Where should it go? Siting with maps, from layers to a suitability score

Screen Prairie City’s region for a solar farm, a wind farm, a new line and a data-center campus: exclusions, weights, a least-cost path, and what the map cannot tell you

A19 of Energy Systems, Step by Step: stack map layers for the fictional region around Prairie City, exclude land by rule, weight the criteria, rank sites for a solar farm, a wind farm and a data-center campus, route a line along a least-cost path, and see where the screen’s weights, resolution and data date change the answer; an optional NumPy notebook rebuilds it.
  • Analyst to Modeler · L2–L3
  • 25 min guided · 60 min lab · about 30 min practice
  • After A11

The default runs, from the same model (Prairie City is fictional; its numbers are illustrative.)

Solar farm, 150 MW: the five best 2 km × 2 km blocks. Land left after exclusions: 954 km².

Site Where (from the centre) Score Up front Substation Energy a year
A 12 km E, 8 km N 73 $7.1M East hub (S4) 313 GWh
B 9 km E, 8 km N 70 $9.2M East hub (S4) 312 GWh
C 12 km W, 1 km S 68 $13.0M West (S3) 317 GWh
D 11 km E, 11 km N 67 $9.2M East hub (S4) 311 GWh
E 15 km W, 2 km S 66 $15.1M West (S3) 316 GWh

Wind farm, 150 MW: the five best 7 km × 7 km blocks. Land left after exclusions: 960 km².

Site Where (from the centre) Score Up front Substation Energy a year
A 17 km W, 3 km N 70 $16.6M West (S3) 470 GWh
B 22 km W, 9 km S 66 $26.2M West (S3) 509 GWh
C 25 km W, 1 km S 66 $21.3M West (S3) 508 GWh
D 25 km W, 11 km N 64 $30.0M West (S3) 513 GWh
E 15 km W, 4 km S 62 $12.4M West (S3) 458 GWh

Data-center campus, 150 MW: the five best 1 km cells, at least 2 km apart. Land left after exclusions: 912 km².

Site Where (from the centre) Score Up front Substation Water main
A 7 km W, 1 km S 67 $5.5M West (S3) 1.0 km
B 11 km E, 9 km N 63 $14.6M East hub (S4) 8.5 km
C 10 km W, 1 km S 60 $10.5M West (S3) 4.0 km
D 8 km E, 9 km N 58 $14.7M East hub (S4) 6.7 km
E 7 km W, 2 km N 55 $8.5M West (S3) 1.0 km

Corridor from wind site A to West (S3): 13.7 km for $15.3 million with protected land forbidden; 10.2 km for $11.7 million when it is allowed at NREL's mitigation cost (3.4 km through it).

Up-front costs are 2025 dollars (NREL's 2022 line costs converted with BEA's GDP deflator): the new line, an access road, land and, for the data center, a water main. Energy is a year's output; the two are never added.

With JavaScript on, you run the screen yourself: a map with its layers, switches for the exclusions, sliders for the weights, a routed line and a table of the best sites.

  • Fictional: Prairie City is fictional; its numbers are illustrative. Every layer is generated from a seed and a stated recipe; nothing was downloaded.
  • Sourced: tracking solar arrays take 4.2 acres per MW of panels (DC); 2019's tracking plants had 1.29 MW of panels per MW of inverters (AC, the grid side) (LBNL, slides 13 and 8).
  • Sourced: wind farms spread over 34.5 ha per MW, where 1 ha = 10,000 m², about 2.5 acres (NREL, 2009, Table 1).
  • Sourced: setbacks, slope rules and the line's cost per mile come from NREL's 2024 supply curves (Tables 7, 9, 11 and 12), in 2022 dollars, shown in 2025 dollars with BEA's GDP deflator.
  • Shared city data: the data-center campus's 150 MW contract comes from Prairie City's data, as on the course's other city pages.
  • Set for teaching: the four transmission substations S1–S4 and their spare capacity, land prices, access roads, water mains, block sizes and the capacity-factor rules.

About 25 guided minutes in eight steps. Practice, about 30 more: the set under Check.

Optional · build it in Python · notebook lab, about 60 minutes

The same region in NumPy, from the same seed, checked against this page with assert cells.

Inside: masks, scores and the best blocks; a least-cost path; a cost model; cleaning and aligning simulated raw layers; the public layers a real screen would use.

  1. Download the notebook and the lesson data into one folder (in Colab, upload the data with the folder icon).
  2. Open it in Google Colab (File → Upload notebook) or locally with Python 3 and NumPy.
  3. Run all cells (a few seconds on a 2025 laptop), then try the practice tasks at the end.

