundations, earthing and drainage

The survey program decides whether the civil and structural design can be lean or has to be conservative everywhere. The IFC guide recommends boreholes or trial pits at regular intervals, usually 2.5 m to 3 m deep, assessing groundwater level, soil resistivity, load-bearing properties, obstructions and pile drivability, plus soil pH and chemistry to set the corrosion protection; in highly corrosive soil, pile coating thickness should be derived by calculation ^[7]. IEA PVPS adds seismic and flooding risk to the site assessment ^[6].

Piles show the yield-and-CAPEX logic most clearly. DNV describes pre-production pile load tests, often based on ASTM D1143 (compression), D3689 (tension) and D3966 (lateral), as the way to establish design pile capacities, and notes that a 250 MW plant may have about 100,000 piles ^[9]. At that scale, a few hundred millimeters of embedment or one coating class repeats across the whole site. DNV also describes the opposite failure: pre-drilling applied site-wide where clay would have allowed direct driving, and pre-drilled piles with less capacity than designed that sink under the tracker and module weight and need costly remediation ^[9]. My recommendation is to run pull-out and lateral tests in every distinct soil zone, including pre-drilled conditions where they are likely, and to set embedment zone by zone.

Measured soil resistivity lets the earthing system be designed for the ground actually present, while groundwater and hydrological data set drainage, cable routes and station elevations. Flooding can damage electrical equipment ^[7], so a hydrological assessment protects energy as well as assets. These surveys also feed the risk register and insurance discussion, but their direct economic effect is that the design stops paying for uncertainty.

Module selection: certificates are the floor, test data is the decision

IEC 61215-1 qualifies module designs for long-term operation in open-air climates, but states that its test results are not a quantitative prediction of module lifetime ^[10]. Kiwa PVEL, which tests factory-witnessed samples, says certification tests are typically not rigorous enough to reveal the effects of design and component changes, and that modules sold under one model name can be built from different bills of materials (BOMs) ^[11].

Recent data supports that view. In Kiwa PVEL’s 2026 Scorecard, 87% of manufacturers and 61% of BOMs had at least one failure, no BOM was a Top Performer in every test, 26% of BOMs in mechanical stress testing had a broken module, and 45% of manufacturers had at least one major delamination defect ^[12]. RETC’s 2026 PV Module Index found that 11% of BOMs failed damp heat testing with more than 5% power loss, up from 6% a year earlier, and 8% failed UV-induced degradation testing ^[13].

Energy performance is just as BOM-specific. Kiwa PVEL attributes rising energy yields in its PAN testing to better temperature coefficients and bifaciality ^[12], and IEA PVPS gives typical bifaciality factors of about 70% for PERC, 80% for TOPCon and approaching 90% for some heterojunction cells ^[6]. These differences only show their value when the shortlisted BOMs are simulated in the same project, with the same weather file and layout, using third-party PAN files; RETC’s test program includes PAN file characterization for PVsyst for this purpose ^[13].

Degradation closes the loop. IEA PVPS notes that current practice often assumes linear degradation of about 0.5% to 0.6% per year for crystalline silicon modules, cites an NREL review putting long-term degradation at about 0.5% per year with an uncertainty of about ±0.25% per year, and warns that the assumed rate and its behavior over time can have significant financial consequences ^[14]. LID and LETID, UV and damp heat results should set a degradation scenario per BOM, and the tested BOM belongs in the supply contract.

Making the simulation comparison honest

The PVsyst loss diagram is the common language for these comparisons. PVsyst orders the array and system losses as incidence angle, soiling, irradiance level, temperature, LID, module quality, mismatch, degradation, DC and AC wiring, transformer, auxiliaries and unavailability, sets them to reasonable defaults, and advises defining each one for the specific system after the first run ^[15]. Defaults are placeholders, not a design. Each lever moves specific lines:

• Row pitch, terrain zoning and backtracking move the near-shading and electrical shading losses.
• The module BOM moves the incidence angle, low-irradiance, thermal, LID, quality and degradation losses, plus the bifacial gain.
• Stringing and cable layout move the mismatch and ohmic losses.
• The Pnom ratio trades inverter overload loss against the energy from the extra DC.

Comparisons stay meaningful when only the lever changes. In the IEA PVPS bifacial round-robin, total energy predicted by different participants for the same scenarios varied by as much as +5% to −10% from the mean ^[6], so a 0.5% gain between two layouts is credible as a relative result within one model, not as an absolute forecast. The final design should then be carried through to P50 and P90. IEA PVPS recommends stating yield results with an exceedance probability and notes that lower uncertainty raises the value at a given exceedance probability, which strengthens the business case ^[1]. Site-measured albedo, soiling and terrain data therefore improve the P90 as well as the P50.

Design-stage optimization checklist

The table summarizes the levers, their typical effect, what to simulate and the data each simulation needs.

Conclusion

Design is the only point in a project’s life when yield and CAPEX can be traded freely against each other. The Pnom ratio, the row pitch, the terrain strategy, the foundations and the module BOM are decided once, and each can be tested in simulation for the price of engineering time. Done with site data and a consistent model, that work delivers a lower LCOE, a more defensible P90 and a method that repeats across a portfolio. It does not replace good construction or good O&M, but it sets the ceiling both of them work under.