Land management: data availability and process understanding for global change studies

In the light of daunting global sustainability challenges such as climate change, biodiversity loss and food security, improving our understanding of the complex dynamics of the Earth system is crucial. However, large knowledge gaps related to the effects of land management persist, in particular those human‐induced changes in terrestrial ecosystems that do not result in land‐cover conversions. Here, we review the current state of knowledge of ten common land management activities for their biogeochemical and biophysical impacts, the level of process understanding and data availability. Our review shows that ca. one‐tenth of the ice‐free land surface is under intense human management, half under medium and one‐fifth under extensive management. Based on our review, we cluster these ten management activities into three groups: (i) management activities for which data sets are available, and for which a good knowledge base exists (cropland harvest and irrigation); (ii) management activities for which sufficient knowledge on biogeochemical and biophysical effects exists but robust global data sets are lacking (forest harvest, tree species selection, grazing and mowing harvest, N fertilization); and (iii) land management practices with severe data gaps concomitant with an unsatisfactory level of process understanding (crop species selection, artificial wetland drainage, tillage and fire management and crop residue management, an element of crop harvest). Although we identify multiple impediments to progress, we conclude that the current status of process understanding and data availability is sufficient to advance with incorporating management in, for example, Earth system or dynamic vegetation models in order to provide a systematic assessment of their role in the Earth system. This review contributes to a strategic prioritization of research efforts across multiple disciplines, including land system research, ecological research and Earth system modelling.


74
We have entered a proposed new geologic epoch, the Anthropocene, characterized by a surging human 75 population and the accumulation of human-made artefacts resulting in grand sustainability challenges 76 such as climate change, biodiversity loss and threats to food security (Steffen et al., 2015). Finding 77 solutions to these challenges is a central task for policy makers and scientists (Reid et al., 2010;

124
For each management activity we compiled information on: the current global extent; past, ongoing and 2.1.
Forestry harvest 2.1.1. Extent and data availability Forests cover 32.7-40.8 Mkm 2 or 30% of the ice-free land surface and 2/3 -3/4 of global forests (26,4 Mkm²) are under some form of management FAO, 2010;Pan et al., 2013;Luyssaert 137 et al., 2014;Birdsey & Pan, 2015). Forest use reaches back to the cradle of civilization (Perlin, 2005;138 Hosonuma et al., 2012), while scientific forest management, i.e. management schemes that involve 139 careful planning based on empirical observations and forest-ecological process understanding (Mårald et 140 al., 2016), originated in the late 18 th century (Farrell et al., 2000). The share of managed forests and 141 management intensity are expected to increase further along with global demand for wood products 142 (Eggers et al., 2008;Meyfroidt & Lambin, 2011;Levers et al., 2014). Virtually all temperate and southern 143 boreal forests in the northern hemisphere are already managed for wood production (Farrell et al., 2000). 144 Northern boreal forest are at present largely unused for wood production  and could 145 become increasingly managed in the future due to growing global demand for wood products and 146 comparative advantages in boreal forestry compared to other regions (Westholm et al., 2015). Temperate 147 forests are mostly under some version of age class-based management. In contrast, wood extraction from 148 tropical forest often targets selected species, resulting in forest degradation. Significant parts of tropical 149 forest (5.5 Mkm 2 ) are in different stages of recovery from prior logging and/or agricultural use (Pan et al.,150 2.1.2. Effects of forestry harvest 160 The knowledge on biogeochemical effects of wood harvest is relatively advanced, although considerable 161 uncertainties still persist, and biogeochemical as well as biophysical effects are strong. Around 2000, 162 forest harvest amounted to 1 Pg C (carbon) yr -1 consisting of around 0.5 Pg C yr -1 for wood fuel and 163 another 0.5 Pg C yr -1 as timber (Krausmann et al., 2008;FAOSTAT, 2015). Forest harvest mobilizes 164 annually less than 0.5% of the global standing biomass (Saugier et al., 2001;Pan et al., 2011), but the flux 165 represents around 7% of the global forest net primary production (NPP) , reaching 166 15% in highly managed regions such as Europe (Luyssaert et al., 2010). Uncertainty ranges in wood flows 167 are large (Krausmann et al., 2008;Bais et al., 2015). In general, harvest reduces standing biomass 168 compared to intact forest (Harmon et al., 1990;McGarvey et al., 2014), with the notable exception of 169 coppices (Luyssaert et al., 2011). Soil and litter carbon pools generally decrease only slightly, but 170 deadwood decreases in managed forests by 95% compared to old-growth forests (McGarvey et al., 2014). 171 Nevertheless, the net effect of forest management on carbon stock reductions on the one hand, and 172 wood use for fossil fuel substitution on the other, remain unclear, due to complex legacy effects (Marland 173 & Schlamadinger, 1997;Lippke et al., 2011;Holtsmark, 2012). The effects of forest management on CH4 174 and N2O emissions are considered negligible, with the exception of fertilized short-rotation coppices 175 (Robertson et al., 2000;Zona et al., 2013). Predicted intensification of forest management by means of 176 short-rotation coppicing or total-tree harvest may require frequent fertilization, potentially resulting in 177 increased N2O emissions (Schulze et al., 2012).
subcanopy (Miller et al., 2007) which in turn could increase fire susceptibility. In temperate and boreal 187 sites, biophysical effects of forest management on surface temperature were shown to be of a similar 188 magnitude (e.g., around 2K at the vegetation surface) as the effects of land-cover changes (Luyssaert et 189 al., 2014). 190 2.2.
