03 - Land use, water, soil, N and SOC review of cleaned v0.7.0

Reviewer role: sustainable land-use and agro-environmental modeller (adversarial review) Source reviewed (read-only): CIAT/cleaned @ 800d53a (v0.7.0) Scope: land_requirement.R, land_productivity.R, biomass_calculation.R, water_requirement.R, soil_health.R, soc.R, nitrogen_balance.R, compare_scenario.R, differences.R, merge_outputs.R, inst/extdata/*.json, vignettes, README.

Verdict

The land, water, erosion, N-balance and SOC modules should not be used for quantitative ex-ante claims yet. The land footprint silently drops feeds (any Inf/NaN area becomes 0), so up to 74% of the ration's dry matter got no land in a test run. It is also inflated by up to about 70% because feed demand is set to the larger of the energy-based and protein-based DMI. The N balance adds per-hectare terms to farm totals, has a scalar ifelse bug that throws away grazing-manure N, silently ignores fertiliser types missing from the fertiliser table, and calls outputs/inputs including losses "NUE". Water is reported in mm·ha labelled as m³ (10x too low), with no green/blue split and no season length. SOC applies IPCC factors to a measured stock of arbitrary depth, so any baseline shows a permanent loss, and tree-SOC is always zero. There is no uncertainty treatment. The scenario % changes flip sign for negative baselines and report 0% for undefined ones.

How numbers were checked

• The documented README run fails on the shipped inst/extdata/example_input.json. It also fails on data/qt_example.json and on data(mufindi), because crop_name is missing (see D-01).
• I ran the pipeline from source with Rscript (dplyr/tidyr/jsonlite; nothing installed) after a minimal patch in a scratch copy:
  • added crop_name
  • derived dry_yield/residue_dry_yield from fresh_yield, harvest_index and water_content using the commented formula at land_requirement.R:84-88
  • aliased fat_milkcontent, cp_lys_* and n_manure_content
  • set percentage_n = 34 where it was missing
• Runs: "EX" = example_input.json (Muheza) and "QT" = data/qt_example.json. The numbers quoted below come from these runs.

Findings table

Counts: Critical 2, High 18, Medium 13, Low 3 (36 in total).

Detailed findings

Land requirement

LU-01 - Silent zero land for feeds (Critical, CONFIRMED)

• Evidence:
  • area_total = feed_item_dm/(crop_yield*crop_removal) ... for source_type "Main", and feed_item_dm/(cr_yield*crop_residue_removal) for "Residue". Any other source_type gets 0 (L99-102).
  • Then mutate_if(is.numeric, list(~na_if(.,Inf))) %>% replace(is.na(.), 0) (L120-121).
  • QT run: cowpea (Main, main_product_removal=0, 7,480 kg DM) and rice straw (Residue, residue_dry_yield=0, 7,203 kg DM) both get 0 ha. That is 74% of the 19,765 kg DM ration. Only oats get land (0.169 ha).
  • EX run: maize stover (residue_removal=0, 2,043 kg DM, 12% of DM) gets 0 ha. "Concentrate (commercial)" has source_type "Purchased", so it always gets 0 ha.
  • The zero area then flows into erosion, N, water, SOC and GHG, and into every per-ha and per-kg indicator.
• Expected: fed DM must map to land or to an explicit "no land attributed / purchased with external footprint" category. A zero removal fraction for a fed item is logically inconsistent and should be an error.
• Impact: land and all area-scaled impacts are biased downward by an unknown, input-dependent amount (up to about 75% in the test). Scenarios that change a feed's removal fraction from 0 look like large land "savings".
• Fix:
  • Validate: stop if feed_item_dm > 0 and the denominator is ≤ 0.
  • Handle "Purchased" explicitly, with a default yield or an external land footprint factor.
  • Report the unallocated DM.

