Global Gridded Nitrogen Indicators: Influence of Crop Maps

Abstract Displaying Nitrogen (N) indicators on a global grid poses unique opportunities to quantify environmental impacts from N application in different world regions under a variety of conditions. Such calculations require the use of maps showing the geo‐spatial distribution of crop production. Although there are several crop maps in the scientific literature to choose from, the consequences of this choice for the calculation of N indicators still need to be evaluated. In this study we analyze the differences in results for N Use Efficiency (NUE) and N surplus calculated on the global scale using two different crop maps (SPAM and M3). For our calculations we used publicly available statistical and literature data combined with each crop map and carefully traced the origins of the differences in the results. Our results showed that the regions most affected by discrepancies caused by differences in crop maps (yields and physical area) are Central Asia and the Russian Federation, Australia and Oceania, and North Africa. However, we also found that the inclusion or exclusion of grass crops influences the results, as does the aggregation of crops to categories. Considering all these differences, we note that M3 seems to provide the more plausible results for the calculation of N indicators. Our analysis not only highlights the importance of determining the critical parameters for N indicator calculation, but also allows key parameters connected with N use and overuse to be identified on the global scale.


Introduction 14
This file contains detailed information on the calculation of all nitrogen (N) inputs and outputs to 15 and from soil surfaces of two crop maps (SPAM and M3) which were used for the calculation of 16 soil surface N budgets and NUEs. Most data used for these calculations was collected from several 17 sources from autumn 2018 to spring 2019. Data was processed upon receival. This processing 18 differed between the data types received and will be described in more detail throughout the 19 following sections. The programming language "Python" was used for all calculations.

22
Additional data related to this paper and datasets mentioned in this text can be accessed 23 through the IIASA Data Repository (DARE): https://doi.org/10.22022/air/10-2020.109..

25
To calculate nitrogen input to soil from manure excretion and management, data on total 26 livestock numbers of cattle and small ruminants (sheep and goats) and the distribution of 27 ruminant livestock production systems was taken from "Gridded Livestock of the World" (GLW) 28 which was a project initiated by the Food and Agriculture Organization of the United Nations 29 (FAO) and the Environmental Research Group Oxford (ERGO) (Robinson et al., 2014a(Robinson et al., , 2014b(Robinson et al., , 30 2014c(Robinson et al., , 2014d. The grid containing total animal numbers is produced by 31 combining several aspects. First a GIS map is developed containing sub-national statistical data 32 on Livestock numbers per administrative unit (FAO-GAUL). Then a suitability mask containing 33 information on elevation, slope gradient, protected areas and biophysical characteristics 34 (elevations higher than 4750 m above sea level, areas with a slope gradient higher than 40%, 35 protected areas and urban areas or areas permanently covered in snow or ice are excluded) was 36 developed. Data was taken from several models (GTOPO30 model, WDPA and GLC2000).

37
Additionally, a layer containing predictor variables such as length of plant growth (LPG), 38 population density and travel times to areas with a population of more than 50000 people, 39 temperature, precipitation, green-up, senescence was created as was a layer containing agro-40 ecological zones due to the circumstance that different predictor variables and different zones

46
The data was available as a 5' grid in tiff format and was converted to a 0.5-degree grid by 47 summing up the respective 36 cells between 0.5-degree latitude and 0.5-degree longitude. The 48 dataset using the dasymetric method was chosen, meaning that the distribution of livestock is 49 based on population, vegetation and topographic information.

51
Manure nitrogen excretion for cattle and small ruminants was calculated using the beforehand 52 calculated livestock grid and nitrogen excretion rates per GAINS (Greenhouse Gas -Air Pollution

53
Interactions and Synergies) region, taken from the GAINS model (International Institute for 54 Applied Systems Analysis AIR Group [IIASA AIR Group], 2018a) (1). Since nitrogen excretion is 55 higher for dairy cattle, a differentiation was introduced between dairy cattle and other cattle by 56 using a weighted average of milk cows per GAINS region calculated from FAOSTAT livestock data 57 available per country since this differentiation was not included in the gridded data. The milk 58 cow ratio was weighted using shares of manure nitrogen excretion of dairy cattle and other 59 cattle per country (FAO, 2019e). The procedure to calculate the average of milk cows per GAINS 60 regions was chosen for consistency reasons because data on milk yield influencing nitrogen 61 excretion rates was later taken from GAINS and was only available for each GAINS region (IIASA 62 AIR Group, 2018d).  with the difference being that in the latter more than 10 percent of the animal feed comes from 112 crop by-products or stubble or "more than 10 percent of the total value production comes from

119
The data on livestock systems was available as a 0.5' grid and was again converted to a 0.5-degree 120 grid. Whereas the ruminant livestock system grid showed which livestock system was dominant 121 in each grid cell, the monogastric livestock system grid displayed the total number of animals held 122 by each livestock system.