Tree species selection coniferous forests has warmed the lower boundary layer by 0.08K between 1750 and 2010 (Naudts et al., 240 2016 Grazing and mowing harvest is the most spatially extensive land management activity worldwide, 245 covering 28-56 Mkm 2 or 21-40 % of the terrestrial, ice-free surface, with a wide range of grazing intensity 246 (Herrero et al., 2013;Luyssaert et al., 2014;Petz et al., 2014;FAOSTAT, 2015). Grazing is one of the oldest 247 land management activities, reaching back 7k-10k years (Blondel, 2006;Dunne et al., 2012), and occurs 248 across practically all biomes: from arid to wet climates and over soils with varying fertility (Asner et al., 249 2004;Steinfeld et al., 2006;Erb et al., 2007). Livestock fulfils many functions beyond the provision of food 250 (FAO, 2011), but animal-based food production almost increased exponentially since the 1950s, due to 251 increasing population and more meat-and dairy-rich diets (Naylor et al., 2005;Kastner et al., 2012;Tilman 252 & Clark, 2014). These trends are expected to continue , but depending on the degree of intensification of 253 livestock production systems, the uncertainties on future net changes in grazing lands area are very large 254 Krausmann et al., 2008;Herrero et al., 2013), which is up to one third of the total global socio-economic 265 biomass harvest (Krausmann et al., 2008). Grazing is a key factor for many ecosystem properties, 266 including plant biomass and diversity. Grazing can both deplete and enhance soil C stocks, depending on 267 grazing intensity. For example, in arid lands, overgrazing is a pervasive driver of loss of soil function 268 (Bridges & Oldeman, 1999), resulting in reductions in soil organic carbon (SOC) and aboveground biomass 269 (Gallardo & Schlesinger, 1992;Asner et al., 2004). In semiarid regions, high grazing pressures could lead to 270 woody encroachment (Eldridge et al., 2011;Anadón et al., 2014), and thus to an increase in both above-271 and belowground carbon stocks. A global meta-analysis of grazing effects on belowground C revealed 272 large differences in the response of C3-and C4-dominated grasslands under different rainfall regimes 273 (McSherry & Ritchie, 2013). Globally, the response of plant traits to grazing is influenced by climate and 274 herbivore history (Díaz et al., 2007). At the same time, grazing can influence ecosystem C uptake in the 275 Arctic tundra, with implications for response to a warming climate (Väisänen et al., 2014). Incorporation 276 of current grazing and grazing history into climate models will improve predictions of terrestrial C sinks 277 and sources. 278 Forest grazing (e.g., reindeer grazing in the boreal zone) directly affects the understorey and indirectly 279 forest growth through nutrient export, recruitment, and the promotion of grazing tolerant species 280 (Adams, 1975;Erb et al., 2013b) but comprehensive assessments are lacking. The production of methane 281 is an important biogeochemical effect of ruminant grazers, strongly determined by the fraction of 282 roughage (grass biomass) in feedstuff (Steinfeld et al., 2006;Thornton & Herrero, 2010;Herrero et al., 283 2013), but large uncertainties related to quantities remain (Lassey, 2007). Soil compaction, induced, e.g., 284 by trampling, can contribute to anaerobic microsites, reducing the CH4 oxidation potential of the soil (Luo 285 et al., 1999). Nitrogen cycling is strongly affected by the addition of manure and urine (Allard et al., 2007). 286 The effect of animal waste N inputs interacts with poor drainage, influenced also by topography, to result 287 in localized greater N2O fluxes (Saggar et al., 2015). Biogeochemical effects of grazing are influenced by 288 livestock density. Some modelling and site-specific studies have found that a reduction of livestock 289 densities results in increased soil C storage and decreased N2O and CH4 (Baron et al., 2002;Chang et al., 290 2015). A study of year-round measurements of N2O in the Mongolian steppe found that while animal 291 stocking rate was positively correlated with growing-season emissions, grazing decreased overall annual N2O emissions (Wolf et al., 2010). Sites with little and no grazing showed large pulses of N2O release 293 during spring snowmelt compared to high grazing sites, suggesting that grazing may influence N cycling 294 response to changes in climate in high-altitude ecosystems. Biophysical effects of grazing mainly depend 295 on ecosystem type and soil properties. In local contexts, grazing has been reported to reduce plant 296 biomass; thus increasing albedo by about 0.04 compared to unmanaged grassland (Rosset et al., 2001;297 Hammerle et al., 2008). However, the effect of soil exposure resulting from canopy decreases is 298 ambiguous, resulting in an albedo reduction on dark soils (Rosset et al., 1997;Fan et al., 2010), and in an 299 albedo increase on bright soils (Li et al., 2000). Reindeer grazing has been reported to reduce albedo due 300 to a reduction of the light-colored lichen layer (Cohen et al., 2013). Reductions in roughness length due to 301 grazing are expected to have a small affect on turbulent fluxes (i.e. surface fluxes of energy, moisture and 302 momentum), but can lead to enhanced soil erosion (Li et al., 2000). The observed effect of mowing on 303 the cumulative evapotranspiration was small (10% increase, about 40 mm), although sufficient to 304 decrease soil water content in a managed field (Rosset et al., 2001). The integrated climate effect from 305 excluding grazing by bison in the Great Plains was modelled to be a 0.7K decrease in maximum 306 temperatures and a small increase in minimum temperatures (Eastman et al., 2001). 