LU-02 - Land driven by the larger of energy- and CP-based DMI (High, CONFIRMED mechanism)

• Evidence:
  • dmi_s = ifelse(dmi_required_cp>dmi_required_e, dmi_required_cp, dmi_required_e) (energy_requirement.R:174), then feed_item_dm = fraction_dry_matter*dmi_s (land_requirement.R:93).
  • Every season and category in EX and QT is CP-limiting.
  • EX improved cow, S1: 2,794 vs 1,654 kg DM (+69%). Annual DMI is 9,714 kg, or 26.6 kg/d (4.4% of 600 kg BW). The energy-based 15.6 kg/d is realistic for 3,660 kg milk.
  • QT: +2% to +35%.
• Expected: land should follow actual or feasible intake (energy-based DMI capped by intake capacity). A CP deficit should be reported as a deficit, not converted into extra feed.
• Impact: land, water, N and erosion are overstated by up to about 70% on low-CP diets. Diet improvements that raise CP look like large land savings.
• Fix: use energy DMI for land, cap by an intake-capacity check (for example about 2.5-3.5% BW for cattle), and output the CP gap separately.

LU-03 - Allocation of crop land between food and residue feed (Medium, PLAUSIBLE)

• Evidence:
  • For residues, area_feed = area_total * cr_yield*res_removal/(crop_yield*crop_removal + cr_yield*res_removal) (L103-107). This is DM-mass allocation.
  • Main items: crop_residue_removal is forced to 0 (L97-98), so a grain or fodder item takes 100% of its crop's area.
  • If both "maize grain" (Main) and "maize stover" (Residue) are fed, the two areas are computed independently from different DM demands and never reconciled.
  • QT test with rice straw residue_dry_yield=5: straw gets 21% of 1.8 ha.
• Expected: a documented allocation choice. FAO LEAP feed guidelines use economic allocation by default, which typically gives stover a much smaller share than mass allocation. Also a single crop-level area when several co-products of one crop are fed.
• Impact: residue land share is likely overstated several-fold relative to economic allocation (my inference; depends on prices). There is a double-count risk for crops whose grain and residue are both fed.
• Fix: make the allocation method a parameter (mass/economic/none), and compute crop area once per crop with co-product shares.
• Refutation attempt: mass allocation is a defensible choice, but it is undocumented and non-configurable, so the finding stands.

LU-04 - Feed-source classification by name substring (Medium, CONFIRMED)

• Evidence: str_detect(feed, "OFR"), "OFC", "IP" in land_requirement.R:108-116 and repeated in the water, N, merge and GHG code. A name like "…PIPER…" or "SHIP…" would be classed as imported.
• Off-farm and imported land is assessed with the farm's own annual_prec, soil_*, R, K, et, cropland_system and so on (soil_health.R:26-27; water_requirement.R:33; soc.R:106).
• Fix: add an explicit feed_origin field. Use separate (or default regional) soil and climate parameters for off-farm land, or report off-farm land as area only.

LU-05 - Intercropping reduction (Low, PLAUSIBLE)

• area*(1 - intercrop*intercrop_fraction) (L100-102). Whether yields are sole-crop or intercrop yields, and whether the "saved" land belongs to the companion crop, is undocumented.
• A land-equivalent-ratio approach would be more transparent.

LU-06 - Land intensity metrics (Medium, CONFIRMED)

• total_land_requirement_ha_per_kg_fpcm <- safe_div(ha, kg)*1000 (differences.R:211) is ha per tonne, but it is plotted as "ha/kg" (compare_scenario.R:328-342).
• The same total land is divided separately by milk, by meat and by protein, so there is no co-product allocation. Per-kg-meat and per-kg-milk footprints both carry 100% of the land. The same applies to water, N, erosion and GHG.
• land_productivity() computes no land productivity at all. It returns livestock outputs.
• AME days use 2100 kcal (differences.R:195-196) but 2500 kcal in merge_outputs.R:530.
• Fix:
  • Relabel the unit, or report m²/kg.
  • Add biophysical or economic milk/meat allocation (IDF 2015 for dairy).
  • Use one AME constant.

