Online-only Table 2.
Data sources, constraint thresholds, and processing steps to produce constraints and resource yield criteria maps.
| Sector | Data sources (n; units; resolution) | Constraints (exclusions)* | Data layer(s) processing steps† |
|---|---|---|---|
| CSP | Average annual direct normal irradiance (DNI)141 (W/m2∙y; 3.7 km) | <125 W/m2∙y based on ref.26 |
i.→ For every ith cell calculated resource potential in MWh/km2 as PD ∙ CFi ∙ 8760, where PD = power density of 17 MW/km2 56, CFi = spatially explicit capacity factor using DNI values and assuming 6-hr storage based on ref.56, and 8760 = number of hours in a year. ii.→ Limited analysis cells by constraints including removal of any operating CSP plants. |
| Slope142 (%; 90 m) | >5% based on ref.57 | ||
| Landcover143 (1 km) | wetlands, rock/ice, artificial areas, forested areas based on ref.26 | ||
| CSP plant locations from SolarPACES144 (n = 96)§ | any operating plants | ||
| PV | Average annual global horizontal irradiation (GHI)141 (W/m2∙y; 3.7 km) | NA |
i.→ For every ith cell calculated resource potential in MWh/km2 as PD ∙ CFi ∙ 8760, where PD = power density of 30 MW/km2 56, CFi = spatially explicit capacity factor using GHI values and based on eq. 1 in. in ref.56 (while similarly assuming 4% outage rate and 4% inefficiency rate), and 8760 = number of hours in a year. ii.→ Limited analysis cells by constraints. |
| Slope142 (%; 90 m) | >30% based on ref.26 ‡ | ||
| Landcover143 (1 km) | wetlands, rock/ice, artificial areas, forested areas based on ref.26 | ||
| Wind | Average annual wind speed at 80 meters145 (m/s; 5 km) | <6 m/s based on refs26,56,60 |
i.→ For every ith cell calculated air density factor (ADi) by dividing air density (ρi) by sea-level density of 1.225 kg/m3, where for each elevation (Zi): ρi = 1.225–1.94 * 10−4 * Zi. Values of ADi ranged from 1.00 at sea level to 0.475 at 3,000 m elevation. ii.→ Calculated resource potential in MWh/km2 as PD ∙ CFi ∙ ADi ∙ 8760, where PD = power density of 5 MW/km2, CFi = spatially explicit capacity factor based fitting a local polynomial regression (loess) to average annual wind speed data reported by ref.56, ADi = air density factor calculated as described above, and 8760 = number of hours in a year. iii.→ Limited analysis cells by constraints including removing cells with ≥3 existing turbines. |
| Slope142 (%; 90 m) | >30% based on ref.26 ‡ | ||
| Elevation146 (m; 1 km) | >3,000 m based on ref.60 | ||
| Landcover143 (1 km) | wetlands, rock/ice, artificial areas based on refs23,26 | ||
| Wind turbine locations** (n = 90,106)§ | any cell with ≥3 turbines based on refs. 59,147,148 indications of wind development | ||
| Hydro | Hydropower potential locations18 (n = 11,839,398; kWh/y) | cells with <1 MW potential to accommodate utility-scale hydropower development130 |
i.→ Limited hydropower potential locations to those generating ≥8,760,000 kWh (i.e., 1 MW) and removed any locations ≤1 km of existing hydroelectric dams, consistent with average distance of existing dams (1.14 km; this study based on GRanD data). ii.→ Hydropower potential locations spanned fully inundated, or partially inundated cells; attribute values of fully inundated cells to the closest terrestrial cell (n = 935). iii.→ Divided kWh by 1000 to calculate MWh and rasterize locations. |
| Existing hydropower dams149 (n = 2,134)§ | any cell within 1km of a dam identified as hydroelectric for main, major, or secondary use | ||
| Coal | Global coal basins maps20 (n = 2,053) | NA |
i.→ For each jurisdiction (country or state): a.→ Clipped basins by jurisdiction and calculate basin area within b.→ Divided estimates of technically recoverable coal by the area of the coal basins within the jurisdiction to obtain million short ton/km2 c.→ Rasterized basins and attribute million short ton/km2 to every coal basin cell within the jurisdiction. ii.→ Merged jurisdictional results into a single raster of coal yield. iii.→ Limited analysis cells by constraints including removal of any cells with existing active coal mine. |
| Country or state-level estimates of technically recoverable coal20 (n = 305; million short tons) | NA | ||
| Existing coal mines from 8 datasets150–157 (n = 2,301)§ | any cell with active coal mine | ||