307 alteration of short cultivation and long fallow periods, which was a particularly widespread form of 319 cropland management in many regions of the world (Emanuelsson, 2009) which illustrates the highly 320 interconnected nature of management and land-cover change. Today, this form of land use is declining at 321 the global scale, although it remains important in many frontier areas characterized by e.g. unequal or 322 insecure access to investment and market opportunities or in areas with low incentives to intensify 323 cropland production (van Vliet et al., 2012). Cropland expansion is tied to human population growth, but 324 moderated by technological development that allowed for substantial yield increases per cropland area, 325 in particular after 1950 (Pongratz et al., 2008;Kaplan et al., 2010;Ellis et al., 2013;Krausmann et al., 326 2013). The dynamics of cropland expansion and contraction in different regions of the world are caused 327 by complex interactions between endogenous factors such as population dynamics, consumption 328 patterns, technologies and political decisions, and exogenous forces related to international trade and 329 other manifestations of globalization, in interplay with intensification dynamics (Krausmann et al., 2008(Krausmann et al., , 330 2013Meyfroidt & Lambin, 2011;Kastner et al., 2012;Kissinger et al., 2012;Ray et al., 2012;Ray & Foley, 331 2013). Cropland shows the highest land-use intensity, compared to grazing land or forest, in terms of 332 inputs to land (capital, energy, material) as well as outputs from land (Kuemmerle et al., 2013;333 Niedertscheider et al., 2016). The spatial extent of cropland is probably the best-described land-use 334 feature at the global scale, with many datasets existing (see Table 2).).. Nevertheless, major uncertainties 335 remain related to cropland patterns in some world regions, particularly across large swaths of Central, 336 removal into a dynamic vegetation model significantly increased the amount of historical land-use change 346 based C emissions estimated by the most common agricultural scenarios, which do not include 347 management information (Pugh et al., 2015). . Cropland harvest amounted to 3.2 PgC yr -1 in 2000, around 348 half of total biomass harvest, or around 5% of global terrestrial NPP (Wirsenius, 2003;Krausmann et al., 349 2008). Primary products (e.g. grains) cover 45%, secondary products (e.g. straw, stover and roots) 46%, 350 and 9% are fodder crops. The majority of cropland produce is used directly as food, but a non-negligible 351 amount of around 1.3 PgC yr -1 is used as feed for livestock (fodder crops and concentrates). In 2004, crop 352 harvest for bioenergy amounted to 1.6 EJ yr -1 from agricultural by-products and 1.1 EJ yr -1 from fuel crops, 353 which is roughly equivalent to 0.043 and 0.03 PgC yr -1 , respectively (Sims et al., 2007). 0.7 PgC yr -1 of 354 secondary products remain on site, possibly ploughed to the soil or burned subsequently (Wirsenius, 355 2003;Krausmann et al., 2008). Cropland systems, mainly consisting of annual, herbaceous plants, usually 356 contain little carbon in vegetation and soil per m² (Saugier et al., 2001). Thus, crop residues left on field 357 add only small amounts of carbon to soil pools (Bolinder et al., 2007;Anderson-Teixeira et al., 2012). 358 Information on local impact of crop residue removal (or retention) on GHG emissions, soil carbon and 359 yields is available (Bationo & Mokwunye, 1991;Lal, 2004Lal, , 2005Lehtinen et al., 2014;Pittelkow et al., 360 2015). Also national data on emissions from crop residues is available (FAOSTAT, 2015). However, the lack 361 of primary data such as from long-term field studies and the use of crude factor introduces large 362 uncertainties related to estimates of crop residue management effects. Large uncertainties also relate to 363 the contribution of crop residue, including roots and exudates, to the build-up of soil organic carbon 364 (Bolinder et al., 2007;Kätterer et al., 2012). This limits our ability to assess its impact at the global scale. 365 With current policies for increasing biomass use for bioenergy, crop residue harvest can result in 366 additional SOC losses, proportional to residue removal (Gollany et al., 2011). Synergistic effects are also 367 frequent: Negative effects of crop residue removal on soil carbon are enhanced with N fertilization (Smith 368 et al., 2012). 369 Biophysical effects of crop harvest are well documented, in particular related to changes in albedo, 370 roughness and evapotranspiration. When crops are harvested, soil becomes exposed and albedo (Davin et 371 al., 2014) as well as roughness drop (Oke, 1987). Evapotranspiration was estimated to decrease by 23% in on the presence and intensity of post-harvest management practices, e.g. ploughing, tillage, after-374 cropping or mulching. Evapotranspiration partly depends on soil water holding capacity, which in turn is 375 affected by tillage (Cresswell et al., 1993) and crop residue management (Horton et al., 1996). Crop 376 residue management is an important factor, but information is scarce. Compared to bare soil, crop 377 residues reduce extremes of heat and water fluxes at the soil surface when crops residues are left on-site 378 (Horton et al., 1996;Davin et al., 2014). 379

2.5.