Nitrogen balance

N-01 - Leaching term mixes per-ha and total units (Critical, CONFIRMED)

• Evidence:
  • out3a/b/c = (n_mineralized_kg_ha_year + fertilizer_rate + in2) * (...)/100 (L249-252).
  • n_mineralized and fertilizer_rate are kg/ha, but in2 = organic_n_kg_total is kg per farm (L245).
  • out3 is never multiplied by area_total, yet nout = out1+out2+out3+out4+out5 adds it to farm totals (L256).
  • By contrast, out4 uses organic_n_kg_per_ha and * area_total (L253-254).
  • EX: out3 = 2.55 kg for every feed, including zero-area feeds.
  • QT oats (0.169 ha): out3 = 46.4 kg total, where the correct value is about 0.169 x the per-ha rate using the per-ha manure.
• Expected: out3 = (Nmin + Nfert + Norg_per_ha) x frac x area_total, as out4 already does.
• Impact: leaching is underestimated by a factor equal to the area when area > 1 ha, and overestimated when area < 1 ha. The bias flows into nbalance, "NUE", the leaching area and the per-ha summaries.
• Fix: use organic_n_kg_per_ha and multiply by area_total.

N-02 - Leaching regression mis-parenthesised (High, PLAUSIBLE)

• Evidence: 0.021 * (annual_precipitation - 3.9) (L249), while the other clay classes use the form a*P + b (0.014*P + 0.71, 0.0071*P + 5.4).
• Expected: De Willigen (2000), as used in the NUTMON toolbox, gives 0.021*P - 3.9 (%) for clay < 35%.
• Impact: leaching fraction is +3.8 percentage points too high: 20.9% vs 17.1% at 1000 mm, and 8.3% vs 4.5% at 400 mm.
• Caveat: verdict is PLAUSIBLE because I checked against the published form from memory. The pattern of the other two clay classes strongly supports it.

N-03 - Scalar ifelse discards grazing-manure N (High, CONFIRMED)

• Evidence: animal_manure_grazing = ifelse(denom > 0, sum_n_content_manure_grazing*(...)/denom, 0) (L234-238). denom is a scalar, so ifelse returns only element 1 of the vector, recycled to every row. Checked: ifelse(5082>0, c(0,126.6,0), 0) returns 0.
• QT: 126.6 kg of grazing-manure N becomes 0 for all feeds. If feed 1 had had a non-zero value, that value would have been copied to every feed (over-count instead).
• The allocation key itself (harvested DM, including off-farm and imported feeds) is also questionable: grazing manure falls on grazed land, not on purchased-concentrate land.
• Fix: use if (denom > 0) ... else 0 (vector form), and allocate grazing manure only to on-farm grazed areas.

N-04 - Missing fertiliser types or N content silently give 0 N (High, CONFIRMED)

• Evidence: ifelse(feed_selected$urea == 0, 0, as.numeric(para$fertilizer[which(desc=="Urea"),]$percentage_n)/100), then !is.finite(...) becomes 0 (L116-120). If "Urea" is not in para$fertilizer, or percentage_n is absent (as in example_input.json, whose fertiliser row has no percentage_n), the N input is 0.
• QT: 400 kg urea/ha on oats gives in1 = 0.85 kg total (only N-solutions counted). Adding a Urea row gives in1 = 32.0 kg, so about 184 kg N/ha was ignored.
• ghg_emission.R:517-535 still counts the urea mass in fertiliser manufacturing emissions, while soil N2O (which uses fertilizer_rate from n_balance) does not.
• Fix: use a fixed internal N-content table (urea 46%, AN 33-34%, AS 21%, DAP 18%, etc.) with user override, and stop on unknown types.

N-05 - Manure and purchased organic N multiplied across feeds (High, CONFIRMED)

• Evidence: animal_manure_collected <- n_content_manure_collected * manure_fraction and organic_n_imported <- manure_fraction*(purchased_...) inside the per-feed loop (L148-155). There is no division by the sum of fractions and no area weighting.
• QT: 82.66 kg N is given to both oats and rice (fraction_as_fertilizer = 1 each), so 200% of the available manure N is applied. organic_n_kg_per_ha = 488 kg N/ha on oats.
• Fix: normalise fraction_as_fertilizer to sum to ≤ 1 across feeds (with a warning), and optionally distribute by area.