126
NAppl… nitrogen application for a livestock type in one grid cell

142
To calculate nitrogen input from manure on cropland, only manure managed and applied was 143 included due to the assumption of unmanaged animal droppings being excreted on pasture-and 144 rangeland. Managed manure on cropland was calculated using different fractions for different 145 countries following the procedure described in Liu et al. (2013) that allows to exclude the amount 146 of manure applied to pastureland. As no differentiation between different US states was made in 147 our calculations, we used the average of 87% of manure going to cropland. For developing 148 countries, it was assumed that 90% of manure is applied to cropland, as described by Smil (1999

173
Due to the existence of grass crops (alfalfa, clover, vetches, mixed grass, fornes (forage not 174 elsewhere specified) and grass nes (grass not elsewhere specified)) in the M3 crop map, the IFA

185
Although the global total of mineral fertilizer use found in FAOSTAT data matched the global total 186 found in IFA data quite well, there were significant differences between regional data which again 187 differed between M3 and SPAM. Due to these differences, each grid cell was updated so that the 188 sum of all grid cells belonging to a country would match the FAOSTAT data (FAO, 2019a).

190
FAOSTAT data is described to include mineral fertilizer used for pastures and aquacultures but

203
We compared M3 and SPAM calculations with IFA fertilizer to M3 and SPAM calculations using 204 the FAOSTAT adjusted fertilizer use to identify areas where the crop and region-specific allocation 205 of IFA data leads to an over-or underestimation of mineral fertilizer use (see Figure S1 for regional 206 differences in mineral fertilizer application between these two sources).

312 S5. Nitrogen Harvest 313
To calculate the nitrogen in harvest, information on total production per crop was taken M3 or 314 SPAM. Since the M3 data was representative for the year 2000, each grid cell was updated using 315 FAOSTAT production data for 2010 using the same procedure as for the update of harvested areas

343
However, since the data provided by Ramankutty et al. (2008) was from the year 2000, it was 344 updated to 2010 using FAO data (FAO, 2019d). Total areas were updated using FAOSTAT country 345 data and distributing it the same way harvested areas and yields were updated (see paper (3)).

347 S7. Complimentary Results and Details Used for Analysis 348
To arrive at the results presented in the manuscript, all budget terms and especially their crop 349 map dependent variation for each region but also country and crop category was analyzed as 350 described below.

353
Discrepancies in cropland area found in M3 and SPAM is high for most regions ( Figure S3). This

365
Manure that is managed and recycled to cropland is filtered to include only cells on which 366 cropland bigger than 5% of the land area is found to exclude outliers. As can be seen in Figure S3 367 and S4, not all cells where manure N can be found according to FAO GLW contain more than 5% 368 cropland. However, as can be seen in Figure S5

384
This was to be expected as the share of N deposition allocated to a country depends on that 385 country's share of cropland area (see 'Methodology'). However, as manure N, cropland allocation 386 also effects these results as cells where N deposition is shown are excluded when the respective 387 crop map allocates less than 5% cropland to this cell.

389
Harvested Area

390
Harvested areas taken from M3 and SPAM are globally very similar. On a regional basis, a higher 391 discrepancy of harvested area can be found in Western Industrial Europe ( Figure S6)

405
Western Industrial Europe and North Africa. These differences in crop category composition 406 mostly concern Residuals, where all pulses are included and M3 generally shows higher 407 production values for this category. The finer crop resolution for this crop category found in M3, 408 allows a more detailed allocation of BNF (see supplementary material S4 (Table S2 and Table S3)).

409
This means that using M3, rather high BNF rates (e.g. 88 kg/ha for peas and 115 kg/ha for 410 Fababeans) are assigned to crops that are not explicitly mentioned in SPAM but are expected to 411 be included in the crop category other pulses, which is assigned an average BNF rate of 23 kg/ha.

413
Crop production 414 More regional discrepancies and a slight global discrepancy can be found when looking at crop

446
The following figures compliment the results presented in the main script.

482
Additional material used to derive at the results described in the paper, focusing on the role of