Crop species selection sedentary subsistence, with species selected according to human needs (e.g. food, health, stimulants, 385 fiber). Recently, biomass energy production from dedicated oil, starch or sugar plants, but also fast-386 growing grasses, has increased rapidly and is anticipated to accelerate in the future (Beringer et al., 2011;387 Haberl et al., 2013). Data availability for recent crop type distribution is similar to that on cropland 388 harvest, however, spatially explicit time series and global data on inter-annual dynamics, such as 389 rotational schemes, are lacking (Table 1; SI). 390

391
While Information on biophysical effects of crop species selection is available, much less is available on 392 biogeochemical effects. Both effects seem to be relatively weak in comparison to other management 393 types, probably also owing to comparatively small knowledge base. In particular, effects of species 394 selection on individual carbon pools are largely unknown. Crop type is known to affect SOC accumulation 395 and decomposition rates, and the allocation of carbon to shoots or roots. For example, shoot to root 396 ratios were found to increase in the order natural grasses < forages < soybean < corn (Bolinder et al., 397 2007). A shift from annual to perennial crops and the introduction of cover crops can significantly increase 398 belowground biomass under perennial bioenergy grasses (switchgrass, Miscanthus, native prairie mix) 400 compared to a corn-corn-soy rotation agricultural system. Increasing crop rotational diversity can also 401 positively influence SOC storage (McDaniel et al., 2013;Tiemann et al., 2015). Strong difficulties to assess 402 species-selection effects arise from legacy effects, which render systematic long-term studies necessary. 403 For instance, in a 22 year experiment, comparing maize, wheat and soybean cultivation, SOC content was 404 found to be about 7% higher under soybean as compared to wheat and maize. Other GHG emissions are 405 also crop-specific. For example, N2O emissions factors from fertilization vary from 0.77% of added 406 nitrogen for rice to 2.76% for maize (Stehfest, 2005). Effects of crop species on CH4 balances are less clear, 407 except for paddy rice, where high emissions occur. 408 Cropland albedo varies significantly among crops, ranging between 0.15 for sugarcane and 0.26 for sugar 409 beet, with significant variations even among related species, e.g. 0.04 higher for wheat compared to 410 barley (Piggin & Schwerdtfeger, 1973;Monteith & Unsworth, 2013). Even within a species, cultivars show 411 differences in albedo of up to 0.03 units. Differences in planting and harvesting dates for different crop 412 species and cultivars, and associated changes in leaf phenology, also affect biophysical conditions. More 413 productive cultivars and earlier planting dates lead, for example, to an earlier harvest and to enhanced 414 exposure of dark soil in the fall, resulting in lower end-of-season albedo and an increase in net radiation 415 (Sacks & Kucharik, 2011). Whether the end-of-season albedo increases or decreases depends on the ratio 416 between the soil and vegetation albedo. In many regions of the world soil albedo is lower than plant 417 albedo, but not in some (semi-)arid regions where soils may have a similar or even higher albedo than the 418 vegetation. Similarly, water-use efficiency and evapotranspiration between crop species differs widely 419 (Yoo et al., 2009), even for the same cultivars (Anda & Løke, 2005). Although crop heights are limited, 420 roughness can be expected to vary similarly as for grasslands (Li et al., 2000). 421 2.6.
N-Fertilization of cropland and grazing land 422 2.6.1. Extent and data availability 423 Fertilizers are used to enhance plant growth by controlling the level of nutrients in soils. Nitrogen (N) 424 plays a prominent role as one of the most important plant nutrients which is often limited in agriculture (LeBauer & Treseder, 2008). N-Fertilizers are either organic fertilizer derived from manure (livestock 426 feces), sewage sludge or mineral fertilizer. Reactive nitrogen was a scarce resource in preindustrial 427 agriculture, mainly in the form of animal manure, leading to sophisticated management schemes to 428 balance the N-withdrawals associated with harvest (Sutton et al., 2011). The invention of the Haber-Bosch 429 process and the availability of fossil energy triggered a process of innovation in agriculture with surging 430 levels of N-fertilization. Today, the transformation of N to reactive forms and its use as fertilizer on 431 agricultural lands represent one of the most important human-induced environmental changes (Gruber & 432 Galloway, 2008;Davidson, 2009). The use of synthetic fertilizers is projected to increase in response to 433 growing human population, increases in food consumption and crop-based biofuel production (IFA, 2007). 434 Practically all croplands are under N-fertilization schemes, with strong regional variations in intensity of 435 input volumes and composition (Gruber & Galloway, 2008;Vitousek et al., 2009), but also grasslands and 436 forests (the latter not discussed here) can be under N-fertilization schemes. The highest cropland 437 fertilization levels surpass 200 kg N ha -1 yr -1 e.g. in the Nile delta and 90 kg N ha -1 yr -1 in New Zealand 438 (Potter et al., 2010;Mueller et al., 2012), and 14% of cropland are fertilized with levels above 100kgN ha -1 439 yr -1 . Globally, much lower intensity level prevail, 59% of the global cropland area show application rates 440 below 5010kgN ha -1 yr -1 , and around one quarter of global croplands show fertilization rates below 10kgN 441 ha -1 yr -1 (SI). Grasslands often do not receive any N fertilization (except for manure inputs from grazing 442 animals) but some grasslands are also heavily fertilized with rates put to 100 (Haas et al., 2001) and even 443 300 kg N ha-1 yr-1 (Flechard et al., 2007). Globally, animal manure makes up approximately 65% of N 444 inputs to cropland (Potter et al., 2010), and is the dominant N source in the Southern hemisphere. 445 Regionally, mainly in concentrated industrial livestock production, manure availability can exceed local 446 fertilizer demand, resulting in substantial environmental problems such as groundwater pollution 447 (IAASTD, 2009). The status of data availability is intermediate. National time series data as well as 448 spatially-explicit assessments are available (Table 1), but characterized by large gaps and uncertainties, 449 particularly relating to spatial patterns and livestock manure. Global data on N fertilization of grasslands, 450 albeit a wide-spread activity in many region, is scarce and crude-model derived (SI).