N-06 - "NUE" is not NUE (High, CONFIRMED)

• Evidence: nue = nout/nin (L257), where nout includes leaching, gaseous loss and erosion. Mining is nue > 0.9 and leaching is nue < 0.5 (L285-294).
• EX: NUE 24.2 and 54.4, so the pasture is classed as "mining". Higher leaching raises this "NUE", which pushes a plot toward "mining", the opposite of the intended signal.
• Expected: NUE = N in harvested products / N inputs (EU Nitrogen Expert Panel, 2015). The 0.5/0.9 thresholds come from that framework and only make sense with that numerator.
• Fix: nue = (out1+out2)/nin, and keep the balance separate.

N-07 - Two bases for soil_n (Medium, CONFIRMED)

• ntot_kg_ha_20cm = soil_n*20*soil_bulk*10 (L164). If soil_n is in g/kg, the correct stock is soil_n*0.2 m*1e4 m²*BD = 2000*BD*soil_n kg/ha, so the code is 10x low. If it is in %, the code is 100x low.
• out5 = soil_loss_plot(t)*soil_n*1.5 (L184) is kg N only if soil_n is in g/kg.
• EX: mineralisation 15.6 kg N/ha/yr vs 156 on the g/kg basis. Mineralised N feeds both the leaching and gaseous pools, so both are likely underestimated.
• Units are undocumented (input-contract.Rmd:49: "package basis").

N-08 - Negative inputs and losses (Medium, CONFIRMED)

• in4a = 2 + (P-1350)*0.005 (Smaling-type relation) is < 0 for P < 950 mm, which covers much of semi-arid SSA.
• out4 fraction -9.4 + 0.13*clay + 0.01*P is < 0 for, for example, clay 20% and P 600 mm.
• Fix: pmax(0, …).

N-09 - Cross-module N consistency (Medium, PLAUSIBLE)

• The N balance applies excreted N in collected manure (n_content_manure_collected) with no housing or storage losses. The GHG module models NH3 and N2O losses from the same manure, so the same N is both lost and fully applied.
• ghg_emission.R:374-375 sets conc_of_n_from_crop_residue_managed_soil <- sum(nitrogen_balance$conc_of_min_fert_n): fertiliser N is used as residue N.
• L373 reads nitrogen_balance$rough_of_n_from_crop_residue, which does not exist, so it returns 0.
• Main-product N removal uses main_n as a fraction of DM, while n_fixing uses 0.5 x shoot N (no roots, with a fixed %Ndfa). Both are undocumented.

Soil erosion

SE-01 - Erosion silently zero when slope inputs are missing (High, CONFIRMED)

• The nested ifelse returns NA_real_ for any unmatched class (L58), so soil_loss_ha_year is NA.
• n_balance then uses sum(..., na.rm=TRUE), which gives 0 (nitrogen_balance.R:179-182), and summaries do the same.
• EX: every feed has slope_desc: "", so LS = NA and erosion = 0 with no warning. QT has slope_length 0, which is accepted as the class "≤1".
• Fix: validate slope_desc against the allowed set, and stop or warn on NA.

SE-02 - Rainfall erosivity formula (High, PLAUSIBLE)

• R = 0.55*(annual_prec/rain_length) - 4.7 (L26) has no reference, and rain_length has no unit (6 and 5 in the examples, apparently months). This gives R = 98 (EX) and 160 (QT).
• Widely used SSA relations give far larger values:
  • Roose (1977): R ≈ 0.5·P in US units, often converted as about 0.5·P·1.73 in SI, i.e. about 970 at 1119 mm.
  • Hurni (1985, Ethiopia): R = 0.562·P - 8.12, i.e. about 620.
• Unless K is on a matching non-SI basis, soil loss is likely underestimated 5-10x.
• K = 0.19-0.25 in the examples looks like US-customary K (SI K is typically 0.01-0.07 t·h·MJ⁻¹·mm⁻¹).
• Verdict PLAUSIBLE: the bias could partly cancel through unit mixing, but that cannot be verified because units are undocumented.
• Fix: cite the formula, state the SI units for R and K, and consider regional R options.