| CO | Global158,159, U.S.160 and Australia161 assessment units (AUs) of median recoverable oil reserve (n = 778; barrels of oil or bbl), and dry gas (n = 705; ft3), and liquid natural gas (NGL) (n = 823; bbl) | NA |
i.→ Geo-referenced and digitized all world shale prospective area (PA) maps and assigned technically recoverable values to each PA. ii.→ For dry gas, converted reserve estimates from ft3 of dry gas to BOE using a conversion factor of 6000 ft3 to 1 BOE, then summed for each PA or AU the total gas estimates in BOE from the converted dry gas and NGL. iii.→ For each sector AU or PA: a.→ Divided the technically recoverable oil or gas reserve estimates by the area of the AU/PA to obtain BOE/km2. b.→ Rasterized AU/PA and attribute BOE/km2 to every cell. iv.→ Combined all AUs/PAs for each sector, summarizing overlapping BOE/km2 cell values. |
| CG | |||
| UO | World shale prospective areas (PAs)162 risked recoverable oil (n = 109; bbl) and gas (n = 192; ft3), and U.S. unconventional AUs of median recoverable oil reserves163,164 (n = 22; bbl), dry gas (n = 130; ft3), and liquid natural gas (NGL) (n = 113; bbl) | NA | |
| UG | |||
| MM | Mineral deposit point locations and categorical deposit sizes (e.g., very larger, large, medium) from refs165–172 combined into a global dataset consisting of 167 minerals (n = 320,843)†† | 84 metallic minerals (e.g, gold, silver, iron), plus gemstone and uranium (n = 207,258)†† |
i.→ For any given mineral, removed spatial duplicates and assign highest deposit value to deposit location§§. ii.→ Split data to U.S. and non-U.S. regions‡‡. iii.→ Ran Kernel Density (KD) using: a.→ Radii of 60 km for metallic, as used for gold deposits in Australia75,76, and 20 km for non-metallic, as used for aggregate minerals in Poland173. b.→ Weights based on classifications and criteria in ref.74 as follows: 1 – occurrences, 4 – small deposits, 9 – medium deposits, 16 – large deposits, and 25 – very large deposits (volume classifications in Appendix 1 of ref.74). iv.→ Selected only cells with KD > 0.001 deposits per km2 based on ref.76. v.→ Limited to analysis constraints where we defined cells with existing mines as any cell containing past and current mines in the collated database, or cells with ≥ 50% overlap with mapped mineral or industrial areas**. vi.→ Standardized and merged U.S. and non-U.S. regions into one global map. |
| NMM | 83 non-metallic minerals (e.g., aggregates, gravel, and sand; n = 113,498)†† | ||
| Crop | 2012 yield and area data by political unit (national or sub-national)174 | Included barley (n = 5,389), cassava (n = 6,053), maize (n = 13,681), oil palm (n = 2,007), rapeseed (n = 3,496), rice (n = 3,984), sugarcane (n = 2,912), sorghum (n = 5,421 soybean (n = 5,707), and wheat (n = 7,377) |
i.→ For each political unit, obtained the area-weighted average yield in ton/km2 (response variable). ii.→ Processed explanatory variables (abbreviations): a.→ Used WORLDCLIM to calculate crop-specific annual growing degree days (GDD) and mean annual precipitation (Prec) following ref.79 and converted temp of coldest month into a binary variable of 1 if coldest month of the year is between −8 °C and +5 °C, else 0 (VF). b.→ Used WATCH input data to calculate number of days with temp >34 °C (KDD34). c.→ Used Harmonized World Soil Database V1.2 slope datasets to identify percent of 10 km cell greater than 10% and 30% (slgt10 and slgt30). d.→ Used ISRIC world soil information database to extract the soil water holding capacity for top 20 cm (awc). e.→ Used irrigation fraction data to linearly extrapolate data for 2000 and 2005 while constraining fraction 0–1 (Irr). iii.→ For each crop, modeled ton/km2 as a function of climate, slope, soil, and irrigation data using a 95th percentile quantile regression on standardized variables. Started with general model: GDD+Prec+ Irr+VF + KDD34+slgt10+slgt30 +awc +GDD2+Prec2+GDD*Prec+ Prec*Irr and conducted backward selection with bootstrapping methods††† to select crop-specific models and coefficients (Table 3). iv.→ For every crop, created global predictive maps using model results: a.→ Used input explanatory variables data and remove cells with values <2.5th and >97.5th percentile CIs based on model coefficient results. b.→ Predicted global yield map based on model coefficients. c.→ Removed any cells with predicted values < the 10th percentile observed area-averaged yield (ton/km2). v.→ Limited analysis cells by constraints. |