452
The biogeochemical effects of N fertilization, of both cropland and grazing land, are strong and relatively 453 well documented and understood. Cropland fertilization is a strong driver of anthropogenic GHG 454 emissions, in particular of nitrous oxide (N2O), nitric oxide (NO) and ammonia (NH3). A typical fertilized 455 cropland emits 2-3 times more nitrogen than the approximately 0.5 kg N ha -1 yr -1 emitted under non-456 fertilized conditions (Stehfest & Bouwman, 2006), while fertilized grasslands emit 3-4 times more N2O 457 than unfertilized ones (Flechard et al., 2007). The global N2O emissions on fertilized croplands and grazing 458 lands sum to 4.1 to 5.3 Tg N yr in the beginning of the century (Stehfest & Bouwman, 2006;Syakila & 459 Kroeze, 2011), one fifth of it occurring on grazing lands (Stehfest & Bouwman, 2006). Beyond N 460 application rates, N2O emissions are determined by crop type, fertilizer type, soil water content, SOC 461 content, soil pH and texture, soil mineral N content and climate. NH3 emissions are determined by 462 fertilizer type, temperature, wind speed, rain and pH (Sommer et al., 2004). Acidification from N fertilizers 463 can lead to increased abiotic CO2 emissions from calcareous soils (Matocha et al., 2016). Fertilization also 464 affects ecological processes, including productivity, C inputs to the soil, and SOC storage in croplands by 465 affecting the shoot to root ratio (Müller et al., 2000), influences the efficiency of photosynthesis, and 466 ultimately the exchange of C between land and the atmosphere, as fertilization studies in forests reveal 467 (Vicca et al., 2012;Fernández-Martínez et al., 2014). Long-term studies from Sweden suggest that each kg 468 N fertilizer increased SOC stocks by 1 to 2 kg (Kätterer et al., 2012). Fertilization effects on SOC were 469 particularly strong with organic fertilization (Körschens et al., 2013). Fertilization also increases 470 atmospheric N and thus deposition (Ciais et al., 2013a) and results in N leakage (Galloway et al., 2003). 471 Fluxes of total anthropogenic N from land to the ocean via leaching from soils and riverine transport have 472 been estimated at 40-70 Tg N yr −1 (Boyer et al., 2006;Fowler et al., 2013). Increased nutrient input to 473 rivers and freshwater systems impact on water quality and biodiversity (Settele et al., 2014)and the 474 subsequent increased nutrient loading of coastal oceans is believed to be the primary cause of hypoxia 475 (Wong et al., 2014). 476 Few direct effects of fertilization on biophysical properties -besides indirect effects of changes in crop 477 biomass or height due to altered productivity -have been documented, and the magnitude of impacts is probably not strong. Forest-site studies suggest that enhanced leaf nitrogen concentrations increase 479 canopy albedo (Ollinger et al., 2008), presumably through changes in canopy structure rather than in leaf-480 level albedo (Wicklein et al., 2012). Also, nitrogen fertilization improved grassland water use efficiency but 481 simultaneously increased absolute evapotranspiration, and thus the latent heat flux, from 280 to 310 mm 482 (Brown, 1971;Rose et al., 2012). N-driven increases in plant height and leaf mass will be reflected in 483 With the mechanization of agriculture, arable land became regularly tilled to suppress weeds and 487 enhance soil structure and nutrient availability. Archeological findings suggest that humans manipulated 488 soil structure through some form of tillage with ards and hoes already some 4500 years ago (Postan et al., 489 1987). From the 1950s, with the advent of modern herbicides no-till systems became more prominent, 490 mainly in the U.S. (IAASTD, 2009). To date, continental or global data on the area, distribution or intensity 491 of tillage is sparse. It can be assumed, however, that all croplands that are permanently used are regularly 492 tilled, except for (1) perennial crops, which cover approximately 10% of cropland area or 1.5 Mkm² 493 (FAOSTAT, 2014) and (2) no-till agriculture (or reduced tillage) on 1.11 million km 2 (Derpsch et al., 2010), 494 which is around 8% of the global arable land. No-tillage systems are particularly widespread in Brazil and 495 the U.S., where 70% respectively 30% of the total cultivated area is under no-tillage management. 496 However, most of these lands are not permanently under zero tillage but are still ploughed from time to 497 time. Global maps of zero-tillage are missing, as do maps on qualitative aspects of tillage, such as type and 498 depth of tillage. 499 2.7.2. Effects of tillage 500 Tillage effects remain weakly understood. Ploughing of native grassland upon conversion to croplands 501 drastically depleted SOC (Mann, 1986). Such ploughing disrupts aggregate structure, aerating the soil and 502 activating microbial decomposition (Rovira & Greacen, 1957). No-tillage practices promised to 503 significantly mitigate carbon emissions from SOC (IAASTD, 2009). However, some evidence is available indicating that on most soil types and in most climate regimes adoption of no-tillage practices after 505 tillage-based management does not significantly increase SOC stocks (Baker et al., 2007;Hermle et al., 506 2008;Govaerts et al., 2009), but there is still controversy on this aspect of the adaption of no-tillage 507 (Powlson et al., 2014(Powlson et al., , 2015Neufeldt et al., 2015). These findings and studies looking deeper into the soil 508 profile suggest that conventional tillage may not result in net losses of soil C, but rather results in a 509 redistribution of carbon in the soil profile. Other findings are inconclusive, e.g. on the impacts of 510 conservation tillage on productivity of cropland. While no-tillage is often reducing crop yields, other 511 activities such as crop residue management of crop rotations play a decisive role for the overall effects 512 (Pittelkow et al., 2015). Other key factors are the depth and type of tillage, which vary worldwide. 