SE-03 - LS table and P default (Medium, PLAUSIBLE)

• The LS lookup (L34-58) has no source and no unit for slope length (metres? 1-30 looks too short for field slopes), and uses four very wide steepness classes ("Hilly 5-20%" covers a 4x range in S).
• slope_p_factor = 0.11 appears for every feed in both examples, including pasture and purchased concentrate. That value corresponds to well-maintained terraces; P = 1 for no practice. Erosion is about 9x lower than without conservation practices.
• Fix: use a continuous LS (for example Wischmeier-Smith or Moore-Burch), document units, and default P to 1.

SE-04 - Erosion summary rows shifted (High, CONFIRMED)

• sources = c("total","on-farm","rough of","conc of","conc ip") (merge_outputs.R:620).
• erosion_t_soil_year = c(soil_loss_plot, rough_of_soil_loss, conc_of_soil_loss, conc_ip_soil_loss, farm_soil_loss) (L649-655).
• So the "on-farm" row shows off-farm roughage erosion, and so on. erosion_t_soil_ha then divides those shifted values by the mislabelled areas (L659-665).
• Fix: reorder so farm comes second.

SE-05 - Per-ha N summary summed across feeds (Medium, CONFIRMED)

• balance_N_kg_N_ha = sum_col(nitrogen_balance, "nbalance_feed_only_kg_n_ha") (merge_outputs.R:628-634) adds kg/ha values across feeds. It should be the total divided by the area.
• differences.R:231 uses nbalance_kg_n_total (whole crop, including the food share), while merge_outputs uses the feed-only balance. The two headline N balances differ in scope.

Water

W-01 - mm·ha reported as m³ (High, CONFIRMED)

• ET = kc_frac*et (mm) and water_use = ET*sum(area_feed) (L48-49). kc_water_use_m3_per_ha = feed_water_use/area_feed (L57) is therefore just Kc·ET0 in mm. EX pasture shows 815 "m³/ha" = 0.6 x 1359 mm.
• 1 mm over 1 ha = 10 m³, so every reported water volume and intensity (total_water_use, water_use_fpcm, _meat, _protein, and total_water_use_m3* in differences.R) is 10x too low.
• EX water_use_fpcm = 0.235 "m³/kg" should be 2.35 m³/kg under the model's own assumptions.
• Fix: multiply by 10.

W-02 - ET method incomplete (High, CONFIRMED)

• ET0 (para$et, annual) x the mean of Kc_ini/mid/late is applied as if every crop transpires all year. There is no crop season length, even though cultivation_period exists in QT feed items.
• No effective rainfall, no irrigation or blue-water term, no green/blue/grey separation. The wfp_green/blue/grey fields in feed_items are ignored.
• Livestock drinking and servicing water (livestock$water_requirement) is ignored, even though the function is called water_requirement.
• For annual crops (about 120-150 d) ET is overestimated about 2-3x. This partly offsets W-01, which is not a correction.
• fraction_of_precipitation_used_for_feed_production = ET/annual_prec is an area-weighted crop ET/P ratio, not a share of farm rainfall.
• Fix: use CROPWAT-style stage-length-weighted Kc x ET0 over the season, split into green water (min of ETc and Peff) and blue water, and add drinking water.

W-03 - Double allocation for residues (Medium, CONFIRMED)

• water_use is already computed on area_feed (the allocated share). Then feed_water_use = water_use*(1 - area_non_feed/area_total) (L50-51) applies the share a second time.
• QT test (rice straw with 21% share): 307 vs the expected 1460 mm/ha.
• total_water_use = ET*sum(area_feed) (L74) does not include the second allocation, so the per-feed and total water outputs disagree.