| WORLDCLIM175 (1-km) | Thornethwaite Moisture Index <0.25 without irrigation fraction >0.05 | ||
| Water and Global Change (WATCH) Forcing Data176 (50-km) | NA | ||
| Slope142 (%; 90 m) | >30% based on ref.8 | ||
| Average water holding capacity177 (10 km) | NA | ||
| Irrigation fraction178 (10 km) | |||
| Percent cropland179 (250 m) | Cropland ≥ 95%180 | ||
| Bio | see above | see above |
i.→ Used cropland yield values (tons/km2) derived above for the five first generation biofuel crops (listed below) and multiplied yield by gallon of gasoline equivalent (GGE) conversion rates20 (maize-162.47, oil palm-56.55, rapeseed-99.41, sugarcane-32.1, soybean-46.41). ii.→ Limited analysis cells by constraints identified in crop expansion above. |
Table includes references, justifications, and rational for producing resource yield layers and constraints for the 13 development sectors. Data sources column identifies data used in yield estimates in bold (e.g., Average annual direct normal irradiance) and includes data on the number of points or polygons used for yield estimates and/or constraint mapping (n), original input units provided by data (units), and cell size of raster data (resolution). Constraints column identifies the threshold value of the data source used to map non-suitable lands that were excluded from each development potential index (DPI). Abbreviations of sector are as follows: CSP – concentrated solar power, PV – photovoltaic solar power, Wind – wind power, Hydro – hydropower, CO – conventional oil, CG – conventional gas, UO – unconventional oil, UG - unconventional gas, MM – metallic minerals, NMM – nonmetallic minerals, Crop – cropland expansion, Bio – biofuels expansion.
*Urban areas defined as human-built environments created by ref.181 were excluded from resource yield maps for all sectors.
†Online-only Table 1 provides a detailed literature review that informed parameter selection and/or threshold values.
§Point locations of existing development used for excluding cells from resource yield maps.
‡Reference26 used a slope of 27%, however because the Harmonized World Soil slope data used here, which were derived from 90-m resolution digital elevation data, limit users to binned slope breaks (e.g, 2, 5, 10, 15, 30), we used the next closest bin break of 30%.
**Data from OpenStreetMap.org (©OpenStreetMap contributors, CC-BY-SA https://www.openstreetmap.org/copyright).
††87 locations had undefined mineral type and were therefore removed from the analysis; these data were used to derive resource yield and final mining DPIs based on kernel density analyses specified in processing steps.
§§To avoid inflating kernel density values, we removed spatial duplicates for each mineral (e.g., gold, sand, etc.) and assigned the highest deposit value to that location. We acknowledge that density values will still be higher if the same location was sampled for different minerals, or if multiple locations in close proximity were sampled for a given mineral. However, higher density values in these cases are justified as it is more likely that these areas will be developed given sampling intensity of the deposits.
‡‡82% of deposit locations was located within the U.S., hence we created mining density maps separately for the U.S. and non-U.S. regions, which we later normalized separately and then merged for a final global map of resource yield potential.
†††We used the built-in confidence intervals (CIs) in R (using rq() with the ci = TRUE option) unless either the CI upper and or lower bounds were numerically infinite. In this case, we determined CIs with a bootstrap method by constructing an NxM array of parameters where M is the number of parameters and N is the number of bootstrap samples (using sampling with replacement). The 2.5% and 97.5% quantiles of the M values in the distribution of each parameter represent the lower and upper bounds. We used N = 200 when determining the least significant parameter in the model simplification step discussed above, and we used N = 1000 when determining confidence intervals.