513 Evidence on the effects of no-tillage on N2O emissions is site-specific and inconclusive (Rochette, 2008). A 514 recent meta-analysis reported that no-till reduced N2O emissions after 10 years of adoption and when 515 fertilizer was added below the soil surface, especially in humid climates (van Kessel et al., 2013). No-tillage 516 generally reduces soil erosion, but regional-to global-scale effects are uncertain, because most eroded 517 soil carbon is deposited in nearby ecosystems (Van Oost et al., 2007). 518 Tillage has small biophysical effects. Through a decreased soil water holding capacity, excess tillage 519 increased the shortwave albedo from 0.12 under minimum tillage to 0.15 under excess tillage (Cresswell 520 et al., 1993). Furthermore, soil water holding capacity, which is affected by tillage (Cresswell et al., 1993) 521 and crop residue management (Horton et al., 1996), also controls evapotranspiration. Soils covered with 522 crop residues after harvest evaporate less than tilled soils (Horton et al., 1996)  Paddy rice cultivation is particularly important in East, South and Southeast Asia where its history reaches 536 back at least 6k years, originating probably in China (Cao et al., 2006;Fuller, 2012;Kalbitz et al., 2013). 537 Small-scale crop irrigation dates back to the origins of agriculture (Postel, 2001), while large-scale 538 irrigation is a recent outcome of the Green Revolution. Nowadays, 30% of the global wheat fields (0.7 539 Mkm 2 ), 20% of the maize fields (0.3 Mkm 2 ), and half of the global citrus, sugar cane, and cotton crops are 540 irrigated . Moreover, cropland irrigation accounts for approximately 70% of global 541 freshwater consumption (Wisser et al., 2008). Rice cultivation requires a particularly intensive form of 542 irrigation, involving regular flooding of fields for longer periods (Salmon et al., 2015). Irrigation datasets 543 exist and are relatively robust, in particular for rice, but large similar problems of uncertainties prevail as 544 with cropland maps (see above; Salmon et al., 2015). Furthermore, Earth system effects depend on 545 actually applied irrigation, which is much less documented than area equipped for irrigation. 546 is limiting plant growth, in particular in semi-arid and arid regions. Irrigation affects soil moisture, 550 temperature, and N availability, which are all drivers for the production and evolution of GHG emissions 551 from soils (Dobbie et al., 1999;Dobbie & Smith, 2003). Accelerated soil carbon decomposition under 552 irrigation is typically offset by higher NPP and greater carbon inputs into the soil (Liebig et al., 2005;Smith 553 et al., 2008). A global review of irrigation effects concluded that irrigated cropping systems in arid and the size of the effect is highly dependent on climate and initial SOC content (Liebig et al., 2005;Trost et 556 al., 2013). Furthermore, irrigated soils are more often affected by anoxic soil conditions which in turn 557 favour denitrification and N2O production, especially when fertilized (Verma et al., 2006). This is 558 particularly the case in paddy fields, where emission factors range between 341 and 993 gN ha -1 , 559 depending on the length of the irrigation scheme, corresponding to irrigation-induced emission factors of 560 0.22-0.37% of the added nitrogen (Akiyama et al., 2005). Soil texture and climate can mediate these 561 effects of irrigation on biogeochemical processes, but the statistical evidence is weak (Scheer et al., 2012;562 Trost et al., 2013;Jamali et al., 2015). According to the review by Trost et al. (2013) there is no consistent 563 effect of irrigation on N2O emissions. The capacity of soils to oxidize atmospheric CH4 may be reduced 564 under irrigation (Ellert & Janzen, 1999;Sainju et al., 2012). Irrigated rice fields alone are emitting 565 approximately 30-40 TgCH4 yr -1 (Kirschke et al., 2013). 566

Effects of cropland irrigation
Changes in ecosystem water availability significantly alter the surface albedo and roughness through their 567 impact on plant growth and ecosystem conditions (Cresswell et al., 1993;Wang & Davidson, 2007). 568 Because water surfaces have lower reflectance, flooding reduces the albedo of dry soil of about 0.2 to a 569 level of 0.03 -0.1 (Kozlowski, 1984). A modelling study over the Great Plains in the USA has shown that 570 irrigation can alter atmospheric circulation and precipitation patterns (Huber et al., 2014). Despite its 571 surface cooling effect (about 0.8 K), irrigation was simulated to increase global radiative forcing in the 572 range of 0.03 to 0.1 Wm -2 (Boucher et al., 2004). 573

2.9.