W-04 - OFC included in on-farm water (Medium, CONFIRMED)

• kc_water_use_on_farm = feed_water_use - roughages - ip_concentrates (L56) omits kc_water_use_of_concentrates.

W-05 - Plain-mean Kc (Low, CONFIRMED)

• (kc_initial + kc_midseason + kc_late)/3 (L46). FAO-56 weights each Kc by stage length.

Soil organic carbon and biomass

S-01 - Tree SOC always zero (High, CONFIRMED)

• biomass_calculation() returns a plain tibble (L141-144). soil_organic_carbon() reads biomass[["trees_non_feed_biomass"]]$c_increase_soc (soc.R:234), which is NULL, so the sum is 0. Verified in both runs.
• differences.R adds biomass$co2_increase (above-ground only), so the 25% below-ground/soil increment is dropped everywhere.

S-02 - Permanent SOC loss for steady-state land (High, CONFIRMED)

• ((SOC*FLU*FMG*FI) - SOC)/20 * area (soc.R:125-132), where SOC is the measured field stock.
• IPCC Tier 1 computes SOC0 and SOC(0-T) from SOCref x F for the current and previous management, and ΔC = (SOC0 - SOC(0-T))/D. Land under the same management for more than 20 years has ΔC = 0.
• The code instead reports an annual loss for every baseline as if native land had been converted within the last 20 years. It also applies the factors to a measured stock that already reflects current management, which double-counts the effect.
• QT: -2.93 t C/yr on 0.169 ha, i.e. -17 t C/ha/yr. That is physically implausible.
• Scenario differences are partly meaningful (ΔF x SOC/20) only when areas are equal.
• Fix: take baseline management as the prior state and scenario management as the current state (or explicit land-use-change inputs), and use SOCref from IPCC Table 2.3 or the measured 0-30 cm stock as SOC(0-T).

S-03 - Depth and units (High, CONFIRMED)

• soil_amount = 1e6*(soil_depth/100)*soil_bulk = 1e4·depth·BD. This is t/ha only if depth is in metres.
• EX soil_depth = 2 gives 2 m and 832 t C/ha. QT soil_bulk = 6 is impossible but accepted, giving 1440 t C/ha.
• IPCC factors are defined for 0-30 cm. At 2 m depth the flux is inflated about 3-7x.
• Fix: fix the depth at 0.3 m (or document cm and convert), and range-check BD (0.8-1.8 g/cm³) and soil_c.

S-04 - Non-IPCC grassland factor (High, CONFIRMED)

• stock_change_parameters.json has grassland.management."Medium": 1.5. IPCC 2006 Vol 4 Table 6.2 has nominally managed 1.0, moderately degraded 0.95-0.97, severely degraded 0.7 and improved 1.14-1.17. There is no 1.5.
• Selecting "Medium" gives +50%, i.e. (40 x 0.5)/20 = +1 t C/ha/yr.
• The cropland factors match IPCC 2006 Table 5.5. The 2019 Refinement revised several of them; which version to use is an author decision.

S-05 - Land categories (Medium, CONFIRMED)

• Cropland area = sum(area_feed) - grasses - tree_legume (L106-107). This includes off-farm and imported concentrate land at farm SOC and farm management.
• Grassland uses a fixed 40 t C/ha (L142, L176), not field_soc or IPCC SOCref.
• Tree-legume and tree-crop land is excluded from every SOC pool.
• Organic soils are hard-coded to 0 (L210-216).

S-06 - Doc mismatch (Low, CONFIRMED)

• output-contract.Rmd:258-262 lists field_soc and per-land-use rows. The function returns only total_annual_change_carbon_soils and total_change_co2_soils.

Scenario comparison and uncertainty

C-01 - % change arithmetic (High, CONFIRMED)

• Every indicator uses ifelse(!is.finite((S-B)/B*100), 0, (S-B)/B*100) (compare_scenario.R:49-150).
• A baseline N balance of -100 improving to -50 is reported as -50%, which reads as "worse".
• A baseline of 0 (common for SOC, erosion or meat after LU-01/SE-01) gives 0%.
• Fix: report absolute differences, use (S-B)/|B|, and return NA when B = 0.