Notebook (.ipynb) Lesson data

What you just did, as diagrams

Layers
A raster layer gives every cell of one grid a number or a class: slope, wind, land price, protected or not.
Exclusions
An exclusion layer removes land a rule forbids, whatever else it offers.
Weights
A weighted overlay adds criterion scores; the weights are value choices, so state them and test others.
Sites
Blocks with enough buildable land are ranked by score; a score ranks, a cost model prices.
Lines
A least-cost path over a cost surface routes a new line to a substation with room.
Limits
Cell size, the date of each layer and whose weights were used can each change the short list.
Go deeper · AnalystThe arithmetic behind the map
Step Formula In this lesson
A criterion’s score \(s = 1 - \min\!\big(1, \max(0, \tfrac{d - d_0}{d_1 - d_0})\big)\) near a substation: 1 at 0 km, 0 from 15 km (solar)
Suitability \(S = \sum_k w_k s_k \,/\, \sum_k w_k\) default solar weights 20, 35, 15, 20, 10
Room in a block \(B = \sum_c a_c f_c \ge\) the land needed 150 × 1.29 × 4.2 = 812.7 acres ≈ 3.29 km² of a 4 km² block
A line’s cost \(\sum_\text{steps} \tfrac{c_u + c_v}{2}\,\ell\) \(c\) ≈ $1,095,979 per km on farmland (2025 dollars), × 1.8 in flood zones
  • Scores. \(d\) is a distance in km and \(d_0\), \(d_1\) the ramp’s ends; \(w_k\) is a weight and \(s_k\) a score from 0 to 1.
  • Blocks. \(a_c\) is a cell’s area in km² and \(f_c\) its share outside setbacks, checked at 16 points.
  • Lines. \(c_u\) and \(c_v\) are two neighbouring cells’ costs per km; the step \(\ell\) is 1 km, or \(\sqrt{2}\) km on a diagonal.
  • Resources are rescaled from the region’s lowest (0) to its highest (1); land price falls from 1 at $2,500 an acre to 0 at $15,000.
  • The line’s cost per km is NREL’s $1,615,100 a mile (2022 dollars) × 128.893 / 118.026 (BEA’s GDP deflator, to 2025 dollars) ÷ 1.609344 km a mile.
  • Ties. Equal scores go to the more northern, then the more western block. Equal-cost paths may differ in their steps but share their length and cost; the page draws the first found.
Go deeper · ModelerFrom a screen to a supply curve
  • The notebook (above) rebuilds every layer as a NumPy array from the same seed, then masks, scores, ranks blocks and runs Dijkstra’s algorithm for the line.
  • Real layers come as rasters and vectors in different coordinate systems, with gaps and errors; the notebook cleans and aligns simulated ones onto one grid.
  • Areas and distances. An equal-area projection keeps areas true, not distances: check its distance error where you measure setbacks and routes, or measure on the curved Earth (geodesic distances).
  • A supply curve sorts many sites by their cost, as NREL’s supply curves do for about 67,000 cells of 11.5 km × 11.5 km, and hands it to a planning model such as M3.
  • Out of this course’s scope: remote sensing, geostatistics, cartography and field surveys.
Go deeperThe data-center card: spare capacity and the cost of waiting

A campus near a substation without spare capacity waits for an upgrade. In step 7, that wait has no price; one study prices it.

A late grid connection costs the developer lost margin and carrying costs; ratepayers face utility carrying and stranding costs; society bears external costs.

The study keeps these present-value accounts separate.

Approved excerpt of: Li et al. (2026), accepted paper, 2026 North American Power Symposium (NAPS); DOI pending.

Holds when: the delay is given and the inputs are illustrative. Leaves out: why service is late.

The full case: case file CF7; how a large load asks for service: A6.

Go deeperWhat this lesson leaves out
  • Rights-of-way of existing roads and lines, airports, radar, wetlands beyond the flood zone, and wildlife surveys.
  • Two projects on one substation: each site here is screened alone, so they could both claim its spare capacity.
  • Grid studies: whether the network beyond the substation can take the power or the load.
  • Lease payments to landowners, taxes, and the plant's own cost, which is the same at every site.
  • Where the data center's water comes from, and how much its cooling uses.

Check yourself

Myth or fact? “The site with the highest suitability score is the cheapest to build.”

  • Myth
  • Fact
Answer

Myth. Wind site A costs $16.6 million up front, site E $12.4 million. A score mixes criteria by chosen weights; a cost model prices them.

Myth or fact? “Excluding land and giving it a low weight do the same job.”

  • Myth
  • Fact
Answer

Myth. Priced at about $41,500 a mile (NREL's $38,000 in 2022 dollars), the new line crossed 3.4 km of protected land; only the exclusion kept it out.

Myth or fact? “Sunshine differs by only 4.5% across this region, so it barely moves a solar site's score.”

  • Myth
  • Fact
Answer

Myth. Min–max scaling stretches the 4.5% to a full 0 to 1: at weight 20, sunshine spreads blocks' scores by about 19 points, more than cheap land's 17. Scaling and weights, not physics, set that.

Myth or fact? “A map of 1 km cells can check a 220 m setback.”

  • Myth
  • Fact
Answer

Myth. A setback is smaller than a cell: this map checks 16 points in each cell, and NREL's supply curves check setbacks on 90 m cells.