Artificial drainage of wetlands 574 2.9.1. Extent and data availability 575 Drainage aims at improving soil characteristics for agriculture and at facilitating the use of machinery. 576 While historically drainage relied on channels and sewers, currently prevailing drainage systems often also 577 use subsurface hollow-pipes or similar technologies (FAO, 1985). Approximately 11% of global croplands, 578 or 1.6 Mkm², are subject to artificial drainage (Feick et al., 2005), but the strongest biogeochemical and 579 biophysical effects of drainage are expected when wetlands are drained, e.g., peatlands, inland flood 580 plains, coastal wetlands, or lakes. Wetlands are estimated to cover 5.3-26.9 Mkm 2 (Melton et al., 2013), of which 0.18 Mkm 2 are probably drained (SI), but data are scarce. Wetland drainage dates back for 582 millennia, e.g., in lowland Europe (Emanuelsson, 2009), but accelerated especially between 1830 and 583 1950 with the drainage of over 30% of the Scandinavian peatlands and large-scale drainage projects in 584 Russia, Canada and the US (Brinson & Malvárez, 2002). Despite attempts for wetland conservation (see 585 e.g. (Dugan, 1990), or the international RAMSAR treaty (www.ramsar.org), large-scale new drainage 586 installation is still ongoing (Brinson & Malvárez, 2002;Lähteenoja et al., 2009), in particular in Asia , for 587 instance in relation with palm oil expansion (Davidson, 2014). Consistent data on wetland drainage are 588 practically inexistent. 589

590
The biogeochemical and biophysical effects of drainage are not well documented, partly because most 591 studies aim at assessing the effects of associated land use and cover changes, rather than the effects of 592 drainage itself. While the sparse evidence suggests that biogeochemical effects are strong, biophysical 593 effects are probably only of medium size. On forest sites, drainage can increase biomass through 594 increased NPP (Trettin & Jurgensen, 2003). Drained peatlands are, however, hotspots of GHG emissions 595 (Hiraishi et al., 2014). When expressed in units of radiative forcing, the soil emissions of CO2, CH4 and N2O 596 in drained forested peatlands decrease or even offset the carbon sink in aboveground biomass (Schils et 597 al., 2008). The cultivation of drained wetlands leads to rapid losses of large stocks of soil carbon 598 accumulated over thousands of years (Drösler et al., 2013). A 50% increase in fluvial carbon losses 2005) which in turn results in lower minimum night-time temperatures (Marshall et al., 2003). The 609 relationship between evapotranspiration and night-time temperatures has been modelled (Venäläinen et 610 al., 1999;Marshall et al., 2003), suggesting considerable temperature drops of up to 10 K. Although the 611 direct effect of drainage on albedo and roughness length is not clear, increasing plant growth is likely to 612 increase the surface roughness and decrease spring-time albedo (Lohila et al., 2010). 613 2.10. Fire management 614 2.10.1. Extent and data availability 615 Fire began to be used by humans around 50k to 100k years ago (James, 1989;Bar-Yosef, 2002), and while 616 it is unclear when it was first employed to shape ecosystems, today is a versatile land management tool 617 (Lauk & Erb, 2009;Bowman et al., 2011), e.g., for plant selection or agricultural waste removal. Note that 618 fire use for land clearing, including swidden agriculture, represents a land-cover change and is thus not 619 discussed here. Fire occurs naturally in most ecosystems, while in many regions natural fires today are 620 suppressed (Hurtt et al., 2002;Andela & van der Werf, 2014), population density playing an important 621 role (Archibald et al., 2009). Yet, prescribed fires are, next to mechanical thinning, a widespread practice 622 to reduce or retard wildfire spread and intensity (Fernandes & Botelho, 2003). As fire frequency is 623 expected to increase in the future due to climate change, fire prevention might increase in importance. 624 Globally, the annual area burned through human-induced and natural fires is estimated at 3.0-5.1 Mkm² 625 in the last decades (Wiedinmyer et al., 2011;Giglio et al., 2013). The proportion of human-induced fires is 626 difficult to assess (van der Werf et al., 2008), and in particular the ratio between fires that lead to land-627 cover change and fires used to manage ecosystems is unknown. No specific global, spatially explicit 628 information on fire as a management tool (including fire prevention and prescribed fires), exists (Table 1). 629

630
The effects of fire management on biogeochemical and biophysical properties of ecosystems are well-631 documented and mainly biogeochemical. However, these studies do not systematically separate natural 632 from anthropogenic fires. Globally, fire-induced carbon emissions are estimated to range from 1.6 to 2.8 2009). The large uncertainties owe to large differences in the assumptions of fuel loads (Granier et al., 635 2011) and the difficulty to assess smaller fires. Fire emissions also include aerosols and trace gases (Akagi 636 et al., 2011), which impact atmospheric chemistry and significantly contribute to overall aerosol direct 637 and indirect radiative forcing (Ward et al., 2012). Fires result in short-term carbon losses from the direct 638 combustion of biomass and lagged losses from the decomposition of dead biomass (Hurteau & Brooks, 639 2011). Fires affect nutrient supply (Mahowald et al., 2005) and soil carbon dynamics (Knicker, 2007). The 640 storage of carbon in long-lived pools such as SOC is influenced by fires through the accumulation of char 641 or pyrogenic carbon (Santín et al., 2008). Repeated burning in the process of agricultural land 642 management (e.g. residue burning) reduces carbon accumulation rates (Zarin et al., 2005). The effects of 643 fire suppression (Archibald et al., 2009;Wang et al., 2010) or management activities that indirectly alter 644 fire regimes (van Wilgen et al., 2014), however, represent a knowledge gap. Despite the direct carbon 645 stock increases resulting from fire prevention and similar measures (Bond-Lamberty et al., 2007), such 646 activities can lead to greater future ecosystem carbon losses through the accumulation of large fuel loads 647 that potentially increase the risk of severe fires (Hurteau & Brooks, 2011;O'Connor et al., 2014). Indirect 648 biogeochemical effects of fire, e.g. post-fire degradation, are not systematically quantified. 649 Various observational studies scrutinized the effects of specific fires on surface energy fluxes. Immediately 650 after a boreal forest fire, albedo decreased to 0.05, increasing to 0.12 over a period of 30 years and then 651 averaging to 0.08 similar to a pre-fire state (Amiro et al., 2006). Effects of fire aerosols might also be 652 important, although uncertainty is high (Landry et al., 2015). Also latent heat energy fluxes and overall 653 radiative forcing are affected (Randerson et al., 2006). Randerson et al. (2006) estimated a radiative 654 forcing of -5 W/m 2 immediately after a boreal forest fire, which remained high at -4 W/m 2 over 80 years 655 after the fire. In a savannah, a halving of the albedo (0.12 to 0.07) was observed, followed by a recovery 3. Discussion and conclusions 659 The ten land management practices selected for this review affect a considerable proportion of the global 660 terrestrial surface (Fig. 2). Grazing and forest harvest and tree species selection are largest in terms of 661 extent, covering almost 60% of the terrestrial, ice-free global land surface. However, the importance of a 662 management practice depends not only on its spatial extent and effects on the Earth system, but also on 663 the intensity of management, which differs markedly in extent across management practice (Fig. 2). 