C-02 - Zero-denominator intensities (Medium, CONFIRMED)

• safe_div(..., default = 0) (differences.R:39) and ifelse(!is.finite(...), 0, ...) (water_requirement.R:76-81) turn undefined intensities into 0. A dairy-only farm then shows "0 m³/kg meat" and "0 t CO2e/kg meat".

C-03 - No uncertainty or sensitivity (High, CONFIRMED by absence)

• No module carries ranges, Monte Carlo, or one-at-a-time sensitivity.
• Results depend strongly on unobserved or default inputs: yield and HI (land varies inversely), removal fractions (division, LU-01), Kc, R/LS/P (multiplicative), soil_n and BD (N-07, S-03).
• The comparison view presents single-point % changes with no indication of the noise floor.
• Fix: at minimum, ship a sensitivity wrapper over key parameters (±20-50%) and show ranges in compare_scenario.

Documentation

D-01 - Shipped examples do not run (High, CONFIRMED)

• The README "Standard Model Run" fails on inst/extdata/example_input.json, data/qt_example.json and data(mufindi) at unnest(cols = c(crop_name)).
• After adding crop_name, it also needs dry_yield, residue_dry_yield (EX), percentage_n (EX), fat_milkcontent, cp_lys_pregnancy, cp_lys_growth, n_manure_content, lw_gain_piglets and proportion_growth_piglets_milk.
• The key agronomic derivations (dry yield, residue yield, residue DM) live outside the package; the code at land_requirement.R:84-88 that did this is commented out. Those derivations therefore cannot be reviewed or tested.
• The commented formula uses the feed's dm_content for the main product's DM, which is wrong for residue items.

D-02 - Units unstated or inconsistent (Medium, CONFIRMED)

• input-contract.Rmd:42-53 gives soil_n, soil_c, soil_depth, et, rain_length and slope_length only on the "package basis".
• dm_content is 0.3 (a fraction) in EX but 89 (a percent) in QT, and the README says percent.
• There are no range checks.

Done well

• The IPCC 2006 cropland stock-change factors (FLU/FMG/FI) in stock_change_parameters.json are transcribed correctly and are looked up by explicit key, with a clear error on a missing key (soc.R:39-71).
• The NUTMON/De Willigen structure is recognisable: deposition 0.14·√P, clay-dependent leaching classes, and the gaseous-loss regression. The N balance also separates off-farm and on-farm sources.
• Land is disaggregated by source (on-farm, off-farm roughage, off-farm concentrate, imported), which is the right structure for footprint shifting, even though the classification is fragile.
• FPCM correction (0.337 + 0.116F + 0.06P) is applied consistently in the productivity and water modules.
• The DBH-based tree biomass allometry and the C-to-CO2 conversions are dimensionally consistent (kg to t via /1000, 0.48 C fraction, 44/12).

Open questions for authors

1. What are the intended units of soil_n, soil_c, soil_depth, soil_k_value, rain_length, slope_length and et, and what is the source of the R formula and the LS table?
2. Where (app or batch) are dry_yield, residue_dry_yield and crop_name derived, and can that code be moved into the package and tested?
3. Is mass allocation of crop land to residues a deliberate choice? Would you accept economic allocation (FAO LEAP) as an option?
4. Should land and feed demand follow energy-limited intake rather than max(energy, CP)?
5. Is SOC change meant to be baseline-to-scenario (management change) or native-to-current? Is the grassland "Medium" = 1.5 factor intentional?
6. What should a zero removal fraction on a fed item mean? And how should "Purchased" feeds get a land and water footprint?
7. Is any validation dataset (for example NUTMON farm studies or measured erosion plots in East Africa) available to benchmark the N balance and erosion magnitudes?
8. Should milk and meat footprints be allocated (IDF 2015 or economic) rather than each carrying 100% of the impacts?