A county forbids turbines within 400 m of any home. In a screening study, that rule is best treated as…

  • an exclusion layer
  • a low weight on “away from homes”
  • a cost per km
Answer

An exclusion: it is a rule, so no other advantage of a site may outvote it.

Your homes layer is dated 2019 and it is now 2025. What can go wrong?

  • Newer homes inside setbacks or next to a site are missed
  • The wind speeds change
  • Nothing: homes rarely change
Answer

Newer homes are missed. In step 8, 36 homes built since 2019 push solar site A out of the top five.

Practice · about 30 minutes on your own

  1. A score by hand. One cell's scores and the default weights are below. What is its suitability, out of 100?

    Criterion Score, 0 to 1 Weight Neighbours' weight
    Sunshine 0.6 20 10
    Near a substation 0.9 35 10
    Near a road 1.0 15 5
    Cheap land 0.4 20 5
    Away from homes 0.5 10 70
    Answer

    (0.6 × 20 + 0.9 × 35 + 1.0 × 15 + 0.4 × 20 + 0.5 × 10) ÷ 100 = 0.715, or 71.5 out of 100.

  2. Other weights. Score the same cell with the neighbours' weights, the table's last column.

    Answer

    (0.6 × 10 + 0.9 × 10 + 1.0 × 5 + 0.4 × 5 + 0.5 × 70) ÷ 100 = 0.57, or 57 out of 100: lower, because this cell is close to homes.

  3. Land for solar. A 200 MW (AC) farm has 1.3 MW of panels per MW of inverters. At LBNL's 4.2 acres per MW of panels, how much land do the arrays need?

    Answer

    200 × 1.3 × 4.2 = 1,092 acres, about 4.42 km²; the whole site is larger (roads, fences, setbacks).

  4. Land for wind. A 100 MW wind farm takes NREL's average 34.5 ha per MW. How large is it, and how much do turbines, roads and substations occupy for good, at 0.3 ha per MW?

    Answer

    3,450 ha (34.5 km²) in all; 30 ha for good, under 1%: farming goes on between the turbines.

  5. Setbacks. 40 farmsteads, none within 800 m of another. How much land does a 400 m setback remove, and how much a 220 m one?

    Answer

    40 × π × 0.4² ≈ 20.1 km² against 40 × π × 0.22² ≈ 6.1 km²: the area grows with the square of the setback.

  6. A line's cost. A 12 km route crosses 8 km of farmland and 4 km of flood zone (× 1.8). Price it in 2025 dollars: NREL's $1,615,100 a mile (2022), deflator 128.893 ÷ 118.026.

    Answer

    $1,615,100 × 128.893 ÷ 118.026 ÷ 1.609344 ≈ $1,095,979 a km; × (8 + 4 × 1.8) ≈ $16.7 million in 2025 dollars.

  7. A least-cost path by hand. Each step goes to a side-by-side cell and costs the two cells' average (costs per km below). What is the cheapest path from top left to bottom right?

    Row Left Middle Right
    Top 1 1 4
    Middle 1 5 1
    Bottom 2 1 1
    Answer

    Down the left column, then along the bottom row: 1 + 1.5 + 1.5 + 1 = 5. Along the top row and down the right column costs 7.

  8. Read a claim critically. A report says: “Our map shows the best solar site in the county.” What four questions would you ask?

    Answer

    Whose weights, and would other weights pick another site; which land was excluded, by what rule; how fine and how recent the layers are; who was asked, and what a visit found.

Numbers, sources and what’s next

Numbers to remember

Number What it means Source
4.2 acres per MW Land under single-axis tracking solar arrays per MW of panels (DC), the median of U.S. plants built in 2019; the whole site is larger. LBNL slides, slide 13 (PDF) · 1 February 2022
34.5 ha per MW Average total area of U.S. wind plants (1 ha = 10,000 m²); turbines, roads and substations occupy about 0.3 ha per MW of it for good. NREL (2009), Table 1, p. 10 (PDF) · August 2009
220 m · 61 m Setbacks from buildings for wind and for solar in NREL's reference case: the median of existing local rules. NREL (2025), Tables 7 and 9 (PDF) · revised June 2025
$1.6 million a mile A new 138 kV line over pasture with MISO's costs (also used for SPP), in 2022 dollars, before land-cover multipliers. NREL (2025), Table 11, p. 34 (PDF) · revised June 2025

Keep going

Back to the course map

Related lab: an optional NumPy notebook that rebuilds the map from the same seed, checks the page's sites and line, and passes the sites to a cost model, about 60 minutes.

In the real world

NREL's 2024 supply curves screen the contiguous U.S. the same way: exclusions and local setbacks on 90 m cells, about 67,000 candidate sites of 11.5 km × 11.5 km, and least-cost paths to the grid.

Where its median setbacks become the 90th percentile of local rules, much less land is left: the rules are local choices.

Sources and further reading

Spotted an error, or stuck? Write to hlli@ksu.edu. Corrections are made in place and dated here.

Text licensed CC BY 4.0, code MIT: teachers and students may reuse and adapt it, with credit to Honglin Li.