664 Management intensity has shown pronounced increases at the global scale in recent decades, yet is 665 currently largely overlooked (Rounsevell et al., 2012;Erb et al., 2013a;Luyssaert et al., 2014). According Despite these knowledge gaps, some insights in the relative weight of biogeochemical and biophysical 691 impacts of individual management activities emerged from our review. For instance, while grazing is 692 associated with strong biogeochemical, but relatively small biophysical effects, tree species selection is 693 characterized by strong biophysical, but limited biogeochemical effects. In contrast, forest harvest is 694 important in both respects (Figure 3). Similarly, strong biophysical as well as biogeochemical effects 695 originate from irrigation, cropland harvest and wetland drainage, although affecting much smaller areas. 696 Other agricultural activities, such as fertilization, tillage, residue management are associated mainly with 697 biogeochemical impacts. Crop species selection, in contrast, ranks low with regard to biogeochemical and 698 biophysical effects. But, as most land management activities are not isolated from each other, but 699 intricately linked (e.g. crop harvest, irrigation and fertilization), robust assessment on their relative 700 significance require the application of Earth System models and, as our review reveals, improved 701 databases. 702 Our review focused on documented Earth system effects of land management that have occurred over 703 the past decades. Yet land management plays an increasing role in discussions on mitigating future 704 climate change (Foley et al., 2005). This makes it particularly important to consider that management 705 effects act on a range of timescales: While changes in land surface properties impose immediate effects 706 on the atmosphere, changes in carbon and nitrogen fluxes invokes counter-fluxes in the coupled land-707 atmosphere-ocean system, causing a distinct temporal evolution and a delayed response of the Earth 708 system (Ciais et al., 2013b). The emergence of biogeochemical effects can also typically include longer 709 timescales than that of biogeophysical effects, as they can alter slow-responding system components such 710 as SOC. While biogeophysical effects and greenhouse gas fluxes due to management are persistent once 711 the new management system is in equilibrium, changes in carbon stocks cease to cause fluxes over time. 712 Assessment of a land use activity in the mitigation context thus depends not just on the spatial scale, with 713 fluxes of the well-mixed greenhouse gases causing a global signal, while biogeophysical effects act 714 predominantly on the local scale, but crucially also on an integrated assessment of the various effects and 715 their different timescales in relation to the time horizon of interest (Cherubini et al., 2012). 716  (Siebert et al., 2015) Many data, e.g. those by FAO, relate to area equipped for irrigation, while the amount of water actually used is difficult to assess. Higher quality for paddy rice. Artificial wetland drainage (Feick et al., 2005) Poor data availability. Gridded assessments cover all drainage, not only wetlands.
A mixed picture emerges regarding data availability and robustness of global, long-term land management information (Table 1). This is a consequence of the history of research and past investments in generating the datasets. Remote sensing, while particularly well-suited to assess certain land uses at the global level (e.g. cropping, irrigation, or the outbreak of fires), encounters severe difficulties in depicting other uses such as grazing Kuemmerle et al., 2013). Furthermore, statistical reporting schemes focus mainly on management activities of economic interest, such as crop and forest harvest and ignore others, e.g. crop residue management. In addition, inconsistent definitions affect data robustness (FAOSTAT, 2015;See et al., 2015).
While a comprehensive assessment of Earth system impacts induced by management requires more data and ultimately their integration in a modelling environment, as well as the inclusion of other management activites not discussed here, we conclude that management is a key factor in the Earth system, severely influencing many biogeochemical and biophysical processes and parameters. We also conclude that the current status of process understanding and data availability is sufficient to advance with the integration of land management in Earth system models in order to assess their overall impacts. Hence, we are able to classify the ten land management activities into groups along the two dimensions, i.e. data availability and process understanding (Table 2), and thus identify the most pressing research priorities.
A first group is characterized by relatively advanced data availability and process understanding. This group contains irrigation and cropland harvest. For these activities the the state of knowledge is sufficient for implementing these activities in integrative assessment environments such as Earth System Models.
Electronic copy available at: https://ssrn.com/abstract=3305968 The second group is characterized by severe data gaps, but relatively advanced process understanding.
This includes wood harvest, tree species selection, grazing, and N-fertilization, motivating calls for fostered research efforts from the global land use data community (e.g. Verburg et al., 2016) to develop improved datasets, e.g. by taking advantage of the increasingly available data from satellite observations (Kuemmerle et al., 2013;Joshi et al., 2016), or crowd sourcing (See et al., 2015), but also alternative approaches that exploit existing databases. These management activities could be included in Earth system models but global parameterisation and validation may be difficult for now. A third group is characterized by concomitant data and knowledge gaps. The management types in this group require an intensification of efforts of both the data and the ecological communities, in order to advance the understanding of the impact of these management practices on the Earth system. No activity was classified as a combination "advanced data" and "poor understanding".
Advancing the current state of process understanding and data availability on land management is a central undertaking to improve the understanding of land-use induced impacts on the Earth system and their feedbacks in the coupled socio-ecological system, central for e.g. the recently published Sustainability Development Goals (Costanza et al., 2016). In addition to enhancing data availability and process understanding, data access, usability, and quality control will become essential for transferring these achievements into beneficial information across multiple disciplines to tackle the grand sustainability challenges relate to land management.